LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* [git patches] ocfs2 fixes
@ 2007-01-22 19:02 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-01-22 19:02 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, ocfs2-devel, linux-kernel

Hi Linus,
	This update is just a small set of ocfs2 fixes suitable for merging
late in the cycle as they're all either straightforward or trivial.

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/export.c   |    5 ---
 fs/ocfs2/inode.c    |   11 ++------
 fs/ocfs2/namei.c    |   69 +++++++++++++++++++++++++++-------------------------
 fs/ocfs2/ocfs2_fs.h |   43 +++++++++++++++++++++++++++++++-
 fs/ocfs2/symlink.c  |    3 --
 5 files changed, 84 insertions(+), 47 deletions(-)

Mark Fasheh:
      ocfs2: Don't print errors when following symlinks
      ocfs2: Directory c/mtime update fixes
      ocfs2: cleanup ocfs2_iget() errors
      ocfs2: Add backup superblock info to ocfs2_fs.h

diff --git a/fs/ocfs2/export.c b/fs/ocfs2/export.c
index 06be6e7..56e1fef 100644
--- a/fs/ocfs2/export.c
+++ b/fs/ocfs2/export.c
@@ -60,14 +60,11 @@ static struct dentry *ocfs2_get_dentry(s
 
 	inode = ocfs2_iget(OCFS2_SB(sb), handle->ih_blkno, 0);
 
-	if (IS_ERR(inode)) {
-		mlog_errno(PTR_ERR(inode));
+	if (IS_ERR(inode))
 		return (void *)inode;
-	}
 
 	if (handle->ih_generation != inode->i_generation) {
 		iput(inode);
-		mlog_errno(-ESTALE);
 		return ERR_PTR(-ESTALE);
 	}
 
diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
index e4d9149..28ab56f 100644
--- a/fs/ocfs2/inode.c
+++ b/fs/ocfs2/inode.c
@@ -146,7 +146,6 @@ struct inode *ocfs2_iget(struct ocfs2_su
 	if (is_bad_inode(inode)) {
 		iput(inode);
 		inode = ERR_PTR(-ESTALE);
-		mlog_errno(PTR_ERR(inode));
 		goto bail;
 	}
 
@@ -155,8 +154,7 @@ bail:
 		mlog(0, "returning inode with number %llu\n",
 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
 		mlog_exit_ptr(inode);
-	} else
-		mlog_errno(PTR_ERR(inode));
+	}
 
 	return inode;
 }
@@ -247,7 +245,7 @@ int ocfs2_populate_inode(struct inode *i
 	 * today.  change if needed. */
 	if (!OCFS2_IS_VALID_DINODE(fe) ||
 	    !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
-		mlog(ML_ERROR, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
+		mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
 		     "signature = %.*s, flags = 0x%x\n",
 		     inode->i_ino,
 		     (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
@@ -478,11 +476,8 @@ static int ocfs2_read_locked_inode(struc
 	    S_ISBLK(le16_to_cpu(fe->i_mode)))
     		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
 
-	if (ocfs2_populate_inode(inode, fe, 0) < 0) {
-		mlog(ML_ERROR, "populate failed! i_blkno=%llu, i_ino=%lu\n",
-		     (unsigned long long)fe->i_blkno, inode->i_ino);
+	if (ocfs2_populate_inode(inode, fe, 0) < 0)
 		goto bail;
-	}
 
 	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
 
diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
index 9637039..f3d7803 100644
--- a/fs/ocfs2/namei.c
+++ b/fs/ocfs2/namei.c
@@ -932,14 +932,15 @@ static int ocfs2_unlink(struct inode *di
 		goto leave;
 	}
 
-	if (S_ISDIR(inode->i_mode)) {
+	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
+	if (S_ISDIR(inode->i_mode))
 		drop_nlink(dir);
-		status = ocfs2_mark_inode_dirty(handle, dir,
-						parent_node_bh);
-		if (status < 0) {
-			mlog_errno(status);
+
+	status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		if (S_ISDIR(inode->i_mode))
 			inc_nlink(dir);
-		}
 	}
 
 leave:
@@ -1068,6 +1069,7 @@ static int ocfs2_rename(struct inode *ol
 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 	struct buffer_head *orphan_entry_bh = NULL;
 	struct buffer_head *newfe_bh = NULL;
+	struct buffer_head *old_inode_bh = NULL;
 	struct buffer_head *insert_entry_bh = NULL;
 	struct ocfs2_super *osb = NULL;
 	u64 newfe_blkno;
@@ -1079,7 +1081,7 @@ static int ocfs2_rename(struct inode *ol
 	struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
 	struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
 						    // this is the 1st dirent bh
-	nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
+	nlink_t old_dir_nlink = old_dir->i_nlink;
 
 	/* At some point it might be nice to break this function up a
 	 * bit. */
@@ -1139,12 +1141,11 @@ static int ocfs2_rename(struct inode *ol
 	}
 
 	/*
-	 * Though we don't require an inode meta data update if
-	 * old_inode is not a directory, we lock anyway here to ensure
-	 * the vote thread on other nodes won't have to concurrently
-	 * downconvert the inode and the dentry locks.
+	 * Aside from allowing a meta data update, the locking here
+	 * also ensures that the vote thread on other nodes won't have
+	 * to concurrently downconvert the inode and the dentry locks.
 	 */
-	status = ocfs2_meta_lock(old_inode, NULL, 1);
+	status = ocfs2_meta_lock(old_inode, &old_inode_bh, 1);
 	if (status < 0) {
 		if (status != -ENOENT)
 			mlog_errno(status);
@@ -1355,6 +1356,7 @@ static int ocfs2_rename(struct inode *ol
 
 	old_inode->i_ctime = CURRENT_TIME;
 	mark_inode_dirty(old_inode);
+	ocfs2_mark_inode_dirty(handle, old_inode, old_inode_bh);
 
 	/* now that the name has been added to new_dir, remove the old name */
 	status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
@@ -1384,27 +1386,22 @@ static int ocfs2_rename(struct inode *ol
 		}
 	}
 	mark_inode_dirty(old_dir);
-	if (new_inode)
+	ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
+	if (new_inode) {
 		mark_inode_dirty(new_inode);
+		ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
+	}
 
-	if (old_dir != new_dir)
-		if (new_dir_nlink != new_dir->i_nlink) {
-			if (!new_dir_bh) {
-				mlog(ML_ERROR, "need to change nlink for new "
-				     "dir %llu from %d to %d but bh is NULL\n",
-				     (unsigned long long)OCFS2_I(new_dir)->ip_blkno,
-				     (int)new_dir_nlink, new_dir->i_nlink);
-			} else {
-				struct ocfs2_dinode *fe;
-				status = ocfs2_journal_access(handle,
-							      new_dir,
-							      new_dir_bh,
-							      OCFS2_JOURNAL_ACCESS_WRITE);
-				fe = (struct ocfs2_dinode *) new_dir_bh->b_data;
-				fe->i_links_count = cpu_to_le16(new_dir->i_nlink);
-				status = ocfs2_journal_dirty(handle, new_dir_bh);
-			}
-		}
+	if (old_dir != new_dir) {
+		/* Keep the same times on both directories.*/
+		new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
+
+		/*
+		 * This will also pick up the i_nlink change from the
+		 * block above.
+		 */
+		ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
+	}
 
 	if (old_dir_nlink != old_dir->i_nlink) {
 		if (!old_dir_bh) {
@@ -1455,6 +1452,8 @@ bail:
 		iput(new_inode);
 	if (newfe_bh)
 		brelse(newfe_bh);
+	if (old_inode_bh)
+		brelse(old_inode_bh);
 	if (old_dir_bh)
 		brelse(old_dir_bh);
 	if (new_dir_bh)
@@ -1826,6 +1825,13 @@ static int __ocfs2_add_entry(handle_t *h
 		     (le16_to_cpu(de->rec_len) >= rec_len)) ||
 		    (le16_to_cpu(de->rec_len) >=
 		     (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
+			dir->i_mtime = dir->i_ctime = CURRENT_TIME;
+			retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
+			if (retval < 0) {
+				mlog_errno(retval);
+				goto bail;
+			}
+
 			status = ocfs2_journal_access(handle, dir, insert_bh,
 						      OCFS2_JOURNAL_ACCESS_WRITE);
 			/* By now the buffer is marked for journaling */
@@ -1848,7 +1854,6 @@ static int __ocfs2_add_entry(handle_t *h
 			de->name_len = namelen;
 			memcpy(de->name, name, namelen);
 
-			dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 			dir->i_version++;
 			status = ocfs2_journal_dirty(handle, insert_bh);
 			retval = 0;
diff --git a/fs/ocfs2/ocfs2_fs.h b/fs/ocfs2/ocfs2_fs.h
index b5c6856..c99e905 100644
--- a/fs/ocfs2/ocfs2_fs.h
+++ b/fs/ocfs2/ocfs2_fs.h
@@ -85,7 +85,7 @@ #define OCFS2_CLEAR_RO_COMPAT_FEATURE(sb
 #define OCFS2_CLEAR_INCOMPAT_FEATURE(sb,mask)			\
 	OCFS2_SB(sb)->s_feature_incompat &= ~(mask)
 
-#define OCFS2_FEATURE_COMPAT_SUPP	0
+#define OCFS2_FEATURE_COMPAT_SUPP	OCFS2_FEATURE_COMPAT_BACKUP_SB
 #define OCFS2_FEATURE_INCOMPAT_SUPP	OCFS2_FEATURE_INCOMPAT_LOCAL_MOUNT
 #define OCFS2_FEATURE_RO_COMPAT_SUPP	0
 
@@ -110,6 +110,20 @@ #define OCFS2_FEATURE_INCOMPAT_LOCAL_MOU
 #define OCFS2_FEATURE_INCOMPAT_SPARSE_ALLOC	0x0010
 
 /*
+ * backup superblock flag is used to indicate that this volume
+ * has backup superblocks.
+ */
+#define OCFS2_FEATURE_COMPAT_BACKUP_SB		0x0001
+
+/* The byte offset of the first backup block will be 1G.
+ * The following will be 4G, 16G, 64G, 256G and 1T.
+ */
+#define OCFS2_BACKUP_SB_START			1 << 30
+
+/* the max backup superblock nums */
+#define OCFS2_MAX_BACKUP_SUPERBLOCKS	6
+
+/*
  * Flags on ocfs2_dinode.i_flags
  */
 #define OCFS2_VALID_FL		(0x00000001)	/* Inode is valid */
@@ -566,6 +580,20 @@ static inline int ocfs2_truncate_recs_pe
 
 	return size / sizeof(struct ocfs2_truncate_rec);
 }
+
+static inline u64 ocfs2_backup_super_blkno(struct super_block *sb, int index)
+{
+	u64 offset = OCFS2_BACKUP_SB_START;
+
+	if (index >= 0 && index < OCFS2_MAX_BACKUP_SUPERBLOCKS) {
+		offset <<= (2 * index);
+		offset /= sb->s_blocksize;
+		return offset;
+	}
+
+	return 0;
+
+}
 #else
 static inline int ocfs2_fast_symlink_chars(int blocksize)
 {
@@ -631,6 +659,19 @@ static inline int ocfs2_truncate_recs_pe
 
 	return size / sizeof(struct ocfs2_truncate_rec);
 }
+
+static inline uint64_t ocfs2_backup_super_blkno(int blocksize, int index)
+{
+	uint64_t offset = OCFS2_BACKUP_SB_START;
+
+	if (index >= 0 && index < OCFS2_MAX_BACKUP_SUPERBLOCKS) {
+		offset <<= (2 * index);
+		offset /= blocksize;
+		return offset;
+	}
+
+	return 0;
+}
 #endif  /* __KERNEL__ */
 
 
diff --git a/fs/ocfs2/symlink.c b/fs/ocfs2/symlink.c
index 957d687..03b0191 100644
--- a/fs/ocfs2/symlink.c
+++ b/fs/ocfs2/symlink.c
@@ -158,8 +158,7 @@ static void *ocfs2_follow_link(struct de
 	}
 
 	status = vfs_follow_link(nd, link);
-	if (status && status != -ENOENT)
-		mlog_errno(status);
+
 bail:
 	if (page) {
 		kunmap(page);

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] Ocfs2 fixes
@ 2009-03-13 17:14 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2009-03-13 17:14 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: linux-kernel, ocfs2-devel, Joel Becker, Andrew Morton

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/alloc.c    |    3 ++-
 fs/ocfs2/aops.c     |    7 +++++--
 fs/ocfs2/namei.c    |    3 ++-
 fs/ocfs2/ocfs2_fs.h |    6 ------
 fs/ocfs2/xattr.c    |   30 ++++++++++++++++++++++++------
 5 files changed, 33 insertions(+), 16 deletions(-)

Tao Ma (2):
      ocfs2: Fix a bug found by sparse check.
      ocfs2: Use xs->bucket to set xattr value outside

Tiger Yang (2):
      ocfs2: reserve xattr block for new directory with inline data
      ocfs2: tweak to get the maximum inline data size with xattr

--
Mark Fasheh

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] Ocfs2 fixes
@ 2009-02-26 19:57 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2009-02-26 19:57 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: linux-kernel, ocfs2-devel, Joel Becker, Andrew Morton

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/alloc.c         |   27 ++++++++++++++++++++++++++-
 fs/ocfs2/dlm/dlmmaster.c |   12 ++++++------
 fs/ocfs2/dlm/dlmthread.c |    3 +--
 fs/ocfs2/dlm/dlmunlock.c |    4 ++--
 fs/ocfs2/dlmglue.c       |   11 ++++++++---
 fs/ocfs2/ocfs2.h         |    3 +++
 fs/ocfs2/super.c         |    8 ++++++++
 fs/ocfs2/xattr.c         |   27 +++++++++++++++++----------
 8 files changed, 71 insertions(+), 24 deletions(-)

Sunil Mushran (4):
      ocfs2/dlm: Retract fix for race between purge and migrate
      ocfs2: Cleanup the lockname print in dlmglue.c
      ocfs2/dlm: Use ast_lock to protect ast_list
      ocfs2/dlm: Make dlm_assert_master_handler() kill itself instead of the asserter

Tao Ma (3):
      ocfs2: Access and dirty the buffer_head in mark_written.
      ocfs2: Use the right access_* method in ctime update of xattr.
      ocfs2: lock the metaecc process for xattr bucket

Tiger Yang (1):
      ocfs2: set gap to seperate entry and value when xattr in bucket

wengang wang (1):
      ocfs2: add IO error check in ocfs2_get_sector()

--
Mark Fasheh

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] Ocfs2 fixes
@ 2008-12-17 19:34 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2008-12-17 19:34 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: linux-kernel, ocfs2-devel, Joel Becker, Andrew Morton

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/ocfs2_fs.h |    8 +++++++-
 fs/ocfs2/xattr.c    |    4 ++--
 2 files changed, 9 insertions(+), 3 deletions(-)

Joel Becker (1):
      ocfs2: Add JBD2 compat feature bit.

Tao Ma (1):
      ocfs2: Always update xattr search when creating bucket.

diff --git a/fs/ocfs2/ocfs2_fs.h b/fs/ocfs2/ocfs2_fs.h
index 5f180cf..5e0c0d0 100644
--- a/fs/ocfs2/ocfs2_fs.h
+++ b/fs/ocfs2/ocfs2_fs.h
@@ -86,7 +86,8 @@
 #define OCFS2_CLEAR_INCOMPAT_FEATURE(sb,mask)			\
 	OCFS2_SB(sb)->s_feature_incompat &= ~(mask)
 
-#define OCFS2_FEATURE_COMPAT_SUPP	OCFS2_FEATURE_COMPAT_BACKUP_SB
+#define OCFS2_FEATURE_COMPAT_SUPP	(OCFS2_FEATURE_COMPAT_BACKUP_SB	\
+					 | OCFS2_FEATURE_COMPAT_JBD2_SB)
 #define OCFS2_FEATURE_INCOMPAT_SUPP	(OCFS2_FEATURE_INCOMPAT_LOCAL_MOUNT \
 					 | OCFS2_FEATURE_INCOMPAT_SPARSE_ALLOC \
 					 | OCFS2_FEATURE_INCOMPAT_INLINE_DATA \
@@ -153,6 +154,11 @@
 #define OCFS2_FEATURE_COMPAT_BACKUP_SB		0x0001
 
 /*
+ * The filesystem will correctly handle journal feature bits.
+ */
+#define OCFS2_FEATURE_COMPAT_JBD2_SB		0x0002
+
+/*
  * Unwritten extents support.
  */
 #define OCFS2_FEATURE_RO_COMPAT_UNWRITTEN	0x0001
diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
index 054e2ef..74d7367 100644
--- a/fs/ocfs2/xattr.c
+++ b/fs/ocfs2/xattr.c
@@ -2645,9 +2645,9 @@ static int ocfs2_xattr_update_xattr_search(struct inode *inode,
 				return ret;
 			}
 
-			i = xs->here - old_xh->xh_entries;
-			xs->here = &xs->header->xh_entries[i];
 		}
+		i = xs->here - old_xh->xh_entries;
+		xs->here = &xs->header->xh_entries[i];
 	}
 
 	return ret;

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] Ocfs2 fixes
@ 2008-12-01 22:54 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2008-12-01 22:54 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: linux-kernel, ocfs2-devel, Joel Becker, Andrew Morton

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 Documentation/filesystems/ocfs2.txt |    3 ---
 fs/ocfs2/buffer_head_io.c           |   15 ++++-----------
 fs/ocfs2/dlm/dlmfs.c                |    4 +++-
 fs/ocfs2/dlm/userdlm.h              |    2 +-
 fs/ocfs2/dlmglue.c                  |    3 +--
 fs/ocfs2/ocfs2.h                    |    2 +-
 fs/ocfs2/stack_user.c               |    3 +++
 7 files changed, 13 insertions(+), 19 deletions(-)

Coly Li (2):
      ocfs2: comments typo fix
      ocfs2: fix return value set in init_dlmfs_fs()

David Teigland (2):
      ocfs2: initialize stack_user lvbptr
      ocfs2: fix wake_up in unlock_ast

Mark Fasheh (2):
      ocfs2: Small documentation update
      ocfs2: fix regression in ocfs2_read_blocks_sync()

diff --git a/Documentation/filesystems/ocfs2.txt b/Documentation/filesystems/ocfs2.txt
index 4340cc8..67310fb 100644
--- a/Documentation/filesystems/ocfs2.txt
+++ b/Documentation/filesystems/ocfs2.txt
@@ -28,10 +28,7 @@ Manish Singh  <manish.singh@oracle.com>
 Caveats
 =======
 Features which OCFS2 does not support yet:
-	- extended attributes
 	- quotas
-	- cluster aware flock
-	- cluster aware lockf
 	- Directory change notification (F_NOTIFY)
 	- Distributed Caching (F_SETLEASE/F_GETLEASE/break_lease)
 	- POSIX ACLs
diff --git a/fs/ocfs2/buffer_head_io.c b/fs/ocfs2/buffer_head_io.c
index 7e947c6..3a178ec 100644
--- a/fs/ocfs2/buffer_head_io.c
+++ b/fs/ocfs2/buffer_head_io.c
@@ -112,7 +112,7 @@ int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
 		bh = bhs[i];
 
 		if (buffer_jbd(bh)) {
-			mlog(ML_ERROR,
+			mlog(ML_BH_IO,
 			     "trying to sync read a jbd "
 			     "managed bh (blocknr = %llu), skipping\n",
 			     (unsigned long long)bh->b_blocknr);
@@ -147,15 +147,10 @@ int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
 	for (i = nr; i > 0; i--) {
 		bh = bhs[i - 1];
 
-		if (buffer_jbd(bh)) {
-			mlog(ML_ERROR,
-			     "the journal got the buffer while it was "
-			     "locked for io! (blocknr = %llu)\n",
-			     (unsigned long long)bh->b_blocknr);
-			BUG();
-		}
+		/* No need to wait on the buffer if it's managed by JBD. */
+		if (!buffer_jbd(bh))
+			wait_on_buffer(bh);
 
-		wait_on_buffer(bh);
 		if (!buffer_uptodate(bh)) {
 			/* Status won't be cleared from here on out,
 			 * so we can safely record this and loop back
@@ -251,8 +246,6 @@ int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
 			ignore_cache = 1;
 		}
 
-		/* XXX: Can we ever get this and *not* have the cached
-		 * flag set? */
 		if (buffer_jbd(bh)) {
 			if (ignore_cache)
 				mlog(ML_BH_IO, "trying to sync read a jbd "
diff --git a/fs/ocfs2/dlm/dlmfs.c b/fs/ocfs2/dlm/dlmfs.c
index 533a789..ba962d7 100644
--- a/fs/ocfs2/dlm/dlmfs.c
+++ b/fs/ocfs2/dlm/dlmfs.c
@@ -608,8 +608,10 @@ static int __init init_dlmfs_fs(void)
 				0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 					SLAB_MEM_SPREAD),
 				dlmfs_init_once);
-	if (!dlmfs_inode_cache)
+	if (!dlmfs_inode_cache) {
+		status = -ENOMEM;
 		goto bail;
+	}
 	cleanup_inode = 1;
 
 	user_dlm_worker = create_singlethread_workqueue("user_dlm");
diff --git a/fs/ocfs2/dlm/userdlm.h b/fs/ocfs2/dlm/userdlm.h
index 39ec277..0c3cc03 100644
--- a/fs/ocfs2/dlm/userdlm.h
+++ b/fs/ocfs2/dlm/userdlm.h
@@ -33,7 +33,7 @@
 #include <linux/workqueue.h>
 
 /* user_lock_res->l_flags flags. */
-#define USER_LOCK_ATTACHED      (0x00000001) /* have we initialized
+#define USER_LOCK_ATTACHED      (0x00000001) /* we have initialized
 					       * the lvb */
 #define USER_LOCK_BUSY          (0x00000002) /* we are currently in
 					       * dlm_lock */
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
index ec68442..6e6cc0a 100644
--- a/fs/ocfs2/dlmglue.c
+++ b/fs/ocfs2/dlmglue.c
@@ -2841,9 +2841,8 @@ static void ocfs2_unlock_ast(void *opaque, int error)
 
 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
 	lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
-	spin_unlock_irqrestore(&lockres->l_lock, flags);
-
 	wake_up(&lockres->l_event);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
 
 	mlog_exit_void();
 }
diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
index fef7ece..3fed9e3 100644
--- a/fs/ocfs2/ocfs2.h
+++ b/fs/ocfs2/ocfs2.h
@@ -85,7 +85,7 @@ enum ocfs2_unlock_action {
 };
 
 /* ocfs2_lock_res->l_flags flags. */
-#define OCFS2_LOCK_ATTACHED      (0x00000001) /* have we initialized
+#define OCFS2_LOCK_ATTACHED      (0x00000001) /* we have initialized
 					       * the lvb */
 #define OCFS2_LOCK_BUSY          (0x00000002) /* we are currently in
 					       * dlm_lock */
diff --git a/fs/ocfs2/stack_user.c b/fs/ocfs2/stack_user.c
index faec2d8..9b76d41 100644
--- a/fs/ocfs2/stack_user.c
+++ b/fs/ocfs2/stack_user.c
@@ -740,6 +740,9 @@ static int user_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
 
 static void *user_dlm_lvb(union ocfs2_dlm_lksb *lksb)
 {
+	if (!lksb->lksb_fsdlm.sb_lvbptr)
+		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
+					     sizeof(struct dlm_lksb);
 	return (void *)(lksb->lksb_fsdlm.sb_lvbptr);
 }
 

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] Ocfs2 fixes
@ 2008-11-04 18:32 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2008-11-04 18:32 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: linux-kernel, ocfs2-devel, Joel Becker, Andrew Morton

More fixes than usual - that's my fault for missing an update that should
have happened last week. These are all fixes or trivial cleanups though. The
xattr ones are a bit large-ish, but it's a new feature which is getting
banged on a ton so that's to be expected.

Thanks,
	--Mark

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/file.c     |   27 +++--
 fs/ocfs2/inode.c    |    6 +
 fs/ocfs2/journal.c  |    1 +
 fs/ocfs2/mmap.c     |    6 +-
 fs/ocfs2/namei.c    |    8 +-
 fs/ocfs2/ocfs2.h    |    3 +
 fs/ocfs2/ocfs2_fs.h |   17 ++-
 fs/ocfs2/xattr.c    |  372 ++++++++++++++++++++++++++++-----------------------
 fs/ocfs2/xattr.h    |   38 +-----
 9 files changed, 256 insertions(+), 222 deletions(-)

Dmitri Monakhov (1):
      ocfs2: truncate outstanding block after direct io failure

Jan Kara (3):
      ocfs2: Fix check of return value of ocfs2_start_trans()
      ocfs2: Fix checking of return value of new_inode()
      ocfs2: Let inode be really deleted when ocfs2_mknod_locked() fails

Joel Becker (5):
      ocfs2: Check xattr block signatures properly.
      ocfs2: Don't return -EFAULT from a corrupt xattr entry.
      ocfs2: Check errors from ocfs2_xattr_update_xattr_search()
      ocfs2: Specify appropriate journal access for new xattr buckets.
      ocfs2: Don't repeat ocfs2_xattr_block_find()

Mark Fasheh (1):
      ocfs2: fix printk related build warnings in xattr.c

Sunil Mushran (1):
      ocfs2: Set journal descriptor to NULL after journal shutdown

Tao Ma (5):
      ocfs2: Remove unused ocfs2_restore_xattr_block().
      ocfs2: Fix some typos in xattr annotations.
      ocfs2: Fix check of return value of ocfs2_start_trans() in xattr.c.
      ocfs2: return 0 in page_mkwrite to let VFS retry.
      ocfs2/xattr: Proper hash collision handle in bucket division

Tiger Yang (5):
      ocfs2: fix license in xattr
      ocfs2: fix function declaration and definition in xattr
      ocfs2: remove duplicate definition in xattr
      ocfs2: add handler_map array bounds checking
      ocfs2: Check search result in ocfs2_xattr_block_get()

diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 7efe937..e2570a3 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -247,8 +247,8 @@ int ocfs2_update_inode_atime(struct inode *inode,
 	mlog_entry_void();
 
 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
-	if (handle == NULL) {
-		ret = -ENOMEM;
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
 		mlog_errno(ret);
 		goto out;
 	}
@@ -312,8 +312,8 @@ static int ocfs2_simple_size_update(struct inode *inode,
 	handle_t *handle = NULL;
 
 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
-	if (handle == NULL) {
-		ret = -ENOMEM;
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
 		mlog_errno(ret);
 		goto out;
 	}
@@ -1055,8 +1055,8 @@ static int __ocfs2_write_remove_suid(struct inode *inode,
 		   (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
 
 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
-	if (handle == NULL) {
-		ret = -ENOMEM;
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
 		mlog_errno(ret);
 		goto out;
 	}
@@ -1259,8 +1259,8 @@ static int __ocfs2_remove_inode_range(struct inode *inode,
 	}
 
 	handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
-	if (handle == NULL) {
-		ret = -ENOMEM;
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
 		mlog_errno(ret);
 		goto out;
 	}
@@ -1352,8 +1352,8 @@ static int ocfs2_zero_partial_clusters(struct inode *inode,
 		goto out;
 
 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
-	if (handle == NULL) {
-		ret = -ENOMEM;
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
 		mlog_errno(ret);
 		goto out;
 	}
@@ -1866,6 +1866,13 @@ relock:
 		written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
 						    ppos, count, ocount);
 		if (written < 0) {
+			/*
+			 * direct write may have instantiated a few
+			 * blocks outside i_size. Trim these off again.
+			 * Don't need i_size_read because we hold i_mutex.
+			 */
+			if (*ppos + count > inode->i_size)
+				vmtruncate(inode, inode->i_size);
 			ret = written;
 			goto out_dio;
 		}
diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
index 4903688..7aa00d5 100644
--- a/fs/ocfs2/inode.c
+++ b/fs/ocfs2/inode.c
@@ -1106,6 +1106,12 @@ void ocfs2_clear_inode(struct inode *inode)
 	oi->ip_last_trans = 0;
 	oi->ip_dir_start_lookup = 0;
 	oi->ip_blkno = 0ULL;
+
+	/*
+	 * ip_jinode is used to track txns against this inode. We ensure that
+	 * the journal is flushed before journal shutdown. Thus it is safe to
+	 * have inodes get cleaned up after journal shutdown.
+	 */
 	jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
 				       &oi->ip_jinode);
 
diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
index 81e4067..99fe9d5 100644
--- a/fs/ocfs2/journal.c
+++ b/fs/ocfs2/journal.c
@@ -690,6 +690,7 @@ void ocfs2_journal_shutdown(struct ocfs2_super *osb)
 
 	/* Shutdown the kernel journal system */
 	jbd2_journal_destroy(journal->j_journal);
+	journal->j_journal = NULL;
 
 	OCFS2_I(inode)->ip_open_count--;
 
diff --git a/fs/ocfs2/mmap.c b/fs/ocfs2/mmap.c
index 3dc18d6..eea1d24 100644
--- a/fs/ocfs2/mmap.c
+++ b/fs/ocfs2/mmap.c
@@ -113,7 +113,11 @@ static int __ocfs2_page_mkwrite(struct inode *inode, struct buffer_head *di_bh,
 	 * ocfs2_write_begin_nolock().
 	 */
 	if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
-		ret = -EINVAL;
+		/*
+		 * the page has been umapped in ocfs2_data_downconvert_worker.
+		 * So return 0 here and let VFS retry.
+		 */
+		ret = 0;
 		goto out;
 	}
 
diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
index 485a6aa..f4967e6 100644
--- a/fs/ocfs2/namei.c
+++ b/fs/ocfs2/namei.c
@@ -378,8 +378,8 @@ static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 	}
 
 	inode = new_inode(dir->i_sb);
-	if (IS_ERR(inode)) {
-		status = PTR_ERR(inode);
+	if (!inode) {
+		status = -ENOMEM;
 		mlog(ML_ERROR, "new_inode failed!\n");
 		goto leave;
 	}
@@ -491,8 +491,10 @@ leave:
 			brelse(*new_fe_bh);
 			*new_fe_bh = NULL;
 		}
-		if (inode)
+		if (inode) {
+			clear_nlink(inode);
 			iput(inode);
+		}
 	}
 
 	mlog_exit(status);
diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
index a21a465..fef7ece 100644
--- a/fs/ocfs2/ocfs2.h
+++ b/fs/ocfs2/ocfs2.h
@@ -473,6 +473,9 @@ static inline int ocfs2_uses_extended_slot_map(struct ocfs2_super *osb)
 		(____gd)->bg_signature);				\
 } while (0)
 
+#define OCFS2_IS_VALID_XATTR_BLOCK(ptr)					\
+	(!strcmp((ptr)->xb_signature, OCFS2_XATTR_BLOCK_SIGNATURE))
+
 static inline unsigned long ino_from_blkno(struct super_block *sb,
 					   u64 blkno)
 {
diff --git a/fs/ocfs2/ocfs2_fs.h b/fs/ocfs2/ocfs2_fs.h
index f24ce3d..5f180cf 100644
--- a/fs/ocfs2/ocfs2_fs.h
+++ b/fs/ocfs2/ocfs2_fs.h
@@ -742,12 +742,12 @@ struct ocfs2_group_desc
  */
 struct ocfs2_xattr_entry {
 	__le32	xe_name_hash;    /* hash value of xattr prefix+suffix. */
-	__le16	xe_name_offset;  /* byte offset from the 1st etnry in the local
+	__le16	xe_name_offset;  /* byte offset from the 1st entry in the
 				    local xattr storage(inode, xattr block or
 				    xattr bucket). */
 	__u8	xe_name_len;	 /* xattr name len, does't include prefix. */
-	__u8	xe_type;         /* the low 7 bits indicates the name prefix's
-				  * type and the highest 1 bits indicate whether
+	__u8	xe_type;         /* the low 7 bits indicate the name prefix
+				  * type and the highest bit indicates whether
 				  * the EA is stored in the local storage. */
 	__le64	xe_value_size;	 /* real xattr value length. */
 };
@@ -766,9 +766,10 @@ struct ocfs2_xattr_header {
 						   xattr. */
 	__le16	xh_name_value_len;              /* total length of name/value
 						   length in this bucket. */
-	__le16	xh_num_buckets;                 /* bucket nums in one extent
-						   record, only valid in the
-						   first bucket. */
+	__le16	xh_num_buckets;                 /* Number of xattr buckets
+						   in this extent record,
+						   only valid in the first
+						   bucket. */
 	__le64  xh_csum;
 	struct ocfs2_xattr_entry xh_entries[0]; /* xattr entry list. */
 };
@@ -776,8 +777,8 @@ struct ocfs2_xattr_header {
 /*
  * On disk structure for xattr value root.
  *
- * It is used when one extended attribute's size is larger, and we will save it
- * in an outside cluster. It will stored in a b-tree like file content.
+ * When an xattr's value is large enough, it is stored in an external
+ * b-tree like file data.  The xattr value root points to this structure.
  */
 struct ocfs2_xattr_value_root {
 /*00*/	__le32	xr_clusters;              /* clusters covered by xattr value. */
diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
index 802c414..054e2ef 100644
--- a/fs/ocfs2/xattr.c
+++ b/fs/ocfs2/xattr.c
@@ -3,25 +3,20 @@
  *
  * xattr.c
  *
- * Copyright (C) 2008 Oracle.  All rights reserved.
+ * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
  *
  * CREDITS:
- * Lots of code in this file is taken from ext3.
+ * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
+ * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public
- * License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
+ * License version 2 as published by the Free Software Foundation.
  *
  * This program is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * General Public License for more details.
- *
- * You should have received a copy of the GNU General Public
- * License along with this program; if not, write to the
- * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
- * Boston, MA 021110-1307, USA.
  */
 
 #include <linux/capability.h>
@@ -83,7 +78,7 @@ struct xattr_handler *ocfs2_xattr_handlers[] = {
 	NULL
 };
 
-static struct xattr_handler *ocfs2_xattr_handler_map[] = {
+static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 	[OCFS2_XATTR_INDEX_USER]	= &ocfs2_xattr_user_handler,
 	[OCFS2_XATTR_INDEX_TRUSTED]	= &ocfs2_xattr_trusted_handler,
 };
@@ -116,6 +111,10 @@ static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
 					     int *block_off,
 					     int *new_offset);
 
+static int ocfs2_xattr_block_find(struct inode *inode,
+				  int name_index,
+				  const char *name,
+				  struct ocfs2_xattr_search *xs);
 static int ocfs2_xattr_index_block_find(struct inode *inode,
 					struct buffer_head *root_bh,
 					int name_index,
@@ -137,6 +136,24 @@ static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 					  struct buffer_head *xb_bh);
 
+static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
+{
+	return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
+}
+
+static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
+{
+	return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
+}
+
+static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
+{
+	u16 len = sb->s_blocksize -
+		 offsetof(struct ocfs2_xattr_header, xh_entries);
+
+	return len / sizeof(struct ocfs2_xattr_entry);
+}
+
 static inline const char *ocfs2_xattr_prefix(int name_index)
 {
 	struct xattr_handler *handler = NULL;
@@ -542,14 +559,12 @@ static int ocfs2_xattr_block_list(struct inode *inode,
 		mlog_errno(ret);
 		return ret;
 	}
-	/*Verify the signature of xattr block*/
-	if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
-		   strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
-		ret = -EFAULT;
-		goto cleanup;
-	}
 
 	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
+	if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
+		ret = -EIO;
+		goto cleanup;
+	}
 
 	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 		struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
@@ -749,47 +764,25 @@ static int ocfs2_xattr_block_get(struct inode *inode,
 				 size_t buffer_size,
 				 struct ocfs2_xattr_search *xs)
 {
-	struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
-	struct buffer_head *blk_bh = NULL;
 	struct ocfs2_xattr_block *xb;
 	struct ocfs2_xattr_value_root *xv;
 	size_t size;
 	int ret = -ENODATA, name_offset, name_len, block_off, i;
 
-	if (!di->i_xattr_loc)
-		return ret;
-
 	memset(&xs->bucket, 0, sizeof(xs->bucket));
 
-	ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
-	if (ret < 0) {
+	ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
+	if (ret) {
 		mlog_errno(ret);
-		return ret;
-	}
-	/*Verify the signature of xattr block*/
-	if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
-		   strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
-		ret = -EFAULT;
 		goto cleanup;
 	}
 
-	xs->xattr_bh = blk_bh;
-	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
-
-	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
-		xs->header = &xb->xb_attrs.xb_header;
-		xs->base = (void *)xs->header;
-		xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
-		xs->here = xs->header->xh_entries;
-
-		ret = ocfs2_xattr_find_entry(name_index, name, xs);
-	} else
-		ret = ocfs2_xattr_index_block_find(inode, blk_bh,
-						   name_index,
-						   name, xs);
-
-	if (ret)
+	if (xs->not_found) {
+		ret = -ENODATA;
 		goto cleanup;
+	}
+
+	xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
 	size = le64_to_cpu(xs->here->xe_value_size);
 	if (buffer) {
 		ret = -ERANGE;
@@ -828,7 +821,8 @@ cleanup:
 		brelse(xs->bucket.bhs[i]);
 	memset(&xs->bucket, 0, sizeof(xs->bucket));
 
-	brelse(blk_bh);
+	brelse(xs->xattr_bh);
+	xs->xattr_bh = NULL;
 	return ret;
 }
 
@@ -837,11 +831,11 @@ cleanup:
  * Copy an extended attribute into the buffer provided.
  * Buffer is NULL to compute the size of buffer required.
  */
-int ocfs2_xattr_get(struct inode *inode,
-		    int name_index,
-		    const char *name,
-		    void *buffer,
-		    size_t buffer_size)
+static int ocfs2_xattr_get(struct inode *inode,
+			   int name_index,
+			   const char *name,
+			   void *buffer,
+			   size_t buffer_size)
 {
 	int ret;
 	struct ocfs2_dinode *di = NULL;
@@ -871,7 +865,7 @@ int ocfs2_xattr_get(struct inode *inode,
 	down_read(&oi->ip_xattr_sem);
 	ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
 				    buffer_size, &xis);
-	if (ret == -ENODATA)
+	if (ret == -ENODATA && di->i_xattr_loc)
 		ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
 					    buffer_size, &xbs);
 	up_read(&oi->ip_xattr_sem);
@@ -1229,7 +1223,7 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
 
 	free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
 	if (free < 0)
-		return -EFAULT;
+		return -EIO;
 
 	if (!xs->not_found) {
 		size_t size = 0;
@@ -1514,10 +1508,9 @@ static int ocfs2_xattr_free_block(struct inode *inode,
 		goto out;
 	}
 
-	/*Verify the signature of xattr block*/
-	if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
-		   strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
-		ret = -EFAULT;
+	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
+	if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
+		ret = -EIO;
 		goto out;
 	}
 
@@ -1527,7 +1520,6 @@ static int ocfs2_xattr_free_block(struct inode *inode,
 		goto out;
 	}
 
-	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 	blk = le64_to_cpu(xb->xb_blkno);
 	bit = le16_to_cpu(xb->xb_suballoc_bit);
 	bg_blkno = ocfs2_which_suballoc_group(blk, bit);
@@ -1771,15 +1763,14 @@ static int ocfs2_xattr_block_find(struct inode *inode,
 		mlog_errno(ret);
 		return ret;
 	}
-	/*Verify the signature of xattr block*/
-	if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
-		   strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
-			ret = -EFAULT;
-			goto cleanup;
+
+	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
+	if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
+		ret = -EIO;
+		goto cleanup;
 	}
 
 	xs->xattr_bh = blk_bh;
-	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 
 	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 		xs->header = &xb->xb_attrs.xb_header;
@@ -1806,52 +1797,6 @@ cleanup:
 }
 
 /*
- * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
- * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
- * re-initialized.
- */
-static int ocfs2_restore_xattr_block(struct inode *inode,
-				     struct ocfs2_xattr_search *xs)
-{
-	int ret;
-	handle_t *handle;
-	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-	struct ocfs2_xattr_block *xb =
-		(struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
-	struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
-	u16 xb_flags = le16_to_cpu(xb->xb_flags);
-
-	BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
-		le16_to_cpu(el->l_next_free_rec) != 0);
-
-	handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
-	if (IS_ERR(handle)) {
-		ret = PTR_ERR(handle);
-		handle = NULL;
-		goto out;
-	}
-
-	ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
-				   OCFS2_JOURNAL_ACCESS_WRITE);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out_commit;
-	}
-
-	memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
-	       offsetof(struct ocfs2_xattr_block, xb_attrs));
-
-	xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
-
-	ocfs2_journal_dirty(handle, xs->xattr_bh);
-
-out_commit:
-	ocfs2_commit_trans(osb, handle);
-out:
-	return ret;
-}
-
-/*
  * ocfs2_xattr_block_set()
  *
  * Set, replace or remove an extended attribute into external block.
@@ -1961,8 +1906,6 @@ out:
 	}
 
 	ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
-	if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
-		ret = ocfs2_restore_xattr_block(inode, xs);
 
 end:
 
@@ -2398,7 +2341,8 @@ static int ocfs2_xattr_index_block_find(struct inode *inode,
 	BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
 
 	mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
-	     "in the rec is %u\n", num_clusters, p_blkno, first_hash);
+	     "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
+	     first_hash);
 
 	ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
 				      p_blkno, first_hash, num_clusters, xs);
@@ -2422,7 +2366,7 @@ static int ocfs2_iterate_xattr_buckets(struct inode *inode,
 	memset(&bucket, 0, sizeof(bucket));
 
 	mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
-	     clusters, blkno);
+	     clusters, (unsigned long long)blkno);
 
 	for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
 		ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket,
@@ -2440,7 +2384,8 @@ static int ocfs2_iterate_xattr_buckets(struct inode *inode,
 		if (i == 0)
 			num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
 
-		mlog(0, "iterating xattr bucket %llu, first hash %u\n", blkno,
+		mlog(0, "iterating xattr bucket %llu, first hash %u\n",
+		     (unsigned long long)blkno,
 		     le32_to_cpu(bucket.xh->xh_entries[0].xe_name_hash));
 		if (func) {
 			ret = func(inode, &bucket, para);
@@ -2776,7 +2721,8 @@ static int ocfs2_xattr_create_index_block(struct inode *inode,
 	 */
 	blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
 
-	mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
+	mlog(0, "allocate 1 cluster from %llu to xattr block\n",
+	     (unsigned long long)blkno);
 
 	xh_bh = sb_getblk(inode->i_sb, blkno);
 	if (!xh_bh) {
@@ -2818,7 +2764,11 @@ static int ocfs2_xattr_create_index_block(struct inode *inode,
 	if (data_bh)
 		ocfs2_journal_dirty(handle, data_bh);
 
-	ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
+	ret = ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
 
 	/* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
 	memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
@@ -2941,8 +2891,8 @@ static int ocfs2_defrag_xattr_bucket(struct inode *inode,
 
 	mlog(0, "adjust xattr bucket in %llu, count = %u, "
 	     "xh_free_start = %u, xh_name_value_len = %u.\n",
-	     blkno, le16_to_cpu(xh->xh_count), xh_free_start,
-	     le16_to_cpu(xh->xh_name_value_len));
+	     (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
+	     xh_free_start, le16_to_cpu(xh->xh_name_value_len));
 
 	/*
 	 * sort all the entries by their offset.
@@ -3058,7 +3008,7 @@ static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
 	prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
 
 	mlog(0, "move half of xattrs in cluster %llu to %llu\n",
-	     prev_blkno, new_blkno);
+	     (unsigned long long)prev_blkno, (unsigned long long)new_blkno);
 
 	/*
 	 * We need to update the 1st half of the new cluster and
@@ -3168,26 +3118,74 @@ static int ocfs2_read_xattr_bucket(struct inode *inode,
 }
 
 /*
- * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
+ * Find the suitable pos when we divide a bucket into 2.
+ * We have to make sure the xattrs with the same hash value exist
+ * in the same bucket.
+ *
+ * If this ocfs2_xattr_header covers more than one hash value, find a
+ * place where the hash value changes.  Try to find the most even split.
+ * The most common case is that all entries have different hash values,
+ * and the first check we make will find a place to split.
+ */
+static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
+{
+	struct ocfs2_xattr_entry *entries = xh->xh_entries;
+	int count = le16_to_cpu(xh->xh_count);
+	int delta, middle = count / 2;
+
+	/*
+	 * We start at the middle.  Each step gets farther away in both
+	 * directions.  We therefore hit the change in hash value
+	 * nearest to the middle.  Note that this loop does not execute for
+	 * count < 2.
+	 */
+	for (delta = 0; delta < middle; delta++) {
+		/* Let's check delta earlier than middle */
+		if (cmp_xe(&entries[middle - delta - 1],
+			   &entries[middle - delta]))
+			return middle - delta;
+
+		/* For even counts, don't walk off the end */
+		if ((middle + delta + 1) == count)
+			continue;
+
+		/* Now try delta past middle */
+		if (cmp_xe(&entries[middle + delta],
+			   &entries[middle + delta + 1]))
+			return middle + delta + 1;
+	}
+
+	/* Every entry had the same hash */
+	return count;
+}
+
+/*
+ * Move some xattrs in old bucket(blk) to new bucket(new_blk).
  * first_hash will record the 1st hash of the new bucket.
+ *
+ * Normally half of the xattrs will be moved.  But we have to make
+ * sure that the xattrs with the same hash value are stored in the
+ * same bucket. If all the xattrs in this bucket have the same hash
+ * value, the new bucket will be initialized as an empty one and the
+ * first_hash will be initialized as (hash_value+1).
  */
-static int ocfs2_half_xattr_bucket(struct inode *inode,
-				   handle_t *handle,
-				   u64 blk,
-				   u64 new_blk,
-				   u32 *first_hash,
-				   int new_bucket_head)
+static int ocfs2_divide_xattr_bucket(struct inode *inode,
+				    handle_t *handle,
+				    u64 blk,
+				    u64 new_blk,
+				    u32 *first_hash,
+				    int new_bucket_head)
 {
 	int ret, i;
-	u16 count, start, len, name_value_len, xe_len, name_offset;
+	int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
 	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 	struct buffer_head **s_bhs, **t_bhs = NULL;
 	struct ocfs2_xattr_header *xh;
 	struct ocfs2_xattr_entry *xe;
 	int blocksize = inode->i_sb->s_blocksize;
 
-	mlog(0, "move half of xattrs from bucket %llu to %llu\n",
-	     blk, new_blk);
+	mlog(0, "move some of xattrs from bucket %llu to %llu\n",
+	     (unsigned long long)blk, (unsigned long long)new_blk);
 
 	s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
 	if (!s_bhs)
@@ -3220,21 +3218,44 @@ static int ocfs2_half_xattr_bucket(struct inode *inode,
 
 	for (i = 0; i < blk_per_bucket; i++) {
 		ret = ocfs2_journal_access(handle, inode, t_bhs[i],
-					   OCFS2_JOURNAL_ACCESS_CREATE);
+					   new_bucket_head ?
+					   OCFS2_JOURNAL_ACCESS_CREATE :
+					   OCFS2_JOURNAL_ACCESS_WRITE);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
 		}
 	}
 
+	xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
+	count = le16_to_cpu(xh->xh_count);
+	start = ocfs2_xattr_find_divide_pos(xh);
+
+	if (start == count) {
+		xe = &xh->xh_entries[start-1];
+
+		/*
+		 * initialized a new empty bucket here.
+		 * The hash value is set as one larger than
+		 * that of the last entry in the previous bucket.
+		 */
+		for (i = 0; i < blk_per_bucket; i++)
+			memset(t_bhs[i]->b_data, 0, blocksize);
+
+		xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
+		xh->xh_free_start = cpu_to_le16(blocksize);
+		xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
+		le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
+
+		goto set_num_buckets;
+	}
+
 	/* copy the whole bucket to the new first. */
 	for (i = 0; i < blk_per_bucket; i++)
 		memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
 
 	/* update the new bucket. */
 	xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
-	count = le16_to_cpu(xh->xh_count);
-	start = count / 2;
 
 	/*
 	 * Calculate the total name/value len and xh_free_start for
@@ -3291,6 +3312,7 @@ static int ocfs2_half_xattr_bucket(struct inode *inode,
 			xh->xh_free_start = xe->xe_name_offset;
 	}
 
+set_num_buckets:
 	/* set xh->xh_num_buckets for the new xh. */
 	if (new_bucket_head)
 		xh->xh_num_buckets = cpu_to_le16(1);
@@ -3308,9 +3330,13 @@ static int ocfs2_half_xattr_bucket(struct inode *inode,
 		*first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
 
 	/*
-	 * Now only update the 1st block of the old bucket.
-	 * Please note that the entry has been sorted already above.
+	 * Now only update the 1st block of the old bucket.  If we
+	 * just added a new empty bucket, there is no need to modify
+	 * it.
 	 */
+	if (start == count)
+		goto out;
+
 	xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
 	memset(&xh->xh_entries[start], 0,
 	       sizeof(struct ocfs2_xattr_entry) * (count - start));
@@ -3358,7 +3384,8 @@ static int ocfs2_cp_xattr_bucket(struct inode *inode,
 	BUG_ON(s_blkno == t_blkno);
 
 	mlog(0, "cp bucket %llu to %llu, target is %d\n",
-	     s_blkno, t_blkno, t_is_new);
+	     (unsigned long long)s_blkno, (unsigned long long)t_blkno,
+	     t_is_new);
 
 	s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
 			GFP_NOFS);
@@ -3382,6 +3409,8 @@ static int ocfs2_cp_xattr_bucket(struct inode *inode,
 
 	for (i = 0; i < blk_per_bucket; i++) {
 		ret = ocfs2_journal_access(handle, inode, t_bhs[i],
+					   t_is_new ?
+					   OCFS2_JOURNAL_ACCESS_CREATE :
 					   OCFS2_JOURNAL_ACCESS_WRITE);
 		if (ret)
 			goto out;
@@ -3428,7 +3457,8 @@ static int ocfs2_cp_xattr_cluster(struct inode *inode,
 	struct ocfs2_xattr_header *xh;
 	u64 to_blk_start = to_blk;
 
-	mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
+	mlog(0, "cp xattrs from cluster %llu to %llu\n",
+	     (unsigned long long)src_blk, (unsigned long long)to_blk);
 
 	/*
 	 * We need to update the new cluster and 1 more for the update of
@@ -3493,15 +3523,15 @@ out:
 }
 
 /*
- * Move half of the xattrs in this cluster to the new cluster.
+ * Move some xattrs in this cluster to the new cluster.
  * This function should only be called when bucket size == cluster size.
  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
  */
-static int ocfs2_half_xattr_cluster(struct inode *inode,
-				    handle_t *handle,
-				    u64 prev_blk,
-				    u64 new_blk,
-				    u32 *first_hash)
+static int ocfs2_divide_xattr_cluster(struct inode *inode,
+				      handle_t *handle,
+				      u64 prev_blk,
+				      u64 new_blk,
+				      u32 *first_hash)
 {
 	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 	int ret, credits = 2 * blk_per_bucket;
@@ -3515,8 +3545,8 @@ static int ocfs2_half_xattr_cluster(struct inode *inode,
 	}
 
 	/* Move half of the xattr in start_blk to the next bucket. */
-	return  ocfs2_half_xattr_bucket(inode, handle, prev_blk,
-					new_blk, first_hash, 1);
+	return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
+					  new_blk, first_hash, 1);
 }
 
 /*
@@ -3559,7 +3589,8 @@ static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
 	int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 
 	mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
-	     prev_blk, prev_clusters, new_blk);
+	     (unsigned long long)prev_blk, prev_clusters,
+	     (unsigned long long)new_blk);
 
 	if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
 		ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
@@ -3578,9 +3609,9 @@ static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
 						     last_blk, new_blk,
 						     v_start);
 		else {
-			ret = ocfs2_half_xattr_cluster(inode, handle,
-						       last_blk, new_blk,
-						       v_start);
+			ret = ocfs2_divide_xattr_cluster(inode, handle,
+							 last_blk, new_blk,
+							 v_start);
 
 			if ((*header_bh)->b_blocknr == last_blk && extend)
 				*extend = 0;
@@ -3629,7 +3660,7 @@ static int ocfs2_add_new_xattr_cluster(struct inode *inode,
 	mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
 	     "previous xattr blkno = %llu\n",
 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
-	     prev_cpos, prev_blkno);
+	     prev_cpos, (unsigned long long)prev_blkno);
 
 	ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
 
@@ -3716,7 +3747,7 @@ static int ocfs2_add_new_xattr_cluster(struct inode *inode,
 		}
 	}
 	mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
-	     num_bits, block, v_start);
+	     num_bits, (unsigned long long)block, v_start);
 	ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
 				  num_bits, 0, meta_ac);
 	if (ret < 0) {
@@ -3761,7 +3792,7 @@ static int ocfs2_extend_xattr_bucket(struct inode *inode,
 	u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
 
 	mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
-	     "from %llu, len = %u\n", start_blk,
+	     "from %llu, len = %u\n", (unsigned long long)start_blk,
 	     (unsigned long long)first_bh->b_blocknr, num_clusters);
 
 	BUG_ON(bucket >= num_buckets);
@@ -3797,8 +3828,8 @@ static int ocfs2_extend_xattr_bucket(struct inode *inode,
 	}
 
 	/* Move half of the xattr in start_blk to the next bucket. */
-	ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
-				      start_blk + blk_per_bucket, NULL, 0);
+	ret = ocfs2_divide_xattr_bucket(inode, handle, start_blk,
+					start_blk + blk_per_bucket, NULL, 0);
 
 	le16_add_cpu(&first_xh->xh_num_buckets, 1);
 	ocfs2_journal_dirty(handle, first_bh);
@@ -4146,7 +4177,7 @@ static int ocfs2_xattr_value_update_size(struct inode *inode,
 	handle_t *handle = NULL;
 
 	handle = ocfs2_start_trans(osb, 1);
-	if (handle == NULL) {
+	if (IS_ERR(handle)) {
 		ret = -ENOMEM;
 		mlog_errno(ret);
 		goto out;
@@ -4313,7 +4344,7 @@ static int ocfs2_rm_xattr_cluster(struct inode *inode,
 	}
 
 	handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
-	if (handle == NULL) {
+	if (IS_ERR(handle)) {
 		ret = -ENOMEM;
 		mlog_errno(ret);
 		goto out;
@@ -4489,11 +4520,21 @@ out:
 	return ret;
 }
 
-/* check whether the xattr bucket is filled up with the same hash value. */
+/*
+ * check whether the xattr bucket is filled up with the same hash value.
+ * If we want to insert the xattr with the same hash, return -ENOSPC.
+ * If we want to insert a xattr with different hash value, go ahead
+ * and ocfs2_divide_xattr_bucket will handle this.
+ */
 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
-					      struct ocfs2_xattr_bucket *bucket)
+					      struct ocfs2_xattr_bucket *bucket,
+					      const char *name)
 {
 	struct ocfs2_xattr_header *xh = bucket->xh;
+	u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
+
+	if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
+		return 0;
 
 	if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
 	    xh->xh_entries[0].xe_name_hash) {
@@ -4616,7 +4657,9 @@ try_again:
 		 * one bucket's worth, so check it here whether we need to
 		 * add a new bucket for the insert.
 		 */
-		ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
+		ret = ocfs2_check_xattr_bucket_collision(inode,
+							 &xs->bucket,
+							 xi->name);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
@@ -4727,14 +4770,11 @@ out:
 /*
  * 'trusted' attributes support
  */
-
-#define XATTR_TRUSTED_PREFIX "trusted."
-
 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
 				       size_t list_size, const char *name,
 				       size_t name_len)
 {
-	const size_t prefix_len = sizeof(XATTR_TRUSTED_PREFIX) - 1;
+	const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
 	const size_t total_len = prefix_len + name_len + 1;
 
 	if (list && total_len <= list_size) {
@@ -4771,18 +4811,14 @@ struct xattr_handler ocfs2_xattr_trusted_handler = {
 	.set	= ocfs2_xattr_trusted_set,
 };
 
-
 /*
  * 'user' attributes support
  */
-
-#define XATTR_USER_PREFIX "user."
-
 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
 				    size_t list_size, const char *name,
 				    size_t name_len)
 {
-	const size_t prefix_len = sizeof(XATTR_USER_PREFIX) - 1;
+	const size_t prefix_len = XATTR_USER_PREFIX_LEN;
 	const size_t total_len = prefix_len + name_len + 1;
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 
diff --git a/fs/ocfs2/xattr.h b/fs/ocfs2/xattr.h
index c25c7c6..1d8314c 100644
--- a/fs/ocfs2/xattr.h
+++ b/fs/ocfs2/xattr.h
@@ -3,24 +3,16 @@
  *
  * xattr.h
  *
- * Function prototypes
- *
- * Copyright (C) 2008 Oracle.  All rights reserved.
+ * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public
- * License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
+ * License version 2 as published by the Free Software Foundation.
  *
  * This program is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * General Public License for more details.
- *
- * You should have received a copy of the GNU General Public
- * License along with this program; if not, write to the
- * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
- * Boston, MA 021110-1307, USA.
  */
 
 #ifndef OCFS2_XATTR_H
@@ -40,29 +32,11 @@ enum ocfs2_xattr_type {
 
 extern struct xattr_handler ocfs2_xattr_user_handler;
 extern struct xattr_handler ocfs2_xattr_trusted_handler;
-
-extern ssize_t ocfs2_listxattr(struct dentry *, char *, size_t);
-extern int ocfs2_xattr_get(struct inode *, int, const char *, void *, size_t);
-extern int ocfs2_xattr_set(struct inode *, int, const char *, const void *,
-			   size_t, int);
-extern int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh);
 extern struct xattr_handler *ocfs2_xattr_handlers[];
 
-static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
-{
-	return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
-}
-
-static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
-{
-	return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
-}
-
-static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
-{
-	u16 len = sb->s_blocksize -
-		 offsetof(struct ocfs2_xattr_header, xh_entries);
+ssize_t ocfs2_listxattr(struct dentry *, char *, size_t);
+int ocfs2_xattr_set(struct inode *, int, const char *, const void *,
+		    size_t, int);
+int ocfs2_xattr_remove(struct inode *, struct buffer_head *);
 
-	return len / sizeof(struct ocfs2_xattr_entry);
-}
 #endif /* OCFS2_XATTR_H */

--
Mark Fasheh

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] Ocfs2 fixes
@ 2008-06-16 20:12 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2008-06-16 20:12 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: linux-kernel, ocfs2-devel, Joel Becker, Andrew Morton

This reverts a portion of the 'user-stack' unmount behavior back to how we
have traditionally handled unmount notification - via usermode callout to a
small helper application. Well, 'revert' is probably the wrong term. The
user-stack code explicitly never called the helper because we incorrectly
thought that an umount.ocfs2 might be able to do the work of tracking
final-unmount. As it turns out, that's not really possible (bind-mounts in
particular made this difficult) so we're going to go with the tried-and-true
umount handling for all Ocfs2 cluster stacks.


Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/cluster/nodemanager.c |   74 +------------------------
 fs/ocfs2/cluster/nodemanager.h |    4 -
 fs/ocfs2/stack_o2cb.c          |   41 +-------------
 fs/ocfs2/stack_user.c          |    3 +-
 fs/ocfs2/stackglue.c           |  119 +++++++++++++++++++++++++++++++++++++++-
 fs/ocfs2/stackglue.h           |   19 +-----
 6 files changed, 123 insertions(+), 137 deletions(-)

Joel Becker (3):
      ocfs2: Move the hb_ctl_path sysctl into the stack glue.
      ocfs2: Move the call of ocfs2_hb_ctl into the stack glue.
      ocfs2: Remove ->hangup() from stack glue operations.

diff --git a/fs/ocfs2/cluster/nodemanager.c b/fs/ocfs2/cluster/nodemanager.c
index cf9401e..cfdb08b 100644
--- a/fs/ocfs2/cluster/nodemanager.c
+++ b/fs/ocfs2/cluster/nodemanager.c
@@ -21,7 +21,6 @@
 
 #include <linux/kernel.h>
 #include <linux/module.h>
-#include <linux/sysctl.h>
 #include <linux/configfs.h>
 
 #include "tcp.h"
@@ -36,65 +35,6 @@
  * cluster references throughout where nodes are looked up */
 struct o2nm_cluster *o2nm_single_cluster = NULL;
 
-#define OCFS2_MAX_HB_CTL_PATH 256
-static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
-
-static ctl_table ocfs2_nm_table[] = {
-	{
-		.ctl_name	= 1,
-		.procname	= "hb_ctl_path",
-		.data		= ocfs2_hb_ctl_path,
-		.maxlen		= OCFS2_MAX_HB_CTL_PATH,
-		.mode		= 0644,
-		.proc_handler	= &proc_dostring,
-		.strategy	= &sysctl_string,
-	},
-	{ .ctl_name = 0 }
-};
-
-static ctl_table ocfs2_mod_table[] = {
-	{
-		.ctl_name	= FS_OCFS2_NM,
-		.procname	= "nm",
-		.data		= NULL,
-		.maxlen		= 0,
-		.mode		= 0555,
-		.child		= ocfs2_nm_table
-	},
-	{ .ctl_name = 0}
-};
-
-static ctl_table ocfs2_kern_table[] = {
-	{
-		.ctl_name	= FS_OCFS2,
-		.procname	= "ocfs2",
-		.data		= NULL,
-		.maxlen		= 0,
-		.mode		= 0555,
-		.child		= ocfs2_mod_table
-	},
-	{ .ctl_name = 0}
-};
-
-static ctl_table ocfs2_root_table[] = {
-	{
-		.ctl_name	= CTL_FS,
-		.procname	= "fs",
-		.data		= NULL,
-		.maxlen		= 0,
-		.mode		= 0555,
-		.child		= ocfs2_kern_table
-	},
-	{ .ctl_name = 0 }
-};
-
-static struct ctl_table_header *ocfs2_table_header = NULL;
-
-const char *o2nm_get_hb_ctl_path(void)
-{
-	return ocfs2_hb_ctl_path;
-}
-EXPORT_SYMBOL_GPL(o2nm_get_hb_ctl_path);
 
 struct o2nm_node *o2nm_get_node_by_num(u8 node_num)
 {
@@ -941,9 +881,6 @@ void o2nm_undepend_this_node(void)
 
 static void __exit exit_o2nm(void)
 {
-	if (ocfs2_table_header)
-		unregister_sysctl_table(ocfs2_table_header);
-
 	/* XXX sync with hb callbacks and shut down hb? */
 	o2net_unregister_hb_callbacks();
 	configfs_unregister_subsystem(&o2nm_cluster_group.cs_subsys);
@@ -964,16 +901,9 @@ static int __init init_o2nm(void)
 	if (ret)
 		goto out;
 
-	ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
-	if (!ocfs2_table_header) {
-		printk(KERN_ERR "nodemanager: unable to register sysctl\n");
-		ret = -ENOMEM; /* or something. */
-		goto out_o2net;
-	}
-
 	ret = o2net_register_hb_callbacks();
 	if (ret)
-		goto out_sysctl;
+		goto out_o2net;
 
 	config_group_init(&o2nm_cluster_group.cs_subsys.su_group);
 	mutex_init(&o2nm_cluster_group.cs_subsys.su_mutex);
@@ -990,8 +920,6 @@ static int __init init_o2nm(void)
 	configfs_unregister_subsystem(&o2nm_cluster_group.cs_subsys);
 out_callbacks:
 	o2net_unregister_hb_callbacks();
-out_sysctl:
-	unregister_sysctl_table(ocfs2_table_header);
 out_o2net:
 	o2net_exit();
 out:
diff --git a/fs/ocfs2/cluster/nodemanager.h b/fs/ocfs2/cluster/nodemanager.h
index 7c86036..c992ea0 100644
--- a/fs/ocfs2/cluster/nodemanager.h
+++ b/fs/ocfs2/cluster/nodemanager.h
@@ -33,10 +33,6 @@
 #include <linux/configfs.h>
 #include <linux/rbtree.h>
 
-#define FS_OCFS2_NM		1
-
-const char *o2nm_get_hb_ctl_path(void);
-
 struct o2nm_node {
 	spinlock_t		nd_lock;
 	struct config_item	nd_item;
diff --git a/fs/ocfs2/stack_o2cb.c b/fs/ocfs2/stack_o2cb.c
index bbd1667..fcd120f 100644
--- a/fs/ocfs2/stack_o2cb.c
+++ b/fs/ocfs2/stack_o2cb.c
@@ -317,8 +317,7 @@ out:
 	return rc;
 }
 
-static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn,
-				   int hangup_pending)
+static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
 {
 	struct dlm_ctxt *dlm = conn->cc_lockspace;
 	struct o2dlm_private *priv = conn->cc_private;
@@ -333,43 +332,6 @@ static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn,
 	return 0;
 }
 
-static void o2hb_stop(const char *group)
-{
-	int ret;
-	char *argv[5], *envp[3];
-
-	argv[0] = (char *)o2nm_get_hb_ctl_path();
-	argv[1] = "-K";
-	argv[2] = "-u";
-	argv[3] = (char *)group;
-	argv[4] = NULL;
-
-	mlog(0, "Run: %s %s %s %s\n", argv[0], argv[1], argv[2], argv[3]);
-
-	/* minimal command environment taken from cpu_run_sbin_hotplug */
-	envp[0] = "HOME=/";
-	envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
-	envp[2] = NULL;
-
-	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
-	if (ret < 0)
-		mlog_errno(ret);
-}
-
-/*
- * Hangup is a hack for tools compatibility.  Older ocfs2-tools software
- * expects the filesystem to call "ocfs2_hb_ctl" during unmount.  This
- * happens regardless of whether the DLM got started, so we can't do it
- * in ocfs2_cluster_disconnect().  We bring the o2hb_stop() function into
- * the glue and provide a "hangup" API for super.c to call.
- *
- * Other stacks will eventually provide a NULL ->hangup() pointer.
- */
-static void o2cb_cluster_hangup(const char *group, int grouplen)
-{
-	o2hb_stop(group);
-}
-
 static int o2cb_cluster_this_node(unsigned int *node)
 {
 	int node_num;
@@ -388,7 +350,6 @@ static int o2cb_cluster_this_node(unsigned int *node)
 static struct ocfs2_stack_operations o2cb_stack_ops = {
 	.connect	= o2cb_cluster_connect,
 	.disconnect	= o2cb_cluster_disconnect,
-	.hangup		= o2cb_cluster_hangup,
 	.this_node	= o2cb_cluster_this_node,
 	.dlm_lock	= o2cb_dlm_lock,
 	.dlm_unlock	= o2cb_dlm_unlock,
diff --git a/fs/ocfs2/stack_user.c b/fs/ocfs2/stack_user.c
index 6b97d11..c021280 100644
--- a/fs/ocfs2/stack_user.c
+++ b/fs/ocfs2/stack_user.c
@@ -816,8 +816,7 @@ out:
 	return rc;
 }
 
-static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn,
-				   int hangup_pending)
+static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
 {
 	dlm_release_lockspace(conn->cc_lockspace, 2);
 	conn->cc_lockspace = NULL;
diff --git a/fs/ocfs2/stackglue.c b/fs/ocfs2/stackglue.c
index 119f60c..10e149a 100644
--- a/fs/ocfs2/stackglue.c
+++ b/fs/ocfs2/stackglue.c
@@ -26,6 +26,7 @@
 #include <linux/fs.h>
 #include <linux/kobject.h>
 #include <linux/sysfs.h>
+#include <linux/sysctl.h>
 
 #include "ocfs2_fs.h"
 
@@ -33,11 +34,13 @@
 
 #define OCFS2_STACK_PLUGIN_O2CB		"o2cb"
 #define OCFS2_STACK_PLUGIN_USER		"user"
+#define OCFS2_MAX_HB_CTL_PATH		256
 
 static struct ocfs2_locking_protocol *lproto;
 static DEFINE_SPINLOCK(ocfs2_stack_lock);
 static LIST_HEAD(ocfs2_stack_list);
 static char cluster_stack_name[OCFS2_STACK_LABEL_LEN + 1];
+static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
 
 /*
  * The stack currently in use.  If not null, active_stack->sp_count > 0,
@@ -349,7 +352,7 @@ int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
 
 	BUG_ON(conn == NULL);
 
-	ret = active_stack->sp_ops->disconnect(conn, hangup_pending);
+	ret = active_stack->sp_ops->disconnect(conn);
 
 	/* XXX Should we free it anyway? */
 	if (!ret) {
@@ -362,13 +365,48 @@ int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
 }
 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
 
+/*
+ * Leave the group for this filesystem.  This is executed by a userspace
+ * program (stored in ocfs2_hb_ctl_path).
+ */
+static void ocfs2_leave_group(const char *group)
+{
+	int ret;
+	char *argv[5], *envp[3];
+
+	argv[0] = ocfs2_hb_ctl_path;
+	argv[1] = "-K";
+	argv[2] = "-u";
+	argv[3] = (char *)group;
+	argv[4] = NULL;
+
+	/* minimal command environment taken from cpu_run_sbin_hotplug */
+	envp[0] = "HOME=/";
+	envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
+	envp[2] = NULL;
+
+	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
+	if (ret < 0) {
+		printk(KERN_ERR
+		       "ocfs2: Error %d running user helper "
+		       "\"%s %s %s %s\"\n",
+		       ret, argv[0], argv[1], argv[2], argv[3]);
+	}
+}
+
+/*
+ * Hangup is a required post-umount.  ocfs2-tools software expects the
+ * filesystem to call "ocfs2_hb_ctl" during unmount.  This happens
+ * regardless of whether the DLM got started, so we can't do it
+ * in ocfs2_cluster_disconnect().  The ocfs2_leave_group() function does
+ * the actual work.
+ */
 void ocfs2_cluster_hangup(const char *group, int grouplen)
 {
 	BUG_ON(group == NULL);
 	BUG_ON(group[grouplen] != '\0');
 
-	if (active_stack->sp_ops->hangup)
-		active_stack->sp_ops->hangup(group, grouplen);
+	ocfs2_leave_group(group);
 
 	/* cluster_disconnect() was called with hangup_pending==1 */
 	ocfs2_stack_driver_put();
@@ -548,10 +586,83 @@ error:
 	return ret;
 }
 
+/*
+ * Sysctl bits
+ *
+ * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path.  The 'nm' doesn't
+ * make as much sense in a multiple cluster stack world, but it's safer
+ * and easier to preserve the name.
+ */
+
+#define FS_OCFS2_NM		1
+
+static ctl_table ocfs2_nm_table[] = {
+	{
+		.ctl_name	= 1,
+		.procname	= "hb_ctl_path",
+		.data		= ocfs2_hb_ctl_path,
+		.maxlen		= OCFS2_MAX_HB_CTL_PATH,
+		.mode		= 0644,
+		.proc_handler	= &proc_dostring,
+		.strategy	= &sysctl_string,
+	},
+	{ .ctl_name = 0 }
+};
+
+static ctl_table ocfs2_mod_table[] = {
+	{
+		.ctl_name	= FS_OCFS2_NM,
+		.procname	= "nm",
+		.data		= NULL,
+		.maxlen		= 0,
+		.mode		= 0555,
+		.child		= ocfs2_nm_table
+	},
+	{ .ctl_name = 0}
+};
+
+static ctl_table ocfs2_kern_table[] = {
+	{
+		.ctl_name	= FS_OCFS2,
+		.procname	= "ocfs2",
+		.data		= NULL,
+		.maxlen		= 0,
+		.mode		= 0555,
+		.child		= ocfs2_mod_table
+	},
+	{ .ctl_name = 0}
+};
+
+static ctl_table ocfs2_root_table[] = {
+	{
+		.ctl_name	= CTL_FS,
+		.procname	= "fs",
+		.data		= NULL,
+		.maxlen		= 0,
+		.mode		= 0555,
+		.child		= ocfs2_kern_table
+	},
+	{ .ctl_name = 0 }
+};
+
+static struct ctl_table_header *ocfs2_table_header = NULL;
+
+
+/*
+ * Initialization
+ */
+
 static int __init ocfs2_stack_glue_init(void)
 {
 	strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
 
+	ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
+	if (!ocfs2_table_header) {
+		printk(KERN_ERR
+		       "ocfs2 stack glue: unable to register sysctl\n");
+		return -ENOMEM; /* or something. */
+	}
+
 	return ocfs2_sysfs_init();
 }
 
@@ -559,6 +670,8 @@ static void __exit ocfs2_stack_glue_exit(void)
 {
 	lproto = NULL;
 	ocfs2_sysfs_exit();
+	if (ocfs2_table_header)
+		unregister_sysctl_table(ocfs2_table_header);
 }
 
 MODULE_AUTHOR("Oracle");
diff --git a/fs/ocfs2/stackglue.h b/fs/ocfs2/stackglue.h
index 005e4f1..db56281 100644
--- a/fs/ocfs2/stackglue.h
+++ b/fs/ocfs2/stackglue.h
@@ -134,22 +134,10 @@ struct ocfs2_stack_operations {
 	 * be freed.  Thus, a stack must not return from ->disconnect()
 	 * until it will no longer reference the conn pointer.
 	 *
-	 * If hangup_pending is zero, ocfs2_cluster_disconnect() will also
-	 * be dropping the reference on the module.
+	 * Once this call returns, the stack glue will be dropping this
+	 * connection's reference on the module.
 	 */
-	int (*disconnect)(struct ocfs2_cluster_connection *conn,
-			  int hangup_pending);
-
-	/*
-	 * ocfs2_cluster_hangup() exists for compatibility with older
-	 * ocfs2 tools.  Only the classic stack really needs it.  As such
-	 * ->hangup() is not required of all stacks.  See the comment by
-	 * ocfs2_cluster_hangup() for more details.
-	 *
-	 * Note that ocfs2_cluster_hangup() can only be called if
-	 * hangup_pending was passed to ocfs2_cluster_disconnect().
-	 */
-	void (*hangup)(const char *group, int grouplen);
+	int (*disconnect)(struct ocfs2_cluster_connection *conn);
 
 	/*
 	 * ->this_node() returns the cluster's unique identifier for the
@@ -258,4 +246,5 @@ void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto)
 /* Used by stack plugins */
 int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin);
 void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin);
+
 #endif  /* STACKGLUE_H */

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] Ocfs2 fixes
@ 2008-06-01 16:16 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2008-06-01 16:16 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: linux-kernel, ocfs2-devel, Joel Becker, Andrew Morton

Here's a few Ocfs2 fixes that are ready for upstream. The 'user_stack'
rename in particular fixes some compile errors which folks have hit, due to
clashing symbols.
	--Mark

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/cluster/tcp.c          |   28 +++++++++-------------------
 fs/ocfs2/cluster/tcp.h          |   12 ++++++------
 fs/ocfs2/cluster/tcp_internal.h |   32 ++++++++++++++++++++++++++++++++
 fs/ocfs2/dlm/dlmdebug.h         |   12 ++++++------
 fs/ocfs2/stack_user.c           |   34 +++++++++++++++++-----------------
 5 files changed, 70 insertions(+), 48 deletions(-)

Joel Becker (1):
      ocfs2: Rename 'user_stack' plugin structure to 'ocfs2_user_plugin'

Sunil Mushran (3):
      ocfs2/net: Silence build warnings
      ocfs2/dlm: Silence build warnings
      ocfs2/net: Silence build warnings

diff --git a/fs/ocfs2/cluster/tcp.c b/fs/ocfs2/cluster/tcp.c
index 1e44ad1..a27d615 100644
--- a/fs/ocfs2/cluster/tcp.c
+++ b/fs/ocfs2/cluster/tcp.c
@@ -142,53 +142,43 @@ static void o2net_idle_timer(unsigned long data);
 static void o2net_sc_postpone_idle(struct o2net_sock_container *sc);
 static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc);
 
-static void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
-			   u32 msgkey, struct task_struct *task, u8 node)
-{
 #ifdef CONFIG_DEBUG_FS
+void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
+		    u32 msgkey, struct task_struct *task, u8 node)
+{
 	INIT_LIST_HEAD(&nst->st_net_debug_item);
 	nst->st_task = task;
 	nst->st_msg_type = msgtype;
 	nst->st_msg_key = msgkey;
 	nst->st_node = node;
-#endif
 }
 
-static void o2net_set_nst_sock_time(struct o2net_send_tracking *nst)
+void o2net_set_nst_sock_time(struct o2net_send_tracking *nst)
 {
-#ifdef CONFIG_DEBUG_FS
 	do_gettimeofday(&nst->st_sock_time);
-#endif
 }
 
-static void o2net_set_nst_send_time(struct o2net_send_tracking *nst)
+void o2net_set_nst_send_time(struct o2net_send_tracking *nst)
 {
-#ifdef CONFIG_DEBUG_FS
 	do_gettimeofday(&nst->st_send_time);
-#endif
 }
 
-static void o2net_set_nst_status_time(struct o2net_send_tracking *nst)
+void o2net_set_nst_status_time(struct o2net_send_tracking *nst)
 {
-#ifdef CONFIG_DEBUG_FS
 	do_gettimeofday(&nst->st_status_time);
-#endif
 }
 
-static void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
+void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
 					 struct o2net_sock_container *sc)
 {
-#ifdef CONFIG_DEBUG_FS
 	nst->st_sc = sc;
-#endif
 }
 
-static void o2net_set_nst_msg_id(struct o2net_send_tracking *nst, u32 msg_id)
+void o2net_set_nst_msg_id(struct o2net_send_tracking *nst, u32 msg_id)
 {
-#ifdef CONFIG_DEBUG_FS
 	nst->st_id = msg_id;
-#endif
 }
+#endif /* CONFIG_DEBUG_FS */
 
 static inline int o2net_reconnect_delay(void)
 {
diff --git a/fs/ocfs2/cluster/tcp.h b/fs/ocfs2/cluster/tcp.h
index a705d5d..fd6179e 100644
--- a/fs/ocfs2/cluster/tcp.h
+++ b/fs/ocfs2/cluster/tcp.h
@@ -128,23 +128,23 @@ void o2net_debug_del_nst(struct o2net_send_tracking *nst);
 void o2net_debug_add_sc(struct o2net_sock_container *sc);
 void o2net_debug_del_sc(struct o2net_sock_container *sc);
 #else
-static int o2net_debugfs_init(void)
+static inline int o2net_debugfs_init(void)
 {
 	return 0;
 }
-static void o2net_debugfs_exit(void)
+static inline void o2net_debugfs_exit(void)
 {
 }
-static void o2net_debug_add_nst(struct o2net_send_tracking *nst)
+static inline void o2net_debug_add_nst(struct o2net_send_tracking *nst)
 {
 }
-static void o2net_debug_del_nst(struct o2net_send_tracking *nst)
+static inline void o2net_debug_del_nst(struct o2net_send_tracking *nst)
 {
 }
-static void o2net_debug_add_sc(struct o2net_sock_container *sc)
+static inline void o2net_debug_add_sc(struct o2net_sock_container *sc)
 {
 }
-static void o2net_debug_del_sc(struct o2net_sock_container *sc)
+static inline void o2net_debug_del_sc(struct o2net_sock_container *sc)
 {
 }
 #endif	/* CONFIG_DEBUG_FS */
diff --git a/fs/ocfs2/cluster/tcp_internal.h b/fs/ocfs2/cluster/tcp_internal.h
index 8d58cfe..18307ff 100644
--- a/fs/ocfs2/cluster/tcp_internal.h
+++ b/fs/ocfs2/cluster/tcp_internal.h
@@ -224,10 +224,42 @@ struct o2net_send_tracking {
 	struct timeval			st_send_time;
 	struct timeval			st_status_time;
 };
+
+void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
+		    u32 msgkey, struct task_struct *task, u8 node);
+void o2net_set_nst_sock_time(struct o2net_send_tracking *nst);
+void o2net_set_nst_send_time(struct o2net_send_tracking *nst);
+void o2net_set_nst_status_time(struct o2net_send_tracking *nst);
+void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
+				  struct o2net_sock_container *sc);
+void o2net_set_nst_msg_id(struct o2net_send_tracking *nst, u32 msg_id);
+
 #else
 struct o2net_send_tracking {
 	u32	dummy;
 };
+
+static inline void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
+				  u32 msgkey, struct task_struct *task, u8 node)
+{
+}
+static inline void o2net_set_nst_sock_time(struct o2net_send_tracking *nst)
+{
+}
+static inline void o2net_set_nst_send_time(struct o2net_send_tracking *nst)
+{
+}
+static inline void o2net_set_nst_status_time(struct o2net_send_tracking *nst)
+{
+}
+static inline void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
+						struct o2net_sock_container *sc)
+{
+}
+static inline void o2net_set_nst_msg_id(struct o2net_send_tracking *nst,
+					u32 msg_id)
+{
+}
 #endif	/* CONFIG_DEBUG_FS */
 
 #endif /* O2CLUSTER_TCP_INTERNAL_H */
diff --git a/fs/ocfs2/dlm/dlmdebug.h b/fs/ocfs2/dlm/dlmdebug.h
index d34a62a..8c686d2 100644
--- a/fs/ocfs2/dlm/dlmdebug.h
+++ b/fs/ocfs2/dlm/dlmdebug.h
@@ -60,25 +60,25 @@ void dlm_destroy_debugfs_root(void);
 
 #else
 
-static int dlm_debug_init(struct dlm_ctxt *dlm)
+static inline int dlm_debug_init(struct dlm_ctxt *dlm)
 {
 	return 0;
 }
-static void dlm_debug_shutdown(struct dlm_ctxt *dlm)
+static inline void dlm_debug_shutdown(struct dlm_ctxt *dlm)
 {
 }
-static int dlm_create_debugfs_subroot(struct dlm_ctxt *dlm)
+static inline int dlm_create_debugfs_subroot(struct dlm_ctxt *dlm)
 {
 	return 0;
 }
-static void dlm_destroy_debugfs_subroot(struct dlm_ctxt *dlm)
+static inline void dlm_destroy_debugfs_subroot(struct dlm_ctxt *dlm)
 {
 }
-static int dlm_create_debugfs_root(void)
+static inline int dlm_create_debugfs_root(void)
 {
 	return 0;
 }
-static void dlm_destroy_debugfs_root(void)
+static inline void dlm_destroy_debugfs_root(void)
 {
 }
 
diff --git a/fs/ocfs2/stack_user.c b/fs/ocfs2/stack_user.c
index b503772..6b97d11 100644
--- a/fs/ocfs2/stack_user.c
+++ b/fs/ocfs2/stack_user.c
@@ -61,7 +61,7 @@
  * negotiated by the client.  The client negotiates based on the maximum
  * version advertised in /sys/fs/ocfs2/max_locking_protocol.  The major
  * number from the "SETV" message must match
- * user_stack.sp_proto->lp_max_version.pv_major, and the minor number
+ * ocfs2_user_plugin.sp_proto->lp_max_version.pv_major, and the minor number
  * must be less than or equal to ...->lp_max_version.pv_minor.
  *
  * Once this information has been set, mounts will be allowed.  From this
@@ -153,7 +153,7 @@ union ocfs2_control_message {
 	struct ocfs2_control_message_down	u_down;
 };
 
-static struct ocfs2_stack_plugin user_stack;
+static struct ocfs2_stack_plugin ocfs2_user_plugin;
 
 static atomic_t ocfs2_control_opened;
 static int ocfs2_control_this_node = -1;
@@ -399,7 +399,7 @@ static int ocfs2_control_do_setversion_msg(struct file *file,
 	char *ptr = NULL;
 	struct ocfs2_control_private *p = file->private_data;
 	struct ocfs2_protocol_version *max =
-		&user_stack.sp_proto->lp_max_version;
+		&ocfs2_user_plugin.sp_proto->lp_max_version;
 
 	if (ocfs2_control_get_handshake_state(file) !=
 	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
@@ -680,7 +680,7 @@ static void fsdlm_lock_ast_wrapper(void *astarg)
 	struct dlm_lksb *lksb = fsdlm_astarg_to_lksb(astarg);
 	int status = lksb->sb_status;
 
-	BUG_ON(user_stack.sp_proto == NULL);
+	BUG_ON(ocfs2_user_plugin.sp_proto == NULL);
 
 	/*
 	 * For now we're punting on the issue of other non-standard errors
@@ -693,16 +693,16 @@ static void fsdlm_lock_ast_wrapper(void *astarg)
 	 */
 
 	if (status == -DLM_EUNLOCK || status == -DLM_ECANCEL)
-		user_stack.sp_proto->lp_unlock_ast(astarg, 0);
+		ocfs2_user_plugin.sp_proto->lp_unlock_ast(astarg, 0);
 	else
-		user_stack.sp_proto->lp_lock_ast(astarg);
+		ocfs2_user_plugin.sp_proto->lp_lock_ast(astarg);
 }
 
 static void fsdlm_blocking_ast_wrapper(void *astarg, int level)
 {
-	BUG_ON(user_stack.sp_proto == NULL);
+	BUG_ON(ocfs2_user_plugin.sp_proto == NULL);
 
-	user_stack.sp_proto->lp_blocking_ast(astarg, level);
+	ocfs2_user_plugin.sp_proto->lp_blocking_ast(astarg, level);
 }
 
 static int user_dlm_lock(struct ocfs2_cluster_connection *conn,
@@ -838,7 +838,7 @@ static int user_cluster_this_node(unsigned int *this_node)
 	return 0;
 }
 
-static struct ocfs2_stack_operations user_stack_ops = {
+static struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
 	.connect	= user_cluster_connect,
 	.disconnect	= user_cluster_disconnect,
 	.this_node	= user_cluster_this_node,
@@ -849,20 +849,20 @@ static struct ocfs2_stack_operations user_stack_ops = {
 	.dump_lksb	= user_dlm_dump_lksb,
 };
 
-static struct ocfs2_stack_plugin user_stack = {
+static struct ocfs2_stack_plugin ocfs2_user_plugin = {
 	.sp_name	= "user",
-	.sp_ops		= &user_stack_ops,
+	.sp_ops		= &ocfs2_user_plugin_ops,
 	.sp_owner	= THIS_MODULE,
 };
 
 
-static int __init user_stack_init(void)
+static int __init ocfs2_user_plugin_init(void)
 {
 	int rc;
 
 	rc = ocfs2_control_init();
 	if (!rc) {
-		rc = ocfs2_stack_glue_register(&user_stack);
+		rc = ocfs2_stack_glue_register(&ocfs2_user_plugin);
 		if (rc)
 			ocfs2_control_exit();
 	}
@@ -870,14 +870,14 @@ static int __init user_stack_init(void)
 	return rc;
 }
 
-static void __exit user_stack_exit(void)
+static void __exit ocfs2_user_plugin_exit(void)
 {
-	ocfs2_stack_glue_unregister(&user_stack);
+	ocfs2_stack_glue_unregister(&ocfs2_user_plugin);
 	ocfs2_control_exit();
 }
 
 MODULE_AUTHOR("Oracle");
 MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks");
 MODULE_LICENSE("GPL");
-module_init(user_stack_init);
-module_exit(user_stack_exit);
+module_init(ocfs2_user_plugin_init);
+module_exit(ocfs2_user_plugin_exit);

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] Ocfs2 fixes
@ 2008-05-02 18:35 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2008-05-02 18:35 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: linux-kernel, ocfs2-devel, Joel Becker, Andrew Morton

Some fixes/cleanups that I pulled into ocfs2.git after we sent our feature merge.
The /sys/o2cb fix in particular has been asked about a couple times.
	--Mark

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/cluster/sys.c  |    2 +-
 fs/ocfs2/dlm/dlmdebug.c |    8 +++++---
 fs/ocfs2/file.c         |    4 ++++
 fs/ocfs2/localalloc.c   |    4 ++--
 fs/ocfs2/stack_o2cb.c   |    2 +-
 fs/ocfs2/stack_user.c   |    2 +-
 fs/ocfs2/symlink.c      |    2 ++
 7 files changed, 16 insertions(+), 8 deletions(-)

Adrian Bunk (3):
      ocfs2: make struct ocfs2_control_device static
      ocfs2: make struct o2cb_stack_ops static
      ocfs2/dlm: dlmdebug.c: make 2 functions static

Joel Becker (1):
      ocfs2: Correct merge of 52f7c21 (Move /sys/o2cb to /sys/fs/o2cb)

Sunil Mushran (2):
      ocfs2: Allow uid/gid/perm changes of symlinks
      ocfs2: Use GFP_NOFS in kmalloc during localalloc window move

diff --git a/fs/ocfs2/cluster/sys.c b/fs/ocfs2/cluster/sys.c
index 98429fd..bc702da 100644
--- a/fs/ocfs2/cluster/sys.c
+++ b/fs/ocfs2/cluster/sys.c
@@ -65,7 +65,7 @@ int o2cb_sys_init(void)
 {
 	int ret;
 
-	o2cb_kset = kset_create_and_add("o2cb", NULL, NULL);
+	o2cb_kset = kset_create_and_add("o2cb", NULL, fs_kobj);
 	if (!o2cb_kset)
 		return -ENOMEM;
 
diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
index 5f6d858..1b81dcb 100644
--- a/fs/ocfs2/dlm/dlmdebug.c
+++ b/fs/ocfs2/dlm/dlmdebug.c
@@ -44,7 +44,8 @@
 #define MLOG_MASK_PREFIX ML_DLM
 #include "cluster/masklog.h"
 
-int stringify_lockname(const char *lockname, int locklen, char *buf, int len);
+static int stringify_lockname(const char *lockname, int locklen, char *buf,
+			      int len);
 
 void dlm_print_one_lock_resource(struct dlm_lock_resource *res)
 {
@@ -251,7 +252,8 @@ EXPORT_SYMBOL_GPL(dlm_errname);
  *
  * For more on lockname formats, please refer to dlmglue.c and ocfs2_lockid.h.
  */
-int stringify_lockname(const char *lockname, int locklen, char *buf, int len)
+static int stringify_lockname(const char *lockname, int locklen, char *buf,
+			      int len)
 {
 	int out = 0;
 	__be64 inode_blkno_be;
@@ -368,7 +370,7 @@ static void dlm_debug_free(struct kref *kref)
 	kfree(dc);
 }
 
-void dlm_debug_put(struct dlm_debug_ctxt *dc)
+static void dlm_debug_put(struct dlm_debug_ctxt *dc)
 {
 	if (dc)
 		kref_put(&dc->debug_refcnt, dlm_debug_free);
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 9154c82..57e0d30 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -1048,6 +1048,10 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
 	mlog_entry("(0x%p, '%.*s')\n", dentry,
 	           dentry->d_name.len, dentry->d_name.name);
 
+	/* ensuring we don't even attempt to truncate a symlink */
+	if (S_ISLNK(inode->i_mode))
+		attr->ia_valid &= ~ATTR_SIZE;
+
 	if (attr->ia_valid & ATTR_MODE)
 		mlog(0, "mode change: %d\n", attr->ia_mode);
 	if (attr->ia_valid & ATTR_UID)
diff --git a/fs/ocfs2/localalloc.c b/fs/ocfs2/localalloc.c
index ce0dc14..be774bd 100644
--- a/fs/ocfs2/localalloc.c
+++ b/fs/ocfs2/localalloc.c
@@ -260,7 +260,7 @@ void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
 	bh = osb->local_alloc_bh;
 	alloc = (struct ocfs2_dinode *) bh->b_data;
 
-	alloc_copy = kmalloc(bh->b_size, GFP_KERNEL);
+	alloc_copy = kmalloc(bh->b_size, GFP_NOFS);
 	if (!alloc_copy) {
 		status = -ENOMEM;
 		goto out_commit;
@@ -931,7 +931,7 @@ static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
 	 * local alloc shutdown won't try to double free main bitmap
 	 * bits. Make a copy so the sync function knows which bits to
 	 * free. */
-	alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_KERNEL);
+	alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS);
 	if (!alloc_copy) {
 		status = -ENOMEM;
 		mlog_errno(status);
diff --git a/fs/ocfs2/stack_o2cb.c b/fs/ocfs2/stack_o2cb.c
index ac1d74c..bbd1667 100644
--- a/fs/ocfs2/stack_o2cb.c
+++ b/fs/ocfs2/stack_o2cb.c
@@ -385,7 +385,7 @@ static int o2cb_cluster_this_node(unsigned int *node)
 	return 0;
 }
 
-struct ocfs2_stack_operations o2cb_stack_ops = {
+static struct ocfs2_stack_operations o2cb_stack_ops = {
 	.connect	= o2cb_cluster_connect,
 	.disconnect	= o2cb_cluster_disconnect,
 	.hangup		= o2cb_cluster_hangup,
diff --git a/fs/ocfs2/stack_user.c b/fs/ocfs2/stack_user.c
index 7428663..b503772 100644
--- a/fs/ocfs2/stack_user.c
+++ b/fs/ocfs2/stack_user.c
@@ -635,7 +635,7 @@ static const struct file_operations ocfs2_control_fops = {
 	.owner   = THIS_MODULE,
 };
 
-struct miscdevice ocfs2_control_device = {
+static struct miscdevice ocfs2_control_device = {
 	.minor		= MISC_DYNAMIC_MINOR,
 	.name		= "ocfs2_control",
 	.fops		= &ocfs2_control_fops,
diff --git a/fs/ocfs2/symlink.c b/fs/ocfs2/symlink.c
index 7134007..ba9dbb5 100644
--- a/fs/ocfs2/symlink.c
+++ b/fs/ocfs2/symlink.c
@@ -167,9 +167,11 @@ const struct inode_operations ocfs2_symlink_inode_operations = {
 	.readlink	= page_readlink,
 	.follow_link	= ocfs2_follow_link,
 	.getattr	= ocfs2_getattr,
+	.setattr	= ocfs2_setattr,
 };
 const struct inode_operations ocfs2_fast_symlink_inode_operations = {
 	.readlink	= ocfs2_readlink,
 	.follow_link	= ocfs2_follow_link,
 	.getattr	= ocfs2_getattr,
+	.setattr	= ocfs2_setattr,
 };


^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2008-03-10 22:44 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2008-03-10 22:44 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

Hi Linus,

Here's a set of Ocfs2 fixes for post-rc5.
	--Mark

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/cluster/tcp.c     |    9 ++--
 fs/ocfs2/dlm/dlmcommon.h   |   21 +++++----
 fs/ocfs2/dlm/dlmconvert.c  |    2 +-
 fs/ocfs2/dlm/dlmdomain.c   |  103 ++++++++++++++++++++++++++++----------------
 fs/ocfs2/dlm/dlmmaster.c   |   18 ++++++-
 fs/ocfs2/dlm/dlmrecovery.c |   57 ++++++++++++++++++++----
 fs/ocfs2/dlm/dlmthread.c   |    6 ++-
 fs/ocfs2/dlmglue.c         |    2 +-
 fs/ocfs2/resize.c          |    2 +-
 9 files changed, 151 insertions(+), 69 deletions(-)

Andrew Morton (1):
      fs/ocfs2/dlm/dlmdomain.c: fix printk warning

Jan Engelhardt (1):
      [OCFS2]: constify function pointer tables

Joel Becker (1):
      ocfs2: Fix endian bug in o2dlm protocol negotiation.

Sunil Mushran (5):
      ocfs2/dlm: Add missing dlm_lock_put()s
      ocfs2/dlm: Add missing dlm_lockres_put()s in migration path
      ocfs2/dlm: Add missing dlm_lockres_put()s
      ocfs2/dlm: Print message showing the recovery master
      ocfs2/dlm: dlm_thread should not sleep while holding the dlm_spinlock

Tao Ma (3):
      ocfs2: Use dlm_print_one_lock_resource for lock resource print
      ocfs2: Fix an endian bug in online resize.
      ocfs2: Fix NULL pointer dereferences in o2net

diff --git a/fs/ocfs2/cluster/tcp.c b/fs/ocfs2/cluster/tcp.c
index ee50c96..b8057c5 100644
--- a/fs/ocfs2/cluster/tcp.c
+++ b/fs/ocfs2/cluster/tcp.c
@@ -451,9 +451,9 @@ static void o2net_set_nn_state(struct o2net_node *nn,
 		/* delay if we're withing a RECONNECT_DELAY of the
 		 * last attempt */
 		delay = (nn->nn_last_connect_attempt +
-			 msecs_to_jiffies(o2net_reconnect_delay(sc->sc_node)))
+			 msecs_to_jiffies(o2net_reconnect_delay(NULL)))
 			- jiffies;
-		if (delay > msecs_to_jiffies(o2net_reconnect_delay(sc->sc_node)))
+		if (delay > msecs_to_jiffies(o2net_reconnect_delay(NULL)))
 			delay = 0;
 		mlog(ML_CONN, "queueing conn attempt in %lu jiffies\n", delay);
 		queue_delayed_work(o2net_wq, &nn->nn_connect_work, delay);
@@ -1552,12 +1552,11 @@ static void o2net_connect_expired(struct work_struct *work)
 
 	spin_lock(&nn->nn_lock);
 	if (!nn->nn_sc_valid) {
-		struct o2nm_node *node = nn->nn_sc->sc_node;
 		mlog(ML_ERROR, "no connection established with node %u after "
 		     "%u.%u seconds, giving up and returning errors.\n",
 		     o2net_num_from_nn(nn),
-		     o2net_idle_timeout(node) / 1000,
-		     o2net_idle_timeout(node) % 1000);
+		     o2net_idle_timeout(NULL) / 1000,
+		     o2net_idle_timeout(NULL) % 1000);
 
 		o2net_set_nn_state(nn, NULL, 0, -ENOTCONN);
 	}
diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h
index 9843ee1..dc8ea66 100644
--- a/fs/ocfs2/dlm/dlmcommon.h
+++ b/fs/ocfs2/dlm/dlmcommon.h
@@ -176,6 +176,7 @@ struct dlm_mig_lockres_priv
 {
 	struct dlm_lock_resource *lockres;
 	u8 real_master;
+	u8 extra_ref;
 };
 
 struct dlm_assert_master_priv
@@ -602,17 +603,19 @@ enum dlm_query_join_response_code {
 	JOIN_PROTOCOL_MISMATCH,
 };
 
+struct dlm_query_join_packet {
+	u8 code;	/* Response code.  dlm_minor and fs_minor
+			   are only valid if this is JOIN_OK */
+	u8 dlm_minor;	/* The minor version of the protocol the
+			   dlm is speaking. */
+	u8 fs_minor;	/* The minor version of the protocol the
+			   filesystem is speaking. */
+	u8 reserved;
+};
+
 union dlm_query_join_response {
 	u32 intval;
-	struct {
-		u8 code;	/* Response code.  dlm_minor and fs_minor
-				   are only valid if this is JOIN_OK */
-		u8 dlm_minor;	/* The minor version of the protocol the
-				   dlm is speaking. */
-		u8 fs_minor;	/* The minor version of the protocol the
-				   filesystem is speaking. */
-		u8 reserved;
-	} packet;
+	struct dlm_query_join_packet packet;
 };
 
 struct dlm_lock_request
diff --git a/fs/ocfs2/dlm/dlmconvert.c b/fs/ocfs2/dlm/dlmconvert.c
index ecb4d99..75997b4 100644
--- a/fs/ocfs2/dlm/dlmconvert.c
+++ b/fs/ocfs2/dlm/dlmconvert.c
@@ -487,7 +487,7 @@ int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data,
 			       "cookie=%u:%llu\n",
 		     dlm_get_lock_cookie_node(be64_to_cpu(cnv->cookie)),
 		     dlm_get_lock_cookie_seq(be64_to_cpu(cnv->cookie)));
-		__dlm_print_one_lock_resource(res);
+		dlm_print_one_lock_resource(res);
 		goto leave;
 	}
 
diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index 638d2eb..0879d86 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -713,14 +713,46 @@ static int dlm_query_join_proto_check(char *proto_type, int node,
 	return rc;
 }
 
+/*
+ * struct dlm_query_join_packet is made up of four one-byte fields.  They
+ * are effectively in big-endian order already.  However, little-endian
+ * machines swap them before putting the packet on the wire (because
+ * query_join's response is a status, and that status is treated as a u32
+ * on the wire).  Thus, a big-endian and little-endian machines will treat
+ * this structure differently.
+ *
+ * The solution is to have little-endian machines swap the structure when
+ * converting from the structure to the u32 representation.  This will
+ * result in the structure having the correct format on the wire no matter
+ * the host endian format.
+ */
+static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
+					  u32 *wire)
+{
+	union dlm_query_join_response response;
+
+	response.packet = *packet;
+	*wire = cpu_to_be32(response.intval);
+}
+
+static void dlm_query_join_wire_to_packet(u32 wire,
+					  struct dlm_query_join_packet *packet)
+{
+	union dlm_query_join_response response;
+
+	response.intval = cpu_to_be32(wire);
+	*packet = response.packet;
+}
+
 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 				  void **ret_data)
 {
 	struct dlm_query_join_request *query;
-	union dlm_query_join_response response = {
-		.packet.code = JOIN_DISALLOW,
+	struct dlm_query_join_packet packet = {
+		.code = JOIN_DISALLOW,
 	};
 	struct dlm_ctxt *dlm = NULL;
+	u32 response;
 	u8 nodenum;
 
 	query = (struct dlm_query_join_request *) msg->buf;
@@ -737,11 +769,11 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 		mlog(0, "node %u is not in our live map yet\n",
 		     query->node_idx);
 
-		response.packet.code = JOIN_DISALLOW;
+		packet.code = JOIN_DISALLOW;
 		goto respond;
 	}
 
-	response.packet.code = JOIN_OK_NO_MAP;
+	packet.code = JOIN_OK_NO_MAP;
 
 	spin_lock(&dlm_domain_lock);
 	dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
@@ -760,7 +792,7 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 				mlog(0, "disallow join as node %u does not "
 				     "have node %u in its nodemap\n",
 				     query->node_idx, nodenum);
-				response.packet.code = JOIN_DISALLOW;
+				packet.code = JOIN_DISALLOW;
 				goto unlock_respond;
 			}
 		}
@@ -780,23 +812,23 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 			/*If this is a brand new context and we
 			 * haven't started our join process yet, then
 			 * the other node won the race. */
-			response.packet.code = JOIN_OK_NO_MAP;
+			packet.code = JOIN_OK_NO_MAP;
 		} else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
 			/* Disallow parallel joins. */
-			response.packet.code = JOIN_DISALLOW;
+			packet.code = JOIN_DISALLOW;
 		} else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
 			mlog(0, "node %u trying to join, but recovery "
 			     "is ongoing.\n", bit);
-			response.packet.code = JOIN_DISALLOW;
+			packet.code = JOIN_DISALLOW;
 		} else if (test_bit(bit, dlm->recovery_map)) {
 			mlog(0, "node %u trying to join, but it "
 			     "still needs recovery.\n", bit);
-			response.packet.code = JOIN_DISALLOW;
+			packet.code = JOIN_DISALLOW;
 		} else if (test_bit(bit, dlm->domain_map)) {
 			mlog(0, "node %u trying to join, but it "
 			     "is still in the domain! needs recovery?\n",
 			     bit);
-			response.packet.code = JOIN_DISALLOW;
+			packet.code = JOIN_DISALLOW;
 		} else {
 			/* Alright we're fully a part of this domain
 			 * so we keep some state as to who's joining
@@ -807,19 +839,15 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 			if (dlm_query_join_proto_check("DLM", bit,
 						       &dlm->dlm_locking_proto,
 						       &query->dlm_proto)) {
-				response.packet.code =
-					JOIN_PROTOCOL_MISMATCH;
+				packet.code = JOIN_PROTOCOL_MISMATCH;
 			} else if (dlm_query_join_proto_check("fs", bit,
 							      &dlm->fs_locking_proto,
 							      &query->fs_proto)) {
-				response.packet.code =
-					JOIN_PROTOCOL_MISMATCH;
+				packet.code = JOIN_PROTOCOL_MISMATCH;
 			} else {
-				response.packet.dlm_minor =
-					query->dlm_proto.pv_minor;
-				response.packet.fs_minor =
-					query->fs_proto.pv_minor;
-				response.packet.code = JOIN_OK;
+				packet.dlm_minor = query->dlm_proto.pv_minor;
+				packet.fs_minor = query->fs_proto.pv_minor;
+				packet.code = JOIN_OK;
 				__dlm_set_joining_node(dlm, query->node_idx);
 			}
 		}
@@ -830,9 +858,10 @@ unlock_respond:
 	spin_unlock(&dlm_domain_lock);
 
 respond:
-	mlog(0, "We respond with %u\n", response.packet.code);
+	mlog(0, "We respond with %u\n", packet.code);
 
-	return response.intval;
+	dlm_query_join_packet_to_wire(&packet, &response);
+	return response;
 }
 
 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
@@ -937,7 +966,7 @@ static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
 			 sizeof(unsigned long))) {
 		mlog(ML_ERROR,
 		     "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
-		     map_size, BITS_TO_LONGS(O2NM_MAX_NODES));
+		     map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
 		return -EINVAL;
 	}
 
@@ -968,7 +997,8 @@ static int dlm_request_join(struct dlm_ctxt *dlm,
 {
 	int status;
 	struct dlm_query_join_request join_msg;
-	union dlm_query_join_response join_resp;
+	struct dlm_query_join_packet packet;
+	u32 join_resp;
 
 	mlog(0, "querying node %d\n", node);
 
@@ -984,11 +1014,12 @@ static int dlm_request_join(struct dlm_ctxt *dlm,
 
 	status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
 				    sizeof(join_msg), node,
-				    &join_resp.intval);
+				    &join_resp);
 	if (status < 0 && status != -ENOPROTOOPT) {
 		mlog_errno(status);
 		goto bail;
 	}
+	dlm_query_join_wire_to_packet(join_resp, &packet);
 
 	/* -ENOPROTOOPT from the net code means the other side isn't
 	    listening for our message type -- that's fine, it means
@@ -997,10 +1028,10 @@ static int dlm_request_join(struct dlm_ctxt *dlm,
 	if (status == -ENOPROTOOPT) {
 		status = 0;
 		*response = JOIN_OK_NO_MAP;
-	} else if (join_resp.packet.code == JOIN_DISALLOW ||
-		   join_resp.packet.code == JOIN_OK_NO_MAP) {
-		*response = join_resp.packet.code;
-	} else if (join_resp.packet.code == JOIN_PROTOCOL_MISMATCH) {
+	} else if (packet.code == JOIN_DISALLOW ||
+		   packet.code == JOIN_OK_NO_MAP) {
+		*response = packet.code;
+	} else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
 		mlog(ML_NOTICE,
 		     "This node requested DLM locking protocol %u.%u and "
 		     "filesystem locking protocol %u.%u.  At least one of "
@@ -1012,14 +1043,12 @@ static int dlm_request_join(struct dlm_ctxt *dlm,
 		     dlm->fs_locking_proto.pv_minor,
 		     node);
 		status = -EPROTO;
-		*response = join_resp.packet.code;
-	} else if (join_resp.packet.code == JOIN_OK) {
-		*response = join_resp.packet.code;
+		*response = packet.code;
+	} else if (packet.code == JOIN_OK) {
+		*response = packet.code;
 		/* Use the same locking protocol as the remote node */
-		dlm->dlm_locking_proto.pv_minor =
-			join_resp.packet.dlm_minor;
-		dlm->fs_locking_proto.pv_minor =
-			join_resp.packet.fs_minor;
+		dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
+		dlm->fs_locking_proto.pv_minor = packet.fs_minor;
 		mlog(0,
 		     "Node %d responds JOIN_OK with DLM locking protocol "
 		     "%u.%u and fs locking protocol %u.%u\n",
@@ -1031,11 +1060,11 @@ static int dlm_request_join(struct dlm_ctxt *dlm,
 	} else {
 		status = -EINVAL;
 		mlog(ML_ERROR, "invalid response %d from node %u\n",
-		     join_resp.packet.code, node);
+		     packet.code, node);
 	}
 
 	mlog(0, "status %d, node %d response is %d\n", status, node,
-		  *response);
+	     *response);
 
 bail:
 	return status;
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index c92d1b1..ea6b895 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -1663,7 +1663,12 @@ way_up_top:
 		dlm_put_mle(tmpmle);
 	}
 send_response:
-
+	/*
+	 * __dlm_lookup_lockres() grabbed a reference to this lockres.
+	 * The reference is released by dlm_assert_master_worker() under
+	 * the call to dlm_dispatch_assert_master().  If
+	 * dlm_assert_master_worker() isn't called, we drop it here.
+	 */
 	if (dispatch_assert) {
 		if (response != DLM_MASTER_RESP_YES)
 			mlog(ML_ERROR, "invalid response %d\n", response);
@@ -1678,7 +1683,11 @@ send_response:
 		if (ret < 0) {
 			mlog(ML_ERROR, "failed to dispatch assert master work\n");
 			response = DLM_MASTER_RESP_ERROR;
+			dlm_lockres_put(res);
 		}
+	} else {
+		if (res)
+			dlm_lockres_put(res);
 	}
 
 	dlm_put(dlm);
@@ -2348,7 +2357,7 @@ int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
 			mlog(ML_ERROR, "%s:%.*s: node %u trying to drop ref "
 		     	"but it is already dropped!\n", dlm->name,
 		     	res->lockname.len, res->lockname.name, node);
-			__dlm_print_one_lock_resource(res);
+			dlm_print_one_lock_resource(res);
 		}
 		ret = 0;
 		goto done;
@@ -2408,7 +2417,7 @@ static void dlm_deref_lockres_worker(struct dlm_work_item *item, void *data)
 		mlog(ML_ERROR, "%s:%.*s: node %u trying to drop ref "
 		     "but it is already dropped!\n", dlm->name,
 		     res->lockname.len, res->lockname.name, node);
-		__dlm_print_one_lock_resource(res);
+		dlm_print_one_lock_resource(res);
 	}
 
 	dlm_lockres_put(res);
@@ -2933,6 +2942,9 @@ static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
 				dlm_lockres_clear_refmap_bit(lock->ml.node, res);
 				list_del_init(&lock->list);
 				dlm_lock_put(lock);
+				/* In a normal unlock, we would have added a
+				 * DLM_UNLOCK_FREE_LOCK action. Force it. */
+				dlm_lock_put(lock);
 			}
 		}
 		queue++;
diff --git a/fs/ocfs2/dlm/dlmrecovery.c b/fs/ocfs2/dlm/dlmrecovery.c
index 91f747b..bcb9260 100644
--- a/fs/ocfs2/dlm/dlmrecovery.c
+++ b/fs/ocfs2/dlm/dlmrecovery.c
@@ -519,9 +519,9 @@ static int dlm_do_recovery(struct dlm_ctxt *dlm)
 	return 0;
 
 master_here:
-	mlog(0, "(%d) mastering recovery of %s:%u here(this=%u)!\n",
-	     task_pid_nr(dlm->dlm_reco_thread_task),
-	     dlm->name, dlm->reco.dead_node, dlm->node_num);
+	mlog(ML_NOTICE, "(%d) Node %u is the Recovery Master for the Dead Node "
+	     "%u for Domain %s\n", task_pid_nr(dlm->dlm_reco_thread_task),
+	     dlm->node_num, dlm->reco.dead_node, dlm->name);
 
 	status = dlm_remaster_locks(dlm, dlm->reco.dead_node);
 	if (status < 0) {
@@ -1191,7 +1191,7 @@ static int dlm_add_lock_to_array(struct dlm_lock *lock,
 			    (ml->type == LKM_EXMODE ||
 			     memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))) {
 				mlog(ML_ERROR, "mismatched lvbs!\n");
-				__dlm_print_one_lock_resource(lock->lockres);
+				dlm_print_one_lock_resource(lock->lockres);
 				BUG();
 			}
 			memcpy(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN);
@@ -1327,6 +1327,7 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
 		(struct dlm_migratable_lockres *)msg->buf;
 	int ret = 0;
 	u8 real_master;
+	u8 extra_refs = 0;
 	char *buf = NULL;
 	struct dlm_work_item *item = NULL;
 	struct dlm_lock_resource *res = NULL;
@@ -1404,16 +1405,28 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
 		__dlm_insert_lockres(dlm, res);
 		spin_unlock(&dlm->spinlock);
 
+		/* Add an extra ref for this lock-less lockres lest the
+		 * dlm_thread purges it before we get the chance to add
+		 * locks to it */
+		dlm_lockres_get(res);
+
+		/* There are three refs that need to be put.
+		 * 1. Taken above.
+		 * 2. kref_init in dlm_new_lockres()->dlm_init_lockres().
+		 * 3. dlm_lookup_lockres()
+		 * The first one is handled at the end of this function. The
+		 * other two are handled in the worker thread after locks have
+		 * been attached. Yes, we don't wait for purge time to match
+		 * kref_init. The lockres will still have atleast one ref
+		 * added because it is in the hash __dlm_insert_lockres() */
+		extra_refs++;
+
 		/* now that the new lockres is inserted,
 		 * make it usable by other processes */
 		spin_lock(&res->spinlock);
 		res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
 		spin_unlock(&res->spinlock);
 		wake_up(&res->wq);
-
-		/* add an extra ref for just-allocated lockres 
-		 * otherwise the lockres will be purged immediately */
-		dlm_lockres_get(res);
 	}
 
 	/* at this point we have allocated everything we need,
@@ -1443,12 +1456,17 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
 	dlm_init_work_item(dlm, item, dlm_mig_lockres_worker, buf);
 	item->u.ml.lockres = res; /* already have a ref */
 	item->u.ml.real_master = real_master;
+	item->u.ml.extra_ref = extra_refs;
 	spin_lock(&dlm->work_lock);
 	list_add_tail(&item->list, &dlm->work_list);
 	spin_unlock(&dlm->work_lock);
 	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
 
 leave:
+	/* One extra ref taken needs to be put here */
+	if (extra_refs)
+		dlm_lockres_put(res);
+
 	dlm_put(dlm);
 	if (ret < 0) {
 		if (buf)
@@ -1464,17 +1482,19 @@ leave:
 
 static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data)
 {
-	struct dlm_ctxt *dlm = data;
+	struct dlm_ctxt *dlm;
 	struct dlm_migratable_lockres *mres;
 	int ret = 0;
 	struct dlm_lock_resource *res;
 	u8 real_master;
+	u8 extra_ref;
 
 	dlm = item->dlm;
 	mres = (struct dlm_migratable_lockres *)data;
 
 	res = item->u.ml.lockres;
 	real_master = item->u.ml.real_master;
+	extra_ref = item->u.ml.extra_ref;
 
 	if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) {
 		/* this case is super-rare. only occurs if
@@ -1517,6 +1537,12 @@ again:
 	}
 
 leave:
+	/* See comment in dlm_mig_lockres_handler() */
+	if (res) {
+		if (extra_ref)
+			dlm_lockres_put(res);
+		dlm_lockres_put(res);
+	}
 	kfree(data);
 	mlog_exit(ret);
 }
@@ -1644,7 +1670,8 @@ int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
 				/* retry!? */
 				BUG();
 			}
-		}
+		} else /* put.. incase we are not the master */
+			dlm_lockres_put(res);
 		spin_unlock(&res->spinlock);
 	}
 	spin_unlock(&dlm->spinlock);
@@ -1921,6 +1948,7 @@ void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
 		     "Recovering res %s:%.*s, is already on recovery list!\n",
 		     dlm->name, res->lockname.len, res->lockname.name);
 		list_del_init(&res->recovering);
+		dlm_lockres_put(res);
 	}
 	/* We need to hold a reference while on the recovery list */
 	dlm_lockres_get(res);
@@ -2130,11 +2158,16 @@ static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
 	assert_spin_locked(&dlm->spinlock);
 	assert_spin_locked(&res->spinlock);
 
+	/* We do two dlm_lock_put(). One for removing from list and the other is
+	 * to force the DLM_UNLOCK_FREE_LOCK action so as to free the locks */
+
 	/* TODO: check pending_asts, pending_basts here */
 	list_for_each_entry_safe(lock, next, &res->granted, list) {
 		if (lock->ml.node == dead_node) {
 			list_del_init(&lock->list);
 			dlm_lock_put(lock);
+			/* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+			dlm_lock_put(lock);
 			freed++;
 		}
 	}
@@ -2142,6 +2175,8 @@ static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
 		if (lock->ml.node == dead_node) {
 			list_del_init(&lock->list);
 			dlm_lock_put(lock);
+			/* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+			dlm_lock_put(lock);
 			freed++;
 		}
 	}
@@ -2149,6 +2184,8 @@ static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
 		if (lock->ml.node == dead_node) {
 			list_del_init(&lock->list);
 			dlm_lock_put(lock);
+			/* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+			dlm_lock_put(lock);
 			freed++;
 		}
 	}
diff --git a/fs/ocfs2/dlm/dlmthread.c b/fs/ocfs2/dlm/dlmthread.c
index cebd089..4060bb3 100644
--- a/fs/ocfs2/dlm/dlmthread.c
+++ b/fs/ocfs2/dlm/dlmthread.c
@@ -176,12 +176,14 @@ static int dlm_purge_lockres(struct dlm_ctxt *dlm,
 	     res->lockname.name, master);
 
 	if (!master) {
+		/* drop spinlock...  retake below */
+		spin_unlock(&dlm->spinlock);
+
 		spin_lock(&res->spinlock);
 		/* This ensures that clear refmap is sent after the set */
 		__dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG);
 		spin_unlock(&res->spinlock);
-		/* drop spinlock to do messaging, retake below */
-		spin_unlock(&dlm->spinlock);
+
 		/* clear our bit from the master's refmap, ignore errors */
 		ret = dlm_drop_lockres_ref(dlm, res);
 		if (ret < 0) {
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
index f779430..1f1873b 100644
--- a/fs/ocfs2/dlmglue.c
+++ b/fs/ocfs2/dlmglue.c
@@ -2409,7 +2409,7 @@ static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
 	return 0;
 }
 
-static struct seq_operations ocfs2_dlm_seq_ops = {
+static const struct seq_operations ocfs2_dlm_seq_ops = {
 	.start =	ocfs2_dlm_seq_start,
 	.stop =		ocfs2_dlm_seq_stop,
 	.next =		ocfs2_dlm_seq_next,
diff --git a/fs/ocfs2/resize.c b/fs/ocfs2/resize.c
index 37835ff..8166968 100644
--- a/fs/ocfs2/resize.c
+++ b/fs/ocfs2/resize.c
@@ -597,7 +597,7 @@ int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input)
 		memset(cr, 0, sizeof(struct ocfs2_chain_rec));
 	}
 
-	cr->c_blkno = le64_to_cpu(input->group);
+	cr->c_blkno = cpu_to_le64(input->group);
 	le32_add_cpu(&cr->c_total, input->clusters * cl_bpc);
 	le32_add_cpu(&cr->c_free, input->frees * cl_bpc);
 

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2008-03-04 18:46 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2008-03-04 18:46 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

Hi Linus,

Here are some Ocfs2 fixes and minor cleanups for 2.6.24-rc4.

Full patch attached since this is all pretty small.
	--Mark


Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/aops.c          |    2 +-
 fs/ocfs2/dir.c           |    5 ++---
 fs/ocfs2/dlm/dlmmaster.c |    6 +++---
 fs/ocfs2/dlmglue.c       |    8 ++++----
 fs/ocfs2/dlmglue.h       |    2 --
 fs/ocfs2/heartbeat.c     |   26 +++++++++++++-------------
 fs/ocfs2/heartbeat.h     |    5 -----
 fs/ocfs2/localalloc.c    |    6 +-----
 8 files changed, 24 insertions(+), 36 deletions(-)

Adrian Bunk (3):
      [2.6 patch] fs/ocfs2/: possible cleanups
      [2.6 patch] make ocfs2_downconvert_thread() static
      [2.6 patch] ocfs2: make dlm_do_assert_master() static

Julia Lawall (1):
      fs/ocfs2/aops.c: Correct use of ! and &

Marcin Slusarz (1):
      ocfs2: le*_add_cpu conversion

Mark Fasheh (1):
      ocfs2: Fix writeout in ocfs2_data_convert_worker()

Sunil Mushran (1):
      ocfs2: Enable localalloc for local mounts

diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 8224312..90383ed 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -257,7 +257,7 @@ static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 
 	BUG_ON(!PageLocked(page));
-	BUG_ON(!OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL);
+	BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
 
 	ret = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &di_bh,
 			       OCFS2_BH_CACHED, inode);
diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
index e280833..8a18758 100644
--- a/fs/ocfs2/dir.c
+++ b/fs/ocfs2/dir.c
@@ -390,9 +390,8 @@ static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
 				goto bail;
 			}
 			if (pde)
-				pde->rec_len =
-					cpu_to_le16(le16_to_cpu(pde->rec_len) +
-						    le16_to_cpu(de->rec_len));
+				le16_add_cpu(&pde->rec_len,
+						le16_to_cpu(de->rec_len));
 			else
 				de->inode = 0;
 			dir->i_version++;
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index a54d33d..c92d1b1 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -1695,9 +1695,9 @@ send_response:
  * can periodically run all locks owned by this node
  * and re-assert across the cluster...
  */
-int dlm_do_assert_master(struct dlm_ctxt *dlm,
-			 struct dlm_lock_resource *res,
-			 void *nodemap, u32 flags)
+static int dlm_do_assert_master(struct dlm_ctxt *dlm,
+				struct dlm_lock_resource *res,
+				void *nodemap, u32 flags)
 {
 	struct dlm_assert_master assert;
 	int to, tmpret;
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
index 351130c..f779430 100644
--- a/fs/ocfs2/dlmglue.c
+++ b/fs/ocfs2/dlmglue.c
@@ -3042,7 +3042,7 @@ static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
        	inode = ocfs2_lock_res_inode(lockres);
 	mapping = inode->i_mapping;
 
-	if (S_ISREG(inode->i_mode))
+	if (!S_ISREG(inode->i_mode))
 		goto out;
 
 	/*
@@ -3219,8 +3219,8 @@ static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
 	return UNBLOCK_CONTINUE_POST;
 }
 
-void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
-				struct ocfs2_lock_res *lockres)
+static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
+				       struct ocfs2_lock_res *lockres)
 {
 	int status;
 	struct ocfs2_unblock_ctl ctl = {0, 0,};
@@ -3356,7 +3356,7 @@ static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
 	return should_wake;
 }
 
-int ocfs2_downconvert_thread(void *arg)
+static int ocfs2_downconvert_thread(void *arg)
 {
 	int status = 0;
 	struct ocfs2_super *osb = arg;
diff --git a/fs/ocfs2/dlmglue.h b/fs/ocfs2/dlmglue.h
index 1d5b069..e3cf902 100644
--- a/fs/ocfs2/dlmglue.h
+++ b/fs/ocfs2/dlmglue.h
@@ -109,8 +109,6 @@ void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
 			       struct ocfs2_lock_res *lockres);
 
 /* for the downconvert thread */
-void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
-				struct ocfs2_lock_res *lockres);
 void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb);
 
 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void);
diff --git a/fs/ocfs2/heartbeat.c b/fs/ocfs2/heartbeat.c
index c0efd94..0758daf 100644
--- a/fs/ocfs2/heartbeat.c
+++ b/fs/ocfs2/heartbeat.c
@@ -49,10 +49,15 @@ static inline void __ocfs2_node_map_set_bit(struct ocfs2_node_map *map,
 static inline void __ocfs2_node_map_clear_bit(struct ocfs2_node_map *map,
 					      int bit);
 static inline int __ocfs2_node_map_is_empty(struct ocfs2_node_map *map);
-static void __ocfs2_node_map_dup(struct ocfs2_node_map *target,
-				 struct ocfs2_node_map *from);
-static void __ocfs2_node_map_set(struct ocfs2_node_map *target,
-				 struct ocfs2_node_map *from);
+
+/* special case -1 for now
+ * TODO: should *really* make sure the calling func never passes -1!!  */
+static void ocfs2_node_map_init(struct ocfs2_node_map *map)
+{
+	map->num_nodes = OCFS2_NODE_MAP_MAX_NODES;
+	memset(map->map, 0, BITS_TO_LONGS(OCFS2_NODE_MAP_MAX_NODES) *
+	       sizeof(unsigned long));
+}
 
 void ocfs2_init_node_maps(struct ocfs2_super *osb)
 {
@@ -136,15 +141,6 @@ void ocfs2_stop_heartbeat(struct ocfs2_super *osb)
 		mlog_errno(ret);
 }
 
-/* special case -1 for now
- * TODO: should *really* make sure the calling func never passes -1!!  */
-void ocfs2_node_map_init(struct ocfs2_node_map *map)
-{
-	map->num_nodes = OCFS2_NODE_MAP_MAX_NODES;
-	memset(map->map, 0, BITS_TO_LONGS(OCFS2_NODE_MAP_MAX_NODES) *
-	       sizeof(unsigned long));
-}
-
 static inline void __ocfs2_node_map_set_bit(struct ocfs2_node_map *map,
 					    int bit)
 {
@@ -216,6 +212,8 @@ int ocfs2_node_map_is_empty(struct ocfs2_super *osb,
 	return ret;
 }
 
+#if 0
+
 static void __ocfs2_node_map_dup(struct ocfs2_node_map *target,
 				 struct ocfs2_node_map *from)
 {
@@ -254,6 +252,8 @@ static void __ocfs2_node_map_set(struct ocfs2_node_map *target,
 		target->map[i] = from->map[i];
 }
 
+#endif  /*  0  */
+
 /* Returns whether the recovery bit was actually set - it may not be
  * if a node is still marked as needing recovery */
 int ocfs2_recovery_map_set(struct ocfs2_super *osb,
diff --git a/fs/ocfs2/heartbeat.h b/fs/ocfs2/heartbeat.h
index 5685921..eac63ae 100644
--- a/fs/ocfs2/heartbeat.h
+++ b/fs/ocfs2/heartbeat.h
@@ -33,7 +33,6 @@ void ocfs2_stop_heartbeat(struct ocfs2_super *osb);
 
 /* node map functions - used to keep track of mounted and in-recovery
  * nodes. */
-void ocfs2_node_map_init(struct ocfs2_node_map *map);
 int ocfs2_node_map_is_empty(struct ocfs2_super *osb,
 			    struct ocfs2_node_map *map);
 void ocfs2_node_map_set_bit(struct ocfs2_super *osb,
@@ -57,9 +56,5 @@ int ocfs2_recovery_map_set(struct ocfs2_super *osb,
 			   int num);
 void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
 			      int num);
-/* returns 1 if bit is the only bit set in target, 0 otherwise */
-int ocfs2_node_map_is_only(struct ocfs2_super *osb,
-			   struct ocfs2_node_map *target,
-			   int bit);
 
 #endif /* OCFS2_HEARTBEAT_H */
diff --git a/fs/ocfs2/localalloc.c b/fs/ocfs2/localalloc.c
index add1ffd..ab83fd5 100644
--- a/fs/ocfs2/localalloc.c
+++ b/fs/ocfs2/localalloc.c
@@ -120,9 +120,6 @@ int ocfs2_load_local_alloc(struct ocfs2_super *osb)
 
 	mlog_entry_void();
 
-	if (ocfs2_mount_local(osb))
-		goto bail;
-
 	if (osb->local_alloc_size == 0)
 		goto bail;
 
@@ -588,8 +585,7 @@ int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
 	while(bits_wanted--)
 		ocfs2_set_bit(start++, bitmap);
 
-	alloc->id1.bitmap1.i_used = cpu_to_le32(*num_bits +
-				le32_to_cpu(alloc->id1.bitmap1.i_used));
+	le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits);
 
 	status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
 	if (status < 0) {

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-12-17 18:58 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-12-17 18:58 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/alloc.c   |   68 ++++++++++++++++++++++++++++++++++-----------------
 fs/ocfs2/journal.c |   13 +++++++++-
 2 files changed, 57 insertions(+), 24 deletions(-)

Mark Fasheh (4):
      ocfs2: fix exit-while-locked bug in ocfs2_queue_orphans()
      ocfs2: Don't panic when truncating an empty extent
      ocfs2: Allow for debugging of transaction extends
      ocfs2: Re-journal buffers after transaction extend

diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
index ce62c15..23c8cda 100644
--- a/fs/ocfs2/alloc.c
+++ b/fs/ocfs2/alloc.c
@@ -2389,6 +2389,18 @@ static int __ocfs2_rotate_tree_left(struct inode *inode,
 			goto out;
 		}
 
+		/*
+		 * Caller might still want to make changes to the
+		 * tree root, so re-add it to the journal here.
+		 */
+		ret = ocfs2_journal_access(handle, inode,
+					   path_root_bh(left_path),
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
 		ret = ocfs2_rotate_subtree_left(inode, handle, left_path,
 						right_path, subtree_root,
 						dealloc, &deleted);
@@ -3289,16 +3301,6 @@ static int ocfs2_insert_path(struct inode *inode,
 	int ret, subtree_index;
 	struct buffer_head *leaf_bh = path_leaf_bh(right_path);
 
-	/*
-	 * Pass both paths to the journal. The majority of inserts
-	 * will be touching all components anyway.
-	 */
-	ret = ocfs2_journal_access_path(inode, handle, right_path);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out;
-	}
-
 	if (left_path) {
 		int credits = handle->h_buffer_credits;
 
@@ -3323,6 +3325,16 @@ static int ocfs2_insert_path(struct inode *inode,
 		}
 	}
 
+	/*
+	 * Pass both paths to the journal. The majority of inserts
+	 * will be touching all components anyway.
+	 */
+	ret = ocfs2_journal_access_path(inode, handle, right_path);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
 	if (insert->ins_split != SPLIT_NONE) {
 		/*
 		 * We could call ocfs2_insert_at_leaf() for some types
@@ -3331,6 +3343,17 @@ static int ocfs2_insert_path(struct inode *inode,
 		 */
 		ocfs2_split_record(inode, left_path, right_path,
 				   insert_rec, insert->ins_split);
+
+		/*
+		 * Split might have modified either leaf and we don't
+		 * have a guarantee that the later edge insert will
+		 * dirty this for us.
+		 */
+		if (left_path)
+			ret = ocfs2_journal_dirty(handle,
+						  path_leaf_bh(left_path));
+			if (ret)
+				mlog_errno(ret);
 	} else
 		ocfs2_insert_at_leaf(insert_rec, path_leaf_el(right_path),
 				     insert, inode);
@@ -3430,6 +3453,17 @@ static int ocfs2_do_insert_extent(struct inode *inode,
 			mlog_errno(ret);
 			goto out;
 		}
+
+		/*
+		 * ocfs2_rotate_tree_right() might have extended the
+		 * transaction without re-journaling our tree root.
+		 */
+		ret = ocfs2_journal_access(handle, inode, di_bh,
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
 	} else if (type->ins_appending == APPEND_TAIL
 		   && type->ins_contig != CONTIG_LEFT) {
 		ret = ocfs2_append_rec_to_path(inode, handle, insert_rec,
@@ -3941,7 +3975,7 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 {
 	int ret = 0;
 	struct ocfs2_extent_list *el = path_leaf_el(path);
-	struct buffer_head *eb_bh, *last_eb_bh = NULL;
+	struct buffer_head *last_eb_bh = NULL;
 	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
 	struct ocfs2_merge_ctxt ctxt;
 	struct ocfs2_extent_list *rightmost_el;
@@ -3960,14 +3994,6 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 		goto out;
 	}
 
-	eb_bh = path_leaf_bh(path);
-	ret = ocfs2_journal_access(handle, inode, eb_bh,
-				   OCFS2_JOURNAL_ACCESS_WRITE);
-	if (ret) {
-		mlog_errno(ret);
-		goto out;
-	}
-
 	ctxt.c_contig_type = ocfs2_figure_merge_contig_type(inode, el,
 							    split_index,
 							    split_rec);
@@ -4029,8 +4055,6 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 			mlog_errno(ret);
 	}
 
-	ocfs2_journal_dirty(handle, eb_bh);
-
 out:
 	brelse(last_eb_bh);
 	return ret;
@@ -6093,8 +6117,6 @@ start:
 	mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n",
 	     clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr);
 
-	BUG_ON(clusters_to_del == 0);
-
 	mutex_lock(&tl_inode->i_mutex);
 	tl_sem = 1;
 	/* ocfs2_truncate_log_needs_flush guarantees us at least one
diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
index f9d01e2..8d81f6c 100644
--- a/fs/ocfs2/journal.c
+++ b/fs/ocfs2/journal.c
@@ -174,6 +174,12 @@ int ocfs2_commit_trans(struct ocfs2_super *osb,
  * transaction. extend_trans will either extend the current handle by
  * nblocks, or commit it and start a new one with nblocks credits.
  *
+ * This might call journal_restart() which will commit dirty buffers
+ * and then restart the transaction. Before calling
+ * ocfs2_extend_trans(), any changed blocks should have been
+ * dirtied. After calling it, all blocks which need to be changed must
+ * go through another set of journal_access/journal_dirty calls.
+ *
  * WARNING: This will not release any semaphores or disk locks taken
  * during the transaction, so make sure they were taken *before*
  * start_trans or we'll have ordering deadlocks.
@@ -193,11 +199,15 @@ int ocfs2_extend_trans(handle_t *handle, int nblocks)
 
 	mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
 
+#ifdef OCFS2_DEBUG_FS
+	status = 1;
+#else
 	status = journal_extend(handle, nblocks);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
 	}
+#endif
 
 	if (status > 0) {
 		mlog(0, "journal_extend failed, trying journal_restart\n");
@@ -1277,11 +1287,12 @@ static int ocfs2_queue_orphans(struct ocfs2_super *osb,
 				   ocfs2_orphan_filldir);
 	if (status) {
 		mlog_errno(status);
-		goto out;
+		goto out_cluster;
 	}
 
 	*head = priv.head;
 
+out_cluster:
 	ocfs2_meta_unlock(orphan_dir_inode, 0);
 out:
 	mutex_unlock(&orphan_dir_inode->i_mutex);

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-11-28  1:15 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-11-28  1:15 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

Hi Linus,
	Here's some more Ocfs2 patches for 2.6.24. This series also includes
a small performance fixup from Jan Kara. The rest of the patches are bug
fixes or very trivial cleanups.

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/Kconfig                 |    9 +++++++++
 fs/ocfs2/aops.c            |    2 +-
 fs/ocfs2/cluster/masklog.h |    2 +-
 fs/ocfs2/dcache.c          |   20 ++++++++++++++++----
 fs/ocfs2/dlm/dlmmaster.c   |    4 ++--
 fs/ocfs2/file.c            |   19 +++++++++++++++----
 fs/ocfs2/inode.c           |    6 +++---
 fs/ocfs2/localalloc.c      |    5 +++--
 fs/ocfs2/super.c           |    6 +++---
 9 files changed, 53 insertions(+), 20 deletions(-)

Jan Kara (1):
      ocfs2: Remove expensive bitmap scanning

Joe Perches (1):
      fs/ocfs2: Add missing "space"

Mark Fasheh (6):
      ocfs2: Reset journal parameters after s_mount_opt update
      ocfs2: Filter -ENOSPC in mlog_errno()
      ocfs2: log valid inode # on bad inode
      ocfs2: Remove bug statement in ocfs2_dentry_iput()
      ocfs2: Fix comparison in ocfs2_size_fits_inline_data()
      ocfs2: reverse inline-data truncate args

diff --git a/fs/Kconfig b/fs/Kconfig
index 429a002..635f3e2 100644
--- a/fs/Kconfig
+++ b/fs/Kconfig
@@ -459,6 +459,15 @@ config OCFS2_DEBUG_MASKLOG
 	  This option will enlarge your kernel, but it allows debugging of
 	  ocfs2 filesystem issues.
 
+config OCFS2_DEBUG_FS
+	bool "OCFS2 expensive checks"
+	depends on OCFS2_FS
+	default n
+	help
+	  This option will enable expensive consistency checks. Enable
+	  this option for debugging only as it is likely to decrease
+	  performance of the filesystem.
+
 config MINIX_FS
 	tristate "Minix fs support"
 	help
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 556e34c..56f7790 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -1514,7 +1514,7 @@ int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
 {
 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 
-	if (new_size < le16_to_cpu(di->id2.i_data.id_count))
+	if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
 		return 1;
 	return 0;
 }
diff --git a/fs/ocfs2/cluster/masklog.h b/fs/ocfs2/cluster/masklog.h
index cd04606..597e064 100644
--- a/fs/ocfs2/cluster/masklog.h
+++ b/fs/ocfs2/cluster/masklog.h
@@ -212,7 +212,7 @@ extern struct mlog_bits mlog_and_bits, mlog_not_bits;
 #define mlog_errno(st) do {						\
 	int _st = (st);							\
 	if (_st != -ERESTARTSYS && _st != -EINTR &&			\
-	    _st != AOP_TRUNCATED_PAGE)					\
+	    _st != AOP_TRUNCATED_PAGE && _st != -ENOSPC)		\
 		mlog(ML_ERROR, "status = %lld\n", (long long)_st);	\
 } while (0)
 
diff --git a/fs/ocfs2/dcache.c b/fs/ocfs2/dcache.c
index 1957a5e..9923278 100644
--- a/fs/ocfs2/dcache.c
+++ b/fs/ocfs2/dcache.c
@@ -344,12 +344,24 @@ static void ocfs2_dentry_iput(struct dentry *dentry, struct inode *inode)
 {
 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 
-	mlog_bug_on_msg(!dl && !(dentry->d_flags & DCACHE_DISCONNECTED),
-			"dentry: %.*s\n", dentry->d_name.len,
-			dentry->d_name.name);
+	if (!dl) {
+		/*
+		 * No dentry lock is ok if we're disconnected or
+		 * unhashed.
+		 */
+		if (!(dentry->d_flags & DCACHE_DISCONNECTED) &&
+		    !d_unhashed(dentry)) {
+			unsigned long long ino = 0ULL;
+			if (inode)
+				ino = (unsigned long long)OCFS2_I(inode)->ip_blkno;
+			mlog(ML_ERROR, "Dentry is missing cluster lock. "
+			     "inode: %llu, d_flags: 0x%x, d_name: %.*s\n",
+			     ino, dentry->d_flags, dentry->d_name.len,
+			     dentry->d_name.name);
+		}
 
-	if (!dl)
 		goto out;
+	}
 
 	mlog_bug_on_msg(dl->dl_count == 0, "dentry: %.*s, count: %u\n",
 			dentry->d_name.len, dentry->d_name.name,
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index 62e4a7d..a54d33d 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -908,7 +908,7 @@ lookup:
 		 * but they might own this lockres.  wait on them. */
 		bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
 		if (bit < O2NM_MAX_NODES) {
-			mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to"
+			mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to "
 			     "recover before lock mastery can begin\n",
 			     dlm->name, namelen, (char *)lockid, bit);
 			wait_on_recovery = 1;
@@ -962,7 +962,7 @@ redo_request:
 		spin_lock(&dlm->spinlock);
 		bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
 		if (bit < O2NM_MAX_NODES) {
-			mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to"
+			mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to "
 			     "recover before lock mastery can begin\n",
 			     dlm->name, namelen, (char *)lockid, bit);
 			wait_on_recovery = 1;
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index bbac7cd..b75b2e1 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -399,7 +399,7 @@ static int ocfs2_truncate_file(struct inode *inode,
 
 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 		status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
-					       i_size_read(inode), 0);
+					       i_size_read(inode), 1);
 		if (status)
 			mlog_errno(status);
 
@@ -1521,6 +1521,7 @@ static int ocfs2_remove_inode_range(struct inode *inode,
 	u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct address_space *mapping = inode->i_mapping;
 
 	ocfs2_init_dealloc_ctxt(&dealloc);
 
@@ -1529,10 +1530,20 @@ static int ocfs2_remove_inode_range(struct inode *inode,
 
 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 		ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
-					    byte_start + byte_len, 1);
-		if (ret)
+					    byte_start + byte_len, 0);
+		if (ret) {
 			mlog_errno(ret);
-		return ret;
+			goto out;
+		}
+		/*
+		 * There's no need to get fancy with the page cache
+		 * truncate of an inline-data inode. We're talking
+		 * about less than a page here, which will be cached
+		 * in the dinode buffer anyway.
+		 */
+		unmap_mapping_range(mapping, 0, 0, 0);
+		truncate_inode_pages(mapping, 0);
+		goto out;
 	}
 
 	trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
index 1d5e0cb..ebb2bbe 100644
--- a/fs/ocfs2/inode.c
+++ b/fs/ocfs2/inode.c
@@ -455,8 +455,8 @@ static int ocfs2_read_locked_inode(struct inode *inode,
 	status = -EINVAL;
 	fe = (struct ocfs2_dinode *) bh->b_data;
 	if (!OCFS2_IS_VALID_DINODE(fe)) {
-		mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
-		     (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
+		mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
+		     (unsigned long long)args->fi_blkno, 7,
 		     fe->i_signature);
 		goto bail;
 	}
@@ -863,7 +863,7 @@ static int ocfs2_query_inode_wipe(struct inode *inode,
 	status = ocfs2_try_open_lock(inode, 1);
 	if (status == -EAGAIN) {
 		status = 0;
-		mlog(0, "Skipping delete of %llu because it is in use on"
+		mlog(0, "Skipping delete of %llu because it is in use on "
 		     "other nodes\n", (unsigned long long)oi->ip_blkno);
 		goto bail;
 	}
diff --git a/fs/ocfs2/localalloc.c b/fs/ocfs2/localalloc.c
index d272847..58ea88b 100644
--- a/fs/ocfs2/localalloc.c
+++ b/fs/ocfs2/localalloc.c
@@ -484,6 +484,7 @@ int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
 
 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
 
+#ifdef OCFS2_DEBUG_FS
 	if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
 	    ocfs2_local_alloc_count_bits(alloc)) {
 		ocfs2_error(osb->sb, "local alloc inode %llu says it has "
@@ -494,6 +495,7 @@ int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
 		status = -EIO;
 		goto bail;
 	}
+#endif
 
 	free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
 		le32_to_cpu(alloc->id1.bitmap1.i_used);
@@ -712,9 +714,8 @@ static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
 	void *bitmap;
 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
 
-	mlog_entry("total = %u, COUNT = %u, used = %u\n",
+	mlog_entry("total = %u, used = %u\n",
 		   le32_to_cpu(alloc->id1.bitmap1.i_total),
-		   ocfs2_local_alloc_count_bits(alloc),
 		   le32_to_cpu(alloc->id1.bitmap1.i_used));
 
 	if (!alloc->id1.bitmap1.i_total) {
diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
index be562ac..5ee7754 100644
--- a/fs/ocfs2/super.c
+++ b/fs/ocfs2/super.c
@@ -438,14 +438,14 @@ unlock_osb:
 	}
 
 	if (!ret) {
-		if (!ocfs2_is_hard_readonly(osb))
-			ocfs2_set_journal_params(osb);
-
 		/* Only save off the new mount options in case of a successful
 		 * remount. */
 		osb->s_mount_opt = parsed_options.mount_opt;
 		osb->s_atime_quantum = parsed_options.atime_quantum;
 		osb->preferred_slot = parsed_options.slot;
+
+		if (!ocfs2_is_hard_readonly(osb))
+			ocfs2_set_journal_params(osb);
 	}
 out:
 	return ret;

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-11-07 17:41 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-11-07 17:41 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

Hi Linus,
	Here are some Ocfs2 patches - bug fixes, performance fixes and small
cleanups.
	--Mark

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/alloc.c             |    2 +-
 fs/ocfs2/aops.c              |   22 ++++++++++++++++++++++
 fs/ocfs2/cluster/heartbeat.c |    2 +-
 fs/ocfs2/dcache.c            |    2 +-
 fs/ocfs2/dir.c               |    6 +++---
 fs/ocfs2/dlmglue.c           |   25 ++++++++++---------------
 fs/ocfs2/file.c              |   26 +++++++++++++++++++++++++-
 fs/ocfs2/namei.c             |   13 ++++++++++---
 8 files changed, 73 insertions(+), 25 deletions(-)

Adrian Bunk (1):
      [2.6 patch] make ocfs2_find_entry_el() static

Jan Kara (1):
      Fix possibly too long write in o2hb_setup_one_bio()

Mark Fasheh (4):
      ocfs2: Create locks at initially requested level
      ocfs2: Re-order iput in ocfs2_drop_dentry_lock
      ocfs2: Commit journal on sync writes
      ocfs2: fix write() performance regression

Roel Kluin (1):
      Fix priority mistakes in fs/ocfs2/{alloc.c, dlmglue.c}

Srinivas Eeda (1):
      ocfs2: fix rename vs unlink race

diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
index 4ba7f0b..ce62c15 100644
--- a/fs/ocfs2/alloc.c
+++ b/fs/ocfs2/alloc.c
@@ -3946,7 +3946,7 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 	struct ocfs2_merge_ctxt ctxt;
 	struct ocfs2_extent_list *rightmost_el;
 
-	if (!rec->e_flags & OCFS2_EXT_UNWRITTEN) {
+	if (!(rec->e_flags & OCFS2_EXT_UNWRITTEN)) {
 		ret = -EIO;
 		mlog_errno(ret);
 		goto out;
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index c69c1b3..556e34c 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -729,6 +729,27 @@ static void ocfs2_clear_page_regions(struct page *page,
 }
 
 /*
+ * Nonsparse file systems fully allocate before we get to the write
+ * code. This prevents ocfs2_write() from tagging the write as an
+ * allocating one, which means ocfs2_map_page_blocks() might try to
+ * read-in the blocks at the tail of our file. Avoid reading them by
+ * testing i_size against each block offset.
+ */
+static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
+				 unsigned int block_start)
+{
+	u64 offset = page_offset(page) + block_start;
+
+	if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
+		return 1;
+
+	if (i_size_read(inode) > offset)
+		return 1;
+
+	return 0;
+}
+
+/*
  * Some of this taken from block_prepare_write(). We already have our
  * mapping by now though, and the entire write will be allocating or
  * it won't, so not much need to use BH_New.
@@ -781,6 +802,7 @@ int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
 				set_buffer_uptodate(bh);
 		} else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
 			   !buffer_new(bh) &&
+			   ocfs2_should_read_blk(inode, page, block_start) &&
 			   (block_start < from || block_end > to)) {
 			ll_rw_block(READ, 1, &bh);
 			*wait_bh++=bh;
diff --git a/fs/ocfs2/cluster/heartbeat.c b/fs/ocfs2/cluster/heartbeat.c
index 9cc7c04..f02ccb3 100644
--- a/fs/ocfs2/cluster/heartbeat.c
+++ b/fs/ocfs2/cluster/heartbeat.c
@@ -267,7 +267,7 @@ static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
 		current_page = cs / spp;
 		page = reg->hr_slot_data[current_page];
 
-		vec_len = min(PAGE_CACHE_SIZE,
+		vec_len = min(PAGE_CACHE_SIZE - vec_start,
 			      (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
 
 		mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
diff --git a/fs/ocfs2/dcache.c b/fs/ocfs2/dcache.c
index 3094ddb..1957a5e 100644
--- a/fs/ocfs2/dcache.c
+++ b/fs/ocfs2/dcache.c
@@ -318,9 +318,9 @@ out_attach:
 static void ocfs2_drop_dentry_lock(struct ocfs2_super *osb,
 				   struct ocfs2_dentry_lock *dl)
 {
+	iput(dl->dl_inode);
 	ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 	ocfs2_lock_res_free(&dl->dl_lockres);
-	iput(dl->dl_inode);
 	kfree(dl);
 }
 
diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
index 6a2f143..63b28fd 100644
--- a/fs/ocfs2/dir.c
+++ b/fs/ocfs2/dir.c
@@ -208,9 +208,9 @@ out:
 	return NULL;
 }
 
-struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
-					struct inode *dir,
-					struct ocfs2_dir_entry **res_dir)
+static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
+					       struct inode *dir,
+					       struct ocfs2_dir_entry **res_dir)
 {
 	struct super_block *sb;
 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
index 41c76ff..4e97dcc 100644
--- a/fs/ocfs2/dlmglue.c
+++ b/fs/ocfs2/dlmglue.c
@@ -670,7 +670,7 @@ static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *loc
 {
 	mlog_entry_void();
 
-	BUG_ON((!lockres->l_flags & OCFS2_LOCK_BUSY));
+	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
 
 	if (lockres->l_requested > LKM_NLMODE &&
@@ -980,18 +980,6 @@ again:
 		goto unlock;
 	}
 
-	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
-		/* lock has not been created yet. */
-		spin_unlock_irqrestore(&lockres->l_lock, flags);
-
-		ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0);
-		if (ret < 0) {
-			mlog_errno(ret);
-			goto out;
-		}
-		goto again;
-	}
-
 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
 	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
 		/* is the lock is currently blocked on behalf of
@@ -1006,7 +994,14 @@ again:
 			mlog(ML_ERROR, "lockres %s has action %u pending\n",
 			     lockres->l_name, lockres->l_action);
 
-		lockres->l_action = OCFS2_AST_CONVERT;
+		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
+			lockres->l_action = OCFS2_AST_ATTACH;
+			lkm_flags &= ~LKM_CONVERT;
+		} else {
+			lockres->l_action = OCFS2_AST_CONVERT;
+			lkm_flags |= LKM_CONVERT;
+		}
+
 		lockres->l_requested = level;
 		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
 		spin_unlock_irqrestore(&lockres->l_lock, flags);
@@ -1021,7 +1016,7 @@ again:
 		status = dlmlock(osb->dlm,
 				 level,
 				 &lockres->l_lksb,
-				 lkm_flags|LKM_CONVERT,
+				 lkm_flags,
 				 lockres->l_name,
 				 OCFS2_LOCK_ID_MAX_LEN - 1,
 				 ocfs2_locking_ast,
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index f92fe91..bbac7cd 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -1891,9 +1891,11 @@ static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
 	ssize_t written = 0;
 	size_t ocount;		/* original count */
 	size_t count;		/* after file limit checks */
-	loff_t *ppos = &iocb->ki_pos;
+	loff_t old_size, *ppos = &iocb->ki_pos;
+	u32 old_clusters;
 	struct file *file = iocb->ki_filp;
 	struct inode *inode = file->f_path.dentry->d_inode;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 
 	mlog_entry("(0x%p, %u, '%.*s')\n", file,
 		   (unsigned int)nr_segs,
@@ -1949,6 +1951,13 @@ relock:
 		goto relock;
 	}
 
+	/*
+	 * To later detect whether a journal commit for sync writes is
+	 * necessary, we sample i_size, and cluster count here.
+	 */
+	old_size = i_size_read(inode);
+	old_clusters = OCFS2_I(inode)->ip_clusters;
+
 	/* communicate with ocfs2_dio_end_io */
 	ocfs2_iocb_set_rw_locked(iocb, rw_level);
 
@@ -1978,6 +1987,21 @@ out_dio:
 	/* buffered aio wouldn't have proper lock coverage today */
 	BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
 
+	if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
+		/*
+		 * The generic write paths have handled getting data
+		 * to disk, but since we don't make use of the dirty
+		 * inode list, a manual journal commit is necessary
+		 * here.
+		 */
+		if (old_size != i_size_read(inode) ||
+		    old_clusters != OCFS2_I(inode)->ip_clusters) {
+			ret = journal_force_commit(osb->journal->j_journal);
+			if (ret < 0)
+				written = ret;
+		}
+	}
+
 	/* 
 	 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
 	 * function pointer which is called when o_direct io completes so that
diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
index 7292590..989ac27 100644
--- a/fs/ocfs2/namei.c
+++ b/fs/ocfs2/namei.c
@@ -1105,9 +1105,16 @@ static int ocfs2_rename(struct inode *old_dir,
 		goto bail;
 	}
 
-	if (!new_de && new_inode)
-		mlog(ML_ERROR, "inode %lu does not exist in it's parent "
-		     "directory!", new_inode->i_ino);
+	if (!new_de && new_inode) {
+		/*
+		 * Target was unlinked by another node while we were
+		 * waiting to get to ocfs2_rename(). There isn't
+		 * anything we can do here to help the situation, so
+		 * bubble up the appropriate error.
+		 */
+		status = -ENOENT;
+		goto bail;
+	}
 
 	/* In case we need to overwrite an existing file, we blow it
 	 * away first */

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-09-21 16:36 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-09-21 16:36 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

Here are 4 ocfs2 fixes for post -rc7. All fix regressions, 3 related to
running on kernels greater than 4k page size, and one related to our mount
protocol failing on clusters with a mix of 32 and 64 bit systems.
        --Mark

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/aops.c       |   33 +++++++++++++++++++--------------
 fs/ocfs2/file.c       |    4 ++--
 fs/ocfs2/localalloc.c |    4 +---
 fs/ocfs2/localalloc.h |    2 +-
 fs/ocfs2/suballoc.c   |   29 +++++++++++++++++++++--------
 fs/ocfs2/suballoc.h   |   11 +++++++++++
 fs/ocfs2/vote.c       |    4 ++--
 7 files changed, 57 insertions(+), 30 deletions(-)

Mark Fasheh (3):
      ocfs2: Allow smaller allocations during large writes
      ocfs2: Fix pos/len passed to ocfs2_write_cluster
      ocfs2: Don't double set write parameters

Sunil Mushran (1):
      ocfs2: Pack vote message and response structures

diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 50cd8a2..f37f25c 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -930,18 +930,11 @@ static void ocfs2_write_failure(struct inode *inode,
 				loff_t user_pos, unsigned user_len)
 {
 	int i;
-	unsigned from, to;
+	unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
+		to = user_pos + user_len;
 	struct page *tmppage;
 
-	ocfs2_zero_new_buffers(wc->w_target_page, user_pos, user_len);
-
-	if (wc->w_large_pages) {
-		from = wc->w_target_from;
-		to = wc->w_target_to;
-	} else {
-		from = 0;
-		to = PAGE_CACHE_SIZE;
-	}
+	ocfs2_zero_new_buffers(wc->w_target_page, from, to);
 
 	for(i = 0; i < wc->w_num_pages; i++) {
 		tmppage = wc->w_pages[i];
@@ -991,9 +984,6 @@ static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
 			map_from = cluster_start;
 			map_to = cluster_end;
 		}
-
-		wc->w_target_from = map_from;
-		wc->w_target_to = map_to;
 	} else {
 		/*
 		 * If we haven't allocated the new page yet, we
@@ -1211,18 +1201,33 @@ static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
 				       loff_t pos, unsigned len)
 {
 	int ret, i;
+	loff_t cluster_off;
+	unsigned int local_len = len;
 	struct ocfs2_write_cluster_desc *desc;
+	struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
 
 	for (i = 0; i < wc->w_clen; i++) {
 		desc = &wc->w_desc[i];
 
+		/*
+		 * We have to make sure that the total write passed in
+		 * doesn't extend past a single cluster.
+		 */
+		local_len = len;
+		cluster_off = pos & (osb->s_clustersize - 1);
+		if ((cluster_off + local_len) > osb->s_clustersize)
+			local_len = osb->s_clustersize - cluster_off;
+
 		ret = ocfs2_write_cluster(mapping, desc->c_phys,
 					  desc->c_unwritten, data_ac, meta_ac,
-					  wc, desc->c_cpos, pos, len);
+					  wc, desc->c_cpos, pos, local_len);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
 		}
+
+		len -= local_len;
+		pos += local_len;
 	}
 
 	ret = 0;
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 7e34e66..f3bc365 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -491,8 +491,8 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 		goto leave;
 	}
 
-	status = ocfs2_claim_clusters(osb, handle, data_ac, 1,
-				      &bit_off, &num_bits);
+	status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
+					clusters_to_add, &bit_off, &num_bits);
 	if (status < 0) {
 		if (status != -ENOSPC)
 			mlog_errno(status);
diff --git a/fs/ocfs2/localalloc.c b/fs/ocfs2/localalloc.c
index 545f789..de984d2 100644
--- a/fs/ocfs2/localalloc.c
+++ b/fs/ocfs2/localalloc.c
@@ -524,13 +524,12 @@ bail:
 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
 				 handle_t *handle,
 				 struct ocfs2_alloc_context *ac,
-				 u32 min_bits,
+				 u32 bits_wanted,
 				 u32 *bit_off,
 				 u32 *num_bits)
 {
 	int status, start;
 	struct inode *local_alloc_inode;
-	u32 bits_wanted;
 	void *bitmap;
 	struct ocfs2_dinode *alloc;
 	struct ocfs2_local_alloc *la;
@@ -538,7 +537,6 @@ int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
 	mlog_entry_void();
 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
 
-	bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
 	local_alloc_inode = ac->ac_inode;
 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
 	la = OCFS2_LOCAL_ALLOC(alloc);
diff --git a/fs/ocfs2/localalloc.h b/fs/ocfs2/localalloc.h
index 385a101..3f76631 100644
--- a/fs/ocfs2/localalloc.h
+++ b/fs/ocfs2/localalloc.h
@@ -48,7 +48,7 @@ int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
 				 handle_t *handle,
 				 struct ocfs2_alloc_context *ac,
-				 u32 min_bits,
+				 u32 bits_wanted,
 				 u32 *bit_off,
 				 u32 *num_bits);
 
diff --git a/fs/ocfs2/suballoc.c b/fs/ocfs2/suballoc.c
index d9c5c9f..8f09f52 100644
--- a/fs/ocfs2/suballoc.c
+++ b/fs/ocfs2/suballoc.c
@@ -1486,21 +1486,21 @@ static inline void ocfs2_block_to_cluster_group(struct inode *inode,
  * contig. allocation, set to '1' to indicate we can deal with extents
  * of any size.
  */
-int ocfs2_claim_clusters(struct ocfs2_super *osb,
-			 handle_t *handle,
-			 struct ocfs2_alloc_context *ac,
-			 u32 min_clusters,
-			 u32 *cluster_start,
-			 u32 *num_clusters)
+int __ocfs2_claim_clusters(struct ocfs2_super *osb,
+			   handle_t *handle,
+			   struct ocfs2_alloc_context *ac,
+			   u32 min_clusters,
+			   u32 max_clusters,
+			   u32 *cluster_start,
+			   u32 *num_clusters)
 {
 	int status;
-	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
+	unsigned int bits_wanted = max_clusters;
 	u64 bg_blkno = 0;
 	u16 bg_bit_off;
 
 	mlog_entry_void();
 
-	BUG_ON(!ac);
 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
 
 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
@@ -1557,6 +1557,19 @@ bail:
 	return status;
 }
 
+int ocfs2_claim_clusters(struct ocfs2_super *osb,
+			 handle_t *handle,
+			 struct ocfs2_alloc_context *ac,
+			 u32 min_clusters,
+			 u32 *cluster_start,
+			 u32 *num_clusters)
+{
+	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
+
+	return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
+				      bits_wanted, cluster_start, num_clusters);
+}
+
 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
 					       struct inode *alloc_inode,
 					       struct ocfs2_group_desc *bg,
diff --git a/fs/ocfs2/suballoc.h b/fs/ocfs2/suballoc.h
index f212dc0..cafe937 100644
--- a/fs/ocfs2/suballoc.h
+++ b/fs/ocfs2/suballoc.h
@@ -85,6 +85,17 @@ int ocfs2_claim_clusters(struct ocfs2_super *osb,
 			 u32 min_clusters,
 			 u32 *cluster_start,
 			 u32 *num_clusters);
+/*
+ * Use this variant of ocfs2_claim_clusters to specify a maxiumum
+ * number of clusters smaller than the allocation reserved.
+ */
+int __ocfs2_claim_clusters(struct ocfs2_super *osb,
+			   handle_t *handle,
+			   struct ocfs2_alloc_context *ac,
+			   u32 min_clusters,
+			   u32 max_clusters,
+			   u32 *cluster_start,
+			   u32 *num_clusters);
 
 int ocfs2_free_suballoc_bits(handle_t *handle,
 			     struct inode *alloc_inode,
diff --git a/fs/ocfs2/vote.c b/fs/ocfs2/vote.c
index 66a13ee..c053585 100644
--- a/fs/ocfs2/vote.c
+++ b/fs/ocfs2/vote.c
@@ -66,7 +66,7 @@ struct ocfs2_vote_msg
 {
 	struct ocfs2_msg_hdr v_hdr;
 	__be32 v_reserved1;
-};
+} __attribute__ ((packed));
 
 /* Responses are given these values to maintain backwards
  * compatibility with older ocfs2 versions */
@@ -78,7 +78,7 @@ struct ocfs2_response_msg
 {
 	struct ocfs2_msg_hdr r_hdr;
 	__be32 r_response;
-};
+} __attribute__ ((packed));
 
 struct ocfs2_vote_work {
 	struct list_head   w_list;

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-09-11 22:32 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-09-11 22:32 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

This includes a small doc update, which I missed earlier. It doesn't change
any code. The other three patches are real fixes.
	--Mark

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 Documentation/filesystems/ocfs2.txt |   13 ++++--
 fs/Kconfig                          |    3 -
 fs/ocfs2/alloc.c                    |    1 
 fs/ocfs2/aops.c                     |    4 +-
 fs/ocfs2/file.c                     |    1 
 fs/ocfs2/super.c                    |   69 +++++++++++++++++++-----------------
 6 files changed, 50 insertions(+), 41 deletions(-)

Mark Fasheh (2):
      ocfs2: update docs for new features
      ocfs2: Fix calculation of i_blocks during truncate

Tiger Yang (1):
      ocfs2: fix mount option parsing

tao.ma@oracle.com (1):
      ocfs2: Fix a wrong cluster calculation.

diff --git a/Documentation/filesystems/ocfs2.txt b/Documentation/filesystems/ocfs2.txt
index 8ccf0c1..ed55238 100644
--- a/Documentation/filesystems/ocfs2.txt
+++ b/Documentation/filesystems/ocfs2.txt
@@ -28,11 +28,7 @@ Manish Singh  <manish.singh@oracle.com>
 Caveats
 =======
 Features which OCFS2 does not support yet:
-	- sparse files
 	- extended attributes
-	- shared writable mmap
-	- loopback is supported, but data written will not
-	  be cluster coherent.
 	- quotas
 	- cluster aware flock
 	- cluster aware lockf
@@ -57,3 +53,12 @@ nointr			Do not allow signals to interrupt cluster
 atime_quantum=60(*)	OCFS2 will not update atime unless this number
 			of seconds has passed since the last update.
 			Set to zero to always update atime.
+data=ordered	(*)	All data are forced directly out to the main file
+			system prior to its metadata being committed to the
+			journal.
+data=writeback		Data ordering is not preserved, data may be written
+			into the main file system after its metadata has been
+			committed to the journal.
+preferred_slot=0(*)	During mount, try to use this filesystem slot first. If
+			it is in use by another node, the first empty one found
+			will be chosen. Invalid values will be ignored.
diff --git a/fs/Kconfig b/fs/Kconfig
index 58a0650..f9eed6d 100644
--- a/fs/Kconfig
+++ b/fs/Kconfig
@@ -441,9 +441,6 @@ config OCFS2_FS
 
 	  Note: Features which OCFS2 does not support yet:
 	          - extended attributes
-		  - shared writeable mmap
-	          - loopback is supported, but data written will not
-	            be cluster coherent.
 	          - quotas
 	          - cluster aware flock
 	          - Directory change notification (F_NOTIFY)
diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
index 4f51766..778a850 100644
--- a/fs/ocfs2/alloc.c
+++ b/fs/ocfs2/alloc.c
@@ -5602,6 +5602,7 @@ static int ocfs2_do_truncate(struct ocfs2_super *osb,
 				      clusters_to_del;
 	spin_unlock(&OCFS2_I(inode)->ip_lock);
 	le32_add_cpu(&fe->i_clusters, -clusters_to_del);
+	inode->i_blocks = ocfs2_inode_sector_count(inode);
 
 	status = ocfs2_trim_tree(inode, path, handle, tc,
 				 clusters_to_del, &delete_blk);
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 460d440..50cd8a2 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -855,6 +855,7 @@ static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
 				  struct ocfs2_super *osb, loff_t pos,
 				  unsigned len, struct buffer_head *di_bh)
 {
+	u32 cend;
 	struct ocfs2_write_ctxt *wc;
 
 	wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
@@ -862,7 +863,8 @@ static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
 		return -ENOMEM;
 
 	wc->w_cpos = pos >> osb->s_clustersize_bits;
-	wc->w_clen = ocfs2_clusters_for_bytes(osb->sb, len);
+	cend = (pos + len - 1) >> osb->s_clustersize_bits;
+	wc->w_clen = cend - wc->w_cpos + 1;
 	get_bh(di_bh);
 	wc->w_di_bh = di_bh;
 
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 4ffa715..7e34e66 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -314,7 +314,6 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 	}
 
 	i_size_write(inode, new_i_size);
-	inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
 	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 
 	di = (struct ocfs2_dinode *) fe_bh->b_data;
diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
index f2fc9a7..c034b51 100644
--- a/fs/ocfs2/super.c
+++ b/fs/ocfs2/super.c
@@ -81,8 +81,15 @@ static struct dentry *ocfs2_debugfs_root = NULL;
 MODULE_AUTHOR("Oracle");
 MODULE_LICENSE("GPL");
 
+struct mount_options
+{
+	unsigned long	mount_opt;
+	unsigned int	atime_quantum;
+	signed short	slot;
+};
+
 static int ocfs2_parse_options(struct super_block *sb, char *options,
-			       unsigned long *mount_opt, s16 *slot,
+			       struct mount_options *mopt,
 			       int is_remount);
 static void ocfs2_put_super(struct super_block *sb);
 static int ocfs2_mount_volume(struct super_block *sb);
@@ -367,24 +374,23 @@ static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 {
 	int incompat_features;
 	int ret = 0;
-	unsigned long parsed_options;
-	s16 slot;
+	struct mount_options parsed_options;
 	struct ocfs2_super *osb = OCFS2_SB(sb);
 
-	if (!ocfs2_parse_options(sb, data, &parsed_options, &slot, 1)) {
+	if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
 		ret = -EINVAL;
 		goto out;
 	}
 
 	if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
-	    (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
+	    (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
 		ret = -EINVAL;
 		mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 		goto out;
 	}
 
 	if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
-	    (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
+	    (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 		ret = -EINVAL;
 		mlog(ML_ERROR, "Cannot change data mode on remount\n");
 		goto out;
@@ -435,7 +441,9 @@ unlock_osb:
 
 		/* Only save off the new mount options in case of a successful
 		 * remount. */
-		osb->s_mount_opt = parsed_options;
+		osb->s_mount_opt = parsed_options.mount_opt;
+		osb->s_atime_quantum = parsed_options.atime_quantum;
+		osb->preferred_slot = parsed_options.slot;
 	}
 out:
 	return ret;
@@ -547,8 +555,7 @@ static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 {
 	struct dentry *root;
 	int status, sector_size;
-	unsigned long parsed_opt;
-	s16 slot;
+	struct mount_options parsed_options;
 	struct inode *inode = NULL;
 	struct ocfs2_super *osb = NULL;
 	struct buffer_head *bh = NULL;
@@ -556,14 +563,14 @@ static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 
 	mlog_entry("%p, %p, %i", sb, data, silent);
 
-	if (!ocfs2_parse_options(sb, data, &parsed_opt, &slot, 0)) {
+	if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
 		status = -EINVAL;
 		goto read_super_error;
 	}
 
 	/* for now we only have one cluster/node, make sure we see it
 	 * in the heartbeat universe */
-	if (parsed_opt & OCFS2_MOUNT_HB_LOCAL) {
+	if (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 		if (!o2hb_check_local_node_heartbeating()) {
 			status = -EINVAL;
 			goto read_super_error;
@@ -585,8 +592,9 @@ static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 	}
 	brelse(bh);
 	bh = NULL;
-	osb->s_mount_opt = parsed_opt;
-	osb->preferred_slot = slot;
+	osb->s_mount_opt = parsed_options.mount_opt;
+	osb->s_atime_quantum = parsed_options.atime_quantum;
+	osb->preferred_slot = parsed_options.slot;
 
 	sb->s_magic = OCFS2_SUPER_MAGIC;
 
@@ -728,8 +736,7 @@ static struct file_system_type ocfs2_fs_type = {
 
 static int ocfs2_parse_options(struct super_block *sb,
 			       char *options,
-			       unsigned long *mount_opt,
-			       s16 *slot,
+			       struct mount_options *mopt,
 			       int is_remount)
 {
 	int status;
@@ -738,8 +745,9 @@ static int ocfs2_parse_options(struct super_block *sb,
 	mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
 		   options ? options : "(none)");
 
-	*mount_opt = 0;
-	*slot = OCFS2_INVALID_SLOT;
+	mopt->mount_opt = 0;
+	mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
+	mopt->slot = OCFS2_INVALID_SLOT;
 
 	if (!options) {
 		status = 1;
@@ -749,7 +757,6 @@ static int ocfs2_parse_options(struct super_block *sb,
 	while ((p = strsep(&options, ",")) != NULL) {
 		int token, option;
 		substring_t args[MAX_OPT_ARGS];
-		struct ocfs2_super * osb = OCFS2_SB(sb);
 
 		if (!*p)
 			continue;
@@ -757,10 +764,10 @@ static int ocfs2_parse_options(struct super_block *sb,
 		token = match_token(p, tokens, args);
 		switch (token) {
 		case Opt_hb_local:
-			*mount_opt |= OCFS2_MOUNT_HB_LOCAL;
+			mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
 			break;
 		case Opt_hb_none:
-			*mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
+			mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
 			break;
 		case Opt_barrier:
 			if (match_int(&args[0], &option)) {
@@ -768,27 +775,27 @@ static int ocfs2_parse_options(struct super_block *sb,
 				goto bail;
 			}
 			if (option)
-				*mount_opt |= OCFS2_MOUNT_BARRIER;
+				mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
 			else
-				*mount_opt &= ~OCFS2_MOUNT_BARRIER;
+				mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
 			break;
 		case Opt_intr:
-			*mount_opt &= ~OCFS2_MOUNT_NOINTR;
+			mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
 			break;
 		case Opt_nointr:
-			*mount_opt |= OCFS2_MOUNT_NOINTR;
+			mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
 			break;
 		case Opt_err_panic:
-			*mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
+			mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
 			break;
 		case Opt_err_ro:
-			*mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
+			mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
 			break;
 		case Opt_data_ordered:
-			*mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
+			mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
 			break;
 		case Opt_data_writeback:
-			*mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
+			mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
 			break;
 		case Opt_atime_quantum:
 			if (match_int(&args[0], &option)) {
@@ -796,9 +803,7 @@ static int ocfs2_parse_options(struct super_block *sb,
 				goto bail;
 			}
 			if (option >= 0)
-				osb->s_atime_quantum = option;
-			else
-				osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
+				mopt->atime_quantum = option;
 			break;
 		case Opt_slot:
 			option = 0;
@@ -807,7 +812,7 @@ static int ocfs2_parse_options(struct super_block *sb,
 				goto bail;
 			}
 			if (option)
-				*slot = (s16)option;
+				mopt->slot = (s16)option;
 			break;
 		default:
 			mlog(ML_ERROR,

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-08-10 18:43 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-08-10 18:43 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

Mostly fixes and two very small/obvious cleanups.
	--Mark

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/alloc.c       |    4 --
 fs/ocfs2/cluster/tcp.c |   24 +++++++++++------
 fs/ocfs2/file.c        |   28 +++++++++++++++++--
 fs/ocfs2/namei.c       |   16 ++++++++++-
 fs/ocfs2/ocfs2.h       |    8 ++---
 fs/ocfs2/super.c       |   69 +++++++++++++++++++++++++++++--------------------
 fs/ocfs2/super.h       |    2 -
 7 files changed, 101 insertions(+), 50 deletions(-)

Adrian Bunk (1):
      [2.6 patch] ocfs2_insert_extent(): remove dead code

Mark Fasheh (6):
      ocfs2: Restrict inode changes in ocfs2_update_inode_atime()
      ocfs2: use s_maxbytes directly in ocfs2_change_file_space()
      ocfs2: Fix some casting errors related to file writes
      ocfs2: check ia_size limits in setattr
      ocfs2: Fix max offset calculations
      ocfs2: set non-default s_time_gran during mount

Sunil Mushran (2):
      ocfs2: Fix rename/extend race
      ocfs2: Retry sendpage() if it returns EAGAIN

diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
index f5e11f4..4f51766 100644
--- a/fs/ocfs2/alloc.c
+++ b/fs/ocfs2/alloc.c
@@ -3731,7 +3731,6 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 {
 	int status;
 	struct buffer_head *last_eb_bh = NULL;
-	struct buffer_head *bh = NULL;
 	struct ocfs2_insert_type insert = {0, };
 	struct ocfs2_extent_rec rec;
 
@@ -3783,9 +3782,6 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 		ocfs2_extent_map_insert_rec(inode, &rec);
 
 bail:
-	if (bh)
-		brelse(bh);
-
 	if (last_eb_bh)
 		brelse(last_eb_bh);
 
diff --git a/fs/ocfs2/cluster/tcp.c b/fs/ocfs2/cluster/tcp.c
index f0bdfd9..685c180 100644
--- a/fs/ocfs2/cluster/tcp.c
+++ b/fs/ocfs2/cluster/tcp.c
@@ -854,17 +854,25 @@ static void o2net_sendpage(struct o2net_sock_container *sc,
 	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
 	ssize_t ret;
 
-
-	mutex_lock(&sc->sc_send_lock);
-	ret = sc->sc_sock->ops->sendpage(sc->sc_sock,
-					 virt_to_page(kmalloced_virt),
-					 (long)kmalloced_virt & ~PAGE_MASK,
-					 size, MSG_DONTWAIT);
-	mutex_unlock(&sc->sc_send_lock);
-	if (ret != size) {
+	while (1) {
+		mutex_lock(&sc->sc_send_lock);
+		ret = sc->sc_sock->ops->sendpage(sc->sc_sock,
+						 virt_to_page(kmalloced_virt),
+						 (long)kmalloced_virt & ~PAGE_MASK,
+						 size, MSG_DONTWAIT);
+		mutex_unlock(&sc->sc_send_lock);
+		if (ret == size)
+			break;
+		if (ret == (ssize_t)-EAGAIN) {
+			mlog(0, "sendpage of size %zu to " SC_NODEF_FMT
+			     " returned EAGAIN\n", size, SC_NODEF_ARGS(sc));
+			cond_resched();
+			continue;
+		}
 		mlog(ML_ERROR, "sendpage of size %zu to " SC_NODEF_FMT 
 		     " failed with %zd\n", size, SC_NODEF_ARGS(sc), ret);
 		o2net_ensure_shutdown(nn, sc, 0);
+		break;
 	}
 }
 
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index c4034f6..4ffa715 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -187,6 +187,7 @@ int ocfs2_update_inode_atime(struct inode *inode,
 	int ret;
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 	handle_t *handle;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 
 	mlog_entry_void();
 
@@ -197,11 +198,27 @@ int ocfs2_update_inode_atime(struct inode *inode,
 		goto out;
 	}
 
+	ret = ocfs2_journal_access(handle, inode, bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	/*
+	 * Don't use ocfs2_mark_inode_dirty() here as we don't always
+	 * have i_mutex to guard against concurrent changes to other
+	 * inode fields.
+	 */
 	inode->i_atime = CURRENT_TIME;
-	ret = ocfs2_mark_inode_dirty(handle, inode, bh);
+	di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
+	di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
+
+	ret = ocfs2_journal_dirty(handle, bh);
 	if (ret < 0)
 		mlog_errno(ret);
 
+out_commit:
 	ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 out:
 	mlog_exit(ret);
@@ -1011,6 +1028,11 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
 	}
 
 	if (size_change && attr->ia_size != i_size_read(inode)) {
+		if (attr->ia_size > sb->s_maxbytes) {
+			status = -EFBIG;
+			goto bail_unlock;
+		}
+
 		if (i_size_read(inode) > attr->ia_size)
 			status = ocfs2_truncate_file(inode, bh, attr->ia_size);
 		else
@@ -1516,7 +1538,7 @@ static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 	struct buffer_head *di_bh = NULL;
 	handle_t *handle;
-	unsigned long long max_off = ocfs2_max_file_offset(inode->i_sb->s_blocksize_bits);
+	unsigned long long max_off = inode->i_sb->s_maxbytes;
 
 	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 		return -EROFS;
@@ -1942,7 +1964,7 @@ static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
 		}
 
 		dst = kmap_atomic(page, KM_USER0);
-		memcpy(dst + (pos & (PAGE_CACHE_SIZE - 1)), buf, bytes);
+		memcpy(dst + (pos & (loff_t)(PAGE_CACHE_SIZE - 1)), buf, bytes);
 		kunmap_atomic(dst, KM_USER0);
 		flush_dcache_page(page);
 		ocfs2_put_write_source(user_page);
diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
index d430fda..701e6d0 100644
--- a/fs/ocfs2/namei.c
+++ b/fs/ocfs2/namei.c
@@ -1080,6 +1080,7 @@ static int ocfs2_rename(struct inode *old_dir,
 	struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
 						    // this is the 1st dirent bh
 	nlink_t old_dir_nlink = old_dir->i_nlink;
+	struct ocfs2_dinode *old_di;
 
 	/* At some point it might be nice to break this function up a
 	 * bit. */
@@ -1354,7 +1355,20 @@ static int ocfs2_rename(struct inode *old_dir,
 
 	old_inode->i_ctime = CURRENT_TIME;
 	mark_inode_dirty(old_inode);
-	ocfs2_mark_inode_dirty(handle, old_inode, old_inode_bh);
+
+	status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status >= 0) {
+		old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
+
+		old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
+		old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
+
+		status = ocfs2_journal_dirty(handle, old_inode_bh);
+		if (status < 0)
+			mlog_errno(status);
+	} else
+		mlog_errno(status);
 
 	/* now that the name has been added to new_dir, remove the old name */
 	status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
index 5cc90a4..5830785 100644
--- a/fs/ocfs2/ocfs2.h
+++ b/fs/ocfs2/ocfs2.h
@@ -494,16 +494,16 @@ static inline unsigned int ocfs2_page_index_to_clusters(struct super_block *sb,
 /*
  * Find the 1st page index which covers the given clusters.
  */
-static inline unsigned long ocfs2_align_clusters_to_page_index(struct super_block *sb,
+static inline pgoff_t ocfs2_align_clusters_to_page_index(struct super_block *sb,
 							u32 clusters)
 {
 	unsigned int cbits = OCFS2_SB(sb)->s_clustersize_bits;
-	unsigned long index = clusters;
+        pgoff_t index = clusters;
 
 	if (PAGE_CACHE_SHIFT > cbits) {
-		index = clusters >> (PAGE_CACHE_SHIFT - cbits);
+		index = (pgoff_t)clusters >> (PAGE_CACHE_SHIFT - cbits);
 	} else if (PAGE_CACHE_SHIFT < cbits) {
-		index = clusters << (cbits - PAGE_CACHE_SHIFT);
+		index = (pgoff_t)clusters << (cbits - PAGE_CACHE_SHIFT);
 	}
 
 	return index;
diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
index 200c7d4..f2fc9a7 100644
--- a/fs/ocfs2/super.c
+++ b/fs/ocfs2/super.c
@@ -316,39 +316,51 @@ static void ocfs2_destroy_inode(struct inode *inode)
 	kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 }
 
-/* From xfs_super.c:xfs_max_file_offset
- * Copyright (c) 2000-2004 Silicon Graphics, Inc.
- */
-unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
+static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
+						unsigned int cbits)
 {
-	unsigned int pagefactor = 1;
-	unsigned int bitshift = BITS_PER_LONG - 1;
-
-	/* Figure out maximum filesize, on Linux this can depend on
-	 * the filesystem blocksize (on 32 bit platforms).
-	 * __block_prepare_write does this in an [unsigned] long...
-	 *      page->index << (PAGE_CACHE_SHIFT - bbits)
-	 * So, for page sized blocks (4K on 32 bit platforms),
-	 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
-	 *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
-	 * but for smaller blocksizes it is less (bbits = log2 bsize).
-	 * Note1: get_block_t takes a long (implicit cast from above)
-	 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
-	 * can optionally convert the [unsigned] long from above into
-	 * an [unsigned] long long.
+	unsigned int bytes = 1 << cbits;
+	unsigned int trim = bytes;
+	unsigned int bitshift = 32;
+
+	/*
+	 * i_size and all block offsets in ocfs2 are always 64 bits
+	 * wide. i_clusters is 32 bits, in cluster-sized units. So on
+	 * 64 bit platforms, cluster size will be the limiting factor.
 	 */
 
 #if BITS_PER_LONG == 32
 # if defined(CONFIG_LBD)
 	BUILD_BUG_ON(sizeof(sector_t) != 8);
-	pagefactor = PAGE_CACHE_SIZE;
-	bitshift = BITS_PER_LONG;
+	/*
+	 * We might be limited by page cache size.
+	 */
+	if (bytes > PAGE_CACHE_SIZE) {
+		bytes = PAGE_CACHE_SIZE;
+		trim = 1;
+		/*
+		 * Shift by 31 here so that we don't get larger than
+		 * MAX_LFS_FILESIZE
+		 */
+		bitshift = 31;
+	}
 # else
-	pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
+	/*
+	 * We are limited by the size of sector_t. Use block size, as
+	 * that's what we expose to the VFS.
+	 */
+	bytes = 1 << bbits;
+	trim = 1;
+	bitshift = 31;
 # endif
 #endif
 
-	return (((unsigned long long)pagefactor) << bitshift) - 1;
+	/*
+	 * Trim by a whole cluster when we can actually approach the
+	 * on-disk limits. Otherwise we can overflow i_clusters when
+	 * an extent start is at the max offset.
+	 */
+	return (((unsigned long long)bytes) << bitshift) - trim;
 }
 
 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
@@ -1259,8 +1271,8 @@ static int ocfs2_initialize_super(struct super_block *sb,
 				  int sector_size)
 {
 	int status = 0;
-	int i;
-	struct ocfs2_dinode *di = NULL;
+	int i, cbits, bbits;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
 	struct inode *inode = NULL;
 	struct buffer_head *bitmap_bh = NULL;
 	struct ocfs2_journal *journal;
@@ -1279,9 +1291,12 @@ static int ocfs2_initialize_super(struct super_block *sb,
 	sb->s_fs_info = osb;
 	sb->s_op = &ocfs2_sops;
 	sb->s_export_op = &ocfs2_export_ops;
+	sb->s_time_gran = 1;
 	sb->s_flags |= MS_NOATIME;
 	/* this is needed to support O_LARGEFILE */
-	sb->s_maxbytes = ocfs2_max_file_offset(sb->s_blocksize_bits);
+	cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
+	bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
+	sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
 
 	osb->sb = sb;
 	/* Save off for ocfs2_rw_direct */
@@ -1341,8 +1356,6 @@ static int ocfs2_initialize_super(struct super_block *sb,
 		goto bail;
 	}
 
-	di = (struct ocfs2_dinode *)bh->b_data;
-
 	osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
 	if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
 		mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
diff --git a/fs/ocfs2/super.h b/fs/ocfs2/super.h
index 3b9cb3d..783f527 100644
--- a/fs/ocfs2/super.h
+++ b/fs/ocfs2/super.h
@@ -45,6 +45,4 @@ void __ocfs2_abort(struct super_block *sb,
 
 #define ocfs2_abort(sb, fmt, args...) __ocfs2_abort(sb, __PRETTY_FUNCTION__, fmt, ##args)
 
-unsigned long long ocfs2_max_file_offset(unsigned int blockshift);
-
 #endif /* OCFS2_SUPER_H */

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-06-08 17:48 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-06-08 17:48 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/aops.c            |   20 ++++++++++----------
 fs/ocfs2/cluster/masklog.c |    3 +--
 2 files changed, 11 insertions(+), 12 deletions(-)

Mark Fasheh:
      ocfs2: Fix invalid assertion during write on 64k pages

Tiger Yang:
      ocfs2: Fix masklog breakage

diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 0023b31..a480b09 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -798,6 +798,11 @@ int ocfs2_map_and_write_splice_data(stru
 	}
 	to = from + bytes;
 
+	BUG_ON(from > PAGE_CACHE_SIZE);
+	BUG_ON(to > PAGE_CACHE_SIZE);
+	BUG_ON(from < cluster_start);
+	BUG_ON(to > cluster_end);
+
 	if (wc->w_this_page_new)
 		ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
 					    cluster_start, cluster_end, 1);
@@ -809,11 +814,6 @@ int ocfs2_map_and_write_splice_data(stru
 		goto out;
 	}
 
-	BUG_ON(from > PAGE_CACHE_SIZE);
-	BUG_ON(to > PAGE_CACHE_SIZE);
-	BUG_ON(from > osb->s_clustersize);
-	BUG_ON(to > osb->s_clustersize);
-
 	src = buf->ops->map(sp->s_pipe, buf, 1);
 	dst = kmap_atomic(wc->w_this_page, KM_USER1);
 	memcpy(dst + from, src + src_from, bytes);
@@ -890,6 +890,11 @@ int ocfs2_map_and_write_user_data(struct
 
 	to = from + bytes;
 
+	BUG_ON(from > PAGE_CACHE_SIZE);
+	BUG_ON(to > PAGE_CACHE_SIZE);
+	BUG_ON(from < cluster_start);
+	BUG_ON(to > cluster_end);
+
 	if (wc->w_this_page_new)
 		ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
 					    cluster_start, cluster_end, 1);
@@ -901,11 +906,6 @@ int ocfs2_map_and_write_user_data(struct
 		goto out;
 	}
 
-	BUG_ON(from > PAGE_CACHE_SIZE);
-	BUG_ON(to > PAGE_CACHE_SIZE);
-	BUG_ON(from > osb->s_clustersize);
-	BUG_ON(to > osb->s_clustersize);
-
 	dst = kmap(wc->w_this_page);
 	memcpy(dst + from, bp->b_src_buf + src_from, bytes);
 	kunmap(wc->w_this_page);
diff --git a/fs/ocfs2/cluster/masklog.c b/fs/ocfs2/cluster/masklog.c
index a93620c..2b205f5 100644
--- a/fs/ocfs2/cluster/masklog.c
+++ b/fs/ocfs2/cluster/masklog.c
@@ -144,8 +144,7 @@ static struct kobj_type mlog_ktype = {
 };
 
 static struct kset mlog_kset = {
-	.kobj  = {.name = "logmask"},
-	.ktype = &mlog_ktype
+	.kobj   = {.name = "logmask", .ktype = &mlog_ktype},
 };
 
 int mlog_sys_init(struct kset *o2cb_subsys)

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-05-03 18:57 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-05-03 18:57 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

Here's some various fixes and cleanups I had slated for a later pull, but
our Ocfs2 testing is ahead of schedule.

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/alloc.c             |    6 ++--
 fs/ocfs2/aops.c              |   11 ++++----
 fs/ocfs2/cluster/heartbeat.c |    2 -
 fs/ocfs2/cluster/tcp.c       |   10 +++----
 fs/ocfs2/dir.c               |    7 ++++-
 fs/ocfs2/dlm/dlmast.c        |   12 ++++-----
 fs/ocfs2/dlm/dlmrecovery.c   |    4 +--
 fs/ocfs2/dlm/dlmthread.c     |    2 -
 fs/ocfs2/dlmglue.c           |   54 +++++++++++++++++++++++--------------------
 fs/ocfs2/dlmglue.h           |    7 -----
 fs/ocfs2/export.c            |    6 +++-
 fs/ocfs2/file.c              |   17 +++++++++----
 fs/ocfs2/file.h              |    5 ---
 fs/ocfs2/inode.c             |   35 +++++++++++++++++++++++----
 fs/ocfs2/inode.h             |    1 
 fs/ocfs2/ioctl.c             |   24 +++++++++++++++++++
 fs/ocfs2/ioctl.h             |    1 
 fs/ocfs2/journal.c           |    7 +++--
 fs/ocfs2/namei.c             |    5 ++-
 fs/ocfs2/ocfs2.h             |   12 ++++-----
 fs/ocfs2/ocfs2_fs.h          |    2 +
 fs/ocfs2/suballoc.c          |   10 +++----
 fs/ocfs2/super.c             |    2 -
 23 files changed, 151 insertions(+), 91 deletions(-)

Adrian Bunk:
      fs/ocfs2/: make 3 functions static

Jan Kara:
      Copy i_flags to ocfs2 inode flags on write

Joel Becker:
      ocfs2: Wrap access of directory allocations with ip_alloc_sem.

Mark Fasheh:
      ocfs2: Implement compat_ioctl()
      ocfs2: fix sparse warnings in fs/ocfs2
      ocfs2: fix sparse warnings in fs/ocfs2/dlm
      ocfs2: fix sparse warnings in fs/ocfs2/cluster
      ocfs2: Force use of GFP_NOFS in ocfs2_write()

Milind Arun Choudhary:
      ocfs2: use __set_current_state()

diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
index a0c8667..19712a7 100644
--- a/fs/ocfs2/alloc.c
+++ b/fs/ocfs2/alloc.c
@@ -2869,7 +2869,7 @@ int ocfs2_complete_truncate_log_recovery
 	tl = &tl_copy->id2.i_dealloc;
 	num_recs = le16_to_cpu(tl->tl_used);
 	mlog(0, "cleanup %u records from %llu\n", num_recs,
-	     (unsigned long long)tl_copy->i_blkno);
+	     (unsigned long long)le64_to_cpu(tl_copy->i_blkno));
 
 	mutex_lock(&tl_inode->i_mutex);
 	for(i = 0; i < num_recs; i++) {
@@ -3801,8 +3801,8 @@ int ocfs2_prepare_truncate(struct ocfs2_
 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
 
 	mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size ="
-	     "%llu\n", fe->i_clusters, new_i_clusters,
-	     (unsigned long long)fe->i_size);
+	     "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters,
+	     (unsigned long long)le64_to_cpu(fe->i_size));
 
 	*tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL);
 	if (!(*tc)) {
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 56963e6..8e7cafb 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -78,7 +78,8 @@ static int ocfs2_symlink_get_block(struc
 
 	if (!OCFS2_IS_VALID_DINODE(fe)) {
 		mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
-		     (unsigned long long)fe->i_blkno, 7, fe->i_signature);
+		     (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
+		     fe->i_signature);
 		goto bail;
 	}
 
@@ -939,9 +940,9 @@ out:
  * Returns a negative error code or the number of bytes copied into
  * the page.
  */
-int ocfs2_write_data_page(struct inode *inode, handle_t *handle,
-			  u64 *p_blkno, struct page *page,
-			  struct ocfs2_write_ctxt *wc, int new)
+static int ocfs2_write_data_page(struct inode *inode, handle_t *handle,
+				 u64 *p_blkno, struct page *page,
+				 struct ocfs2_write_ctxt *wc, int new)
 {
 	int ret, copied = 0;
 	unsigned int from = 0, to = 0;
@@ -1086,7 +1087,7 @@ static ssize_t ocfs2_write(struct file *
 	for(i = 0; i < numpages; i++) {
 		index = start + i;
 
-		cpages[i] = grab_cache_page(mapping, index);
+		cpages[i] = find_or_create_page(mapping, index, GFP_NOFS);
 		if (!cpages[i]) {
 			ret = -ENOMEM;
 			mlog_errno(ret);
diff --git a/fs/ocfs2/cluster/heartbeat.c b/fs/ocfs2/cluster/heartbeat.c
index eba282d..9791134 100644
--- a/fs/ocfs2/cluster/heartbeat.c
+++ b/fs/ocfs2/cluster/heartbeat.c
@@ -438,7 +438,7 @@ static inline void o2hb_prepare_block(st
 								   hb_block));
 
 	mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
-	     (long long)cpu_to_le64(generation),
+	     (long long)generation,
 	     le32_to_cpu(hb_block->hb_cksum));
 }
 
diff --git a/fs/ocfs2/cluster/tcp.c b/fs/ocfs2/cluster/tcp.c
index 69caf3e..0b229a9 100644
--- a/fs/ocfs2/cluster/tcp.c
+++ b/fs/ocfs2/cluster/tcp.c
@@ -1496,7 +1496,7 @@ static void o2net_start_connect(struct w
 	sock->sk->sk_allocation = GFP_ATOMIC;
 
 	myaddr.sin_family = AF_INET;
-	myaddr.sin_addr.s_addr = (__force u32)mynode->nd_ipv4_address;
+	myaddr.sin_addr.s_addr = mynode->nd_ipv4_address;
 	myaddr.sin_port = (__force u16)htons(0); /* any port */
 
 	ret = sock->ops->bind(sock, (struct sockaddr *)&myaddr,
@@ -1521,8 +1521,8 @@ static void o2net_start_connect(struct w
 	spin_unlock(&nn->nn_lock);
 
 	remoteaddr.sin_family = AF_INET;
-	remoteaddr.sin_addr.s_addr = (__force u32)node->nd_ipv4_address;
-	remoteaddr.sin_port = (__force u16)node->nd_ipv4_port;
+	remoteaddr.sin_addr.s_addr = node->nd_ipv4_address;
+	remoteaddr.sin_port = node->nd_ipv4_port;
 
 	ret = sc->sc_sock->ops->connect(sc->sc_sock,
 					(struct sockaddr *)&remoteaddr,
@@ -1810,8 +1810,8 @@ static int o2net_open_listening_sock(__b
 	int ret;
 	struct sockaddr_in sin = {
 		.sin_family = PF_INET,
-		.sin_addr = { .s_addr = (__force u32)addr },
-		.sin_port = (__force u16)port,
+		.sin_addr = { .s_addr = addr },
+		.sin_port = port,
 	};
 
 	ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
index 67e6866..c441ef1 100644
--- a/fs/ocfs2/dir.c
+++ b/fs/ocfs2/dir.c
@@ -403,7 +403,7 @@ static int ocfs2_extend_dir(struct ocfs2
 			    struct buffer_head **new_de_bh)
 {
 	int status = 0;
-	int credits, num_free_extents;
+	int credits, num_free_extents, drop_alloc_sem = 0;
 	loff_t dir_i_size;
 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 	struct ocfs2_alloc_context *data_ac = NULL;
@@ -452,6 +452,9 @@ static int ocfs2_extend_dir(struct ocfs2
 		credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
 	}
 
+	down_write(&OCFS2_I(dir)->ip_alloc_sem);
+	drop_alloc_sem = 1;
+
 	handle = ocfs2_start_trans(osb, credits);
 	if (IS_ERR(handle)) {
 		status = PTR_ERR(handle);
@@ -497,6 +500,8 @@ static int ocfs2_extend_dir(struct ocfs2
 	*new_de_bh = new_bh;
 	get_bh(*new_de_bh);
 bail:
+	if (drop_alloc_sem)
+		up_write(&OCFS2_I(dir)->ip_alloc_sem);
 	if (handle)
 		ocfs2_commit_trans(osb, handle);
 
diff --git a/fs/ocfs2/dlm/dlmast.c b/fs/ocfs2/dlm/dlmast.c
index 241cad3..2fd8bde 100644
--- a/fs/ocfs2/dlm/dlmast.c
+++ b/fs/ocfs2/dlm/dlmast.c
@@ -312,8 +312,8 @@ int dlm_proxy_ast_handler(struct o2net_m
 	    past->type != DLM_BAST) {
 		mlog(ML_ERROR, "Unknown ast type! %d, cookie=%u:%llu"
 		     "name=%.*s\n", past->type, 
-		     dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
-		     dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
+		     dlm_get_lock_cookie_node(cookie),
+		     dlm_get_lock_cookie_seq(cookie),
 		     locklen, name);
 		ret = DLM_IVLOCKID;
 		goto leave;
@@ -324,8 +324,8 @@ int dlm_proxy_ast_handler(struct o2net_m
 		mlog(0, "got %sast for unknown lockres! "
 		     "cookie=%u:%llu, name=%.*s, namelen=%u\n",
 		     past->type == DLM_AST ? "" : "b",
-		     dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
-		     dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
+		     dlm_get_lock_cookie_node(cookie),
+		     dlm_get_lock_cookie_seq(cookie),
 		     locklen, name, locklen);
 		ret = DLM_IVLOCKID;
 		goto leave;
@@ -370,8 +370,8 @@ int dlm_proxy_ast_handler(struct o2net_m
 
 	mlog(0, "got %sast for unknown lock!  cookie=%u:%llu, "
 	     "name=%.*s, namelen=%u\n", past->type == DLM_AST ? "" : "b", 
-	     dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
-	     dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
+	     dlm_get_lock_cookie_node(cookie),
+	     dlm_get_lock_cookie_seq(cookie),
 	     locklen, name, locklen);
 
 	ret = DLM_NORMAL;
diff --git a/fs/ocfs2/dlm/dlmrecovery.c b/fs/ocfs2/dlm/dlmrecovery.c
index c1807a4..671c4ed 100644
--- a/fs/ocfs2/dlm/dlmrecovery.c
+++ b/fs/ocfs2/dlm/dlmrecovery.c
@@ -1769,7 +1769,7 @@ static int dlm_process_recovery_data(str
 			/* lock is always created locally first, and
 			 * destroyed locally last.  it must be on the list */
 			if (!lock) {
-				u64 c = ml->cookie;
+				__be64 c = ml->cookie;
 				mlog(ML_ERROR, "could not find local lock "
 					       "with cookie %u:%llu!\n",
 				     dlm_get_lock_cookie_node(be64_to_cpu(c)),
@@ -1878,7 +1878,7 @@ skip_lvb:
 		spin_lock(&res->spinlock);
 		list_for_each_entry(lock, queue, list) {
 			if (lock->ml.cookie == ml->cookie) {
-				u64 c = lock->ml.cookie;
+				__be64 c = lock->ml.cookie;
 				mlog(ML_ERROR, "%s:%.*s: %u:%llu: lock already "
 				     "exists on this lockres!\n", dlm->name,
 				     res->lockname.len, res->lockname.name,
diff --git a/fs/ocfs2/dlm/dlmthread.c b/fs/ocfs2/dlm/dlmthread.c
index 2b264c6..cebd089 100644
--- a/fs/ocfs2/dlm/dlmthread.c
+++ b/fs/ocfs2/dlm/dlmthread.c
@@ -76,7 +76,7 @@ repeat:
 		goto repeat;
 	}
 	remove_wait_queue(&res->wq, &wait);
-	current->state = TASK_RUNNING;
+	__set_current_state(TASK_RUNNING);
 }
 
 int __dlm_lockres_has_locks(struct dlm_lock_resource *res)
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
index 27e43b0..024777a 100644
--- a/fs/ocfs2/dlmglue.c
+++ b/fs/ocfs2/dlmglue.c
@@ -104,6 +104,35 @@ static int ocfs2_dentry_convert_worker(s
 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
 				     struct ocfs2_lock_res *lockres);
 
+
+#define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
+
+/* This aids in debugging situations where a bad LVB might be involved. */
+static void ocfs2_dump_meta_lvb_info(u64 level,
+				     const char *function,
+				     unsigned int line,
+				     struct ocfs2_lock_res *lockres)
+{
+	struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
+
+	mlog(level, "LVB information for %s (called from %s:%u):\n",
+	     lockres->l_name, function, line);
+	mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
+	     lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
+	     be32_to_cpu(lvb->lvb_igeneration));
+	mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
+	     (unsigned long long)be64_to_cpu(lvb->lvb_isize),
+	     be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
+	     be16_to_cpu(lvb->lvb_imode));
+	mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
+	     "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
+	     (long long)be64_to_cpu(lvb->lvb_iatime_packed),
+	     (long long)be64_to_cpu(lvb->lvb_ictime_packed),
+	     (long long)be64_to_cpu(lvb->lvb_imtime_packed),
+	     be32_to_cpu(lvb->lvb_iattr));
+}
+
+
 /*
  * OCFS2 Lock Resource Operations
  *
@@ -3078,28 +3107,3 @@ static void ocfs2_schedule_blocked_lock(
 
 	mlog_exit_void();
 }
-
-/* This aids in debugging situations where a bad LVB might be involved. */
-void ocfs2_dump_meta_lvb_info(u64 level,
-			      const char *function,
-			      unsigned int line,
-			      struct ocfs2_lock_res *lockres)
-{
-	struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
-
-	mlog(level, "LVB information for %s (called from %s:%u):\n",
-	     lockres->l_name, function, line);
-	mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
-	     lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
-	     be32_to_cpu(lvb->lvb_igeneration));
-	mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
-	     (unsigned long long)be64_to_cpu(lvb->lvb_isize),
-	     be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
-	     be16_to_cpu(lvb->lvb_imode));
-	mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
-	     "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
-	     (long long)be64_to_cpu(lvb->lvb_iatime_packed),
-	     (long long)be64_to_cpu(lvb->lvb_ictime_packed),
-	     (long long)be64_to_cpu(lvb->lvb_imtime_packed),
-	     be32_to_cpu(lvb->lvb_iattr));
-}
diff --git a/fs/ocfs2/dlmglue.h b/fs/ocfs2/dlmglue.h
index 59cb566..492bad3 100644
--- a/fs/ocfs2/dlmglue.h
+++ b/fs/ocfs2/dlmglue.h
@@ -119,11 +119,4 @@ void ocfs2_process_blocked_lock(struct o
 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void);
 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug);
 
-/* aids in debugging and tracking lvbs */
-void ocfs2_dump_meta_lvb_info(u64 level,
-			      const char *function,
-			      unsigned int line,
-			      struct ocfs2_lock_res *lockres);
-#define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
-
 #endif	/* DLMGLUE_H */
diff --git a/fs/ocfs2/export.c b/fs/ocfs2/export.c
index 56e1fef..bc48177 100644
--- a/fs/ocfs2/export.c
+++ b/fs/ocfs2/export.c
@@ -140,7 +140,7 @@ bail:
 	return parent;
 }
 
-static int ocfs2_encode_fh(struct dentry *dentry, __be32 *fh, int *max_len,
+static int ocfs2_encode_fh(struct dentry *dentry, u32 *fh_in, int *max_len,
 			   int connectable)
 {
 	struct inode *inode = dentry->d_inode;
@@ -148,6 +148,7 @@ static int ocfs2_encode_fh(struct dentry
 	int type = 1;
 	u64 blkno;
 	u32 generation;
+	__le32 *fh = (__force __le32 *) fh_in;
 
 	mlog_entry("(0x%p, '%.*s', 0x%p, %d, %d)\n", dentry,
 		   dentry->d_name.len, dentry->d_name.name,
@@ -199,7 +200,7 @@ bail:
 	return type;
 }
 
-static struct dentry *ocfs2_decode_fh(struct super_block *sb, __be32 *fh,
+static struct dentry *ocfs2_decode_fh(struct super_block *sb, u32 *fh_in,
 				      int fh_len, int fileid_type,
 				      int (*acceptable)(void *context,
 						        struct dentry *de),
@@ -207,6 +208,7 @@ static struct dentry *ocfs2_decode_fh(st
 {
 	struct ocfs2_inode_handle handle, parent;
 	struct dentry *ret = NULL;
+	__le32 *fh = (__force __le32 *) fh_in;
 
 	mlog_entry("(0x%p, 0x%p, %d, %d, 0x%p, 0x%p)\n",
 		   sb, fh, fh_len, fileid_type, acceptable, context);
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 520a2a6..9395b4f 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -207,10 +207,10 @@ out:
 	return ret;
 }
 
-int ocfs2_set_inode_size(handle_t *handle,
-			 struct inode *inode,
-			 struct buffer_head *fe_bh,
-			 u64 new_i_size)
+static int ocfs2_set_inode_size(handle_t *handle,
+				struct inode *inode,
+				struct buffer_head *fe_bh,
+				u64 new_i_size)
 {
 	int status;
 
@@ -713,7 +713,8 @@ restarted_transaction:
 	}
 
 	mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
-	     fe->i_clusters, (unsigned long long)fe->i_size);
+	     le32_to_cpu(fe->i_clusters),
+	     (unsigned long long)le64_to_cpu(fe->i_size));
 	mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
 	     OCFS2_I(inode)->ip_clusters, i_size_read(inode));
 
@@ -1853,6 +1854,9 @@ const struct file_operations ocfs2_fops 
 	.aio_read	= ocfs2_file_aio_read,
 	.aio_write	= ocfs2_file_aio_write,
 	.ioctl		= ocfs2_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl   = ocfs2_compat_ioctl,
+#endif
 	.splice_read	= ocfs2_file_splice_read,
 	.splice_write	= ocfs2_file_splice_write,
 };
@@ -1862,4 +1866,7 @@ const struct file_operations ocfs2_dops 
 	.readdir	= ocfs2_readdir,
 	.fsync		= ocfs2_sync_file,
 	.ioctl		= ocfs2_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl   = ocfs2_compat_ioctl,
+#endif
 };
diff --git a/fs/ocfs2/file.h b/fs/ocfs2/file.h
index 2c4460f..a4dd1fa 100644
--- a/fs/ocfs2/file.h
+++ b/fs/ocfs2/file.h
@@ -56,11 +56,6 @@ int ocfs2_getattr(struct vfsmount *mnt, 
 int ocfs2_permission(struct inode *inode, int mask,
 		     struct nameidata *nd);
 
-int ocfs2_set_inode_size(handle_t *handle,
-			 struct inode *inode,
-			 struct buffer_head *fe_bh,
-			 u64 new_i_size);
-
 int ocfs2_should_update_atime(struct inode *inode,
 			      struct vfsmount *vfsmnt);
 int ocfs2_update_inode_atime(struct inode *inode,
diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
index 21a6050..bc844bf 100644
--- a/fs/ocfs2/inode.c
+++ b/fs/ocfs2/inode.c
@@ -89,6 +89,25 @@ void ocfs2_set_inode_flags(struct inode 
 		inode->i_flags |= S_DIRSYNC;
 }
 
+/* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
+void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
+{
+	unsigned int flags = oi->vfs_inode.i_flags;
+
+	oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
+			OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
+	if (flags & S_SYNC)
+		oi->ip_attr |= OCFS2_SYNC_FL;
+	if (flags & S_APPEND)
+		oi->ip_attr |= OCFS2_APPEND_FL;
+	if (flags & S_IMMUTABLE)
+		oi->ip_attr |= OCFS2_IMMUTABLE_FL;
+	if (flags & S_NOATIME)
+		oi->ip_attr |= OCFS2_NOATIME_FL;
+	if (flags & S_DIRSYNC)
+		oi->ip_attr |= OCFS2_DIRSYNC_FL;
+}
+
 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
 {
 	struct inode *inode = NULL;
@@ -196,7 +215,7 @@ int ocfs2_populate_inode(struct inode *i
 	int status = -EINVAL;
 
 	mlog_entry("(0x%p, size:%llu)\n", inode,
-		   (unsigned long long)fe->i_size);
+		   (unsigned long long)le64_to_cpu(fe->i_size));
 
 	sb = inode->i_sb;
 	osb = OCFS2_SB(sb);
@@ -248,7 +267,7 @@ int ocfs2_populate_inode(struct inode *i
 		mlog(ML_ERROR,
 		     "ip_blkno %llu != i_blkno %llu!\n",
 		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
-		     (unsigned long long)fe->i_blkno);
+		     (unsigned long long)le64_to_cpu(fe->i_blkno));
 
 	inode->i_nlink = le16_to_cpu(fe->i_links_count);
 
@@ -301,7 +320,7 @@ int ocfs2_populate_inode(struct inode *i
 		 * the generation argument to
 		 * ocfs2_inode_lock_res_init() will have to change.
 		 */
-		BUG_ON(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL));
+		BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
 
 		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
 					  OCFS2_LOCK_TYPE_META, 0, inode);
@@ -437,7 +456,8 @@ static int ocfs2_read_locked_inode(struc
 	fe = (struct ocfs2_dinode *) bh->b_data;
 	if (!OCFS2_IS_VALID_DINODE(fe)) {
 		mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
-		     (unsigned long long)fe->i_blkno, 7, fe->i_signature);
+		     (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
+		     fe->i_signature);
 		goto bail;
 	}
 
@@ -812,8 +832,8 @@ static int ocfs2_query_inode_wipe(struct
 		     "Inode %llu (on-disk %llu) not orphaned! "
 		     "Disk flags  0x%x, inode flags 0x%x\n",
 		     (unsigned long long)oi->ip_blkno,
-		     (unsigned long long)di->i_blkno, di->i_flags,
-		     oi->ip_flags);
+		     (unsigned long long)le64_to_cpu(di->i_blkno),
+		     le32_to_cpu(di->i_flags), oi->ip_flags);
 		goto bail;
 	}
 
@@ -1106,8 +1126,10 @@ struct buffer_head *ocfs2_bread(struct i
 		return NULL;
 	}
 
+	down_read(&OCFS2_I(inode)->ip_alloc_sem);
 	tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
 					     NULL);
+	up_read(&OCFS2_I(inode)->ip_alloc_sem);
 	if (tmperr < 0) {
 		mlog_errno(tmperr);
 		goto fail;
@@ -1197,6 +1219,7 @@ int ocfs2_mark_inode_dirty(handle_t *han
 
 	spin_lock(&OCFS2_I(inode)->ip_lock);
 	fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
+	ocfs2_get_inode_flags(OCFS2_I(inode));
 	fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
 	spin_unlock(&OCFS2_I(inode)->ip_lock);
 
diff --git a/fs/ocfs2/inode.h b/fs/ocfs2/inode.h
index 03ae075..a41d081 100644
--- a/fs/ocfs2/inode.h
+++ b/fs/ocfs2/inode.h
@@ -141,6 +141,7 @@ int ocfs2_aio_read(struct file *file, st
 int ocfs2_aio_write(struct file *file, struct kiocb *req, struct iocb *iocb);
 
 void ocfs2_set_inode_flags(struct inode *inode);
+void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi);
 
 static inline blkcnt_t ocfs2_inode_sector_count(struct inode *inode)
 {
diff --git a/fs/ocfs2/ioctl.c b/fs/ocfs2/ioctl.c
index 4768be5..f3ad21a 100644
--- a/fs/ocfs2/ioctl.c
+++ b/fs/ocfs2/ioctl.c
@@ -31,6 +31,7 @@ static int ocfs2_get_inode_attr(struct i
 		mlog_errno(status);
 		return status;
 	}
+	ocfs2_get_inode_flags(OCFS2_I(inode));
 	*flags = OCFS2_I(inode)->ip_attr;
 	ocfs2_meta_unlock(inode, 0);
 
@@ -134,3 +135,26 @@ int ocfs2_ioctl(struct inode * inode, st
 	}
 }
 
+#ifdef CONFIG_COMPAT
+long ocfs2_compat_ioctl(struct file *file, unsigned cmd, unsigned long arg)
+{
+	struct inode *inode = file->f_path.dentry->d_inode;
+	int ret;
+
+	switch (cmd) {
+	case OCFS2_IOC32_GETFLAGS:
+		cmd = OCFS2_IOC_GETFLAGS;
+		break;
+	case OCFS2_IOC32_SETFLAGS:
+		cmd = OCFS2_IOC_SETFLAGS;
+		break;
+	default:
+		return -ENOIOCTLCMD;
+	}
+
+	lock_kernel();
+	ret = ocfs2_ioctl(inode, file, cmd, arg);
+	unlock_kernel();
+	return ret;
+}
+#endif
diff --git a/fs/ocfs2/ioctl.h b/fs/ocfs2/ioctl.h
index 4a7c829..4d6c4f4 100644
--- a/fs/ocfs2/ioctl.h
+++ b/fs/ocfs2/ioctl.h
@@ -12,5 +12,6 @@ #define OCFS2_IOCTL_H
 
 int ocfs2_ioctl(struct inode * inode, struct file * filp,
 	unsigned int cmd, unsigned long arg);
+long ocfs2_compat_ioctl(struct file *file, unsigned cmd, unsigned long arg);
 
 #endif /* OCFS2_IOCTL_H */
diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
index 5a8a90d..dc11880 100644
--- a/fs/ocfs2/journal.c
+++ b/fs/ocfs2/journal.c
@@ -435,7 +435,8 @@ static int ocfs2_journal_toggle_dirty(st
 		 * handle the errors in a specific manner, so no need
 		 * to call ocfs2_error() here. */
 		mlog(ML_ERROR, "Journal dinode %llu  has invalid "
-		     "signature: %.*s", (unsigned long long)fe->i_blkno, 7,
+		     "signature: %.*s",
+		     (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
 		     fe->i_signature);
 		status = -EIO;
 		goto out;
@@ -742,7 +743,7 @@ void ocfs2_complete_recovery(struct work
 		la_dinode = item->lri_la_dinode;
 		if (la_dinode) {
 			mlog(0, "Clean up local alloc %llu\n",
-			     (unsigned long long)la_dinode->i_blkno);
+			     (unsigned long long)le64_to_cpu(la_dinode->i_blkno));
 
 			ret = ocfs2_complete_local_alloc_recovery(osb,
 								  la_dinode);
@@ -755,7 +756,7 @@ void ocfs2_complete_recovery(struct work
 		tl_dinode = item->lri_tl_dinode;
 		if (tl_dinode) {
 			mlog(0, "Clean up truncate log %llu\n",
-			     (unsigned long long)tl_dinode->i_blkno);
+			     (unsigned long long)le64_to_cpu(tl_dinode->i_blkno));
 
 			ret = ocfs2_complete_truncate_log_recovery(osb,
 								   tl_dinode);
diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
index 2bcf353..36289e6 100644
--- a/fs/ocfs2/namei.c
+++ b/fs/ocfs2/namei.c
@@ -578,8 +578,9 @@ static int ocfs2_mknod_locked(struct ocf
 	if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 		mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 		     "i_blkno=%llu, i_ino=%lu\n",
-		     (unsigned long long) (*new_fe_bh)->b_blocknr,
-		     (unsigned long long)fe->i_blkno, inode->i_ino);
+		     (unsigned long long)(*new_fe_bh)->b_blocknr,
+		     (unsigned long long)le64_to_cpu(fe->i_blkno),
+		     inode->i_ino);
 		BUG();
 	}
 
diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
index 82cc92d..a860633 100644
--- a/fs/ocfs2/ocfs2.h
+++ b/fs/ocfs2/ocfs2.h
@@ -363,9 +363,9 @@ #define OCFS2_RO_ON_INVALID_DINODE(__sb,
 	typeof(__di) ____di = (__di);					\
 	ocfs2_error((__sb), 						\
 		"Dinode # %llu has bad signature %.*s",			\
-		(unsigned long long)(____di)->i_blkno, 7,		\
+		(unsigned long long)le64_to_cpu((____di)->i_blkno), 7, 	\
 		(____di)->i_signature);					\
-} while (0);
+} while (0)
 
 #define OCFS2_IS_VALID_EXTENT_BLOCK(ptr)				\
 	(!strcmp((ptr)->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE))
@@ -374,9 +374,9 @@ #define OCFS2_RO_ON_INVALID_EXTENT_BLOCK
 	typeof(__eb) ____eb = (__eb);					\
 	ocfs2_error((__sb), 						\
 		"Extent Block # %llu has bad signature %.*s",		\
-		(unsigned long long)(____eb)->h_blkno, 7,		\
+		(unsigned long long)le64_to_cpu((____eb)->h_blkno), 7,	\
 		(____eb)->h_signature);					\
-} while (0);
+} while (0)
 
 #define OCFS2_IS_VALID_GROUP_DESC(ptr)					\
 	(!strcmp((ptr)->bg_signature, OCFS2_GROUP_DESC_SIGNATURE))
@@ -385,9 +385,9 @@ #define OCFS2_RO_ON_INVALID_GROUP_DESC(_
 	typeof(__gd) ____gd = (__gd);					\
 		ocfs2_error((__sb),					\
 		"Group Descriptor # %llu has bad signature %.*s",	\
-		(unsigned long long)(____gd)->bg_blkno, 7,		\
+		(unsigned long long)le64_to_cpu((____gd)->bg_blkno), 7, \
 		(____gd)->bg_signature);				\
-} while (0);
+} while (0)
 
 static inline unsigned long ino_from_blkno(struct super_block *sb,
 					   u64 blkno)
diff --git a/fs/ocfs2/ocfs2_fs.h b/fs/ocfs2/ocfs2_fs.h
index 7130647..f0d9eb0 100644
--- a/fs/ocfs2/ocfs2_fs.h
+++ b/fs/ocfs2/ocfs2_fs.h
@@ -166,6 +166,8 @@ #define OCFS2_EXT_UNWRITTEN	(0x01)	/* Ex
  */
 #define OCFS2_IOC_GETFLAGS	_IOR('f', 1, long)
 #define OCFS2_IOC_SETFLAGS	_IOW('f', 2, long)
+#define OCFS2_IOC32_GETFLAGS	_IOR('f', 1, int)
+#define OCFS2_IOC32_SETFLAGS	_IOW('f', 2, int)
 
 /*
  * Journal Flags (ocfs2_dinode.id1.journal1.i_flags)
diff --git a/fs/ocfs2/suballoc.c b/fs/ocfs2/suballoc.c
index 0da655a..e343762 100644
--- a/fs/ocfs2/suballoc.c
+++ b/fs/ocfs2/suballoc.c
@@ -849,9 +849,9 @@ static int ocfs2_relink_block_group(hand
 	}
 
 	mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
-	     (unsigned long long)fe->i_blkno, chain,
-	     (unsigned long long)bg->bg_blkno,
-	     (unsigned long long)prev_bg->bg_blkno);
+	     (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
+	     (unsigned long long)le64_to_cpu(bg->bg_blkno),
+	     (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
 
 	fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
 	bg_ptr = le64_to_cpu(bg->bg_next_group);
@@ -1162,7 +1162,7 @@ static int ocfs2_search_chain(struct ocf
 	}
 
 	mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
-	     tmp_bits, (unsigned long long)bg->bg_blkno);
+	     tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
 
 	*num_bits = tmp_bits;
 
@@ -1227,7 +1227,7 @@ static int ocfs2_search_chain(struct ocf
 	}
 
 	mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
-	     (unsigned long long)fe->i_blkno);
+	     (unsigned long long)le64_to_cpu(fe->i_blkno));
 
 	*bg_blkno = le64_to_cpu(bg->bg_blkno);
 	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
index 5c9e824..f549354 100644
--- a/fs/ocfs2/super.c
+++ b/fs/ocfs2/super.c
@@ -1538,7 +1538,7 @@ static int ocfs2_verify_volume(struct oc
 		} else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
 			mlog(ML_ERROR, "bad block number on superblock: "
 			     "found %llu, should be %llu\n",
-			     (unsigned long long)di->i_blkno,
+			     (unsigned long long)le64_to_cpu(di->i_blkno),
 			     (unsigned long long)bh->b_blocknr);
 		} else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
 			    le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-03-28 20:29 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-03-28 20:29 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

Hi Linus,

	These two fixes are for ocfs2 dlm regressions introduced during the
merge window.

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/dlm/dlmdomain.c |    8 ++-
 fs/ocfs2/dlm/dlmmaster.c |   99 ++++++++++++++++++++++++++++++-----------------
 fs/ocfs2/dlm/dlmthread.c |   10 ----
 3 files changed, 73 insertions(+), 44 deletions(-)

Sunil Mushran:
      ocfs2_dlm: Fix lockres ref counting bug
      ocfs2_dlm: Check for migrateable lockres in dlm_empty_lockres()

diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index 6087c47..c558442 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -138,8 +138,10 @@ static void dlm_unregister_domain_handle
 
 void __dlm_unhash_lockres(struct dlm_lock_resource *lockres)
 {
-	hlist_del_init(&lockres->hash_node);
-	dlm_lockres_put(lockres);
+	if (!hlist_unhashed(&lockres->hash_node)) {
+		hlist_del_init(&lockres->hash_node);
+		dlm_lockres_put(lockres);
+	}
 }
 
 void __dlm_insert_lockres(struct dlm_ctxt *dlm,
@@ -655,6 +657,8 @@ void dlm_unregister_domain(struct dlm_ct
 		dlm_kick_thread(dlm, NULL);
 
 		while (dlm_migrate_all_locks(dlm)) {
+			/* Give dlm_thread time to purge the lockres' */
+			msleep(500);
 			mlog(0, "%s: more migration to do\n", dlm->name);
 		}
 		dlm_mark_domain_leaving(dlm);
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index 9229e04..6edffca 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -2424,6 +2424,57 @@ static void dlm_deref_lockres_worker(str
 	dlm_lockres_put(res);
 }
 
+/* Checks whether the lockres can be migrated. Returns 0 if yes, < 0
+ * if not. If 0, numlocks is set to the number of locks in the lockres.
+ */
+static int dlm_is_lockres_migrateable(struct dlm_ctxt *dlm,
+				      struct dlm_lock_resource *res,
+				      int *numlocks)
+{
+	int ret;
+	int i;
+	int count = 0;
+	struct list_head *queue, *iter;
+	struct dlm_lock *lock;
+
+	assert_spin_locked(&res->spinlock);
+
+	ret = -EINVAL;
+	if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
+		mlog(0, "cannot migrate lockres with unknown owner!\n");
+		goto leave;
+	}
+
+	if (res->owner != dlm->node_num) {
+		mlog(0, "cannot migrate lockres this node doesn't own!\n");
+		goto leave;
+	}
+
+	ret = 0;
+	queue = &res->granted;
+	for (i = 0; i < 3; i++) {
+		list_for_each(iter, queue) {
+			lock = list_entry(iter, struct dlm_lock, list);
+			++count;
+			if (lock->ml.node == dlm->node_num) {
+				mlog(0, "found a lock owned by this node still "
+				     "on the %s queue!  will not migrate this "
+				     "lockres\n", (i == 0 ? "granted" :
+						   (i == 1 ? "converting" :
+						    "blocked")));
+				ret = -ENOTEMPTY;
+				goto leave;
+			}
+		}
+		queue++;
+	}
+
+	*numlocks = count;
+	mlog(0, "migrateable lockres having %d locks\n", *numlocks);
+
+leave:
+	return ret;
+}
 
 /*
  * DLM_MIGRATE_LOCKRES
@@ -2437,14 +2488,12 @@ static int dlm_migrate_lockres(struct dl
 	struct dlm_master_list_entry *mle = NULL;
 	struct dlm_master_list_entry *oldmle = NULL;
  	struct dlm_migratable_lockres *mres = NULL;
-	int ret = -EINVAL;
+	int ret = 0;
 	const char *name;
 	unsigned int namelen;
 	int mle_added = 0;
-	struct list_head *queue, *iter;
-	int i;
-	struct dlm_lock *lock;
-	int empty = 1, wake = 0;
+	int numlocks;
+	int wake = 0;
 
 	if (!dlm_grab(dlm))
 		return -EINVAL;
@@ -2458,42 +2507,16 @@ static int dlm_migrate_lockres(struct dl
 	 * ensure this lockres is a proper candidate for migration
 	 */
 	spin_lock(&res->spinlock);
-	if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
-		mlog(0, "cannot migrate lockres with unknown owner!\n");
-		spin_unlock(&res->spinlock);
-		goto leave;
-	}
-	if (res->owner != dlm->node_num) {
-		mlog(0, "cannot migrate lockres this node doesn't own!\n");
+	ret = dlm_is_lockres_migrateable(dlm, res, &numlocks);
+	if (ret < 0) {
 		spin_unlock(&res->spinlock);
 		goto leave;
 	}
-	mlog(0, "checking queues...\n");
-	queue = &res->granted;
-	for (i=0; i<3; i++) {
-		list_for_each(iter, queue) {
-			lock = list_entry (iter, struct dlm_lock, list);
-			empty = 0;
-			if (lock->ml.node == dlm->node_num) {
-				mlog(0, "found a lock owned by this node "
-				     "still on the %s queue!  will not "
-				     "migrate this lockres\n",
-				     i==0 ? "granted" :
-				     (i==1 ? "converting" : "blocked"));
-				spin_unlock(&res->spinlock);
-				ret = -ENOTEMPTY;
-				goto leave;
-			}
-		}
-		queue++;
-	}
-	mlog(0, "all locks on this lockres are nonlocal.  continuing\n");
 	spin_unlock(&res->spinlock);
 
 	/* no work to do */
-	if (empty) {
+	if (numlocks == 0) {
 		mlog(0, "no locks were found on this lockres! done!\n");
-		ret = 0;
 		goto leave;
 	}
 
@@ -2729,6 +2752,7 @@ int dlm_empty_lockres(struct dlm_ctxt *d
 {
 	int ret;
 	int lock_dropped = 0;
+	int numlocks;
 
 	spin_lock(&res->spinlock);
 	if (res->owner != dlm->node_num) {
@@ -2740,6 +2764,13 @@ int dlm_empty_lockres(struct dlm_ctxt *d
 		spin_unlock(&res->spinlock);
 		goto leave;
 	}
+
+	/* No need to migrate a lockres having no locks */
+	ret = dlm_is_lockres_migrateable(dlm, res, &numlocks);
+	if (ret >= 0 && numlocks == 0) {
+		spin_unlock(&res->spinlock);
+		goto leave;
+	}
 	spin_unlock(&res->spinlock);
 
 	/* Wheee! Migrate lockres here! Will sleep so drop spinlock. */
diff --git a/fs/ocfs2/dlm/dlmthread.c b/fs/ocfs2/dlm/dlmthread.c
index 6421a8f..2b264c6 100644
--- a/fs/ocfs2/dlm/dlmthread.c
+++ b/fs/ocfs2/dlm/dlmthread.c
@@ -256,20 +256,14 @@ static void dlm_run_purge_list(struct dl
 			break;
 		}
 
-		mlog(0, "removing lockres %.*s:%p from purgelist\n",
-		     lockres->lockname.len, lockres->lockname.name, lockres);
-		list_del_init(&lockres->purge);
-		dlm_lockres_put(lockres);
-		dlm->purge_count--;
+		dlm_lockres_get(lockres);
 
 		/* This may drop and reacquire the dlm spinlock if it
 		 * has to do migration. */
-		mlog(0, "calling dlm_purge_lockres!\n");
-		dlm_lockres_get(lockres);
 		if (dlm_purge_lockres(dlm, lockres))
 			BUG();
+
 		dlm_lockres_put(lockres);
-		mlog(0, "DONE calling dlm_purge_lockres!\n");
 
 		/* Avoid adding any scheduling latencies */
 		cond_resched_lock(&dlm->spinlock);

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-02-02 17:28 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-02-02 17:28 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, linux-kernel, ocfs2-devel

One more regression fix for 2.6.20.

For some reason my test pulls aren't going through right now. Just to be
safe, I'll also include the entire patch and commit message below in a
manner suitable for commit from an e-mail.
	--Mark

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following update:


From: Mark Fasheh <mark.fasheh@oracle.com>
[PATCH] ocfs2: ocfs2_link() journal credits update

Commit 592282cf2eaa33409c6511ddd3f3ecaa57daeaaa fixed some missing directory
c/mtime updates in part by introducing a dinode update in ocfs2_add_entry().
Unfortunately, ocfs2_link() (which didn't update the directory inode before)
is now missing a single journal credit. Fix this by doubling the number of
inode updates expected during hard link creation.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
---
 fs/ocfs2/journal.h |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/ocfs2/journal.h b/fs/ocfs2/journal.h
index e121636..d026b4f 100644
--- a/fs/ocfs2/journal.h
+++ b/fs/ocfs2/journal.h
@@ -306,8 +306,8 @@ #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2
  * for the dinode, one for the new block. */
 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
 
-/* file update (nlink, etc) + dir entry block */
-#define OCFS2_LINK_CREDITS  (OCFS2_INODE_UPDATE_CREDITS + 1)
+/* file update (nlink, etc) + directory mtime/ctime + dir entry block */
+#define OCFS2_LINK_CREDITS  (2*OCFS2_INODE_UPDATE_CREDITS + 1)
 
 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
  * dir inode link */
-- 
1.4.2.3


^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2007-01-29 20:04 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2007-01-29 20:04 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, ocfs2-devel, linux-kernel

Hi Linus,
	I made a silly error in my last upstream push. This patch makes
ocfs2 build again.
	--Mark

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/ocfs2_fs.h |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Mark Fasheh:
      ocfs2: fix thinko in ocfs2_backup_super_blkno()

diff --git a/fs/ocfs2/ocfs2_fs.h b/fs/ocfs2/ocfs2_fs.h
index c99e905..e61e218 100644
--- a/fs/ocfs2/ocfs2_fs.h
+++ b/fs/ocfs2/ocfs2_fs.h
@@ -587,7 +587,7 @@ static inline u64 ocfs2_backup_super_blk
 
 	if (index >= 0 && index < OCFS2_MAX_BACKUP_SUPERBLOCKS) {
 		offset <<= (2 * index);
-		offset /= sb->s_blocksize;
+		offset >>= sb->s_blocksize_bits;
 		return offset;
 	}
 

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2006-12-29 23:18 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2006-12-29 23:18 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Andrew Morton, ocfs2-devel, linux-kernel

Hi Linus,
 	Here are some 2.6.20 fixes for ocfs2. The patch by Zhen Wei isn't
really a fix, but a very small amount of support for a feature which is
mostly implemented in ocfs2-tools. Considering it's just a single attribute
export via configfs, I'd say it's pretty safe to merge.

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git upstream-linus

to receive the following updates:

 fs/ocfs2/aops.c              |   24 +++++++++++++++++-------
 fs/ocfs2/cluster/heartbeat.c |   17 +++++++++++++++++
 fs/ocfs2/dlmglue.c           |   10 +++++++++-
 fs/ocfs2/file.c              |   13 +++++++++++--
 4 files changed, 54 insertions(+), 10 deletions(-)

Mark Fasheh:
      ocfs2: don't print error in ocfs2_permission()
      ocfs2: Allow direct I/O read past end of file
      ocfs2: ignore NULL vfsmnt in ocfs2_should_update_atime()
      ocfs2: always unmap in ocfs2_data_convert_worker()

Zhen Wei:
      ocfs2: export heartbeat thread pid via configfs

diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index ef6cd30..93628b0 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -540,8 +540,7 @@ static int ocfs2_direct_IO_get_blocks(st
 				     struct buffer_head *bh_result, int create)
 {
 	int ret;
-	u64 vbo_max; /* file offset, max_blocks from iblock */
-	u64 p_blkno;
+	u64 p_blkno, inode_blocks;
 	int contig_blocks;
 	unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
 	unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
@@ -550,12 +549,23 @@ static int ocfs2_direct_IO_get_blocks(st
 	 * nicely aligned and of the right size, so there's no need
 	 * for us to check any of that. */
 
-	vbo_max = ((u64)iblock + max_blocks) << blocksize_bits;
-
 	spin_lock(&OCFS2_I(inode)->ip_lock);
-	if ((iblock + max_blocks) >
-	    ocfs2_clusters_to_blocks(inode->i_sb,
-				     OCFS2_I(inode)->ip_clusters)) {
+	inode_blocks = ocfs2_clusters_to_blocks(inode->i_sb,
+						OCFS2_I(inode)->ip_clusters);
+
+	/*
+	 * For a read which begins past the end of file, we return a hole.
+	 */
+	if (!create && (iblock >= inode_blocks)) {
+		spin_unlock(&OCFS2_I(inode)->ip_lock);
+		ret = 0;
+		goto bail;
+	}
+
+	/*
+	 * Any write past EOF is not allowed because we'd be extending.
+	 */
+	if (create && (iblock + max_blocks) > inode_blocks) {
 		spin_unlock(&OCFS2_I(inode)->ip_lock);
 		ret = -EIO;
 		goto bail;
diff --git a/fs/ocfs2/cluster/heartbeat.c b/fs/ocfs2/cluster/heartbeat.c
index a25ef5a..277ca67 100644
--- a/fs/ocfs2/cluster/heartbeat.c
+++ b/fs/ocfs2/cluster/heartbeat.c
@@ -1447,6 +1447,15 @@ out:
 	return ret;
 }
 
+static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
+                                      char *page)
+{
+	if (!reg->hr_task)
+		return 0;
+
+	return sprintf(page, "%u\n", reg->hr_task->pid);
+}
+
 struct o2hb_region_attribute {
 	struct configfs_attribute attr;
 	ssize_t (*show)(struct o2hb_region *, char *);
@@ -1485,11 +1494,19 @@ static struct o2hb_region_attribute o2hb
 	.store	= o2hb_region_dev_write,
 };
 
+static struct o2hb_region_attribute o2hb_region_attr_pid = {
+       .attr   = { .ca_owner = THIS_MODULE,
+                   .ca_name = "pid",
+                   .ca_mode = S_IRUGO | S_IRUSR },
+       .show   = o2hb_region_pid_read,
+};
+
 static struct configfs_attribute *o2hb_region_attrs[] = {
 	&o2hb_region_attr_block_bytes.attr,
 	&o2hb_region_attr_start_block.attr,
 	&o2hb_region_attr_blocks.attr,
 	&o2hb_region_attr_dev.attr,
+	&o2hb_region_attr_pid.attr,
 	NULL,
 };
 
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
index e622013..e335541 100644
--- a/fs/ocfs2/dlmglue.c
+++ b/fs/ocfs2/dlmglue.c
@@ -2718,6 +2718,15 @@ static int ocfs2_data_convert_worker(str
        	inode = ocfs2_lock_res_inode(lockres);
 	mapping = inode->i_mapping;
 
+	/*
+	 * We need this before the filemap_fdatawrite() so that it can
+	 * transfer the dirty bit from the PTE to the
+	 * page. Unfortunately this means that even for EX->PR
+	 * downconverts, we'll lose our mappings and have to build
+	 * them up again.
+	 */
+	unmap_mapping_range(mapping, 0, 0, 0);
+
 	if (filemap_fdatawrite(mapping)) {
 		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
@@ -2725,7 +2734,6 @@ static int ocfs2_data_convert_worker(str
 	sync_mapping_buffers(mapping);
 	if (blocking == LKM_EXMODE) {
 		truncate_inode_pages(mapping, 0);
-		unmap_mapping_range(mapping, 0, 0, 0);
 	} else {
 		/* We only need to wait on the I/O if we're not also
 		 * truncating pages because truncate_inode_pages waits
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 9fd590b..10953a5 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -149,6 +149,17 @@ int ocfs2_should_update_atime(struct ino
 	    ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 		return 0;
 
+	/*
+	 * We can be called with no vfsmnt structure - NFSD will
+	 * sometimes do this.
+	 *
+	 * Note that our action here is different than touch_atime() -
+	 * if we can't tell whether this is a noatime mount, then we
+	 * don't know whether to trust the value of s_atime_quantum.
+	 */
+	if (vfsmnt == NULL)
+		return 0;
+
 	if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 	    ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 		return 0;
@@ -966,8 +977,6 @@ int ocfs2_permission(struct inode *inode
 	}
 
 	ret = generic_permission(inode, mask, NULL);
-	if (ret)
-		mlog_errno(ret);
 
 	ocfs2_meta_unlock(inode, 0);
 out:

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2006-06-21 18:43 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2006-06-21 18:43 UTC (permalink / raw)
  To: Andrew Morton, Linus Torvalds; +Cc: ocfs2-devel, linux-kernel

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git

to receive the following updates:

 fs/ocfs2/dlm/dlmast.c      |   12 
 fs/ocfs2/dlm/dlmcommon.h   |   63 ++++
 fs/ocfs2/dlm/dlmconvert.c  |   24 +
 fs/ocfs2/dlm/dlmdebug.c    |    6 
 fs/ocfs2/dlm/dlmdebug.h    |   30 --
 fs/ocfs2/dlm/dlmdomain.c   |  101 +++++--
 fs/ocfs2/dlm/dlmfs.c       |    6 
 fs/ocfs2/dlm/dlmlock.c     |   68 ++++-
 fs/ocfs2/dlm/dlmmaster.c   |  448 +++++++++++++++++++++++-----------
 fs/ocfs2/dlm/dlmrecovery.c |  580 ++++++++++++++++++++++++++++++++++-----------
 fs/ocfs2/dlm/dlmthread.c   |   68 ++++-
 fs/ocfs2/dlm/dlmunlock.c   |   10 
 fs/ocfs2/dlm/userdlm.c     |    2 
 13 files changed, 1045 insertions(+), 373 deletions(-)

Adrian Bunk:
      fs/ocfs2/dlm/: cleanups

Daniel Phillips:
      Clean up ocfs2 hash probe and make it faster
      ocfs2: allocate lockres hash pages in an array

Joel Becker:
      ocfs2: Alloc at least a page for the DLM hash

Kurt Hackel:
      ocfs2: add a small delay after a failed migration
      ocfs2: recheck lockres master before sending an unlock request.
      ocfs2: fix inverted logic in dlm_is_node_dead
      ocfs2: Fix empty lvb check
      ocfs2: Better tracking for recovery state changes
      ocfs2: only recover one dead node at a time
      ocfs2: handle network errors during recovery
      ocfs2: clean up recovery related messages
      ocfs2: better mle debugging
      ocfs2: mle ref counting fixes
      ocfs2: detach mle from heartbeat events
      ocfs2: properly initialize the mle structure
      ocfs2: take mle reference during migration
      ocfs2: allow for an assert message during lock mastery
      ocfs2: mle ref count debugging
      ocfs2: dump lockres info before we BUG() on a bad reference
      ocfs2: better error handling during assert master message
      ocfs2: dlm recovery / lockres reference count fix
      ocfs2: make dlm recovery finalization 2 stage
      ocfs2: dump mismatching migrated lvbs before BUG()
      ocfs2: purge lockres' sooner
      ocfs2: do not send master requests to localhost
      ocfs2: update lvb immediately during recovery
      ocfs2: gracefully handle stale create_lock messages.
      ocfs2: teach dlm_restart_lock_mastery() to wait on recovery
      ocfs2: give the dlm dirty list a reference on the lockres
      ocfs2: have dlm_pre_master_reco_lockres() ignore dead nodes
      ocfs2: increase backoff before waiting for recovery
      ocfs2: do not unconditionally purge the lockres in dlmlock_remote()
      ocfs2: temporarily disable automatic lock migration
      ocfs2: pending mastery asserts and migrations should block each other
      ocfs2: special case recovery lock in dlmlock_remote()
      ocfs2: dlm_remaster_locks() should never exit without completing
      ocfs2: remove unneccesary spin_unlock() in dlm_remaster_locks()
      ocfs2: continue recovery when a dead node is encountered
      ocfs2: wait for recovery when starting lock mastery
      ocfs2: use GFP_NOFS in some dlm operations
      ocfs2: use cond_resched() in dlm_thread()
      ocfs2: retry operations when a lock is marked in recovery
      ocfs2: mlog in dlm_convert_lock_handler() should be ML_ERROR
      ocfs2: display message before waiting for recovery to complete
      ocfs2: tune down some noisy messages during dlm recovery
      ocfs2: fix incorrect error returns
      ocfs2: move dlm work to a private work queue
      ocfs2: remove whitespace in dlmunlock.c

Mark Fasheh:
      ocfs2: move lockres qstr next to hlist_node structure
      ocfs2: calculate lockid hash values outside of the spinlock
      ocfs2: inline dlm_lockres_get()
      ocfs2: silence a compile warning in dlm_alloc_pagevec()
      ocfs2: do LVB puts in place
      ocfs2: dlm_print_one_mle() needs to be defined
      ocfs2: fix compiler warnings in dlm_convert_lock_handler()

diff --git a/fs/ocfs2/dlm/dlmast.c b/fs/ocfs2/dlm/dlmast.c
index 355593d..90ce915 100644
--- a/fs/ocfs2/dlm/dlmast.c
+++ b/fs/ocfs2/dlm/dlmast.c
@@ -197,12 +197,14 @@ static void dlm_update_lvb(struct dlm_ct
 				  lock->ml.node == dlm->node_num ? "master" :
 				  "remote");
 			memcpy(lksb->lvb, res->lvb, DLM_LVB_LEN);
-		} else if (lksb->flags & DLM_LKSB_PUT_LVB) {
-			mlog(0, "setting lvb from lockres for %s node\n",
-				  lock->ml.node == dlm->node_num ? "master" :
-				  "remote");
-			memcpy(res->lvb, lksb->lvb, DLM_LVB_LEN);
 		}
+		/* Do nothing for lvb put requests - they should be done in
+ 		 * place when the lock is downconverted - otherwise we risk
+ 		 * racing gets and puts which could result in old lvb data
+ 		 * being propagated. We leave the put flag set and clear it
+ 		 * here. In the future we might want to clear it at the time
+ 		 * the put is actually done.
+		 */
 		spin_unlock(&res->spinlock);
 	}
 
diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h
index 88cc43d..9bdc9cf 100644
--- a/fs/ocfs2/dlm/dlmcommon.h
+++ b/fs/ocfs2/dlm/dlmcommon.h
@@ -37,7 +37,17 @@ #define DLM_LOCK_RES_OWNER_UNKNOWN     O
 #define DLM_THREAD_SHUFFLE_INTERVAL    5     // flush everything every 5 passes
 #define DLM_THREAD_MS                  200   // flush at least every 200 ms
 
-#define DLM_HASH_BUCKETS     (PAGE_SIZE / sizeof(struct hlist_head))
+#define DLM_HASH_SIZE_DEFAULT	(1 << 14)
+#if DLM_HASH_SIZE_DEFAULT < PAGE_SIZE
+# define DLM_HASH_PAGES		1
+#else
+# define DLM_HASH_PAGES		(DLM_HASH_SIZE_DEFAULT / PAGE_SIZE)
+#endif
+#define DLM_BUCKETS_PER_PAGE	(PAGE_SIZE / sizeof(struct hlist_head))
+#define DLM_HASH_BUCKETS	(DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE)
+
+/* Intended to make it easier for us to switch out hash functions */
+#define dlm_lockid_hash(_n, _l) full_name_hash(_n, _l)
 
 enum dlm_ast_type {
 	DLM_AST = 0,
@@ -61,7 +71,8 @@ static inline int dlm_is_recovery_lock(c
 	return 0;
 }
 
-#define DLM_RECO_STATE_ACTIVE  0x0001
+#define DLM_RECO_STATE_ACTIVE    0x0001
+#define DLM_RECO_STATE_FINALIZE  0x0002
 
 struct dlm_recovery_ctxt
 {
@@ -85,7 +96,7 @@ enum dlm_ctxt_state {
 struct dlm_ctxt
 {
 	struct list_head list;
-	struct hlist_head *lockres_hash;
+	struct hlist_head **lockres_hash;
 	struct list_head dirty_list;
 	struct list_head purge_list;
 	struct list_head pending_asts;
@@ -120,6 +131,7 @@ struct dlm_ctxt
 	struct o2hb_callback_func dlm_hb_down;
 	struct task_struct *dlm_thread_task;
 	struct task_struct *dlm_reco_thread_task;
+	struct workqueue_struct *dlm_worker;
 	wait_queue_head_t dlm_thread_wq;
 	wait_queue_head_t dlm_reco_thread_wq;
 	wait_queue_head_t ast_wq;
@@ -132,6 +144,11 @@ struct dlm_ctxt
 	struct list_head	dlm_eviction_callbacks;
 };
 
+static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i)
+{
+	return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE);
+}
+
 /* these keventd work queue items are for less-frequently
  * called functions that cannot be directly called from the
  * net message handlers for some reason, usually because
@@ -216,20 +233,29 @@ struct dlm_lock_resource
 	/* WARNING: Please see the comment in dlm_init_lockres before
 	 * adding fields here. */
 	struct hlist_node hash_node;
+	struct qstr lockname;
 	struct kref      refs;
 
-	/* please keep these next 3 in this order
-	 * some funcs want to iterate over all lists */
+	/*
+	 * Please keep granted, converting, and blocked in this order,
+	 * as some funcs want to iterate over all lists.
+	 *
+	 * All four lists are protected by the hash's reference.
+	 */
 	struct list_head granted;
 	struct list_head converting;
 	struct list_head blocked;
+	struct list_head purge;
 
+	/*
+	 * These two lists require you to hold an additional reference
+	 * while they are on the list.
+	 */
 	struct list_head dirty;
 	struct list_head recovering; // dlm_recovery_ctxt.resources list
 
 	/* unused lock resources have their last_used stamped and are
 	 * put on a list for the dlm thread to run. */
-	struct list_head purge;
 	unsigned long    last_used;
 
 	unsigned migration_pending:1;
@@ -238,7 +264,6 @@ struct dlm_lock_resource
 	wait_queue_head_t wq;
 	u8  owner;              //node which owns the lock resource, or unknown
 	u16 state;
-	struct qstr lockname;
 	char lvb[DLM_LVB_LEN];
 };
 
@@ -300,6 +325,15 @@ enum dlm_lockres_list {
 	DLM_BLOCKED_LIST
 };
 
+static inline int dlm_lvb_is_empty(char *lvb)
+{
+	int i;
+	for (i=0; i<DLM_LVB_LEN; i++)
+		if (lvb[i])
+			return 0;
+	return 1;
+}
+
 static inline struct list_head *
 dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx)
 {
@@ -609,7 +643,8 @@ struct dlm_finalize_reco
 {
 	u8 node_idx;
 	u8 dead_node;
-	__be16 pad1;
+	u8 flags;
+	u8 pad1;
 	__be32 pad2;
 };
 
@@ -676,6 +711,7 @@ void dlm_wait_for_recovery(struct dlm_ct
 void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
 int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node);
 int dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout);
+int dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout);
 
 void dlm_put(struct dlm_ctxt *dlm);
 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm);
@@ -687,14 +723,20 @@ void dlm_lockres_calc_usage(struct dlm_c
 			    struct dlm_lock_resource *res);
 void dlm_purge_lockres(struct dlm_ctxt *dlm,
 		       struct dlm_lock_resource *lockres);
-void dlm_lockres_get(struct dlm_lock_resource *res);
+static inline void dlm_lockres_get(struct dlm_lock_resource *res)
+{
+	/* This is called on every lookup, so it might be worth
+	 * inlining. */
+	kref_get(&res->refs);
+}
 void dlm_lockres_put(struct dlm_lock_resource *res);
 void __dlm_unhash_lockres(struct dlm_lock_resource *res);
 void __dlm_insert_lockres(struct dlm_ctxt *dlm,
 			  struct dlm_lock_resource *res);
 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
 						const char *name,
-						unsigned int len);
+						unsigned int len,
+						unsigned int hash);
 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
 					      const char *name,
 					      unsigned int len);
@@ -819,6 +861,7 @@ void dlm_clean_master_list(struct dlm_ct
 			   u8 dead_node);
 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock);
 
+int __dlm_lockres_unused(struct dlm_lock_resource *res);
 
 static inline const char * dlm_lock_mode_name(int mode)
 {
diff --git a/fs/ocfs2/dlm/dlmconvert.c b/fs/ocfs2/dlm/dlmconvert.c
index 8285228..d555dec 100644
--- a/fs/ocfs2/dlm/dlmconvert.c
+++ b/fs/ocfs2/dlm/dlmconvert.c
@@ -214,6 +214,9 @@ grant:
 	if (lock->ml.node == dlm->node_num)
 		mlog(0, "doing in-place convert for nonlocal lock\n");
 	lock->ml.type = type;
+	if (lock->lksb->flags & DLM_LKSB_PUT_LVB)
+		memcpy(res->lvb, lock->lksb->lvb, DLM_LVB_LEN);
+
 	status = DLM_NORMAL;
 	*call_ast = 1;
 	goto unlock_exit;
@@ -464,6 +467,12 @@ int dlm_convert_lock_handler(struct o2ne
 	}
 
 	spin_lock(&res->spinlock);
+	status = __dlm_lockres_state_to_status(res);
+	if (status != DLM_NORMAL) {
+		spin_unlock(&res->spinlock);
+		dlm_error(status);
+		goto leave;
+	}
 	list_for_each(iter, &res->granted) {
 		lock = list_entry(iter, struct dlm_lock, list);
 		if (lock->ml.cookie == cnv->cookie &&
@@ -473,6 +482,21 @@ int dlm_convert_lock_handler(struct o2ne
 		}
 		lock = NULL;
 	}
+	if (!lock) {
+		__dlm_print_one_lock_resource(res);
+		list_for_each(iter, &res->granted) {
+			lock = list_entry(iter, struct dlm_lock, list);
+			if (lock->ml.node == cnv->node_idx) {
+				mlog(ML_ERROR, "There is something here "
+				     "for node %u, lock->ml.cookie=%llu, "
+				     "cnv->cookie=%llu\n", cnv->node_idx,
+				     (unsigned long long)lock->ml.cookie,
+				     (unsigned long long)cnv->cookie);
+				break;
+			}
+		}
+		lock = NULL;
+	}
 	spin_unlock(&res->spinlock);
 	if (!lock) {
 		status = DLM_IVLOCKID;
diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
index c7eae5d..3f6c8d8 100644
--- a/fs/ocfs2/dlm/dlmdebug.c
+++ b/fs/ocfs2/dlm/dlmdebug.c
@@ -37,10 +37,8 @@ #include "cluster/tcp.h"
 
 #include "dlmapi.h"
 #include "dlmcommon.h"
-#include "dlmdebug.h"
 
 #include "dlmdomain.h"
-#include "dlmdebug.h"
 
 #define MLOG_MASK_PREFIX ML_DLM
 #include "cluster/masklog.h"
@@ -120,6 +118,7 @@ void dlm_print_one_lock(struct dlm_lock 
 }
 EXPORT_SYMBOL_GPL(dlm_print_one_lock);
 
+#if 0
 void dlm_dump_lock_resources(struct dlm_ctxt *dlm)
 {
 	struct dlm_lock_resource *res;
@@ -136,12 +135,13 @@ void dlm_dump_lock_resources(struct dlm_
 
 	spin_lock(&dlm->spinlock);
 	for (i=0; i<DLM_HASH_BUCKETS; i++) {
-		bucket = &(dlm->lockres_hash[i]);
+		bucket = dlm_lockres_hash(dlm, i);
 		hlist_for_each_entry(res, iter, bucket, hash_node)
 			dlm_print_one_lock_resource(res);
 	}
 	spin_unlock(&dlm->spinlock);
 }
+#endif  /*  0  */
 
 static const char *dlm_errnames[] = {
 	[DLM_NORMAL] =			"DLM_NORMAL",
diff --git a/fs/ocfs2/dlm/dlmdebug.h b/fs/ocfs2/dlm/dlmdebug.h
deleted file mode 100644
index 6858510..0000000
--- a/fs/ocfs2/dlm/dlmdebug.h
+++ /dev/null
@@ -1,30 +0,0 @@
-/* -*- mode: c; c-basic-offset: 8; -*-
- * vim: noexpandtab sw=8 ts=8 sts=0:
- *
- * dlmdebug.h
- *
- * Copyright (C) 2004 Oracle.  All rights reserved.
- *
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public
- * License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * General Public License for more details.
- *
- * You should have received a copy of the GNU General Public
- * License along with this program; if not, write to the
- * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
- * Boston, MA 021110-1307, USA.
- *
- */
-
-#ifndef DLMDEBUG_H
-#define DLMDEBUG_H
-
-void dlm_dump_lock_resources(struct dlm_ctxt *dlm);
-
-#endif
diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index 8f3a9e3..ba27c5c 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -41,7 +41,6 @@ #include "cluster/tcp.h"
 #include "dlmapi.h"
 #include "dlmcommon.h"
 
-#include "dlmdebug.h"
 #include "dlmdomain.h"
 
 #include "dlmver.h"
@@ -49,6 +48,33 @@ #include "dlmver.h"
 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
 #include "cluster/masklog.h"
 
+static void dlm_free_pagevec(void **vec, int pages)
+{
+	while (pages--)
+		free_page((unsigned long)vec[pages]);
+	kfree(vec);
+}
+
+static void **dlm_alloc_pagevec(int pages)
+{
+	void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
+	int i;
+
+	if (!vec)
+		return NULL;
+
+	for (i = 0; i < pages; i++)
+		if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
+			goto out_free;
+
+	mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
+	     pages, DLM_HASH_PAGES, (unsigned long)DLM_BUCKETS_PER_PAGE);
+	return vec;
+out_free:
+	dlm_free_pagevec(vec, i);
+	return NULL;
+}
+
 /*
  *
  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
@@ -90,8 +116,7 @@ void __dlm_insert_lockres(struct dlm_ctx
 	assert_spin_locked(&dlm->spinlock);
 
 	q = &res->lockname;
-	q->hash = full_name_hash(q->name, q->len);
-	bucket = &(dlm->lockres_hash[q->hash % DLM_HASH_BUCKETS]);
+	bucket = dlm_lockres_hash(dlm, q->hash);
 
 	/* get a reference for our hashtable */
 	dlm_lockres_get(res);
@@ -100,34 +125,32 @@ void __dlm_insert_lockres(struct dlm_ctx
 }
 
 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
-					 const char *name,
-					 unsigned int len)
+						const char *name,
+						unsigned int len,
+						unsigned int hash)
 {
-	unsigned int hash;
-	struct hlist_node *iter;
-	struct dlm_lock_resource *tmpres=NULL;
 	struct hlist_head *bucket;
+	struct hlist_node *list;
 
 	mlog_entry("%.*s\n", len, name);
 
 	assert_spin_locked(&dlm->spinlock);
 
-	hash = full_name_hash(name, len);
-
-	bucket = &(dlm->lockres_hash[hash % DLM_HASH_BUCKETS]);
-
-	/* check for pre-existing lock */
-	hlist_for_each(iter, bucket) {
-		tmpres = hlist_entry(iter, struct dlm_lock_resource, hash_node);
-		if (tmpres->lockname.len == len &&
-		    memcmp(tmpres->lockname.name, name, len) == 0) {
-			dlm_lockres_get(tmpres);
-			break;
-		}
+	bucket = dlm_lockres_hash(dlm, hash);
 
-		tmpres = NULL;
+	hlist_for_each(list, bucket) {
+		struct dlm_lock_resource *res = hlist_entry(list,
+			struct dlm_lock_resource, hash_node);
+		if (res->lockname.name[0] != name[0])
+			continue;
+		if (unlikely(res->lockname.len != len))
+			continue;
+		if (memcmp(res->lockname.name + 1, name + 1, len - 1))
+			continue;
+		dlm_lockres_get(res);
+		return res;
 	}
-	return tmpres;
+	return NULL;
 }
 
 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
@@ -135,9 +158,10 @@ struct dlm_lock_resource * dlm_lookup_lo
 				    unsigned int len)
 {
 	struct dlm_lock_resource *res;
+	unsigned int hash = dlm_lockid_hash(name, len);
 
 	spin_lock(&dlm->spinlock);
-	res = __dlm_lookup_lockres(dlm, name, len);
+	res = __dlm_lookup_lockres(dlm, name, len, hash);
 	spin_unlock(&dlm->spinlock);
 	return res;
 }
@@ -194,7 +218,7 @@ static int dlm_wait_on_domain_helper(con
 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
 {
 	if (dlm->lockres_hash)
-		free_page((unsigned long) dlm->lockres_hash);
+		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
 
 	if (dlm->name)
 		kfree(dlm->name);
@@ -278,11 +302,21 @@ int dlm_domain_fully_joined(struct dlm_c
 	return ret;
 }
 
+static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
+{
+	if (dlm->dlm_worker) {
+		flush_workqueue(dlm->dlm_worker);
+		destroy_workqueue(dlm->dlm_worker);
+		dlm->dlm_worker = NULL;
+	}
+}
+
 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
 {
 	dlm_unregister_domain_handlers(dlm);
 	dlm_complete_thread(dlm);
 	dlm_complete_recovery_thread(dlm);
+	dlm_destroy_dlm_worker(dlm);
 
 	/* We've left the domain. Now we can take ourselves out of the
 	 * list and allow the kref stuff to help us free the
@@ -304,8 +338,8 @@ static void dlm_migrate_all_locks(struct
 restart:
 	spin_lock(&dlm->spinlock);
 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
-		while (!hlist_empty(&dlm->lockres_hash[i])) {
-			res = hlist_entry(dlm->lockres_hash[i].first,
+		while (!hlist_empty(dlm_lockres_hash(dlm, i))) {
+			res = hlist_entry(dlm_lockres_hash(dlm, i)->first,
 					  struct dlm_lock_resource, hash_node);
 			/* need reference when manually grabbing lockres */
 			dlm_lockres_get(res);
@@ -1126,6 +1160,13 @@ static int dlm_join_domain(struct dlm_ct
 		goto bail;
 	}
 
+	dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
+	if (!dlm->dlm_worker) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
 	do {
 		unsigned int backoff;
 		status = dlm_try_to_join_domain(dlm);
@@ -1166,6 +1207,7 @@ bail:
 		dlm_unregister_domain_handlers(dlm);
 		dlm_complete_thread(dlm);
 		dlm_complete_recovery_thread(dlm);
+		dlm_destroy_dlm_worker(dlm);
 	}
 
 	return status;
@@ -1191,7 +1233,7 @@ static struct dlm_ctxt *dlm_alloc_ctxt(c
 		goto leave;
 	}
 
-	dlm->lockres_hash = (struct hlist_head *) __get_free_page(GFP_KERNEL);
+	dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
 	if (!dlm->lockres_hash) {
 		mlog_errno(-ENOMEM);
 		kfree(dlm->name);
@@ -1200,8 +1242,8 @@ static struct dlm_ctxt *dlm_alloc_ctxt(c
 		goto leave;
 	}
 
-	for (i=0; i<DLM_HASH_BUCKETS; i++)
-		INIT_HLIST_HEAD(&dlm->lockres_hash[i]);
+	for (i = 0; i < DLM_HASH_BUCKETS; i++)
+		INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
 
 	strcpy(dlm->name, domain);
 	dlm->key = key;
@@ -1231,6 +1273,7 @@ static struct dlm_ctxt *dlm_alloc_ctxt(c
 
 	dlm->dlm_thread_task = NULL;
 	dlm->dlm_reco_thread_task = NULL;
+	dlm->dlm_worker = NULL;
 	init_waitqueue_head(&dlm->dlm_thread_wq);
 	init_waitqueue_head(&dlm->dlm_reco_thread_wq);
 	init_waitqueue_head(&dlm->reco.event);
diff --git a/fs/ocfs2/dlm/dlmfs.c b/fs/ocfs2/dlm/dlmfs.c
index 7e88e24..bb2c202 100644
--- a/fs/ocfs2/dlm/dlmfs.c
+++ b/fs/ocfs2/dlm/dlmfs.c
@@ -116,7 +116,7 @@ static int dlmfs_file_open(struct inode 
 	 * doesn't make sense for LVB writes. */
 	file->f_flags &= ~O_APPEND;
 
-	fp = kmalloc(sizeof(*fp), GFP_KERNEL);
+	fp = kmalloc(sizeof(*fp), GFP_NOFS);
 	if (!fp) {
 		status = -ENOMEM;
 		goto bail;
@@ -196,7 +196,7 @@ static ssize_t dlmfs_file_read(struct fi
 	else
 		readlen = count - *ppos;
 
-	lvb_buf = kmalloc(readlen, GFP_KERNEL);
+	lvb_buf = kmalloc(readlen, GFP_NOFS);
 	if (!lvb_buf)
 		return -ENOMEM;
 
@@ -240,7 +240,7 @@ static ssize_t dlmfs_file_write(struct f
 	else
 		writelen = count - *ppos;
 
-	lvb_buf = kmalloc(writelen, GFP_KERNEL);
+	lvb_buf = kmalloc(writelen, GFP_NOFS);
 	if (!lvb_buf)
 		return -ENOMEM;
 
diff --git a/fs/ocfs2/dlm/dlmlock.c b/fs/ocfs2/dlm/dlmlock.c
index 6fea283..a9fc9a6 100644
--- a/fs/ocfs2/dlm/dlmlock.c
+++ b/fs/ocfs2/dlm/dlmlock.c
@@ -201,6 +201,7 @@ static enum dlm_status dlmlock_remote(st
 				      struct dlm_lock *lock, int flags)
 {
 	enum dlm_status status = DLM_DENIED;
+	int lockres_changed = 1;
 
 	mlog_entry("type=%d\n", lock->ml.type);
 	mlog(0, "lockres %.*s, flags = 0x%x\n", res->lockname.len,
@@ -226,8 +227,25 @@ static enum dlm_status dlmlock_remote(st
 	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
 	lock->lock_pending = 0;
 	if (status != DLM_NORMAL) {
-		if (status != DLM_NOTQUEUED)
+		if (status == DLM_RECOVERING &&
+		    dlm_is_recovery_lock(res->lockname.name,
+					 res->lockname.len)) {
+			/* recovery lock was mastered by dead node.
+			 * we need to have calc_usage shoot down this
+			 * lockres and completely remaster it. */
+			mlog(0, "%s: recovery lock was owned by "
+			     "dead node %u, remaster it now.\n",
+			     dlm->name, res->owner);
+		} else if (status != DLM_NOTQUEUED) {
+			/*
+			 * DO NOT call calc_usage, as this would unhash
+			 * the remote lockres before we ever get to use
+			 * it.  treat as if we never made any change to
+			 * the lockres.
+			 */
+			lockres_changed = 0;
 			dlm_error(status);
+		}
 		dlm_revert_pending_lock(res, lock);
 		dlm_lock_put(lock);
 	} else if (dlm_is_recovery_lock(res->lockname.name, 
@@ -244,7 +262,8 @@ static enum dlm_status dlmlock_remote(st
 	}
 	spin_unlock(&res->spinlock);
 
-	dlm_lockres_calc_usage(dlm, res);
+	if (lockres_changed)
+		dlm_lockres_calc_usage(dlm, res);
 
 	wake_up(&res->wq);
 	return status;
@@ -281,6 +300,14 @@ static enum dlm_status dlm_send_remote_l
 	if (tmpret >= 0) {
 		// successfully sent and received
 		ret = status;  // this is already a dlm_status
+		if (ret == DLM_REJECTED) {
+			mlog(ML_ERROR, "%s:%.*s: BUG.  this is a stale lockres "
+			     "no longer owned by %u.  that node is coming back "
+			     "up currently.\n", dlm->name, create.namelen,
+			     create.name, res->owner);
+			dlm_print_one_lock_resource(res);
+			BUG();
+		}
 	} else {
 		mlog_errno(tmpret);
 		if (dlm_is_host_down(tmpret)) {
@@ -382,13 +409,13 @@ struct dlm_lock * dlm_new_lock(int type,
 	struct dlm_lock *lock;
 	int kernel_allocated = 0;
 
-	lock = kcalloc(1, sizeof(*lock), GFP_KERNEL);
+	lock = kcalloc(1, sizeof(*lock), GFP_NOFS);
 	if (!lock)
 		return NULL;
 
 	if (!lksb) {
 		/* zero memory only if kernel-allocated */
-		lksb = kcalloc(1, sizeof(*lksb), GFP_KERNEL);
+		lksb = kcalloc(1, sizeof(*lksb), GFP_NOFS);
 		if (!lksb) {
 			kfree(lock);
 			return NULL;
@@ -429,11 +456,16 @@ int dlm_create_lock_handler(struct o2net
 	if (!dlm_grab(dlm))
 		return DLM_REJECTED;
 
-	mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
-			"Domain %s not fully joined!\n", dlm->name);
-
 	name = create->name;
 	namelen = create->namelen;
+	status = DLM_REJECTED;
+	if (!dlm_domain_fully_joined(dlm)) {
+		mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
+		     "sending a create_lock message for lock %.*s!\n",
+		     dlm->name, create->node_idx, namelen, name);
+		dlm_error(status);
+		goto leave;
+	}
 
 	status = DLM_IVBUFLEN;
 	if (namelen > DLM_LOCKID_NAME_MAX) {
@@ -669,18 +701,22 @@ retry_lock:
 			msleep(100);
 			/* no waiting for dlm_reco_thread */
 			if (recovery) {
-				if (status == DLM_RECOVERING) {
-					mlog(0, "%s: got RECOVERING "
-					     "for $REOCVERY lock, master "
-					     "was %u\n", dlm->name, 
-					     res->owner);
-					dlm_wait_for_node_death(dlm, res->owner, 
-							DLM_NODE_DEATH_WAIT_MAX);
-				}
+				if (status != DLM_RECOVERING)
+					goto retry_lock;
+
+				mlog(0, "%s: got RECOVERING "
+				     "for $RECOVERY lock, master "
+				     "was %u\n", dlm->name,
+				     res->owner);
+				/* wait to see the node go down, then
+				 * drop down and allow the lockres to
+				 * get cleaned up.  need to remaster. */
+				dlm_wait_for_node_death(dlm, res->owner,
+						DLM_NODE_DEATH_WAIT_MAX);
 			} else {
 				dlm_wait_for_recovery(dlm);
+				goto retry_lock;
 			}
-			goto retry_lock;
 		}
 
 		if (status != DLM_NORMAL) {
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index 940be4c..1b8346d 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -47,7 +47,6 @@ #include "cluster/tcp.h"
 
 #include "dlmapi.h"
 #include "dlmcommon.h"
-#include "dlmdebug.h"
 #include "dlmdomain.h"
 
 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_MASTER)
@@ -74,6 +73,7 @@ struct dlm_master_list_entry
 	wait_queue_head_t wq;
 	atomic_t woken;
 	struct kref mle_refs;
+	int inuse;
 	unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
 	unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
 	unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
@@ -127,18 +127,30 @@ static inline int dlm_mle_equal(struct d
 	return 1;
 }
 
-#if 0
-/* Code here is included but defined out as it aids debugging */
+#define dlm_print_nodemap(m)  _dlm_print_nodemap(m,#m)
+static void _dlm_print_nodemap(unsigned long *map, const char *mapname)
+{
+	int i;
+	printk("%s=[ ", mapname);
+	for (i=0; i<O2NM_MAX_NODES; i++)
+		if (test_bit(i, map))
+			printk("%d ", i);
+	printk("]");
+}
 
-void dlm_print_one_mle(struct dlm_master_list_entry *mle)
+static void dlm_print_one_mle(struct dlm_master_list_entry *mle)
 {
-	int i = 0, refs;
+	int refs;
 	char *type;
 	char attached;
 	u8 master;
 	unsigned int namelen;
 	const char *name;
 	struct kref *k;
+	unsigned long *maybe = mle->maybe_map,
+		      *vote = mle->vote_map,
+		      *resp = mle->response_map,
+		      *node = mle->node_map;
 
 	k = &mle->mle_refs;
 	if (mle->type == DLM_MLE_BLOCK)
@@ -159,18 +171,29 @@ void dlm_print_one_mle(struct dlm_master
 		name = mle->u.res->lockname.name;
 	}
 
-	mlog(ML_NOTICE, "  #%3d: %3s  %3d  %3u   %3u %c    (%d)%.*s\n",
-		  i, type, refs, master, mle->new_master, attached,
-		  namelen, namelen, name);
+	mlog(ML_NOTICE, "%.*s: %3s refs=%3d mas=%3u new=%3u evt=%c inuse=%d ",
+		  namelen, name, type, refs, master, mle->new_master, attached,
+		  mle->inuse);
+	dlm_print_nodemap(maybe);
+	printk(", ");
+	dlm_print_nodemap(vote);
+	printk(", ");
+	dlm_print_nodemap(resp);
+	printk(", ");
+	dlm_print_nodemap(node);
+	printk(", ");
+	printk("\n");
 }
 
+#if 0
+/* Code here is included but defined out as it aids debugging */
+
 static void dlm_dump_mles(struct dlm_ctxt *dlm)
 {
 	struct dlm_master_list_entry *mle;
 	struct list_head *iter;
 	
 	mlog(ML_NOTICE, "dumping all mles for domain %s:\n", dlm->name);
-	mlog(ML_NOTICE, "  ####: type refs owner new events? lockname nodemap votemap respmap maybemap\n");
 	spin_lock(&dlm->master_lock);
 	list_for_each(iter, &dlm->master_list) {
 		mle = list_entry(iter, struct dlm_master_list_entry, list);
@@ -314,6 +337,31 @@ static inline void dlm_mle_detach_hb_eve
 	spin_unlock(&dlm->spinlock);
 }
 
+static void dlm_get_mle_inuse(struct dlm_master_list_entry *mle)
+{
+	struct dlm_ctxt *dlm;
+	dlm = mle->dlm;
+
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&dlm->master_lock);
+	mle->inuse++;
+	kref_get(&mle->mle_refs);
+}
+
+static void dlm_put_mle_inuse(struct dlm_master_list_entry *mle)
+{
+	struct dlm_ctxt *dlm;
+	dlm = mle->dlm;
+
+	spin_lock(&dlm->spinlock);
+	spin_lock(&dlm->master_lock);
+	mle->inuse--;
+	__dlm_put_mle(mle);
+	spin_unlock(&dlm->master_lock);
+	spin_unlock(&dlm->spinlock);
+
+}
+
 /* remove from list and free */
 static void __dlm_put_mle(struct dlm_master_list_entry *mle)
 {
@@ -322,9 +370,14 @@ static void __dlm_put_mle(struct dlm_mas
 
 	assert_spin_locked(&dlm->spinlock);
 	assert_spin_locked(&dlm->master_lock);
-	BUG_ON(!atomic_read(&mle->mle_refs.refcount));
-
-	kref_put(&mle->mle_refs, dlm_mle_release);
+	if (!atomic_read(&mle->mle_refs.refcount)) {
+		/* this may or may not crash, but who cares.
+		 * it's a BUG. */
+		mlog(ML_ERROR, "bad mle: %p\n", mle);
+		dlm_print_one_mle(mle);
+		BUG();
+	} else
+		kref_put(&mle->mle_refs, dlm_mle_release);
 }
 
 
@@ -367,6 +420,7 @@ static void dlm_init_mle(struct dlm_mast
 	memset(mle->response_map, 0, sizeof(mle->response_map));
 	mle->master = O2NM_MAX_NODES;
 	mle->new_master = O2NM_MAX_NODES;
+	mle->inuse = 0;
 
 	if (mle->type == DLM_MLE_MASTER) {
 		BUG_ON(!res);
@@ -564,6 +618,28 @@ static void dlm_lockres_release(struct k
 	mlog(0, "destroying lockres %.*s\n", res->lockname.len,
 	     res->lockname.name);
 
+	if (!hlist_unhashed(&res->hash_node) ||
+	    !list_empty(&res->granted) ||
+	    !list_empty(&res->converting) ||
+	    !list_empty(&res->blocked) ||
+	    !list_empty(&res->dirty) ||
+	    !list_empty(&res->recovering) ||
+	    !list_empty(&res->purge)) {
+		mlog(ML_ERROR,
+		     "Going to BUG for resource %.*s."
+		     "  We're on a list! [%c%c%c%c%c%c%c]\n",
+		     res->lockname.len, res->lockname.name,
+		     !hlist_unhashed(&res->hash_node) ? 'H' : ' ',
+		     !list_empty(&res->granted) ? 'G' : ' ',
+		     !list_empty(&res->converting) ? 'C' : ' ',
+		     !list_empty(&res->blocked) ? 'B' : ' ',
+		     !list_empty(&res->dirty) ? 'D' : ' ',
+		     !list_empty(&res->recovering) ? 'R' : ' ',
+		     !list_empty(&res->purge) ? 'P' : ' ');
+
+		dlm_print_one_lock_resource(res);
+	}
+
 	/* By the time we're ready to blow this guy away, we shouldn't
 	 * be on any lists. */
 	BUG_ON(!hlist_unhashed(&res->hash_node));
@@ -579,11 +655,6 @@ static void dlm_lockres_release(struct k
 	kfree(res);
 }
 
-void dlm_lockres_get(struct dlm_lock_resource *res)
-{
-	kref_get(&res->refs);
-}
-
 void dlm_lockres_put(struct dlm_lock_resource *res)
 {
 	kref_put(&res->refs, dlm_lockres_release);
@@ -603,7 +674,7 @@ static void dlm_init_lockres(struct dlm_
 	memcpy(qname, name, namelen);
 
 	res->lockname.len = namelen;
-	res->lockname.hash = full_name_hash(name, namelen);
+	res->lockname.hash = dlm_lockid_hash(name, namelen);
 
 	init_waitqueue_head(&res->wq);
 	spin_lock_init(&res->spinlock);
@@ -637,11 +708,11 @@ struct dlm_lock_resource *dlm_new_lockre
 {
 	struct dlm_lock_resource *res;
 
-	res = kmalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL);
+	res = kmalloc(sizeof(struct dlm_lock_resource), GFP_NOFS);
 	if (!res)
 		return NULL;
 
-	res->lockname.name = kmalloc(namelen, GFP_KERNEL);
+	res->lockname.name = kmalloc(namelen, GFP_NOFS);
 	if (!res->lockname.name) {
 		kfree(res);
 		return NULL;
@@ -677,19 +748,20 @@ struct dlm_lock_resource * dlm_get_lock_
 	int blocked = 0;
 	int ret, nodenum;
 	struct dlm_node_iter iter;
-	unsigned int namelen;
+	unsigned int namelen, hash;
 	int tries = 0;
 	int bit, wait_on_recovery = 0;
 
 	BUG_ON(!lockid);
 
 	namelen = strlen(lockid);
+	hash = dlm_lockid_hash(lockid, namelen);
 
 	mlog(0, "get lockres %s (len %d)\n", lockid, namelen);
 
 lookup:
 	spin_lock(&dlm->spinlock);
-	tmpres = __dlm_lookup_lockres(dlm, lockid, namelen);
+	tmpres = __dlm_lookup_lockres(dlm, lockid, namelen, hash);
 	if (tmpres) {
 		spin_unlock(&dlm->spinlock);
 		mlog(0, "found in hash!\n");
@@ -704,7 +776,7 @@ lookup:
 		mlog(0, "allocating a new resource\n");
 		/* nothing found and we need to allocate one. */
 		alloc_mle = (struct dlm_master_list_entry *)
-			kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL);
+			kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
 		if (!alloc_mle)
 			goto leave;
 		res = dlm_new_lockres(dlm, lockid, namelen);
@@ -790,10 +862,11 @@ lookup:
 	 * if so, the creator of the BLOCK may try to put the last
 	 * ref at this time in the assert master handler, so we
 	 * need an extra one to keep from a bad ptr deref. */
-	dlm_get_mle(mle);
+	dlm_get_mle_inuse(mle);
 	spin_unlock(&dlm->master_lock);
 	spin_unlock(&dlm->spinlock);
 
+redo_request:
 	while (wait_on_recovery) {
 		/* any cluster changes that occurred after dropping the
 		 * dlm spinlock would be detectable be a change on the mle,
@@ -812,7 +885,7 @@ lookup:
 		} 
 
 		dlm_kick_recovery_thread(dlm);
-		msleep(100);
+		msleep(1000);
 		dlm_wait_for_recovery(dlm);
 
 		spin_lock(&dlm->spinlock);
@@ -825,13 +898,15 @@ lookup:
 		} else
 			wait_on_recovery = 0;
 		spin_unlock(&dlm->spinlock);
+
+		if (wait_on_recovery)
+			dlm_wait_for_node_recovery(dlm, bit, 10000);
 	}
 
 	/* must wait for lock to be mastered elsewhere */
 	if (blocked)
 		goto wait;
 
-redo_request:
 	ret = -EINVAL;
 	dlm_node_iter_init(mle->vote_map, &iter);
 	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
@@ -856,6 +931,7 @@ wait:
 	/* keep going until the response map includes all nodes */
 	ret = dlm_wait_for_lock_mastery(dlm, res, mle, &blocked);
 	if (ret < 0) {
+		wait_on_recovery = 1;
 		mlog(0, "%s:%.*s: node map changed, redo the "
 		     "master request now, blocked=%d\n",
 		     dlm->name, res->lockname.len,
@@ -866,7 +942,7 @@ wait:
 			     dlm->name, res->lockname.len, 
 			     res->lockname.name, blocked);
 			dlm_print_one_lock_resource(res);
-			/* dlm_print_one_mle(mle); */
+			dlm_print_one_mle(mle);
 			tries = 0;
 		}
 		goto redo_request;
@@ -880,7 +956,7 @@ wait:
 	dlm_mle_detach_hb_events(dlm, mle);
 	dlm_put_mle(mle);
 	/* put the extra ref */
-	dlm_put_mle(mle);
+	dlm_put_mle_inuse(mle);
 
 wake_waiters:
 	spin_lock(&res->spinlock);
@@ -921,12 +997,14 @@ recheck:
 		spin_unlock(&res->spinlock);
 		/* this will cause the master to re-assert across
 		 * the whole cluster, freeing up mles */
-		ret = dlm_do_master_request(mle, res->owner);
-		if (ret < 0) {
-			/* give recovery a chance to run */
-			mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret);
-			msleep(500);
-			goto recheck;
+		if (res->owner != dlm->node_num) {
+			ret = dlm_do_master_request(mle, res->owner);
+			if (ret < 0) {
+				/* give recovery a chance to run */
+				mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret);
+				msleep(500);
+				goto recheck;
+			}
 		}
 		ret = 0;
 		goto leave;
@@ -962,6 +1040,12 @@ recheck:
 		     "rechecking now\n", dlm->name, res->lockname.len,
 		     res->lockname.name);
 		goto recheck;
+	} else {
+		if (!voting_done) {
+			mlog(0, "map not changed and voting not done "
+			     "for %s:%.*s\n", dlm->name, res->lockname.len,
+			     res->lockname.name);
+		}
 	}
 
 	if (m != O2NM_MAX_NODES) {
@@ -1129,18 +1213,6 @@ static int dlm_restart_lock_mastery(stru
 			set_bit(node, mle->vote_map);
 		} else {
 			mlog(ML_ERROR, "node down! %d\n", node);
-
-			/* if the node wasn't involved in mastery skip it,
-			 * but clear it out from the maps so that it will
-			 * not affect mastery of this lockres */
-			clear_bit(node, mle->response_map);
-			clear_bit(node, mle->vote_map);
-			if (!test_bit(node, mle->maybe_map))
-				goto next;
-
-			/* if we're already blocked on lock mastery, and the
-			 * dead node wasn't the expected master, or there is
-			 * another node in the maybe_map, keep waiting */
 			if (blocked) {
 				int lowest = find_next_bit(mle->maybe_map,
 						       O2NM_MAX_NODES, 0);
@@ -1148,54 +1220,53 @@ static int dlm_restart_lock_mastery(stru
 				/* act like it was never there */
 				clear_bit(node, mle->maybe_map);
 
-			       	if (node != lowest)
-					goto next;
-
-				mlog(ML_ERROR, "expected master %u died while "
-				     "this node was blocked waiting on it!\n",
-				     node);
-				lowest = find_next_bit(mle->maybe_map,
-						       O2NM_MAX_NODES,
-						       lowest+1);
-				if (lowest < O2NM_MAX_NODES) {
-					mlog(0, "still blocked. waiting "
-					     "on %u now\n", lowest);
-					goto next;
+			       	if (node == lowest) {
+					mlog(0, "expected master %u died"
+					    " while this node was blocked "
+					    "waiting on it!\n", node);
+					lowest = find_next_bit(mle->maybe_map,
+						       	O2NM_MAX_NODES,
+						       	lowest+1);
+					if (lowest < O2NM_MAX_NODES) {
+						mlog(0, "%s:%.*s:still "
+						     "blocked. waiting on %u "
+						     "now\n", dlm->name,
+						     res->lockname.len,
+						     res->lockname.name,
+						     lowest);
+					} else {
+						/* mle is an MLE_BLOCK, but
+						 * there is now nothing left to
+						 * block on.  we need to return
+						 * all the way back out and try
+						 * again with an MLE_MASTER.
+						 * dlm_do_local_recovery_cleanup
+						 * has already run, so the mle
+						 * refcount is ok */
+						mlog(0, "%s:%.*s: no "
+						     "longer blocking. try to "
+						     "master this here\n",
+						     dlm->name,
+						     res->lockname.len,
+						     res->lockname.name);
+						mle->type = DLM_MLE_MASTER;
+						mle->u.res = res;
+					}
 				}
-
-				/* mle is an MLE_BLOCK, but there is now
-				 * nothing left to block on.  we need to return
-				 * all the way back out and try again with
-				 * an MLE_MASTER. dlm_do_local_recovery_cleanup
-				 * has already run, so the mle refcount is ok */
-				mlog(0, "no longer blocking. we can "
-				     "try to master this here\n");
-				mle->type = DLM_MLE_MASTER;
-				memset(mle->maybe_map, 0,
-				       sizeof(mle->maybe_map));
-				memset(mle->response_map, 0,
-				       sizeof(mle->maybe_map));
-				memcpy(mle->vote_map, mle->node_map,
-				       sizeof(mle->node_map));
-				mle->u.res = res;
-				set_bit(dlm->node_num, mle->maybe_map);
-
-				ret = -EAGAIN;
-				goto next;
 			}
 
-			clear_bit(node, mle->maybe_map);
-			if (node > dlm->node_num)
-				goto next;
-
-			mlog(0, "dead node in map!\n");
-			/* yuck. go back and re-contact all nodes
-			 * in the vote_map, removing this node. */
-			memset(mle->response_map, 0,
-			       sizeof(mle->response_map));
+			/* now blank out everything, as if we had never
+			 * contacted anyone */
+			memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
+			memset(mle->response_map, 0, sizeof(mle->response_map));
+			/* reset the vote_map to the current node_map */
+			memcpy(mle->vote_map, mle->node_map,
+			       sizeof(mle->node_map));
+			/* put myself into the maybe map */
+			if (mle->type != DLM_MLE_BLOCK)
+				set_bit(dlm->node_num, mle->maybe_map);
 		}
 		ret = -EAGAIN;
-next:
 		node = dlm_bitmap_diff_iter_next(&bdi, &sc);
 	}
 	return ret;
@@ -1316,7 +1387,7 @@ int dlm_master_request_handler(struct o2
 	struct dlm_master_request *request = (struct dlm_master_request *) msg->buf;
 	struct dlm_master_list_entry *mle = NULL, *tmpmle = NULL;
 	char *name;
-	unsigned int namelen;
+	unsigned int namelen, hash;
 	int found, ret;
 	int set_maybe;
 	int dispatch_assert = 0;
@@ -1331,6 +1402,7 @@ int dlm_master_request_handler(struct o2
 
 	name = request->name;
 	namelen = request->namelen;
+	hash = dlm_lockid_hash(name, namelen);
 
 	if (namelen > DLM_LOCKID_NAME_MAX) {
 		response = DLM_IVBUFLEN;
@@ -1339,7 +1411,7 @@ int dlm_master_request_handler(struct o2
 
 way_up_top:
 	spin_lock(&dlm->spinlock);
-	res = __dlm_lookup_lockres(dlm, name, namelen);
+	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
 	if (res) {
 		spin_unlock(&dlm->spinlock);
 
@@ -1459,21 +1531,18 @@ way_up_top:
 			spin_unlock(&dlm->spinlock);
 
 			mle = (struct dlm_master_list_entry *)
-				kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL);
+				kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
 			if (!mle) {
 				response = DLM_MASTER_RESP_ERROR;
 				mlog_errno(-ENOMEM);
 				goto send_response;
 			}
-			spin_lock(&dlm->spinlock);
-			dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL,
-					 name, namelen);
-			spin_unlock(&dlm->spinlock);
 			goto way_up_top;
 		}
 
 		// mlog(0, "this is second time thru, already allocated, "
 		// "add the block.\n");
+		dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL, name, namelen);
 		set_bit(request->node_idx, mle->maybe_map);
 		list_add(&mle->list, &dlm->master_list);
 		response = DLM_MASTER_RESP_NO;
@@ -1556,6 +1625,8 @@ again:
 	dlm_node_iter_init(nodemap, &iter);
 	while ((to = dlm_node_iter_next(&iter)) >= 0) {
 		int r = 0;
+		struct dlm_master_list_entry *mle = NULL;
+
 		mlog(0, "sending assert master to %d (%.*s)\n", to,
 		     namelen, lockname);
 		memset(&assert, 0, sizeof(assert));
@@ -1567,20 +1638,28 @@ again:
 		tmpret = o2net_send_message(DLM_ASSERT_MASTER_MSG, dlm->key,
 					    &assert, sizeof(assert), to, &r);
 		if (tmpret < 0) {
-			mlog(ML_ERROR, "assert_master returned %d!\n", tmpret);
+			mlog(0, "assert_master returned %d!\n", tmpret);
 			if (!dlm_is_host_down(tmpret)) {
-				mlog(ML_ERROR, "unhandled error!\n");
+				mlog(ML_ERROR, "unhandled error=%d!\n", tmpret);
 				BUG();
 			}
 			/* a node died.  finish out the rest of the nodes. */
-			mlog(ML_ERROR, "link to %d went down!\n", to);
+			mlog(0, "link to %d went down!\n", to);
 			/* any nonzero status return will do */
 			ret = tmpret;
 		} else if (r < 0) {
 			/* ok, something horribly messed.  kill thyself. */
 			mlog(ML_ERROR,"during assert master of %.*s to %u, "
 			     "got %d.\n", namelen, lockname, to, r);
-			dlm_dump_lock_resources(dlm);
+			spin_lock(&dlm->spinlock);
+			spin_lock(&dlm->master_lock);
+			if (dlm_find_mle(dlm, &mle, (char *)lockname,
+					 namelen)) {
+				dlm_print_one_mle(mle);
+				__dlm_put_mle(mle);
+			}
+			spin_unlock(&dlm->master_lock);
+			spin_unlock(&dlm->spinlock);
 			BUG();
 		} else if (r == EAGAIN) {
 			mlog(0, "%.*s: node %u create mles on other "
@@ -1612,7 +1691,7 @@ int dlm_assert_master_handler(struct o2n
 	struct dlm_assert_master *assert = (struct dlm_assert_master *)msg->buf;
 	struct dlm_lock_resource *res = NULL;
 	char *name;
-	unsigned int namelen;
+	unsigned int namelen, hash;
 	u32 flags;
 	int master_request = 0;
 	int ret = 0;
@@ -1622,6 +1701,7 @@ int dlm_assert_master_handler(struct o2n
 
 	name = assert->name;
 	namelen = assert->namelen;
+	hash = dlm_lockid_hash(name, namelen);
 	flags = be32_to_cpu(assert->flags);
 
 	if (namelen > DLM_LOCKID_NAME_MAX) {
@@ -1646,7 +1726,7 @@ int dlm_assert_master_handler(struct o2n
 		if (bit >= O2NM_MAX_NODES) {
 			/* not necessarily an error, though less likely.
 			 * could be master just re-asserting. */
-			mlog(ML_ERROR, "no bits set in the maybe_map, but %u "
+			mlog(0, "no bits set in the maybe_map, but %u "
 			     "is asserting! (%.*s)\n", assert->node_idx,
 			     namelen, name);
 		} else if (bit != assert->node_idx) {
@@ -1658,19 +1738,36 @@ int dlm_assert_master_handler(struct o2n
 				 * number winning the mastery will respond
 				 * YES to mastery requests, but this node
 				 * had no way of knowing.  let it pass. */
-				mlog(ML_ERROR, "%u is the lowest node, "
+				mlog(0, "%u is the lowest node, "
 				     "%u is asserting. (%.*s)  %u must "
 				     "have begun after %u won.\n", bit,
 				     assert->node_idx, namelen, name, bit,
 				     assert->node_idx);
 			}
 		}
+		if (mle->type == DLM_MLE_MIGRATION) {
+			if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
+				mlog(0, "%s:%.*s: got cleanup assert"
+				     " from %u for migration\n",
+				     dlm->name, namelen, name,
+				     assert->node_idx);
+			} else if (!(flags & DLM_ASSERT_MASTER_FINISH_MIGRATION)) {
+				mlog(0, "%s:%.*s: got unrelated assert"
+				     " from %u for migration, ignoring\n",
+				     dlm->name, namelen, name,
+				     assert->node_idx);
+				__dlm_put_mle(mle);
+				spin_unlock(&dlm->master_lock);
+				spin_unlock(&dlm->spinlock);
+				goto done;
+			}	
+		}
 	}
 	spin_unlock(&dlm->master_lock);
 
 	/* ok everything checks out with the MLE
 	 * now check to see if there is a lockres */
-	res = __dlm_lookup_lockres(dlm, name, namelen);
+	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
 	if (res) {
 		spin_lock(&res->spinlock);
 		if (res->state & DLM_LOCK_RES_RECOVERING)  {
@@ -1679,7 +1776,8 @@ int dlm_assert_master_handler(struct o2n
 			goto kill;
 		}
 		if (!mle) {
-			if (res->owner != assert->node_idx) {
+			if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN &&
+			    res->owner != assert->node_idx) {
 				mlog(ML_ERROR, "assert_master from "
 					  "%u, but current owner is "
 					  "%u! (%.*s)\n",
@@ -1732,6 +1830,7 @@ ok:
 	if (mle) {
 		int extra_ref = 0;
 		int nn = -1;
+		int rr, err = 0;
 		
 		spin_lock(&mle->spinlock);
 		if (mle->type == DLM_MLE_BLOCK || mle->type == DLM_MLE_MIGRATION)
@@ -1751,27 +1850,64 @@ ok:
 		wake_up(&mle->wq);
 		spin_unlock(&mle->spinlock);
 
-		if (mle->type == DLM_MLE_MIGRATION && res) {
-			mlog(0, "finishing off migration of lockres %.*s, "
-			     "from %u to %u\n",
-			       res->lockname.len, res->lockname.name,
-			       dlm->node_num, mle->new_master);
+		if (res) {
 			spin_lock(&res->spinlock);
-			res->state &= ~DLM_LOCK_RES_MIGRATING;
-			dlm_change_lockres_owner(dlm, res, mle->new_master);
-			BUG_ON(res->state & DLM_LOCK_RES_DIRTY);
+			if (mle->type == DLM_MLE_MIGRATION) {
+				mlog(0, "finishing off migration of lockres %.*s, "
+			     		"from %u to %u\n",
+			       		res->lockname.len, res->lockname.name,
+			       		dlm->node_num, mle->new_master);
+				res->state &= ~DLM_LOCK_RES_MIGRATING;
+				dlm_change_lockres_owner(dlm, res, mle->new_master);
+				BUG_ON(res->state & DLM_LOCK_RES_DIRTY);
+			} else {
+				dlm_change_lockres_owner(dlm, res, mle->master);
+			}
 			spin_unlock(&res->spinlock);
 		}
-		/* master is known, detach if not already detached */
-		dlm_mle_detach_hb_events(dlm, mle);
-		dlm_put_mle(mle);
-		
+
+		/* master is known, detach if not already detached.
+		 * ensures that only one assert_master call will happen
+		 * on this mle. */
+		spin_lock(&dlm->spinlock);
+		spin_lock(&dlm->master_lock);
+
+		rr = atomic_read(&mle->mle_refs.refcount);
+		if (mle->inuse > 0) {
+			if (extra_ref && rr < 3)
+				err = 1;
+			else if (!extra_ref && rr < 2)
+				err = 1;
+		} else {
+			if (extra_ref && rr < 2)
+				err = 1;
+			else if (!extra_ref && rr < 1)
+				err = 1;
+		}
+		if (err) {
+			mlog(ML_ERROR, "%s:%.*s: got assert master from %u "
+			     "that will mess up this node, refs=%d, extra=%d, "
+			     "inuse=%d\n", dlm->name, namelen, name,
+			     assert->node_idx, rr, extra_ref, mle->inuse);
+			dlm_print_one_mle(mle);
+		}
+		list_del_init(&mle->list);
+		__dlm_mle_detach_hb_events(dlm, mle);
+		__dlm_put_mle(mle);
 		if (extra_ref) {
 			/* the assert master message now balances the extra
 		 	 * ref given by the master / migration request message.
 		 	 * if this is the last put, it will be removed
 		 	 * from the list. */
-			dlm_put_mle(mle);
+			__dlm_put_mle(mle);
+		}
+		spin_unlock(&dlm->master_lock);
+		spin_unlock(&dlm->spinlock);
+	} else if (res) {
+		if (res->owner != assert->node_idx) {
+			mlog(0, "assert_master from %u, but current "
+			     "owner is %u (%.*s), no mle\n", assert->node_idx,
+			     res->owner, namelen, name);
 		}
 	}
 
@@ -1788,12 +1924,12 @@ done:
 
 kill:
 	/* kill the caller! */
+	mlog(ML_ERROR, "Bad message received from another node.  Dumping state "
+	     "and killing the other node now!  This node is OK and can continue.\n");
+	__dlm_print_one_lock_resource(res);
 	spin_unlock(&res->spinlock);
 	spin_unlock(&dlm->spinlock);
 	dlm_lockres_put(res);
-	mlog(ML_ERROR, "Bad message received from another node.  Dumping state "
-	     "and killing the other node now!  This node is OK and can continue.\n");
-	dlm_dump_lock_resources(dlm);
 	dlm_put(dlm);
 	return -EINVAL;
 }
@@ -1803,7 +1939,7 @@ int dlm_dispatch_assert_master(struct dl
 			       int ignore_higher, u8 request_from, u32 flags)
 {
 	struct dlm_work_item *item;
-	item = kcalloc(1, sizeof(*item), GFP_KERNEL);
+	item = kcalloc(1, sizeof(*item), GFP_NOFS);
 	if (!item)
 		return -ENOMEM;
 
@@ -1825,7 +1961,7 @@ int dlm_dispatch_assert_master(struct dl
 	list_add_tail(&item->list, &dlm->work_list);
 	spin_unlock(&dlm->work_lock);
 
-	schedule_work(&dlm->dispatched_work);
+	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
 	return 0;
 }
 
@@ -1866,6 +2002,23 @@ static void dlm_assert_master_worker(str
 		}
 	}
 
+	/*
+	 * If we're migrating this lock to someone else, we are no
+	 * longer allowed to assert out own mastery.  OTOH, we need to
+	 * prevent migration from starting while we're still asserting
+	 * our dominance.  The reserved ast delays migration.
+	 */
+	spin_lock(&res->spinlock);
+	if (res->state & DLM_LOCK_RES_MIGRATING) {
+		mlog(0, "Someone asked us to assert mastery, but we're "
+		     "in the middle of migration.  Skipping assert, "
+		     "the new master will handle that.\n");
+		spin_unlock(&res->spinlock);
+		goto put;
+	} else
+		__dlm_lockres_reserve_ast(res);
+	spin_unlock(&res->spinlock);
+
 	/* this call now finishes out the nodemap
 	 * even if one or more nodes die */
 	mlog(0, "worker about to master %.*s here, this=%u\n",
@@ -1875,9 +2028,14 @@ static void dlm_assert_master_worker(str
 				   nodemap, flags);
 	if (ret < 0) {
 		/* no need to restart, we are done */
-		mlog_errno(ret);
+		if (!dlm_is_host_down(ret))
+			mlog_errno(ret);
 	}
 
+	/* Ok, we've asserted ourselves.  Let's let migration start. */
+	dlm_lockres_release_ast(dlm, res);
+
+put:
 	dlm_lockres_put(res);
 
 	mlog(0, "finished with dlm_assert_master_worker\n");
@@ -1916,6 +2074,7 @@ static int dlm_pre_master_reco_lockres(s
 				BUG();
 			/* host is down, so answer for that node would be
 			 * DLM_LOCK_RES_OWNER_UNKNOWN.  continue. */
+			ret = 0;
 		}
 
 		if (master != DLM_LOCK_RES_OWNER_UNKNOWN) {
@@ -2016,14 +2175,14 @@ int dlm_migrate_lockres(struct dlm_ctxt 
 	 */
 
 	ret = -ENOMEM;
-	mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_KERNEL);
+	mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_NOFS);
 	if (!mres) {
 		mlog_errno(ret);
 		goto leave;
 	}
 
 	mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache,
-								GFP_KERNEL);
+								GFP_NOFS);
 	if (!mle) {
 		mlog_errno(ret);
 		goto leave;
@@ -2117,7 +2276,7 @@ fail:
 	 * take both dlm->spinlock and dlm->master_lock */
 	spin_lock(&dlm->spinlock);
 	spin_lock(&dlm->master_lock);
-	dlm_get_mle(mle);
+	dlm_get_mle_inuse(mle);
 	spin_unlock(&dlm->master_lock);
 	spin_unlock(&dlm->spinlock);
 
@@ -2134,7 +2293,10 @@ fail:
 		/* migration failed, detach and clean up mle */
 		dlm_mle_detach_hb_events(dlm, mle);
 		dlm_put_mle(mle);
-		dlm_put_mle(mle);
+		dlm_put_mle_inuse(mle);
+		spin_lock(&res->spinlock);
+		res->state &= ~DLM_LOCK_RES_MIGRATING;
+		spin_unlock(&res->spinlock);
 		goto leave;
 	}
 
@@ -2164,8 +2326,8 @@ fail:
 			/* avoid hang during shutdown when migrating lockres 
 			 * to a node which also goes down */
 			if (dlm_is_node_dead(dlm, target)) {
-				mlog(0, "%s:%.*s: expected migration target %u "
-				     "is no longer up.  restarting.\n",
+				mlog(0, "%s:%.*s: expected migration "
+				     "target %u is no longer up, restarting\n",
 				     dlm->name, res->lockname.len,
 				     res->lockname.name, target);
 				ret = -ERESTARTSYS;
@@ -2175,7 +2337,10 @@ fail:
 			/* migration failed, detach and clean up mle */
 			dlm_mle_detach_hb_events(dlm, mle);
 			dlm_put_mle(mle);
-			dlm_put_mle(mle);
+			dlm_put_mle_inuse(mle);
+			spin_lock(&res->spinlock);
+			res->state &= ~DLM_LOCK_RES_MIGRATING;
+			spin_unlock(&res->spinlock);
 			goto leave;
 		}
 		/* TODO: if node died: stop, clean up, return error */
@@ -2191,7 +2356,7 @@ fail:
 
 	/* master is known, detach if not already detached */
 	dlm_mle_detach_hb_events(dlm, mle);
-	dlm_put_mle(mle);
+	dlm_put_mle_inuse(mle);
 	ret = 0;
 
 	dlm_lockres_calc_usage(dlm, res);
@@ -2462,7 +2627,7 @@ int dlm_migrate_request_handler(struct o
 	struct dlm_migrate_request *migrate = (struct dlm_migrate_request *) msg->buf;
 	struct dlm_master_list_entry *mle = NULL, *oldmle = NULL;
 	const char *name;
-	unsigned int namelen;
+	unsigned int namelen, hash;
 	int ret = 0;
 
 	if (!dlm_grab(dlm))
@@ -2470,10 +2635,11 @@ int dlm_migrate_request_handler(struct o
 
 	name = migrate->name;
 	namelen = migrate->namelen;
+	hash = dlm_lockid_hash(name, namelen);
 
 	/* preallocate.. if this fails, abort */
 	mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache,
-							 GFP_KERNEL);
+							 GFP_NOFS);
 
 	if (!mle) {
 		ret = -ENOMEM;
@@ -2482,7 +2648,7 @@ int dlm_migrate_request_handler(struct o
 
 	/* check for pre-existing lock */
 	spin_lock(&dlm->spinlock);
-	res = __dlm_lookup_lockres(dlm, name, namelen);
+	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
 	spin_lock(&dlm->master_lock);
 
 	if (res) {
@@ -2580,6 +2746,7 @@ static int dlm_add_migration_mle(struct 
 			/* remove it from the list so that only one
 			 * mle will be found */
 			list_del_init(&tmp->list);
+			__dlm_mle_detach_hb_events(dlm, mle);
 		}
 		spin_unlock(&tmp->spinlock);
 	}
@@ -2601,6 +2768,7 @@ void dlm_clean_master_list(struct dlm_ct
 	struct list_head *iter, *iter2;
 	struct dlm_master_list_entry *mle;
 	struct dlm_lock_resource *res;
+	unsigned int hash;
 
 	mlog_entry("dlm=%s, dead node=%u\n", dlm->name, dead_node);
 top:
@@ -2640,7 +2808,7 @@ top:
 				 * may result in the mle being unlinked and
 				 * freed, but there may still be a process
 				 * waiting in the dlmlock path which is fine. */
-				mlog(ML_ERROR, "node %u was expected master\n",
+				mlog(0, "node %u was expected master\n",
 				     dead_node);
 				atomic_set(&mle->woken, 1);
 				spin_unlock(&mle->spinlock);
@@ -2673,19 +2841,21 @@ top:
 
 		/* remove from the list early.  NOTE: unlinking
 		 * list_head while in list_for_each_safe */
+		__dlm_mle_detach_hb_events(dlm, mle);
 		spin_lock(&mle->spinlock);
 		list_del_init(&mle->list);
 		atomic_set(&mle->woken, 1);
 		spin_unlock(&mle->spinlock);
 		wake_up(&mle->wq);
 
-		mlog(0, "node %u died during migration from "
-		     "%u to %u!\n", dead_node,
+		mlog(0, "%s: node %u died during migration from "
+		     "%u to %u!\n", dlm->name, dead_node,
 		     mle->master, mle->new_master);
 		/* if there is a lockres associated with this
 	 	 * mle, find it and set its owner to UNKNOWN */
+		hash = dlm_lockid_hash(mle->u.name.name, mle->u.name.len);
 		res = __dlm_lookup_lockres(dlm, mle->u.name.name,
-					mle->u.name.len);
+					   mle->u.name.len, hash);
 		if (res) {
 			/* unfortunately if we hit this rare case, our
 		 	 * lock ordering is messed.  we need to drop
diff --git a/fs/ocfs2/dlm/dlmrecovery.c b/fs/ocfs2/dlm/dlmrecovery.c
index 805cbab..b5078fd 100644
--- a/fs/ocfs2/dlm/dlmrecovery.c
+++ b/fs/ocfs2/dlm/dlmrecovery.c
@@ -115,12 +115,37 @@ static u64 dlm_get_next_mig_cookie(void)
 	return c;
 }
 
+static inline void dlm_set_reco_dead_node(struct dlm_ctxt *dlm,
+					  u8 dead_node)
+{
+	assert_spin_locked(&dlm->spinlock);
+	if (dlm->reco.dead_node != dead_node)
+		mlog(0, "%s: changing dead_node from %u to %u\n",
+		     dlm->name, dlm->reco.dead_node, dead_node);
+	dlm->reco.dead_node = dead_node;
+}
+
+static inline void dlm_set_reco_master(struct dlm_ctxt *dlm,
+				       u8 master)
+{
+	assert_spin_locked(&dlm->spinlock);
+	mlog(0, "%s: changing new_master from %u to %u\n",
+	     dlm->name, dlm->reco.new_master, master);
+	dlm->reco.new_master = master;
+}
+
+static inline void __dlm_reset_recovery(struct dlm_ctxt *dlm)
+{
+	assert_spin_locked(&dlm->spinlock);
+	clear_bit(dlm->reco.dead_node, dlm->recovery_map);
+	dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
+	dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM);
+}
+
 static inline void dlm_reset_recovery(struct dlm_ctxt *dlm)
 {
 	spin_lock(&dlm->spinlock);
-	clear_bit(dlm->reco.dead_node, dlm->recovery_map);
-	dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
-	dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
+	__dlm_reset_recovery(dlm);
 	spin_unlock(&dlm->spinlock);
 }
 
@@ -132,12 +157,21 @@ void dlm_dispatch_work(void *data)
 	struct list_head *iter, *iter2;
 	struct dlm_work_item *item;
 	dlm_workfunc_t *workfunc;
+	int tot=0;
+
+	if (!dlm_joined(dlm))
+		return;
 
 	spin_lock(&dlm->work_lock);
 	list_splice_init(&dlm->work_list, &tmp_list);
 	spin_unlock(&dlm->work_lock);
 
 	list_for_each_safe(iter, iter2, &tmp_list) {
+		tot++;
+	}
+	mlog(0, "%s: work thread has %d work items\n", dlm->name, tot);
+
+	list_for_each_safe(iter, iter2, &tmp_list) {
 		item = list_entry(iter, struct dlm_work_item, list);
 		workfunc = item->func;
 		list_del_init(&item->list);
@@ -220,6 +254,52 @@ void dlm_complete_recovery_thread(struct
  *
  */
 
+static void dlm_print_reco_node_status(struct dlm_ctxt *dlm)
+{
+	struct dlm_reco_node_data *ndata;
+	struct dlm_lock_resource *res;
+
+	mlog(ML_NOTICE, "%s(%d): recovery info, state=%s, dead=%u, master=%u\n",
+	     dlm->name, dlm->dlm_reco_thread_task->pid,
+	     dlm->reco.state & DLM_RECO_STATE_ACTIVE ? "ACTIVE" : "inactive",
+	     dlm->reco.dead_node, dlm->reco.new_master);
+
+	list_for_each_entry(ndata, &dlm->reco.node_data, list) {
+		char *st = "unknown";
+		switch (ndata->state) {
+			case DLM_RECO_NODE_DATA_INIT:
+				st = "init";
+				break;
+			case DLM_RECO_NODE_DATA_REQUESTING:
+				st = "requesting";
+				break;
+			case DLM_RECO_NODE_DATA_DEAD:
+				st = "dead";
+				break;
+			case DLM_RECO_NODE_DATA_RECEIVING:
+				st = "receiving";
+				break;
+			case DLM_RECO_NODE_DATA_REQUESTED:
+				st = "requested";
+				break;
+			case DLM_RECO_NODE_DATA_DONE:
+				st = "done";
+				break;
+			case DLM_RECO_NODE_DATA_FINALIZE_SENT:
+				st = "finalize-sent";
+				break;
+			default:
+				st = "bad";
+				break;
+		}
+		mlog(ML_NOTICE, "%s: reco state, node %u, state=%s\n",
+		     dlm->name, ndata->node_num, st);
+	}
+	list_for_each_entry(res, &dlm->reco.resources, recovering) {
+		mlog(ML_NOTICE, "%s: lockres %.*s on recovering list\n",
+		     dlm->name, res->lockname.len, res->lockname.name);
+	}
+}
 
 #define DLM_RECO_THREAD_TIMEOUT_MS (5 * 1000)
 
@@ -267,11 +347,23 @@ int dlm_is_node_dead(struct dlm_ctxt *dl
 {
 	int dead;
 	spin_lock(&dlm->spinlock);
-	dead = test_bit(node, dlm->domain_map);
+	dead = !test_bit(node, dlm->domain_map);
 	spin_unlock(&dlm->spinlock);
 	return dead;
 }
 
+/* returns true if node is no longer in the domain
+ * could be dead or just not joined */
+static int dlm_is_node_recovered(struct dlm_ctxt *dlm, u8 node)
+{
+	int recovered;
+	spin_lock(&dlm->spinlock);
+	recovered = !test_bit(node, dlm->recovery_map);
+	spin_unlock(&dlm->spinlock);
+	return recovered;
+}
+
+
 int dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout)
 {
 	if (timeout) {
@@ -290,6 +382,24 @@ int dlm_wait_for_node_death(struct dlm_c
 	return 0;
 }
 
+int dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout)
+{
+	if (timeout) {
+		mlog(0, "%s: waiting %dms for notification of "
+		     "recovery of node %u\n", dlm->name, timeout, node);
+		wait_event_timeout(dlm->dlm_reco_thread_wq,
+			   dlm_is_node_recovered(dlm, node),
+			   msecs_to_jiffies(timeout));
+	} else {
+		mlog(0, "%s: waiting indefinitely for notification "
+		     "of recovery of node %u\n", dlm->name, node);
+		wait_event(dlm->dlm_reco_thread_wq,
+			   dlm_is_node_recovered(dlm, node));
+	}
+	/* for now, return 0 */
+	return 0;
+}
+
 /* callers of the top-level api calls (dlmlock/dlmunlock) should
  * block on the dlm->reco.event when recovery is in progress.
  * the dlm recovery thread will set this state when it begins
@@ -308,6 +418,13 @@ static int dlm_in_recovery(struct dlm_ct
 
 void dlm_wait_for_recovery(struct dlm_ctxt *dlm)
 {
+	if (dlm_in_recovery(dlm)) {
+		mlog(0, "%s: reco thread %d in recovery: "
+		     "state=%d, master=%u, dead=%u\n",
+		     dlm->name, dlm->dlm_reco_thread_task->pid,
+		     dlm->reco.state, dlm->reco.new_master,
+		     dlm->reco.dead_node);
+	}
 	wait_event(dlm->reco.event, !dlm_in_recovery(dlm));
 }
 
@@ -341,7 +458,7 @@ static int dlm_do_recovery(struct dlm_ct
 		mlog(0, "new master %u died while recovering %u!\n",
 		     dlm->reco.new_master, dlm->reco.dead_node);
 		/* unset the new_master, leave dead_node */
-		dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
+		dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM);
 	}
 
 	/* select a target to recover */
@@ -350,14 +467,14 @@ static int dlm_do_recovery(struct dlm_ct
 
 		bit = find_next_bit (dlm->recovery_map, O2NM_MAX_NODES+1, 0);
 		if (bit >= O2NM_MAX_NODES || bit < 0)
-			dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
+			dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
 		else
-			dlm->reco.dead_node = bit;
+			dlm_set_reco_dead_node(dlm, bit);
 	} else if (!test_bit(dlm->reco.dead_node, dlm->recovery_map)) {
 		/* BUG? */
 		mlog(ML_ERROR, "dead_node %u no longer in recovery map!\n",
 		     dlm->reco.dead_node);
-		dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
+		dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
 	}
 
 	if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) {
@@ -366,7 +483,8 @@ static int dlm_do_recovery(struct dlm_ct
 		/* return to main thread loop and sleep. */
 		return 0;
 	}
-	mlog(0, "recovery thread found node %u in the recovery map!\n",
+	mlog(0, "%s(%d):recovery thread found node %u in the recovery map!\n",
+	     dlm->name, dlm->dlm_reco_thread_task->pid,
 	     dlm->reco.dead_node);
 	spin_unlock(&dlm->spinlock);
 
@@ -389,8 +507,8 @@ static int dlm_do_recovery(struct dlm_ct
 		}
 		mlog(0, "another node will master this recovery session.\n");
 	}
-	mlog(0, "dlm=%s, new_master=%u, this node=%u, dead_node=%u\n",
-	     dlm->name, dlm->reco.new_master,
+	mlog(0, "dlm=%s (%d), new_master=%u, this node=%u, dead_node=%u\n",
+	     dlm->name, dlm->dlm_reco_thread_task->pid, dlm->reco.new_master,
 	     dlm->node_num, dlm->reco.dead_node);
 
 	/* it is safe to start everything back up here
@@ -402,11 +520,13 @@ static int dlm_do_recovery(struct dlm_ct
 	return 0;
 
 master_here:
-	mlog(0, "mastering recovery of %s:%u here(this=%u)!\n",
+	mlog(0, "(%d) mastering recovery of %s:%u here(this=%u)!\n",
+	     dlm->dlm_reco_thread_task->pid,
 	     dlm->name, dlm->reco.dead_node, dlm->node_num);
 
 	status = dlm_remaster_locks(dlm, dlm->reco.dead_node);
 	if (status < 0) {
+		/* we should never hit this anymore */
 		mlog(ML_ERROR, "error %d remastering locks for node %u, "
 		     "retrying.\n", status, dlm->reco.dead_node);
 		/* yield a bit to allow any final network messages
@@ -433,9 +553,16 @@ static int dlm_remaster_locks(struct dlm
 	int destroy = 0;
 	int pass = 0;
 
-	status = dlm_init_recovery_area(dlm, dead_node);
-	if (status < 0)
-		goto leave;
+	do {
+		/* we have become recovery master.  there is no escaping
+		 * this, so just keep trying until we get it. */
+		status = dlm_init_recovery_area(dlm, dead_node);
+		if (status < 0) {
+			mlog(ML_ERROR, "%s: failed to alloc recovery area, "
+			     "retrying\n", dlm->name);
+			msleep(1000);
+		}
+	} while (status != 0);
 
 	/* safe to access the node data list without a lock, since this
 	 * process is the only one to change the list */
@@ -452,16 +579,36 @@ static int dlm_remaster_locks(struct dlm
 			continue;
 		}
 
-		status = dlm_request_all_locks(dlm, ndata->node_num, dead_node);
-		if (status < 0) {
-			mlog_errno(status);
-			if (dlm_is_host_down(status))
-				ndata->state = DLM_RECO_NODE_DATA_DEAD;
-			else {
-				destroy = 1;
-				goto leave;
+		do {
+			status = dlm_request_all_locks(dlm, ndata->node_num,
+						       dead_node);
+			if (status < 0) {
+				mlog_errno(status);
+				if (dlm_is_host_down(status)) {
+					/* node died, ignore it for recovery */
+					status = 0;
+					ndata->state = DLM_RECO_NODE_DATA_DEAD;
+					/* wait for the domain map to catch up
+					 * with the network state. */
+					wait_event_timeout(dlm->dlm_reco_thread_wq,
+							   dlm_is_node_dead(dlm,
+								ndata->node_num),
+							   msecs_to_jiffies(1000));
+					mlog(0, "waited 1 sec for %u, "
+					     "dead? %s\n", ndata->node_num,
+					     dlm_is_node_dead(dlm, ndata->node_num) ?
+					     "yes" : "no");
+				} else {
+					/* -ENOMEM on the other node */
+					mlog(0, "%s: node %u returned "
+					     "%d during recovery, retrying "
+					     "after a short wait\n",
+					     dlm->name, ndata->node_num,
+					     status);
+					msleep(100);
+				}
 			}
-		}
+		} while (status != 0);
 
 		switch (ndata->state) {
 			case DLM_RECO_NODE_DATA_INIT:
@@ -473,10 +620,9 @@ static int dlm_remaster_locks(struct dlm
 				mlog(0, "node %u died after requesting "
 				     "recovery info for node %u\n",
 				     ndata->node_num, dead_node);
-				// start all over
-				destroy = 1;
-				status = -EAGAIN;
-				goto leave;
+				/* fine.  don't need this node's info.
+				 * continue without it. */
+				break;
 			case DLM_RECO_NODE_DATA_REQUESTING:
 				ndata->state = DLM_RECO_NODE_DATA_REQUESTED;
 				mlog(0, "now receiving recovery data from "
@@ -520,35 +666,26 @@ static int dlm_remaster_locks(struct dlm
 					BUG();
 					break;
 				case DLM_RECO_NODE_DATA_DEAD:
-					mlog(ML_NOTICE, "node %u died after "
+					mlog(0, "node %u died after "
 					     "requesting recovery info for "
 					     "node %u\n", ndata->node_num,
 					     dead_node);
-					spin_unlock(&dlm_reco_state_lock);
-					// start all over
-					destroy = 1;
-					status = -EAGAIN;
-					/* instead of spinning like crazy here,
-					 * wait for the domain map to catch up
-					 * with the network state.  otherwise this
-					 * can be hit hundreds of times before
-					 * the node is really seen as dead. */
-					wait_event_timeout(dlm->dlm_reco_thread_wq,
-							   dlm_is_node_dead(dlm,
-								ndata->node_num),
-							   msecs_to_jiffies(1000));
-					mlog(0, "waited 1 sec for %u, "
-					     "dead? %s\n", ndata->node_num,
-					     dlm_is_node_dead(dlm, ndata->node_num) ?
-					     "yes" : "no");
-					goto leave;
+					break;
 				case DLM_RECO_NODE_DATA_RECEIVING:
 				case DLM_RECO_NODE_DATA_REQUESTED:
+					mlog(0, "%s: node %u still in state %s\n",
+					     dlm->name, ndata->node_num,
+					     ndata->state==DLM_RECO_NODE_DATA_RECEIVING ?
+					     "receiving" : "requested");
 					all_nodes_done = 0;
 					break;
 				case DLM_RECO_NODE_DATA_DONE:
+					mlog(0, "%s: node %u state is done\n",
+					     dlm->name, ndata->node_num);
 					break;
 				case DLM_RECO_NODE_DATA_FINALIZE_SENT:
+					mlog(0, "%s: node %u state is finalize\n",
+					     dlm->name, ndata->node_num);
 					break;
 			}
 		}
@@ -578,7 +715,7 @@ static int dlm_remaster_locks(struct dlm
 			     jiffies, dlm->reco.dead_node,
 			     dlm->node_num, dlm->reco.new_master);
 			destroy = 1;
-			status = ret;
+			status = 0;
 			/* rescan everything marked dirty along the way */
 			dlm_kick_thread(dlm, NULL);
 			break;
@@ -591,7 +728,6 @@ static int dlm_remaster_locks(struct dlm
 
 	}
 
-leave:
 	if (destroy)
 		dlm_destroy_recovery_area(dlm, dead_node);
 
@@ -617,7 +753,7 @@ static int dlm_init_recovery_area(struct
 		}
 		BUG_ON(num == dead_node);
 
-		ndata = kcalloc(1, sizeof(*ndata), GFP_KERNEL);
+		ndata = kcalloc(1, sizeof(*ndata), GFP_NOFS);
 		if (!ndata) {
 			dlm_destroy_recovery_area(dlm, dead_node);
 			return -ENOMEM;
@@ -691,16 +827,25 @@ int dlm_request_all_locks_handler(struct
 	if (!dlm_grab(dlm))
 		return -EINVAL;
 
+	if (lr->dead_node != dlm->reco.dead_node) {
+		mlog(ML_ERROR, "%s: node %u sent dead_node=%u, but local "
+		     "dead_node is %u\n", dlm->name, lr->node_idx,
+		     lr->dead_node, dlm->reco.dead_node);
+		dlm_print_reco_node_status(dlm);
+		/* this is a hack */
+		dlm_put(dlm);
+		return -ENOMEM;
+	}
 	BUG_ON(lr->dead_node != dlm->reco.dead_node);
 
-	item = kcalloc(1, sizeof(*item), GFP_KERNEL);
+	item = kcalloc(1, sizeof(*item), GFP_NOFS);
 	if (!item) {
 		dlm_put(dlm);
 		return -ENOMEM;
 	}
 
 	/* this will get freed by dlm_request_all_locks_worker */
-	buf = (char *) __get_free_page(GFP_KERNEL);
+	buf = (char *) __get_free_page(GFP_NOFS);
 	if (!buf) {
 		kfree(item);
 		dlm_put(dlm);
@@ -715,7 +860,7 @@ int dlm_request_all_locks_handler(struct
 	spin_lock(&dlm->work_lock);
 	list_add_tail(&item->list, &dlm->work_list);
 	spin_unlock(&dlm->work_lock);
-	schedule_work(&dlm->dispatched_work);
+	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
 
 	dlm_put(dlm);
 	return 0;
@@ -730,32 +875,34 @@ static void dlm_request_all_locks_worker
 	struct list_head *iter;
 	int ret;
 	u8 dead_node, reco_master;
+	int skip_all_done = 0;
 
 	dlm = item->dlm;
 	dead_node = item->u.ral.dead_node;
 	reco_master = item->u.ral.reco_master;
 	mres = (struct dlm_migratable_lockres *)data;
 
+	mlog(0, "%s: recovery worker started, dead=%u, master=%u\n",
+	     dlm->name, dead_node, reco_master);
+
 	if (dead_node != dlm->reco.dead_node ||
 	    reco_master != dlm->reco.new_master) {
-		/* show extra debug info if the recovery state is messed */
-		mlog(ML_ERROR, "%s: bad reco state: reco(dead=%u, master=%u), "
-		     "request(dead=%u, master=%u)\n",
-		     dlm->name, dlm->reco.dead_node, dlm->reco.new_master,
-		     dead_node, reco_master);
-		mlog(ML_ERROR, "%s: name=%.*s master=%u locks=%u/%u flags=%u "
-		     "entry[0]={c=%u:%llu,l=%u,f=%u,t=%d,ct=%d,hb=%d,n=%u}\n",
-		     dlm->name, mres->lockname_len, mres->lockname, mres->master,
-		     mres->num_locks, mres->total_locks, mres->flags,
-		     dlm_get_lock_cookie_node(mres->ml[0].cookie),
-		     dlm_get_lock_cookie_seq(mres->ml[0].cookie),
-		     mres->ml[0].list, mres->ml[0].flags,
-		     mres->ml[0].type, mres->ml[0].convert_type,
-		     mres->ml[0].highest_blocked, mres->ml[0].node);
-		BUG();
+		/* worker could have been created before the recovery master
+		 * died.  if so, do not continue, but do not error. */
+		if (dlm->reco.new_master == O2NM_INVALID_NODE_NUM) {
+			mlog(ML_NOTICE, "%s: will not send recovery state, "
+			     "recovery master %u died, thread=(dead=%u,mas=%u)"
+			     " current=(dead=%u,mas=%u)\n", dlm->name,
+			     reco_master, dead_node, reco_master,
+			     dlm->reco.dead_node, dlm->reco.new_master);
+		} else {
+			mlog(ML_NOTICE, "%s: reco state invalid: reco(dead=%u, "
+			     "master=%u), request(dead=%u, master=%u)\n",
+			     dlm->name, dlm->reco.dead_node,
+			     dlm->reco.new_master, dead_node, reco_master);
+		}
+		goto leave;
 	}
-	BUG_ON(dead_node != dlm->reco.dead_node);
-	BUG_ON(reco_master != dlm->reco.new_master);
 
 	/* lock resources should have already been moved to the
  	 * dlm->reco.resources list.  now move items from that list
@@ -766,12 +913,20 @@ static void dlm_request_all_locks_worker
 	dlm_move_reco_locks_to_list(dlm, &resources, dead_node);
 
 	/* now we can begin blasting lockreses without the dlm lock */
+
+	/* any errors returned will be due to the new_master dying,
+	 * the dlm_reco_thread should detect this */
 	list_for_each(iter, &resources) {
 		res = list_entry (iter, struct dlm_lock_resource, recovering);
 		ret = dlm_send_one_lockres(dlm, res, mres, reco_master,
 				   	DLM_MRES_RECOVERY);
-		if (ret < 0)
-			mlog_errno(ret);
+		if (ret < 0) {
+			mlog(ML_ERROR, "%s: node %u went down while sending "
+			     "recovery state for dead node %u, ret=%d\n", dlm->name,
+			     reco_master, dead_node, ret);
+			skip_all_done = 1;
+			break;
+		}
 	}
 
 	/* move the resources back to the list */
@@ -779,10 +934,15 @@ static void dlm_request_all_locks_worker
 	list_splice_init(&resources, &dlm->reco.resources);
 	spin_unlock(&dlm->spinlock);
 
-	ret = dlm_send_all_done_msg(dlm, dead_node, reco_master);
-	if (ret < 0)
-		mlog_errno(ret);
-
+	if (!skip_all_done) {
+		ret = dlm_send_all_done_msg(dlm, dead_node, reco_master);
+		if (ret < 0) {
+			mlog(ML_ERROR, "%s: node %u went down while sending "
+			     "recovery all-done for dead node %u, ret=%d\n",
+			     dlm->name, reco_master, dead_node, ret);
+		}
+	}
+leave:
 	free_page((unsigned long)data);
 }
 
@@ -801,8 +961,14 @@ static int dlm_send_all_done_msg(struct 
 
 	ret = o2net_send_message(DLM_RECO_DATA_DONE_MSG, dlm->key, &done_msg,
 				 sizeof(done_msg), send_to, &tmpret);
-	/* negative status is ignored by the caller */
-	if (ret >= 0)
+	if (ret < 0) {
+		if (!dlm_is_host_down(ret)) {
+			mlog_errno(ret);
+			mlog(ML_ERROR, "%s: unknown error sending data-done "
+			     "to %u\n", dlm->name, send_to);
+			BUG();
+		}
+	} else
 		ret = tmpret;
 	return ret;
 }
@@ -822,7 +988,11 @@ int dlm_reco_data_done_handler(struct o2
 	mlog(0, "got DATA DONE: dead_node=%u, reco.dead_node=%u, "
 	     "node_idx=%u, this node=%u\n", done->dead_node,
 	     dlm->reco.dead_node, done->node_idx, dlm->node_num);
-	BUG_ON(done->dead_node != dlm->reco.dead_node);
+
+	mlog_bug_on_msg((done->dead_node != dlm->reco.dead_node),
+			"Got DATA DONE: dead_node=%u, reco.dead_node=%u, "
+			"node_idx=%u, this node=%u\n", done->dead_node,
+			dlm->reco.dead_node, done->node_idx, dlm->node_num);
 
 	spin_lock(&dlm_reco_state_lock);
 	list_for_each(iter, &dlm->reco.node_data) {
@@ -1023,8 +1193,9 @@ static int dlm_add_lock_to_array(struct 
 		    ml->type == LKM_PRMODE) {
 			/* if it is already set, this had better be a PR
 			 * and it has to match */
-			if (mres->lvb[0] && (ml->type == LKM_EXMODE ||
-			    memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))) {
+			if (!dlm_lvb_is_empty(mres->lvb) &&
+			    (ml->type == LKM_EXMODE ||
+			     memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))) {
 				mlog(ML_ERROR, "mismatched lvbs!\n");
 				__dlm_print_one_lock_resource(lock->lockres);
 				BUG();
@@ -1083,22 +1254,25 @@ int dlm_send_one_lockres(struct dlm_ctxt
 			 * we must send it immediately. */
 			ret = dlm_send_mig_lockres_msg(dlm, mres, send_to,
 						       res, total_locks);
-			if (ret < 0) {
-				// TODO
-				mlog(ML_ERROR, "dlm_send_mig_lockres_msg "
-				     "returned %d, TODO\n", ret);
-				BUG();
-			}
+			if (ret < 0)
+				goto error;
 		}
 	}
 	/* flush any remaining locks */
 	ret = dlm_send_mig_lockres_msg(dlm, mres, send_to, res, total_locks);
-	if (ret < 0) {
-		// TODO
-		mlog(ML_ERROR, "dlm_send_mig_lockres_msg returned %d, "
-		     "TODO\n", ret);
+	if (ret < 0)
+		goto error;
+	return ret;
+
+error:
+	mlog(ML_ERROR, "%s: dlm_send_mig_lockres_msg returned %d\n",
+	     dlm->name, ret);
+	if (!dlm_is_host_down(ret))
 		BUG();
-	}
+	mlog(0, "%s: node %u went down while sending %s "
+	     "lockres %.*s\n", dlm->name, send_to,
+	     flags & DLM_MRES_RECOVERY ?  "recovery" : "migration",
+	     res->lockname.len, res->lockname.name);
 	return ret;
 }
 
@@ -1146,8 +1320,8 @@ int dlm_mig_lockres_handler(struct o2net
 		mlog(0, "all done flag.  all lockres data received!\n");
 
 	ret = -ENOMEM;
-	buf = kmalloc(be16_to_cpu(msg->data_len), GFP_KERNEL);
-	item = kcalloc(1, sizeof(*item), GFP_KERNEL);
+	buf = kmalloc(be16_to_cpu(msg->data_len), GFP_NOFS);
+	item = kcalloc(1, sizeof(*item), GFP_NOFS);
 	if (!buf || !item)
 		goto leave;
 
@@ -1238,7 +1412,7 @@ int dlm_mig_lockres_handler(struct o2net
 	spin_lock(&dlm->work_lock);
 	list_add_tail(&item->list, &dlm->work_list);
 	spin_unlock(&dlm->work_lock);
-	schedule_work(&dlm->dispatched_work);
+	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
 
 leave:
 	dlm_put(dlm);
@@ -1406,6 +1580,7 @@ int dlm_master_requery_handler(struct o2
 	struct dlm_ctxt *dlm = data;
 	struct dlm_master_requery *req = (struct dlm_master_requery *)msg->buf;
 	struct dlm_lock_resource *res = NULL;
+	unsigned int hash;
 	int master = DLM_LOCK_RES_OWNER_UNKNOWN;
 	u32 flags = DLM_ASSERT_MASTER_REQUERY;
 
@@ -1415,8 +1590,10 @@ int dlm_master_requery_handler(struct o2
 		return master;
 	}
 
+	hash = dlm_lockid_hash(req->name, req->namelen);
+
 	spin_lock(&dlm->spinlock);
-	res = __dlm_lookup_lockres(dlm, req->name, req->namelen);
+	res = __dlm_lookup_lockres(dlm, req->name, req->namelen, hash);
 	if (res) {
 		spin_lock(&res->spinlock);
 		master = res->owner;
@@ -1483,7 +1660,7 @@ static int dlm_process_recovery_data(str
 	struct dlm_lock *newlock = NULL;
 	struct dlm_lockstatus *lksb = NULL;
 	int ret = 0;
-	int i;
+	int i, bad;
 	struct list_head *iter;
 	struct dlm_lock *lock = NULL;
 
@@ -1553,28 +1730,48 @@ static int dlm_process_recovery_data(str
 		}
 		lksb->flags |= (ml->flags &
 				(DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB));
-			
-		if (mres->lvb[0]) {
+
+		if (ml->type == LKM_NLMODE)
+			goto skip_lvb;
+
+		if (!dlm_lvb_is_empty(mres->lvb)) {
 			if (lksb->flags & DLM_LKSB_PUT_LVB) {
 				/* other node was trying to update
 				 * lvb when node died.  recreate the
 				 * lksb with the updated lvb. */
 				memcpy(lksb->lvb, mres->lvb, DLM_LVB_LEN);
+				/* the lock resource lvb update must happen
+				 * NOW, before the spinlock is dropped.
+				 * we no longer wait for the AST to update
+				 * the lvb. */
+				memcpy(res->lvb, mres->lvb, DLM_LVB_LEN);
 			} else {
 				/* otherwise, the node is sending its 
 				 * most recent valid lvb info */
 				BUG_ON(ml->type != LKM_EXMODE &&
 				       ml->type != LKM_PRMODE);
-				if (res->lvb[0] && (ml->type == LKM_EXMODE ||
-				    memcmp(res->lvb, mres->lvb, DLM_LVB_LEN))) {
-					mlog(ML_ERROR, "received bad lvb!\n");
-					__dlm_print_one_lock_resource(res);
-					BUG();
+				if (!dlm_lvb_is_empty(res->lvb) &&
+ 				    (ml->type == LKM_EXMODE ||
+ 				     memcmp(res->lvb, mres->lvb, DLM_LVB_LEN))) {
+ 					int i;
+ 					mlog(ML_ERROR, "%s:%.*s: received bad "
+ 					     "lvb! type=%d\n", dlm->name,
+ 					     res->lockname.len,
+ 					     res->lockname.name, ml->type);
+ 					printk("lockres lvb=[");
+ 					for (i=0; i<DLM_LVB_LEN; i++)
+ 						printk("%02x", res->lvb[i]);
+ 					printk("]\nmigrated lvb=[");
+ 					for (i=0; i<DLM_LVB_LEN; i++)
+ 						printk("%02x", mres->lvb[i]);
+ 					printk("]\n");
+ 					dlm_print_one_lock_resource(res);
+ 					BUG();
 				}
 				memcpy(res->lvb, mres->lvb, DLM_LVB_LEN);
 			}
 		}
-
+skip_lvb:
 
 		/* NOTE:
 		 * wrt lock queue ordering and recovery:
@@ -1592,9 +1789,33 @@ static int dlm_process_recovery_data(str
 		 * relative to each other, but clearly *not*
 		 * preserved relative to locks from other nodes.
 		 */
+		bad = 0;
 		spin_lock(&res->spinlock);
-		dlm_lock_get(newlock);
-		list_add_tail(&newlock->list, queue);
+		list_for_each_entry(lock, queue, list) {
+			if (lock->ml.cookie == ml->cookie) {
+				u64 c = lock->ml.cookie;
+				mlog(ML_ERROR, "%s:%.*s: %u:%llu: lock already "
+				     "exists on this lockres!\n", dlm->name,
+				     res->lockname.len, res->lockname.name,
+				     dlm_get_lock_cookie_node(c),
+				     dlm_get_lock_cookie_seq(c));
+
+				mlog(ML_NOTICE, "sent lock: type=%d, conv=%d, "
+				     "node=%u, cookie=%u:%llu, queue=%d\n",
+	      			     ml->type, ml->convert_type, ml->node,
+				     dlm_get_lock_cookie_node(ml->cookie),
+				     dlm_get_lock_cookie_seq(ml->cookie),
+				     ml->list);
+
+				__dlm_print_one_lock_resource(res);
+				bad = 1;
+				break;
+			}
+		}
+		if (!bad) {
+			dlm_lock_get(newlock);
+			list_add_tail(&newlock->list, queue);
+		}
 		spin_unlock(&res->spinlock);
 	}
 	mlog(0, "done running all the locks\n");
@@ -1618,8 +1839,14 @@ void dlm_move_lockres_to_recovery_list(s
 	struct dlm_lock *lock;
 
 	res->state |= DLM_LOCK_RES_RECOVERING;
-	if (!list_empty(&res->recovering))
+	if (!list_empty(&res->recovering)) {
+		mlog(0,
+		     "Recovering res %s:%.*s, is already on recovery list!\n",
+		     dlm->name, res->lockname.len, res->lockname.name);
 		list_del_init(&res->recovering);
+	}
+	/* We need to hold a reference while on the recovery list */
+	dlm_lockres_get(res);
 	list_add_tail(&res->recovering, &dlm->reco.resources);
 
 	/* find any pending locks and put them back on proper list */
@@ -1708,9 +1935,11 @@ static void dlm_finish_local_lockres_rec
 			spin_lock(&res->spinlock);
 			dlm_change_lockres_owner(dlm, res, new_master);
 			res->state &= ~DLM_LOCK_RES_RECOVERING;
-			__dlm_dirty_lockres(dlm, res);
+			if (!__dlm_lockres_unused(res))
+				__dlm_dirty_lockres(dlm, res);
 			spin_unlock(&res->spinlock);
 			wake_up(&res->wq);
+			dlm_lockres_put(res);
 		}
 	}
 
@@ -1719,7 +1948,7 @@ static void dlm_finish_local_lockres_rec
 	 * the RECOVERING state and set the owner
 	 * if necessary */
 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
-		bucket = &(dlm->lockres_hash[i]);
+		bucket = dlm_lockres_hash(dlm, i);
 		hlist_for_each_entry(res, hash_iter, bucket, hash_node) {
 			if (res->state & DLM_LOCK_RES_RECOVERING) {
 				if (res->owner == dead_node) {
@@ -1743,11 +1972,13 @@ static void dlm_finish_local_lockres_rec
 					     dlm->name, res->lockname.len,
 					     res->lockname.name, res->owner);
 					list_del_init(&res->recovering);
+					dlm_lockres_put(res);
 				}
 				spin_lock(&res->spinlock);
 				dlm_change_lockres_owner(dlm, res, new_master);
 				res->state &= ~DLM_LOCK_RES_RECOVERING;
-				__dlm_dirty_lockres(dlm, res);
+				if (!__dlm_lockres_unused(res))
+					__dlm_dirty_lockres(dlm, res);
 				spin_unlock(&res->spinlock);
 				wake_up(&res->wq);
 			}
@@ -1884,7 +2115,7 @@ static void dlm_do_local_recovery_cleanu
 	 *    need to be fired as a result.
 	 */
 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
-		bucket = &(dlm->lockres_hash[i]);
+		bucket = dlm_lockres_hash(dlm, i);
 		hlist_for_each_entry(res, iter, bucket, hash_node) {
  			/* always prune any $RECOVERY entries for dead nodes,
  			 * otherwise hangs can occur during later recovery */
@@ -1924,6 +2155,20 @@ static void __dlm_hb_node_down(struct dl
 {
 	assert_spin_locked(&dlm->spinlock);
 
+	if (dlm->reco.new_master == idx) {
+		mlog(0, "%s: recovery master %d just died\n",
+		     dlm->name, idx);
+		if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
+			/* finalize1 was reached, so it is safe to clear
+			 * the new_master and dead_node.  that recovery
+			 * is complete. */
+			mlog(0, "%s: dead master %d had reached "
+			     "finalize1 state, clearing\n", dlm->name, idx);
+			dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE;
+			__dlm_reset_recovery(dlm);
+		}
+	}
+
 	/* check to see if the node is already considered dead */
 	if (!test_bit(idx, dlm->live_nodes_map)) {
 		mlog(0, "for domain %s, node %d is already dead. "
@@ -2087,7 +2332,7 @@ again:	
 
 			/* set the new_master to this node */
 			spin_lock(&dlm->spinlock);
-			dlm->reco.new_master = dlm->node_num;
+			dlm_set_reco_master(dlm, dlm->node_num);
 			spin_unlock(&dlm->spinlock);
 		}
 
@@ -2125,6 +2370,10 @@ again:	
 		mlog(0, "%s: reco master %u is ready to recover %u\n",
 		     dlm->name, dlm->reco.new_master, dlm->reco.dead_node);
 		status = -EEXIST;
+	} else if (ret == DLM_RECOVERING) {
+		mlog(0, "dlm=%s dlmlock says master node died (this=%u)\n",
+		     dlm->name, dlm->node_num);
+		goto again;
 	} else {
 		struct dlm_lock_resource *res;
 
@@ -2156,7 +2405,7 @@ static int dlm_send_begin_reco_message(s
 
 	mlog_entry("%u\n", dead_node);
 
-	mlog(0, "dead node is %u\n", dead_node);
+	mlog(0, "%s: dead node is %u\n", dlm->name, dead_node);
 
 	spin_lock(&dlm->spinlock);
 	dlm_node_iter_init(dlm->domain_map, &iter);
@@ -2214,6 +2463,14 @@ retry:
 			 * another ENOMEM */
 			msleep(100);
 			goto retry;
+		} else if (ret == EAGAIN) {
+			mlog(0, "%s: trying to start recovery of node "
+			     "%u, but node %u is waiting for last recovery "
+			     "to complete, backoff for a bit\n", dlm->name,
+			     dead_node, nodenum);
+			/* TODO Look into replacing msleep with cond_resched() */
+			msleep(100);
+			goto retry;
 		}
 	}
 
@@ -2229,8 +2486,20 @@ int dlm_begin_reco_handler(struct o2net_
 	if (!dlm_grab(dlm))
 		return 0;
 
-	mlog(0, "node %u wants to recover node %u\n",
-		  br->node_idx, br->dead_node);
+	spin_lock(&dlm->spinlock);
+	if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
+		mlog(0, "%s: node %u wants to recover node %u (%u:%u) "
+		     "but this node is in finalize state, waiting on finalize2\n",
+		     dlm->name, br->node_idx, br->dead_node,
+		     dlm->reco.dead_node, dlm->reco.new_master);
+		spin_unlock(&dlm->spinlock);
+		return EAGAIN;
+	}
+	spin_unlock(&dlm->spinlock);
+
+	mlog(0, "%s: node %u wants to recover node %u (%u:%u)\n",
+	     dlm->name, br->node_idx, br->dead_node,
+	     dlm->reco.dead_node, dlm->reco.new_master);
 
 	dlm_fire_domain_eviction_callbacks(dlm, br->dead_node);
 
@@ -2252,8 +2521,8 @@ int dlm_begin_reco_handler(struct o2net_
 		     "node %u changing it to %u\n", dlm->name, 
 		     dlm->reco.dead_node, br->node_idx, br->dead_node);
 	}
-	dlm->reco.new_master = br->node_idx;
-	dlm->reco.dead_node = br->dead_node;
+	dlm_set_reco_master(dlm, br->node_idx);
+	dlm_set_reco_dead_node(dlm, br->dead_node);
 	if (!test_bit(br->dead_node, dlm->recovery_map)) {
 		mlog(0, "recovery master %u sees %u as dead, but this "
 		     "node has not yet.  marking %u as dead\n",
@@ -2272,10 +2541,16 @@ int dlm_begin_reco_handler(struct o2net_
 	spin_unlock(&dlm->spinlock);
 
 	dlm_kick_recovery_thread(dlm);
+
+	mlog(0, "%s: recovery started by node %u, for %u (%u:%u)\n",
+	     dlm->name, br->node_idx, br->dead_node,
+	     dlm->reco.dead_node, dlm->reco.new_master);
+
 	dlm_put(dlm);
 	return 0;
 }
 
+#define DLM_FINALIZE_STAGE2  0x01
 static int dlm_send_finalize_reco_message(struct dlm_ctxt *dlm)
 {
 	int ret = 0;
@@ -2283,25 +2558,31 @@ static int dlm_send_finalize_reco_messag
 	struct dlm_node_iter iter;
 	int nodenum;
 	int status;
+	int stage = 1;
 
-	mlog(0, "finishing recovery for node %s:%u\n",
-	     dlm->name, dlm->reco.dead_node);
+	mlog(0, "finishing recovery for node %s:%u, "
+	     "stage %d\n", dlm->name, dlm->reco.dead_node, stage);
 
 	spin_lock(&dlm->spinlock);
 	dlm_node_iter_init(dlm->domain_map, &iter);
 	spin_unlock(&dlm->spinlock);
 
+stage2:
 	memset(&fr, 0, sizeof(fr));
 	fr.node_idx = dlm->node_num;
 	fr.dead_node = dlm->reco.dead_node;
+	if (stage == 2)
+		fr.flags |= DLM_FINALIZE_STAGE2;
 
 	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
 		if (nodenum == dlm->node_num)
 			continue;
 		ret = o2net_send_message(DLM_FINALIZE_RECO_MSG, dlm->key,
 					 &fr, sizeof(fr), nodenum, &status);
-		if (ret >= 0) {
+		if (ret >= 0)
 			ret = status;
+		if (ret < 0) {
+			mlog_errno(ret);
 			if (dlm_is_host_down(ret)) {
 				/* this has no effect on this recovery 
 				 * session, so set the status to zero to 
@@ -2309,13 +2590,17 @@ static int dlm_send_finalize_reco_messag
 				mlog(ML_ERROR, "node %u went down after this "
 				     "node finished recovery.\n", nodenum);
 				ret = 0;
+				continue;
 			}
-		}
-		if (ret < 0) {
-			mlog_errno(ret);
 			break;
 		}
 	}
+	if (stage == 1) {
+		/* reset the node_iter back to the top and send finalize2 */
+		iter.curnode = -1;
+		stage = 2;
+		goto stage2;
+	}
 
 	return ret;
 }
@@ -2324,14 +2609,19 @@ int dlm_finalize_reco_handler(struct o2n
 {
 	struct dlm_ctxt *dlm = data;
 	struct dlm_finalize_reco *fr = (struct dlm_finalize_reco *)msg->buf;
+	int stage = 1;
 
 	/* ok to return 0, domain has gone away */
 	if (!dlm_grab(dlm))
 		return 0;
 
-	mlog(0, "node %u finalizing recovery of node %u\n",
-	     fr->node_idx, fr->dead_node);
+	if (fr->flags & DLM_FINALIZE_STAGE2)
+		stage = 2;
 
+	mlog(0, "%s: node %u finalizing recovery stage%d of "
+	     "node %u (%u:%u)\n", dlm->name, fr->node_idx, stage,
+	     fr->dead_node, dlm->reco.dead_node, dlm->reco.new_master);
+ 
 	spin_lock(&dlm->spinlock);
 
 	if (dlm->reco.new_master != fr->node_idx) {
@@ -2347,13 +2637,41 @@ int dlm_finalize_reco_handler(struct o2n
 		BUG();
 	}
 
-	dlm_finish_local_lockres_recovery(dlm, fr->dead_node, fr->node_idx);
-
-	spin_unlock(&dlm->spinlock);
+	switch (stage) {
+		case 1:
+			dlm_finish_local_lockres_recovery(dlm, fr->dead_node, fr->node_idx);
+			if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
+				mlog(ML_ERROR, "%s: received finalize1 from "
+				     "new master %u for dead node %u, but "
+				     "this node has already received it!\n",
+				     dlm->name, fr->node_idx, fr->dead_node);
+				dlm_print_reco_node_status(dlm);
+				BUG();
+			}
+			dlm->reco.state |= DLM_RECO_STATE_FINALIZE;
+			spin_unlock(&dlm->spinlock);
+			break;
+		case 2:
+			if (!(dlm->reco.state & DLM_RECO_STATE_FINALIZE)) {
+				mlog(ML_ERROR, "%s: received finalize2 from "
+				     "new master %u for dead node %u, but "
+				     "this node did not have finalize1!\n",
+				     dlm->name, fr->node_idx, fr->dead_node);
+				dlm_print_reco_node_status(dlm);
+				BUG();
+			}
+			dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE;
+			spin_unlock(&dlm->spinlock);
+			dlm_reset_recovery(dlm);
+			dlm_kick_recovery_thread(dlm);
+			break;
+		default:
+			BUG();
+	}
 
-	dlm_reset_recovery(dlm);
+	mlog(0, "%s: recovery done, reco master was %u, dead now %u, master now %u\n",
+	     dlm->name, fr->node_idx, dlm->reco.dead_node, dlm->reco.new_master);
 
-	dlm_kick_recovery_thread(dlm);
 	dlm_put(dlm);
 	return 0;
 }
diff --git a/fs/ocfs2/dlm/dlmthread.c b/fs/ocfs2/dlm/dlmthread.c
index 5be9d14..4641d06 100644
--- a/fs/ocfs2/dlm/dlmthread.c
+++ b/fs/ocfs2/dlm/dlmthread.c
@@ -39,6 +39,7 @@ #include <linux/socket.h>
 #include <linux/inet.h>
 #include <linux/timer.h>
 #include <linux/kthread.h>
+#include <linux/delay.h>
 
 
 #include "cluster/heartbeat.h"
@@ -53,6 +54,8 @@ #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_
 #include "cluster/masklog.h"
 
 static int dlm_thread(void *data);
+static void dlm_purge_lockres_now(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *lockres);
 
 static void dlm_flush_asts(struct dlm_ctxt *dlm);
 
@@ -80,7 +83,7 @@ repeat:
 }
 
 
-static int __dlm_lockres_unused(struct dlm_lock_resource *res)
+int __dlm_lockres_unused(struct dlm_lock_resource *res)
 {
 	if (list_empty(&res->granted) &&
 	    list_empty(&res->converting) &&
@@ -103,6 +106,20 @@ void __dlm_lockres_calc_usage(struct dlm
 	assert_spin_locked(&res->spinlock);
 
 	if (__dlm_lockres_unused(res)){
+		/* For now, just keep any resource we master */
+		if (res->owner == dlm->node_num)
+		{
+			if (!list_empty(&res->purge)) {
+				mlog(0, "we master %s:%.*s, but it is on "
+				     "the purge list.  Removing\n",
+				     dlm->name, res->lockname.len,
+				     res->lockname.name);
+				list_del_init(&res->purge);
+				dlm->purge_count--;
+			}
+			return;
+		}
+
 		if (list_empty(&res->purge)) {
 			mlog(0, "putting lockres %.*s from purge list\n",
 			     res->lockname.len, res->lockname.name);
@@ -110,10 +127,23 @@ void __dlm_lockres_calc_usage(struct dlm
 			res->last_used = jiffies;
 			list_add_tail(&res->purge, &dlm->purge_list);
 			dlm->purge_count++;
+
+			/* if this node is not the owner, there is
+			 * no way to keep track of who the owner could be.
+			 * unhash it to avoid serious problems. */
+			if (res->owner != dlm->node_num) {
+				mlog(0, "%s:%.*s: doing immediate "
+				     "purge of lockres owned by %u\n",
+				     dlm->name, res->lockname.len,
+				     res->lockname.name, res->owner);
+
+				dlm_purge_lockres_now(dlm, res);
+			}
 		}
 	} else if (!list_empty(&res->purge)) {
-		mlog(0, "removing lockres %.*s from purge list\n",
-		     res->lockname.len, res->lockname.name);
+		mlog(0, "removing lockres %.*s from purge list, "
+		     "owner=%u\n", res->lockname.len, res->lockname.name,
+		     res->owner);
 
 		list_del_init(&res->purge);
 		dlm->purge_count--;
@@ -165,6 +195,7 @@ again:
 	} else if (ret < 0) {
 		mlog(ML_NOTICE, "lockres %.*s: migrate failed, retrying\n",
 		     lockres->lockname.len, lockres->lockname.name);
+		msleep(100);
 		goto again;
 	}
 
@@ -178,6 +209,24 @@ finish:
 	__dlm_unhash_lockres(lockres);
 }
 
+/* make an unused lockres go away immediately.
+ * as soon as the dlm spinlock is dropped, this lockres
+ * will not be found. kfree still happens on last put. */
+static void dlm_purge_lockres_now(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *lockres)
+{
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&lockres->spinlock);
+
+	BUG_ON(!__dlm_lockres_unused(lockres));
+
+	if (!list_empty(&lockres->purge)) {
+		list_del_init(&lockres->purge);
+		dlm->purge_count--;
+	}
+	__dlm_unhash_lockres(lockres);
+}
+
 static void dlm_run_purge_list(struct dlm_ctxt *dlm,
 			       int purge_now)
 {
@@ -422,6 +471,8 @@ void __dlm_dirty_lockres(struct dlm_ctxt
 	/* don't shuffle secondary queues */
 	if ((res->owner == dlm->node_num) &&
 	    !(res->state & DLM_LOCK_RES_DIRTY)) {
+		/* ref for dirty_list */
+		dlm_lockres_get(res);
 		list_add_tail(&res->dirty, &dlm->dirty_list);
 		res->state |= DLM_LOCK_RES_DIRTY;
 	}
@@ -606,6 +657,8 @@ static int dlm_thread(void *data)
 			list_del_init(&res->dirty);
 			spin_unlock(&res->spinlock);
 			spin_unlock(&dlm->spinlock);
+			/* Drop dirty_list ref */
+			dlm_lockres_put(res);
 
 		 	/* lockres can be re-dirtied/re-added to the
 			 * dirty_list in this gap, but that is ok */
@@ -642,8 +695,9 @@ static int dlm_thread(void *data)
 			 * spinlock and do NOT have the dlm lock.
 			 * safe to reserve/queue asts and run the lists. */
 
-			mlog(0, "calling dlm_shuffle_lists with dlm=%p, "
-			     "res=%p\n", dlm, res);
+			mlog(0, "calling dlm_shuffle_lists with dlm=%s, "
+			     "res=%.*s\n", dlm->name,
+			     res->lockname.len, res->lockname.name);
 
 			/* called while holding lockres lock */
 			dlm_shuffle_lists(dlm, res);
@@ -657,6 +711,8 @@ in_progress:
 			/* if the lock was in-progress, stick
 			 * it on the back of the list */
 			if (delay) {
+				/* ref for dirty_list */
+				dlm_lockres_get(res);
 				spin_lock(&res->spinlock);
 				list_add_tail(&res->dirty, &dlm->dirty_list);
 				res->state |= DLM_LOCK_RES_DIRTY;
@@ -677,7 +733,7 @@ in_progress:
 
 		/* yield and continue right away if there is more work to do */
 		if (!n) {
-			yield();
+			cond_resched();
 			continue;
 		}
 
diff --git a/fs/ocfs2/dlm/dlmunlock.c b/fs/ocfs2/dlm/dlmunlock.c
index 7b1a275..1fe1112 100644
--- a/fs/ocfs2/dlm/dlmunlock.c
+++ b/fs/ocfs2/dlm/dlmunlock.c
@@ -319,6 +319,16 @@ static enum dlm_status dlm_send_remote_u
 
 	mlog_entry("%.*s\n", res->lockname.len, res->lockname.name);
 
+	if (owner == dlm->node_num) {
+		/* ended up trying to contact ourself.  this means
+		 * that the lockres had been remote but became local
+		 * via a migration.  just retry it, now as local */
+		mlog(0, "%s:%.*s: this node became the master due to a "
+		     "migration, re-evaluate now\n", dlm->name,
+		     res->lockname.len, res->lockname.name);
+		return DLM_FORWARD;
+	}
+
 	memset(&unlock, 0, sizeof(unlock));
 	unlock.node_idx = dlm->node_num;
 	unlock.flags = cpu_to_be32(flags);
diff --git a/fs/ocfs2/dlm/userdlm.c b/fs/ocfs2/dlm/userdlm.c
index 74ca4e5..e641b08 100644
--- a/fs/ocfs2/dlm/userdlm.c
+++ b/fs/ocfs2/dlm/userdlm.c
@@ -672,7 +672,7 @@ struct dlm_ctxt *user_dlm_register_conte
 	u32 dlm_key;
 	char *domain;
 
-	domain = kmalloc(name->len + 1, GFP_KERNEL);
+	domain = kmalloc(name->len + 1, GFP_NOFS);
 	if (!domain) {
 		mlog_errno(-ENOMEM);
 		return ERR_PTR(-ENOMEM);

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2006-06-19 18:46 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2006-06-19 18:46 UTC (permalink / raw)
  To: Andrew Morton, Linus Torvalds; +Cc: ocfs2-devel, linux-kernel

Please pull from 'upstream-linus' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2.git

to receive the following updates:

 fs/ocfs2/dlm/dlmast.c      |   12 
 fs/ocfs2/dlm/dlmcommon.h   |   63 ++++
 fs/ocfs2/dlm/dlmconvert.c  |   24 +
 fs/ocfs2/dlm/dlmdebug.c    |    6 
 fs/ocfs2/dlm/dlmdebug.h    |   30 --
 fs/ocfs2/dlm/dlmdomain.c   |  101 +++++--
 fs/ocfs2/dlm/dlmfs.c       |    6 
 fs/ocfs2/dlm/dlmlock.c     |   68 ++++-
 fs/ocfs2/dlm/dlmmaster.c   |  448 +++++++++++++++++++++++-----------
 fs/ocfs2/dlm/dlmrecovery.c |  580 ++++++++++++++++++++++++++++++++++-----------
 fs/ocfs2/dlm/dlmthread.c   |   68 ++++-
 fs/ocfs2/dlm/dlmunlock.c   |   10 
 fs/ocfs2/dlm/userdlm.c     |    2 
 13 files changed, 1045 insertions(+), 373 deletions(-)

Adrian Bunk:
      fs/ocfs2/dlm/: cleanups

Daniel Phillips:
      Clean up ocfs2 hash probe and make it faster
      ocfs2: allocate lockres hash pages in an array

Joel Becker:
      ocfs2: Alloc at least a page for the DLM hash

Kurt Hackel:
      ocfs2: add a small delay after a failed migration
      ocfs2: recheck lockres master before sending an unlock request.
      ocfs2: fix inverted logic in dlm_is_node_dead
      ocfs2: Fix empty lvb check
      ocfs2: Better tracking for recovery state changes
      ocfs2: only recover one dead node at a time
      ocfs2: handle network errors during recovery
      ocfs2: clean up recovery related messages
      ocfs2: better mle debugging
      ocfs2: mle ref counting fixes
      ocfs2: detach mle from heartbeat events
      ocfs2: properly initialize the mle structure
      ocfs2: take mle reference during migration
      ocfs2: allow for an assert message during lock mastery
      ocfs2: mle ref count debugging
      ocfs2: dump lockres info before we BUG() on a bad reference
      ocfs2: better error handling during assert master message
      ocfs2: dlm recovery / lockres reference count fix
      ocfs2: make dlm recovery finalization 2 stage
      ocfs2: dump mismatching migrated lvbs before BUG()
      ocfs2: purge lockres' sooner
      ocfs2: do not send master requests to localhost
      ocfs2: update lvb immediately during recovery
      ocfs2: gracefully handle stale create_lock messages.
      ocfs2: teach dlm_restart_lock_mastery() to wait on recovery
      ocfs2: give the dlm dirty list a reference on the lockres
      ocfs2: have dlm_pre_master_reco_lockres() ignore dead nodes
      ocfs2: increase backoff before waiting for recovery
      ocfs2: do not unconditionally purge the lockres in dlmlock_remote()
      ocfs2: temporarily disable automatic lock migration
      ocfs2: pending mastery asserts and migrations should block each other
      ocfs2: special case recovery lock in dlmlock_remote()
      ocfs2: dlm_remaster_locks() should never exit without completing
      ocfs2: remove unneccesary spin_unlock() in dlm_remaster_locks()
      ocfs2: continue recovery when a dead node is encountered
      ocfs2: wait for recovery when starting lock mastery
      ocfs2: use GFP_NOFS in some dlm operations
      ocfs2: use cond_resched() in dlm_thread()
      ocfs2: retry operations when a lock is marked in recovery
      ocfs2: mlog in dlm_convert_lock_handler() should be ML_ERROR
      ocfs2: display message before waiting for recovery to complete
      ocfs2: tune down some noisy messages during dlm recovery
      ocfs2: fix incorrect error returns
      ocfs2: move dlm work to a private work queue
      ocfs2: remove whitespace in dlmunlock.c

Mark Fasheh:
      ocfs2: move lockres qstr next to hlist_node structure
      ocfs2: calculate lockid hash values outside of the spinlock
      ocfs2: inline dlm_lockres_get()
      ocfs2: silence a compile warning in dlm_alloc_pagevec()
      ocfs2: do LVB puts in place
      ocfs2: dlm_print_one_mle() needs to be defined
      ocfs2: fix compiler warnings in dlm_convert_lock_handler()

diff --git a/fs/ocfs2/dlm/dlmast.c b/fs/ocfs2/dlm/dlmast.c
index 355593d..90ce915 100644
--- a/fs/ocfs2/dlm/dlmast.c
+++ b/fs/ocfs2/dlm/dlmast.c
@@ -197,12 +197,14 @@ static void dlm_update_lvb(struct dlm_ct
 				  lock->ml.node == dlm->node_num ? "master" :
 				  "remote");
 			memcpy(lksb->lvb, res->lvb, DLM_LVB_LEN);
-		} else if (lksb->flags & DLM_LKSB_PUT_LVB) {
-			mlog(0, "setting lvb from lockres for %s node\n",
-				  lock->ml.node == dlm->node_num ? "master" :
-				  "remote");
-			memcpy(res->lvb, lksb->lvb, DLM_LVB_LEN);
 		}
+		/* Do nothing for lvb put requests - they should be done in
+ 		 * place when the lock is downconverted - otherwise we risk
+ 		 * racing gets and puts which could result in old lvb data
+ 		 * being propagated. We leave the put flag set and clear it
+ 		 * here. In the future we might want to clear it at the time
+ 		 * the put is actually done.
+		 */
 		spin_unlock(&res->spinlock);
 	}
 
diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h
index 88cc43d..9bdc9cf 100644
--- a/fs/ocfs2/dlm/dlmcommon.h
+++ b/fs/ocfs2/dlm/dlmcommon.h
@@ -37,7 +37,17 @@ #define DLM_LOCK_RES_OWNER_UNKNOWN     O
 #define DLM_THREAD_SHUFFLE_INTERVAL    5     // flush everything every 5 passes
 #define DLM_THREAD_MS                  200   // flush at least every 200 ms
 
-#define DLM_HASH_BUCKETS     (PAGE_SIZE / sizeof(struct hlist_head))
+#define DLM_HASH_SIZE_DEFAULT	(1 << 14)
+#if DLM_HASH_SIZE_DEFAULT < PAGE_SIZE
+# define DLM_HASH_PAGES		1
+#else
+# define DLM_HASH_PAGES		(DLM_HASH_SIZE_DEFAULT / PAGE_SIZE)
+#endif
+#define DLM_BUCKETS_PER_PAGE	(PAGE_SIZE / sizeof(struct hlist_head))
+#define DLM_HASH_BUCKETS	(DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE)
+
+/* Intended to make it easier for us to switch out hash functions */
+#define dlm_lockid_hash(_n, _l) full_name_hash(_n, _l)
 
 enum dlm_ast_type {
 	DLM_AST = 0,
@@ -61,7 +71,8 @@ static inline int dlm_is_recovery_lock(c
 	return 0;
 }
 
-#define DLM_RECO_STATE_ACTIVE  0x0001
+#define DLM_RECO_STATE_ACTIVE    0x0001
+#define DLM_RECO_STATE_FINALIZE  0x0002
 
 struct dlm_recovery_ctxt
 {
@@ -85,7 +96,7 @@ enum dlm_ctxt_state {
 struct dlm_ctxt
 {
 	struct list_head list;
-	struct hlist_head *lockres_hash;
+	struct hlist_head **lockres_hash;
 	struct list_head dirty_list;
 	struct list_head purge_list;
 	struct list_head pending_asts;
@@ -120,6 +131,7 @@ struct dlm_ctxt
 	struct o2hb_callback_func dlm_hb_down;
 	struct task_struct *dlm_thread_task;
 	struct task_struct *dlm_reco_thread_task;
+	struct workqueue_struct *dlm_worker;
 	wait_queue_head_t dlm_thread_wq;
 	wait_queue_head_t dlm_reco_thread_wq;
 	wait_queue_head_t ast_wq;
@@ -132,6 +144,11 @@ struct dlm_ctxt
 	struct list_head	dlm_eviction_callbacks;
 };
 
+static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i)
+{
+	return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE);
+}
+
 /* these keventd work queue items are for less-frequently
  * called functions that cannot be directly called from the
  * net message handlers for some reason, usually because
@@ -216,20 +233,29 @@ struct dlm_lock_resource
 	/* WARNING: Please see the comment in dlm_init_lockres before
 	 * adding fields here. */
 	struct hlist_node hash_node;
+	struct qstr lockname;
 	struct kref      refs;
 
-	/* please keep these next 3 in this order
-	 * some funcs want to iterate over all lists */
+	/*
+	 * Please keep granted, converting, and blocked in this order,
+	 * as some funcs want to iterate over all lists.
+	 *
+	 * All four lists are protected by the hash's reference.
+	 */
 	struct list_head granted;
 	struct list_head converting;
 	struct list_head blocked;
+	struct list_head purge;
 
+	/*
+	 * These two lists require you to hold an additional reference
+	 * while they are on the list.
+	 */
 	struct list_head dirty;
 	struct list_head recovering; // dlm_recovery_ctxt.resources list
 
 	/* unused lock resources have their last_used stamped and are
 	 * put on a list for the dlm thread to run. */
-	struct list_head purge;
 	unsigned long    last_used;
 
 	unsigned migration_pending:1;
@@ -238,7 +264,6 @@ struct dlm_lock_resource
 	wait_queue_head_t wq;
 	u8  owner;              //node which owns the lock resource, or unknown
 	u16 state;
-	struct qstr lockname;
 	char lvb[DLM_LVB_LEN];
 };
 
@@ -300,6 +325,15 @@ enum dlm_lockres_list {
 	DLM_BLOCKED_LIST
 };
 
+static inline int dlm_lvb_is_empty(char *lvb)
+{
+	int i;
+	for (i=0; i<DLM_LVB_LEN; i++)
+		if (lvb[i])
+			return 0;
+	return 1;
+}
+
 static inline struct list_head *
 dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx)
 {
@@ -609,7 +643,8 @@ struct dlm_finalize_reco
 {
 	u8 node_idx;
 	u8 dead_node;
-	__be16 pad1;
+	u8 flags;
+	u8 pad1;
 	__be32 pad2;
 };
 
@@ -676,6 +711,7 @@ void dlm_wait_for_recovery(struct dlm_ct
 void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
 int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node);
 int dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout);
+int dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout);
 
 void dlm_put(struct dlm_ctxt *dlm);
 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm);
@@ -687,14 +723,20 @@ void dlm_lockres_calc_usage(struct dlm_c
 			    struct dlm_lock_resource *res);
 void dlm_purge_lockres(struct dlm_ctxt *dlm,
 		       struct dlm_lock_resource *lockres);
-void dlm_lockres_get(struct dlm_lock_resource *res);
+static inline void dlm_lockres_get(struct dlm_lock_resource *res)
+{
+	/* This is called on every lookup, so it might be worth
+	 * inlining. */
+	kref_get(&res->refs);
+}
 void dlm_lockres_put(struct dlm_lock_resource *res);
 void __dlm_unhash_lockres(struct dlm_lock_resource *res);
 void __dlm_insert_lockres(struct dlm_ctxt *dlm,
 			  struct dlm_lock_resource *res);
 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
 						const char *name,
-						unsigned int len);
+						unsigned int len,
+						unsigned int hash);
 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
 					      const char *name,
 					      unsigned int len);
@@ -819,6 +861,7 @@ void dlm_clean_master_list(struct dlm_ct
 			   u8 dead_node);
 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock);
 
+int __dlm_lockres_unused(struct dlm_lock_resource *res);
 
 static inline const char * dlm_lock_mode_name(int mode)
 {
diff --git a/fs/ocfs2/dlm/dlmconvert.c b/fs/ocfs2/dlm/dlmconvert.c
index 8285228..d555dec 100644
--- a/fs/ocfs2/dlm/dlmconvert.c
+++ b/fs/ocfs2/dlm/dlmconvert.c
@@ -214,6 +214,9 @@ grant:
 	if (lock->ml.node == dlm->node_num)
 		mlog(0, "doing in-place convert for nonlocal lock\n");
 	lock->ml.type = type;
+	if (lock->lksb->flags & DLM_LKSB_PUT_LVB)
+		memcpy(res->lvb, lock->lksb->lvb, DLM_LVB_LEN);
+
 	status = DLM_NORMAL;
 	*call_ast = 1;
 	goto unlock_exit;
@@ -464,6 +467,12 @@ int dlm_convert_lock_handler(struct o2ne
 	}
 
 	spin_lock(&res->spinlock);
+	status = __dlm_lockres_state_to_status(res);
+	if (status != DLM_NORMAL) {
+		spin_unlock(&res->spinlock);
+		dlm_error(status);
+		goto leave;
+	}
 	list_for_each(iter, &res->granted) {
 		lock = list_entry(iter, struct dlm_lock, list);
 		if (lock->ml.cookie == cnv->cookie &&
@@ -473,6 +482,21 @@ int dlm_convert_lock_handler(struct o2ne
 		}
 		lock = NULL;
 	}
+	if (!lock) {
+		__dlm_print_one_lock_resource(res);
+		list_for_each(iter, &res->granted) {
+			lock = list_entry(iter, struct dlm_lock, list);
+			if (lock->ml.node == cnv->node_idx) {
+				mlog(ML_ERROR, "There is something here "
+				     "for node %u, lock->ml.cookie=%llu, "
+				     "cnv->cookie=%llu\n", cnv->node_idx,
+				     (unsigned long long)lock->ml.cookie,
+				     (unsigned long long)cnv->cookie);
+				break;
+			}
+		}
+		lock = NULL;
+	}
 	spin_unlock(&res->spinlock);
 	if (!lock) {
 		status = DLM_IVLOCKID;
diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
index c7eae5d..3f6c8d8 100644
--- a/fs/ocfs2/dlm/dlmdebug.c
+++ b/fs/ocfs2/dlm/dlmdebug.c
@@ -37,10 +37,8 @@ #include "cluster/tcp.h"
 
 #include "dlmapi.h"
 #include "dlmcommon.h"
-#include "dlmdebug.h"
 
 #include "dlmdomain.h"
-#include "dlmdebug.h"
 
 #define MLOG_MASK_PREFIX ML_DLM
 #include "cluster/masklog.h"
@@ -120,6 +118,7 @@ void dlm_print_one_lock(struct dlm_lock 
 }
 EXPORT_SYMBOL_GPL(dlm_print_one_lock);
 
+#if 0
 void dlm_dump_lock_resources(struct dlm_ctxt *dlm)
 {
 	struct dlm_lock_resource *res;
@@ -136,12 +135,13 @@ void dlm_dump_lock_resources(struct dlm_
 
 	spin_lock(&dlm->spinlock);
 	for (i=0; i<DLM_HASH_BUCKETS; i++) {
-		bucket = &(dlm->lockres_hash[i]);
+		bucket = dlm_lockres_hash(dlm, i);
 		hlist_for_each_entry(res, iter, bucket, hash_node)
 			dlm_print_one_lock_resource(res);
 	}
 	spin_unlock(&dlm->spinlock);
 }
+#endif  /*  0  */
 
 static const char *dlm_errnames[] = {
 	[DLM_NORMAL] =			"DLM_NORMAL",
diff --git a/fs/ocfs2/dlm/dlmdebug.h b/fs/ocfs2/dlm/dlmdebug.h
deleted file mode 100644
index 6858510..0000000
--- a/fs/ocfs2/dlm/dlmdebug.h
+++ /dev/null
@@ -1,30 +0,0 @@
-/* -*- mode: c; c-basic-offset: 8; -*-
- * vim: noexpandtab sw=8 ts=8 sts=0:
- *
- * dlmdebug.h
- *
- * Copyright (C) 2004 Oracle.  All rights reserved.
- *
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public
- * License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * General Public License for more details.
- *
- * You should have received a copy of the GNU General Public
- * License along with this program; if not, write to the
- * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
- * Boston, MA 021110-1307, USA.
- *
- */
-
-#ifndef DLMDEBUG_H
-#define DLMDEBUG_H
-
-void dlm_dump_lock_resources(struct dlm_ctxt *dlm);
-
-#endif
diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index 8f3a9e3..ba27c5c 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -41,7 +41,6 @@ #include "cluster/tcp.h"
 #include "dlmapi.h"
 #include "dlmcommon.h"
 
-#include "dlmdebug.h"
 #include "dlmdomain.h"
 
 #include "dlmver.h"
@@ -49,6 +48,33 @@ #include "dlmver.h"
 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
 #include "cluster/masklog.h"
 
+static void dlm_free_pagevec(void **vec, int pages)
+{
+	while (pages--)
+		free_page((unsigned long)vec[pages]);
+	kfree(vec);
+}
+
+static void **dlm_alloc_pagevec(int pages)
+{
+	void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
+	int i;
+
+	if (!vec)
+		return NULL;
+
+	for (i = 0; i < pages; i++)
+		if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
+			goto out_free;
+
+	mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
+	     pages, DLM_HASH_PAGES, (unsigned long)DLM_BUCKETS_PER_PAGE);
+	return vec;
+out_free:
+	dlm_free_pagevec(vec, i);
+	return NULL;
+}
+
 /*
  *
  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
@@ -90,8 +116,7 @@ void __dlm_insert_lockres(struct dlm_ctx
 	assert_spin_locked(&dlm->spinlock);
 
 	q = &res->lockname;
-	q->hash = full_name_hash(q->name, q->len);
-	bucket = &(dlm->lockres_hash[q->hash % DLM_HASH_BUCKETS]);
+	bucket = dlm_lockres_hash(dlm, q->hash);
 
 	/* get a reference for our hashtable */
 	dlm_lockres_get(res);
@@ -100,34 +125,32 @@ void __dlm_insert_lockres(struct dlm_ctx
 }
 
 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
-					 const char *name,
-					 unsigned int len)
+						const char *name,
+						unsigned int len,
+						unsigned int hash)
 {
-	unsigned int hash;
-	struct hlist_node *iter;
-	struct dlm_lock_resource *tmpres=NULL;
 	struct hlist_head *bucket;
+	struct hlist_node *list;
 
 	mlog_entry("%.*s\n", len, name);
 
 	assert_spin_locked(&dlm->spinlock);
 
-	hash = full_name_hash(name, len);
-
-	bucket = &(dlm->lockres_hash[hash % DLM_HASH_BUCKETS]);
-
-	/* check for pre-existing lock */
-	hlist_for_each(iter, bucket) {
-		tmpres = hlist_entry(iter, struct dlm_lock_resource, hash_node);
-		if (tmpres->lockname.len == len &&
-		    memcmp(tmpres->lockname.name, name, len) == 0) {
-			dlm_lockres_get(tmpres);
-			break;
-		}
+	bucket = dlm_lockres_hash(dlm, hash);
 
-		tmpres = NULL;
+	hlist_for_each(list, bucket) {
+		struct dlm_lock_resource *res = hlist_entry(list,
+			struct dlm_lock_resource, hash_node);
+		if (res->lockname.name[0] != name[0])
+			continue;
+		if (unlikely(res->lockname.len != len))
+			continue;
+		if (memcmp(res->lockname.name + 1, name + 1, len - 1))
+			continue;
+		dlm_lockres_get(res);
+		return res;
 	}
-	return tmpres;
+	return NULL;
 }
 
 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
@@ -135,9 +158,10 @@ struct dlm_lock_resource * dlm_lookup_lo
 				    unsigned int len)
 {
 	struct dlm_lock_resource *res;
+	unsigned int hash = dlm_lockid_hash(name, len);
 
 	spin_lock(&dlm->spinlock);
-	res = __dlm_lookup_lockres(dlm, name, len);
+	res = __dlm_lookup_lockres(dlm, name, len, hash);
 	spin_unlock(&dlm->spinlock);
 	return res;
 }
@@ -194,7 +218,7 @@ static int dlm_wait_on_domain_helper(con
 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
 {
 	if (dlm->lockres_hash)
-		free_page((unsigned long) dlm->lockres_hash);
+		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
 
 	if (dlm->name)
 		kfree(dlm->name);
@@ -278,11 +302,21 @@ int dlm_domain_fully_joined(struct dlm_c
 	return ret;
 }
 
+static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
+{
+	if (dlm->dlm_worker) {
+		flush_workqueue(dlm->dlm_worker);
+		destroy_workqueue(dlm->dlm_worker);
+		dlm->dlm_worker = NULL;
+	}
+}
+
 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
 {
 	dlm_unregister_domain_handlers(dlm);
 	dlm_complete_thread(dlm);
 	dlm_complete_recovery_thread(dlm);
+	dlm_destroy_dlm_worker(dlm);
 
 	/* We've left the domain. Now we can take ourselves out of the
 	 * list and allow the kref stuff to help us free the
@@ -304,8 +338,8 @@ static void dlm_migrate_all_locks(struct
 restart:
 	spin_lock(&dlm->spinlock);
 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
-		while (!hlist_empty(&dlm->lockres_hash[i])) {
-			res = hlist_entry(dlm->lockres_hash[i].first,
+		while (!hlist_empty(dlm_lockres_hash(dlm, i))) {
+			res = hlist_entry(dlm_lockres_hash(dlm, i)->first,
 					  struct dlm_lock_resource, hash_node);
 			/* need reference when manually grabbing lockres */
 			dlm_lockres_get(res);
@@ -1126,6 +1160,13 @@ static int dlm_join_domain(struct dlm_ct
 		goto bail;
 	}
 
+	dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
+	if (!dlm->dlm_worker) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
 	do {
 		unsigned int backoff;
 		status = dlm_try_to_join_domain(dlm);
@@ -1166,6 +1207,7 @@ bail:
 		dlm_unregister_domain_handlers(dlm);
 		dlm_complete_thread(dlm);
 		dlm_complete_recovery_thread(dlm);
+		dlm_destroy_dlm_worker(dlm);
 	}
 
 	return status;
@@ -1191,7 +1233,7 @@ static struct dlm_ctxt *dlm_alloc_ctxt(c
 		goto leave;
 	}
 
-	dlm->lockres_hash = (struct hlist_head *) __get_free_page(GFP_KERNEL);
+	dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
 	if (!dlm->lockres_hash) {
 		mlog_errno(-ENOMEM);
 		kfree(dlm->name);
@@ -1200,8 +1242,8 @@ static struct dlm_ctxt *dlm_alloc_ctxt(c
 		goto leave;
 	}
 
-	for (i=0; i<DLM_HASH_BUCKETS; i++)
-		INIT_HLIST_HEAD(&dlm->lockres_hash[i]);
+	for (i = 0; i < DLM_HASH_BUCKETS; i++)
+		INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
 
 	strcpy(dlm->name, domain);
 	dlm->key = key;
@@ -1231,6 +1273,7 @@ static struct dlm_ctxt *dlm_alloc_ctxt(c
 
 	dlm->dlm_thread_task = NULL;
 	dlm->dlm_reco_thread_task = NULL;
+	dlm->dlm_worker = NULL;
 	init_waitqueue_head(&dlm->dlm_thread_wq);
 	init_waitqueue_head(&dlm->dlm_reco_thread_wq);
 	init_waitqueue_head(&dlm->reco.event);
diff --git a/fs/ocfs2/dlm/dlmfs.c b/fs/ocfs2/dlm/dlmfs.c
index 7e88e24..bb2c202 100644
--- a/fs/ocfs2/dlm/dlmfs.c
+++ b/fs/ocfs2/dlm/dlmfs.c
@@ -116,7 +116,7 @@ static int dlmfs_file_open(struct inode 
 	 * doesn't make sense for LVB writes. */
 	file->f_flags &= ~O_APPEND;
 
-	fp = kmalloc(sizeof(*fp), GFP_KERNEL);
+	fp = kmalloc(sizeof(*fp), GFP_NOFS);
 	if (!fp) {
 		status = -ENOMEM;
 		goto bail;
@@ -196,7 +196,7 @@ static ssize_t dlmfs_file_read(struct fi
 	else
 		readlen = count - *ppos;
 
-	lvb_buf = kmalloc(readlen, GFP_KERNEL);
+	lvb_buf = kmalloc(readlen, GFP_NOFS);
 	if (!lvb_buf)
 		return -ENOMEM;
 
@@ -240,7 +240,7 @@ static ssize_t dlmfs_file_write(struct f
 	else
 		writelen = count - *ppos;
 
-	lvb_buf = kmalloc(writelen, GFP_KERNEL);
+	lvb_buf = kmalloc(writelen, GFP_NOFS);
 	if (!lvb_buf)
 		return -ENOMEM;
 
diff --git a/fs/ocfs2/dlm/dlmlock.c b/fs/ocfs2/dlm/dlmlock.c
index 6fea283..a9fc9a6 100644
--- a/fs/ocfs2/dlm/dlmlock.c
+++ b/fs/ocfs2/dlm/dlmlock.c
@@ -201,6 +201,7 @@ static enum dlm_status dlmlock_remote(st
 				      struct dlm_lock *lock, int flags)
 {
 	enum dlm_status status = DLM_DENIED;
+	int lockres_changed = 1;
 
 	mlog_entry("type=%d\n", lock->ml.type);
 	mlog(0, "lockres %.*s, flags = 0x%x\n", res->lockname.len,
@@ -226,8 +227,25 @@ static enum dlm_status dlmlock_remote(st
 	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
 	lock->lock_pending = 0;
 	if (status != DLM_NORMAL) {
-		if (status != DLM_NOTQUEUED)
+		if (status == DLM_RECOVERING &&
+		    dlm_is_recovery_lock(res->lockname.name,
+					 res->lockname.len)) {
+			/* recovery lock was mastered by dead node.
+			 * we need to have calc_usage shoot down this
+			 * lockres and completely remaster it. */
+			mlog(0, "%s: recovery lock was owned by "
+			     "dead node %u, remaster it now.\n",
+			     dlm->name, res->owner);
+		} else if (status != DLM_NOTQUEUED) {
+			/*
+			 * DO NOT call calc_usage, as this would unhash
+			 * the remote lockres before we ever get to use
+			 * it.  treat as if we never made any change to
+			 * the lockres.
+			 */
+			lockres_changed = 0;
 			dlm_error(status);
+		}
 		dlm_revert_pending_lock(res, lock);
 		dlm_lock_put(lock);
 	} else if (dlm_is_recovery_lock(res->lockname.name, 
@@ -244,7 +262,8 @@ static enum dlm_status dlmlock_remote(st
 	}
 	spin_unlock(&res->spinlock);
 
-	dlm_lockres_calc_usage(dlm, res);
+	if (lockres_changed)
+		dlm_lockres_calc_usage(dlm, res);
 
 	wake_up(&res->wq);
 	return status;
@@ -281,6 +300,14 @@ static enum dlm_status dlm_send_remote_l
 	if (tmpret >= 0) {
 		// successfully sent and received
 		ret = status;  // this is already a dlm_status
+		if (ret == DLM_REJECTED) {
+			mlog(ML_ERROR, "%s:%.*s: BUG.  this is a stale lockres "
+			     "no longer owned by %u.  that node is coming back "
+			     "up currently.\n", dlm->name, create.namelen,
+			     create.name, res->owner);
+			dlm_print_one_lock_resource(res);
+			BUG();
+		}
 	} else {
 		mlog_errno(tmpret);
 		if (dlm_is_host_down(tmpret)) {
@@ -382,13 +409,13 @@ struct dlm_lock * dlm_new_lock(int type,
 	struct dlm_lock *lock;
 	int kernel_allocated = 0;
 
-	lock = kcalloc(1, sizeof(*lock), GFP_KERNEL);
+	lock = kcalloc(1, sizeof(*lock), GFP_NOFS);
 	if (!lock)
 		return NULL;
 
 	if (!lksb) {
 		/* zero memory only if kernel-allocated */
-		lksb = kcalloc(1, sizeof(*lksb), GFP_KERNEL);
+		lksb = kcalloc(1, sizeof(*lksb), GFP_NOFS);
 		if (!lksb) {
 			kfree(lock);
 			return NULL;
@@ -429,11 +456,16 @@ int dlm_create_lock_handler(struct o2net
 	if (!dlm_grab(dlm))
 		return DLM_REJECTED;
 
-	mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
-			"Domain %s not fully joined!\n", dlm->name);
-
 	name = create->name;
 	namelen = create->namelen;
+	status = DLM_REJECTED;
+	if (!dlm_domain_fully_joined(dlm)) {
+		mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
+		     "sending a create_lock message for lock %.*s!\n",
+		     dlm->name, create->node_idx, namelen, name);
+		dlm_error(status);
+		goto leave;
+	}
 
 	status = DLM_IVBUFLEN;
 	if (namelen > DLM_LOCKID_NAME_MAX) {
@@ -669,18 +701,22 @@ retry_lock:
 			msleep(100);
 			/* no waiting for dlm_reco_thread */
 			if (recovery) {
-				if (status == DLM_RECOVERING) {
-					mlog(0, "%s: got RECOVERING "
-					     "for $REOCVERY lock, master "
-					     "was %u\n", dlm->name, 
-					     res->owner);
-					dlm_wait_for_node_death(dlm, res->owner, 
-							DLM_NODE_DEATH_WAIT_MAX);
-				}
+				if (status != DLM_RECOVERING)
+					goto retry_lock;
+
+				mlog(0, "%s: got RECOVERING "
+				     "for $RECOVERY lock, master "
+				     "was %u\n", dlm->name,
+				     res->owner);
+				/* wait to see the node go down, then
+				 * drop down and allow the lockres to
+				 * get cleaned up.  need to remaster. */
+				dlm_wait_for_node_death(dlm, res->owner,
+						DLM_NODE_DEATH_WAIT_MAX);
 			} else {
 				dlm_wait_for_recovery(dlm);
+				goto retry_lock;
 			}
-			goto retry_lock;
 		}
 
 		if (status != DLM_NORMAL) {
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index 940be4c..1b8346d 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -47,7 +47,6 @@ #include "cluster/tcp.h"
 
 #include "dlmapi.h"
 #include "dlmcommon.h"
-#include "dlmdebug.h"
 #include "dlmdomain.h"
 
 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_MASTER)
@@ -74,6 +73,7 @@ struct dlm_master_list_entry
 	wait_queue_head_t wq;
 	atomic_t woken;
 	struct kref mle_refs;
+	int inuse;
 	unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
 	unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
 	unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
@@ -127,18 +127,30 @@ static inline int dlm_mle_equal(struct d
 	return 1;
 }
 
-#if 0
-/* Code here is included but defined out as it aids debugging */
+#define dlm_print_nodemap(m)  _dlm_print_nodemap(m,#m)
+static void _dlm_print_nodemap(unsigned long *map, const char *mapname)
+{
+	int i;
+	printk("%s=[ ", mapname);
+	for (i=0; i<O2NM_MAX_NODES; i++)
+		if (test_bit(i, map))
+			printk("%d ", i);
+	printk("]");
+}
 
-void dlm_print_one_mle(struct dlm_master_list_entry *mle)
+static void dlm_print_one_mle(struct dlm_master_list_entry *mle)
 {
-	int i = 0, refs;
+	int refs;
 	char *type;
 	char attached;
 	u8 master;
 	unsigned int namelen;
 	const char *name;
 	struct kref *k;
+	unsigned long *maybe = mle->maybe_map,
+		      *vote = mle->vote_map,
+		      *resp = mle->response_map,
+		      *node = mle->node_map;
 
 	k = &mle->mle_refs;
 	if (mle->type == DLM_MLE_BLOCK)
@@ -159,18 +171,29 @@ void dlm_print_one_mle(struct dlm_master
 		name = mle->u.res->lockname.name;
 	}
 
-	mlog(ML_NOTICE, "  #%3d: %3s  %3d  %3u   %3u %c    (%d)%.*s\n",
-		  i, type, refs, master, mle->new_master, attached,
-		  namelen, namelen, name);
+	mlog(ML_NOTICE, "%.*s: %3s refs=%3d mas=%3u new=%3u evt=%c inuse=%d ",
+		  namelen, name, type, refs, master, mle->new_master, attached,
+		  mle->inuse);
+	dlm_print_nodemap(maybe);
+	printk(", ");
+	dlm_print_nodemap(vote);
+	printk(", ");
+	dlm_print_nodemap(resp);
+	printk(", ");
+	dlm_print_nodemap(node);
+	printk(", ");
+	printk("\n");
 }
 
+#if 0
+/* Code here is included but defined out as it aids debugging */
+
 static void dlm_dump_mles(struct dlm_ctxt *dlm)
 {
 	struct dlm_master_list_entry *mle;
 	struct list_head *iter;
 	
 	mlog(ML_NOTICE, "dumping all mles for domain %s:\n", dlm->name);
-	mlog(ML_NOTICE, "  ####: type refs owner new events? lockname nodemap votemap respmap maybemap\n");
 	spin_lock(&dlm->master_lock);
 	list_for_each(iter, &dlm->master_list) {
 		mle = list_entry(iter, struct dlm_master_list_entry, list);
@@ -314,6 +337,31 @@ static inline void dlm_mle_detach_hb_eve
 	spin_unlock(&dlm->spinlock);
 }
 
+static void dlm_get_mle_inuse(struct dlm_master_list_entry *mle)
+{
+	struct dlm_ctxt *dlm;
+	dlm = mle->dlm;
+
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&dlm->master_lock);
+	mle->inuse++;
+	kref_get(&mle->mle_refs);
+}
+
+static void dlm_put_mle_inuse(struct dlm_master_list_entry *mle)
+{
+	struct dlm_ctxt *dlm;
+	dlm = mle->dlm;
+
+	spin_lock(&dlm->spinlock);
+	spin_lock(&dlm->master_lock);
+	mle->inuse--;
+	__dlm_put_mle(mle);
+	spin_unlock(&dlm->master_lock);
+	spin_unlock(&dlm->spinlock);
+
+}
+
 /* remove from list and free */
 static void __dlm_put_mle(struct dlm_master_list_entry *mle)
 {
@@ -322,9 +370,14 @@ static void __dlm_put_mle(struct dlm_mas
 
 	assert_spin_locked(&dlm->spinlock);
 	assert_spin_locked(&dlm->master_lock);
-	BUG_ON(!atomic_read(&mle->mle_refs.refcount));
-
-	kref_put(&mle->mle_refs, dlm_mle_release);
+	if (!atomic_read(&mle->mle_refs.refcount)) {
+		/* this may or may not crash, but who cares.
+		 * it's a BUG. */
+		mlog(ML_ERROR, "bad mle: %p\n", mle);
+		dlm_print_one_mle(mle);
+		BUG();
+	} else
+		kref_put(&mle->mle_refs, dlm_mle_release);
 }
 
 
@@ -367,6 +420,7 @@ static void dlm_init_mle(struct dlm_mast
 	memset(mle->response_map, 0, sizeof(mle->response_map));
 	mle->master = O2NM_MAX_NODES;
 	mle->new_master = O2NM_MAX_NODES;
+	mle->inuse = 0;
 
 	if (mle->type == DLM_MLE_MASTER) {
 		BUG_ON(!res);
@@ -564,6 +618,28 @@ static void dlm_lockres_release(struct k
 	mlog(0, "destroying lockres %.*s\n", res->lockname.len,
 	     res->lockname.name);
 
+	if (!hlist_unhashed(&res->hash_node) ||
+	    !list_empty(&res->granted) ||
+	    !list_empty(&res->converting) ||
+	    !list_empty(&res->blocked) ||
+	    !list_empty(&res->dirty) ||
+	    !list_empty(&res->recovering) ||
+	    !list_empty(&res->purge)) {
+		mlog(ML_ERROR,
+		     "Going to BUG for resource %.*s."
+		     "  We're on a list! [%c%c%c%c%c%c%c]\n",
+		     res->lockname.len, res->lockname.name,
+		     !hlist_unhashed(&res->hash_node) ? 'H' : ' ',
+		     !list_empty(&res->granted) ? 'G' : ' ',
+		     !list_empty(&res->converting) ? 'C' : ' ',
+		     !list_empty(&res->blocked) ? 'B' : ' ',
+		     !list_empty(&res->dirty) ? 'D' : ' ',
+		     !list_empty(&res->recovering) ? 'R' : ' ',
+		     !list_empty(&res->purge) ? 'P' : ' ');
+
+		dlm_print_one_lock_resource(res);
+	}
+
 	/* By the time we're ready to blow this guy away, we shouldn't
 	 * be on any lists. */
 	BUG_ON(!hlist_unhashed(&res->hash_node));
@@ -579,11 +655,6 @@ static void dlm_lockres_release(struct k
 	kfree(res);
 }
 
-void dlm_lockres_get(struct dlm_lock_resource *res)
-{
-	kref_get(&res->refs);
-}
-
 void dlm_lockres_put(struct dlm_lock_resource *res)
 {
 	kref_put(&res->refs, dlm_lockres_release);
@@ -603,7 +674,7 @@ static void dlm_init_lockres(struct dlm_
 	memcpy(qname, name, namelen);
 
 	res->lockname.len = namelen;
-	res->lockname.hash = full_name_hash(name, namelen);
+	res->lockname.hash = dlm_lockid_hash(name, namelen);
 
 	init_waitqueue_head(&res->wq);
 	spin_lock_init(&res->spinlock);
@@ -637,11 +708,11 @@ struct dlm_lock_resource *dlm_new_lockre
 {
 	struct dlm_lock_resource *res;
 
-	res = kmalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL);
+	res = kmalloc(sizeof(struct dlm_lock_resource), GFP_NOFS);
 	if (!res)
 		return NULL;
 
-	res->lockname.name = kmalloc(namelen, GFP_KERNEL);
+	res->lockname.name = kmalloc(namelen, GFP_NOFS);
 	if (!res->lockname.name) {
 		kfree(res);
 		return NULL;
@@ -677,19 +748,20 @@ struct dlm_lock_resource * dlm_get_lock_
 	int blocked = 0;
 	int ret, nodenum;
 	struct dlm_node_iter iter;
-	unsigned int namelen;
+	unsigned int namelen, hash;
 	int tries = 0;
 	int bit, wait_on_recovery = 0;
 
 	BUG_ON(!lockid);
 
 	namelen = strlen(lockid);
+	hash = dlm_lockid_hash(lockid, namelen);
 
 	mlog(0, "get lockres %s (len %d)\n", lockid, namelen);
 
 lookup:
 	spin_lock(&dlm->spinlock);
-	tmpres = __dlm_lookup_lockres(dlm, lockid, namelen);
+	tmpres = __dlm_lookup_lockres(dlm, lockid, namelen, hash);
 	if (tmpres) {
 		spin_unlock(&dlm->spinlock);
 		mlog(0, "found in hash!\n");
@@ -704,7 +776,7 @@ lookup:
 		mlog(0, "allocating a new resource\n");
 		/* nothing found and we need to allocate one. */
 		alloc_mle = (struct dlm_master_list_entry *)
-			kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL);
+			kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
 		if (!alloc_mle)
 			goto leave;
 		res = dlm_new_lockres(dlm, lockid, namelen);
@@ -790,10 +862,11 @@ lookup:
 	 * if so, the creator of the BLOCK may try to put the last
 	 * ref at this time in the assert master handler, so we
 	 * need an extra one to keep from a bad ptr deref. */
-	dlm_get_mle(mle);
+	dlm_get_mle_inuse(mle);
 	spin_unlock(&dlm->master_lock);
 	spin_unlock(&dlm->spinlock);
 
+redo_request:
 	while (wait_on_recovery) {
 		/* any cluster changes that occurred after dropping the
 		 * dlm spinlock would be detectable be a change on the mle,
@@ -812,7 +885,7 @@ lookup:
 		} 
 
 		dlm_kick_recovery_thread(dlm);
-		msleep(100);
+		msleep(1000);
 		dlm_wait_for_recovery(dlm);
 
 		spin_lock(&dlm->spinlock);
@@ -825,13 +898,15 @@ lookup:
 		} else
 			wait_on_recovery = 0;
 		spin_unlock(&dlm->spinlock);
+
+		if (wait_on_recovery)
+			dlm_wait_for_node_recovery(dlm, bit, 10000);
 	}
 
 	/* must wait for lock to be mastered elsewhere */
 	if (blocked)
 		goto wait;
 
-redo_request:
 	ret = -EINVAL;
 	dlm_node_iter_init(mle->vote_map, &iter);
 	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
@@ -856,6 +931,7 @@ wait:
 	/* keep going until the response map includes all nodes */
 	ret = dlm_wait_for_lock_mastery(dlm, res, mle, &blocked);
 	if (ret < 0) {
+		wait_on_recovery = 1;
 		mlog(0, "%s:%.*s: node map changed, redo the "
 		     "master request now, blocked=%d\n",
 		     dlm->name, res->lockname.len,
@@ -866,7 +942,7 @@ wait:
 			     dlm->name, res->lockname.len, 
 			     res->lockname.name, blocked);
 			dlm_print_one_lock_resource(res);
-			/* dlm_print_one_mle(mle); */
+			dlm_print_one_mle(mle);
 			tries = 0;
 		}
 		goto redo_request;
@@ -880,7 +956,7 @@ wait:
 	dlm_mle_detach_hb_events(dlm, mle);
 	dlm_put_mle(mle);
 	/* put the extra ref */
-	dlm_put_mle(mle);
+	dlm_put_mle_inuse(mle);
 
 wake_waiters:
 	spin_lock(&res->spinlock);
@@ -921,12 +997,14 @@ recheck:
 		spin_unlock(&res->spinlock);
 		/* this will cause the master to re-assert across
 		 * the whole cluster, freeing up mles */
-		ret = dlm_do_master_request(mle, res->owner);
-		if (ret < 0) {
-			/* give recovery a chance to run */
-			mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret);
-			msleep(500);
-			goto recheck;
+		if (res->owner != dlm->node_num) {
+			ret = dlm_do_master_request(mle, res->owner);
+			if (ret < 0) {
+				/* give recovery a chance to run */
+				mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret);
+				msleep(500);
+				goto recheck;
+			}
 		}
 		ret = 0;
 		goto leave;
@@ -962,6 +1040,12 @@ recheck:
 		     "rechecking now\n", dlm->name, res->lockname.len,
 		     res->lockname.name);
 		goto recheck;
+	} else {
+		if (!voting_done) {
+			mlog(0, "map not changed and voting not done "
+			     "for %s:%.*s\n", dlm->name, res->lockname.len,
+			     res->lockname.name);
+		}
 	}
 
 	if (m != O2NM_MAX_NODES) {
@@ -1129,18 +1213,6 @@ static int dlm_restart_lock_mastery(stru
 			set_bit(node, mle->vote_map);
 		} else {
 			mlog(ML_ERROR, "node down! %d\n", node);
-
-			/* if the node wasn't involved in mastery skip it,
-			 * but clear it out from the maps so that it will
-			 * not affect mastery of this lockres */
-			clear_bit(node, mle->response_map);
-			clear_bit(node, mle->vote_map);
-			if (!test_bit(node, mle->maybe_map))
-				goto next;
-
-			/* if we're already blocked on lock mastery, and the
-			 * dead node wasn't the expected master, or there is
-			 * another node in the maybe_map, keep waiting */
 			if (blocked) {
 				int lowest = find_next_bit(mle->maybe_map,
 						       O2NM_MAX_NODES, 0);
@@ -1148,54 +1220,53 @@ static int dlm_restart_lock_mastery(stru
 				/* act like it was never there */
 				clear_bit(node, mle->maybe_map);
 
-			       	if (node != lowest)
-					goto next;
-
-				mlog(ML_ERROR, "expected master %u died while "
-				     "this node was blocked waiting on it!\n",
-				     node);
-				lowest = find_next_bit(mle->maybe_map,
-						       O2NM_MAX_NODES,
-						       lowest+1);
-				if (lowest < O2NM_MAX_NODES) {
-					mlog(0, "still blocked. waiting "
-					     "on %u now\n", lowest);
-					goto next;
+			       	if (node == lowest) {
+					mlog(0, "expected master %u died"
+					    " while this node was blocked "
+					    "waiting on it!\n", node);
+					lowest = find_next_bit(mle->maybe_map,
+						       	O2NM_MAX_NODES,
+						       	lowest+1);
+					if (lowest < O2NM_MAX_NODES) {
+						mlog(0, "%s:%.*s:still "
+						     "blocked. waiting on %u "
+						     "now\n", dlm->name,
+						     res->lockname.len,
+						     res->lockname.name,
+						     lowest);
+					} else {
+						/* mle is an MLE_BLOCK, but
+						 * there is now nothing left to
+						 * block on.  we need to return
+						 * all the way back out and try
+						 * again with an MLE_MASTER.
+						 * dlm_do_local_recovery_cleanup
+						 * has already run, so the mle
+						 * refcount is ok */
+						mlog(0, "%s:%.*s: no "
+						     "longer blocking. try to "
+						     "master this here\n",
+						     dlm->name,
+						     res->lockname.len,
+						     res->lockname.name);
+						mle->type = DLM_MLE_MASTER;
+						mle->u.res = res;
+					}
 				}
-
-				/* mle is an MLE_BLOCK, but there is now
-				 * nothing left to block on.  we need to return
-				 * all the way back out and try again with
-				 * an MLE_MASTER. dlm_do_local_recovery_cleanup
-				 * has already run, so the mle refcount is ok */
-				mlog(0, "no longer blocking. we can "
-				     "try to master this here\n");
-				mle->type = DLM_MLE_MASTER;
-				memset(mle->maybe_map, 0,
-				       sizeof(mle->maybe_map));
-				memset(mle->response_map, 0,
-				       sizeof(mle->maybe_map));
-				memcpy(mle->vote_map, mle->node_map,
-				       sizeof(mle->node_map));
-				mle->u.res = res;
-				set_bit(dlm->node_num, mle->maybe_map);
-
-				ret = -EAGAIN;
-				goto next;
 			}
 
-			clear_bit(node, mle->maybe_map);
-			if (node > dlm->node_num)
-				goto next;
-
-			mlog(0, "dead node in map!\n");
-			/* yuck. go back and re-contact all nodes
-			 * in the vote_map, removing this node. */
-			memset(mle->response_map, 0,
-			       sizeof(mle->response_map));
+			/* now blank out everything, as if we had never
+			 * contacted anyone */
+			memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
+			memset(mle->response_map, 0, sizeof(mle->response_map));
+			/* reset the vote_map to the current node_map */
+			memcpy(mle->vote_map, mle->node_map,
+			       sizeof(mle->node_map));
+			/* put myself into the maybe map */
+			if (mle->type != DLM_MLE_BLOCK)
+				set_bit(dlm->node_num, mle->maybe_map);
 		}
 		ret = -EAGAIN;
-next:
 		node = dlm_bitmap_diff_iter_next(&bdi, &sc);
 	}
 	return ret;
@@ -1316,7 +1387,7 @@ int dlm_master_request_handler(struct o2
 	struct dlm_master_request *request = (struct dlm_master_request *) msg->buf;
 	struct dlm_master_list_entry *mle = NULL, *tmpmle = NULL;
 	char *name;
-	unsigned int namelen;
+	unsigned int namelen, hash;
 	int found, ret;
 	int set_maybe;
 	int dispatch_assert = 0;
@@ -1331,6 +1402,7 @@ int dlm_master_request_handler(struct o2
 
 	name = request->name;
 	namelen = request->namelen;
+	hash = dlm_lockid_hash(name, namelen);
 
 	if (namelen > DLM_LOCKID_NAME_MAX) {
 		response = DLM_IVBUFLEN;
@@ -1339,7 +1411,7 @@ int dlm_master_request_handler(struct o2
 
 way_up_top:
 	spin_lock(&dlm->spinlock);
-	res = __dlm_lookup_lockres(dlm, name, namelen);
+	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
 	if (res) {
 		spin_unlock(&dlm->spinlock);
 
@@ -1459,21 +1531,18 @@ way_up_top:
 			spin_unlock(&dlm->spinlock);
 
 			mle = (struct dlm_master_list_entry *)
-				kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL);
+				kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
 			if (!mle) {
 				response = DLM_MASTER_RESP_ERROR;
 				mlog_errno(-ENOMEM);
 				goto send_response;
 			}
-			spin_lock(&dlm->spinlock);
-			dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL,
-					 name, namelen);
-			spin_unlock(&dlm->spinlock);
 			goto way_up_top;
 		}
 
 		// mlog(0, "this is second time thru, already allocated, "
 		// "add the block.\n");
+		dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL, name, namelen);
 		set_bit(request->node_idx, mle->maybe_map);
 		list_add(&mle->list, &dlm->master_list);
 		response = DLM_MASTER_RESP_NO;
@@ -1556,6 +1625,8 @@ again:
 	dlm_node_iter_init(nodemap, &iter);
 	while ((to = dlm_node_iter_next(&iter)) >= 0) {
 		int r = 0;
+		struct dlm_master_list_entry *mle = NULL;
+
 		mlog(0, "sending assert master to %d (%.*s)\n", to,
 		     namelen, lockname);
 		memset(&assert, 0, sizeof(assert));
@@ -1567,20 +1638,28 @@ again:
 		tmpret = o2net_send_message(DLM_ASSERT_MASTER_MSG, dlm->key,
 					    &assert, sizeof(assert), to, &r);
 		if (tmpret < 0) {
-			mlog(ML_ERROR, "assert_master returned %d!\n", tmpret);
+			mlog(0, "assert_master returned %d!\n", tmpret);
 			if (!dlm_is_host_down(tmpret)) {
-				mlog(ML_ERROR, "unhandled error!\n");
+				mlog(ML_ERROR, "unhandled error=%d!\n", tmpret);
 				BUG();
 			}
 			/* a node died.  finish out the rest of the nodes. */
-			mlog(ML_ERROR, "link to %d went down!\n", to);
+			mlog(0, "link to %d went down!\n", to);
 			/* any nonzero status return will do */
 			ret = tmpret;
 		} else if (r < 0) {
 			/* ok, something horribly messed.  kill thyself. */
 			mlog(ML_ERROR,"during assert master of %.*s to %u, "
 			     "got %d.\n", namelen, lockname, to, r);
-			dlm_dump_lock_resources(dlm);
+			spin_lock(&dlm->spinlock);
+			spin_lock(&dlm->master_lock);
+			if (dlm_find_mle(dlm, &mle, (char *)lockname,
+					 namelen)) {
+				dlm_print_one_mle(mle);
+				__dlm_put_mle(mle);
+			}
+			spin_unlock(&dlm->master_lock);
+			spin_unlock(&dlm->spinlock);
 			BUG();
 		} else if (r == EAGAIN) {
 			mlog(0, "%.*s: node %u create mles on other "
@@ -1612,7 +1691,7 @@ int dlm_assert_master_handler(struct o2n
 	struct dlm_assert_master *assert = (struct dlm_assert_master *)msg->buf;
 	struct dlm_lock_resource *res = NULL;
 	char *name;
-	unsigned int namelen;
+	unsigned int namelen, hash;
 	u32 flags;
 	int master_request = 0;
 	int ret = 0;
@@ -1622,6 +1701,7 @@ int dlm_assert_master_handler(struct o2n
 
 	name = assert->name;
 	namelen = assert->namelen;
+	hash = dlm_lockid_hash(name, namelen);
 	flags = be32_to_cpu(assert->flags);
 
 	if (namelen > DLM_LOCKID_NAME_MAX) {
@@ -1646,7 +1726,7 @@ int dlm_assert_master_handler(struct o2n
 		if (bit >= O2NM_MAX_NODES) {
 			/* not necessarily an error, though less likely.
 			 * could be master just re-asserting. */
-			mlog(ML_ERROR, "no bits set in the maybe_map, but %u "
+			mlog(0, "no bits set in the maybe_map, but %u "
 			     "is asserting! (%.*s)\n", assert->node_idx,
 			     namelen, name);
 		} else if (bit != assert->node_idx) {
@@ -1658,19 +1738,36 @@ int dlm_assert_master_handler(struct o2n
 				 * number winning the mastery will respond
 				 * YES to mastery requests, but this node
 				 * had no way of knowing.  let it pass. */
-				mlog(ML_ERROR, "%u is the lowest node, "
+				mlog(0, "%u is the lowest node, "
 				     "%u is asserting. (%.*s)  %u must "
 				     "have begun after %u won.\n", bit,
 				     assert->node_idx, namelen, name, bit,
 				     assert->node_idx);
 			}
 		}
+		if (mle->type == DLM_MLE_MIGRATION) {
+			if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
+				mlog(0, "%s:%.*s: got cleanup assert"
+				     " from %u for migration\n",
+				     dlm->name, namelen, name,
+				     assert->node_idx);
+			} else if (!(flags & DLM_ASSERT_MASTER_FINISH_MIGRATION)) {
+				mlog(0, "%s:%.*s: got unrelated assert"
+				     " from %u for migration, ignoring\n",
+				     dlm->name, namelen, name,
+				     assert->node_idx);
+				__dlm_put_mle(mle);
+				spin_unlock(&dlm->master_lock);
+				spin_unlock(&dlm->spinlock);
+				goto done;
+			}	
+		}
 	}
 	spin_unlock(&dlm->master_lock);
 
 	/* ok everything checks out with the MLE
 	 * now check to see if there is a lockres */
-	res = __dlm_lookup_lockres(dlm, name, namelen);
+	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
 	if (res) {
 		spin_lock(&res->spinlock);
 		if (res->state & DLM_LOCK_RES_RECOVERING)  {
@@ -1679,7 +1776,8 @@ int dlm_assert_master_handler(struct o2n
 			goto kill;
 		}
 		if (!mle) {
-			if (res->owner != assert->node_idx) {
+			if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN &&
+			    res->owner != assert->node_idx) {
 				mlog(ML_ERROR, "assert_master from "
 					  "%u, but current owner is "
 					  "%u! (%.*s)\n",
@@ -1732,6 +1830,7 @@ ok:
 	if (mle) {
 		int extra_ref = 0;
 		int nn = -1;
+		int rr, err = 0;
 		
 		spin_lock(&mle->spinlock);
 		if (mle->type == DLM_MLE_BLOCK || mle->type == DLM_MLE_MIGRATION)
@@ -1751,27 +1850,64 @@ ok:
 		wake_up(&mle->wq);
 		spin_unlock(&mle->spinlock);
 
-		if (mle->type == DLM_MLE_MIGRATION && res) {
-			mlog(0, "finishing off migration of lockres %.*s, "
-			     "from %u to %u\n",
-			       res->lockname.len, res->lockname.name,
-			       dlm->node_num, mle->new_master);
+		if (res) {
 			spin_lock(&res->spinlock);
-			res->state &= ~DLM_LOCK_RES_MIGRATING;
-			dlm_change_lockres_owner(dlm, res, mle->new_master);
-			BUG_ON(res->state & DLM_LOCK_RES_DIRTY);
+			if (mle->type == DLM_MLE_MIGRATION) {
+				mlog(0, "finishing off migration of lockres %.*s, "
+			     		"from %u to %u\n",
+			       		res->lockname.len, res->lockname.name,
+			       		dlm->node_num, mle->new_master);
+				res->state &= ~DLM_LOCK_RES_MIGRATING;
+				dlm_change_lockres_owner(dlm, res, mle->new_master);
+				BUG_ON(res->state & DLM_LOCK_RES_DIRTY);
+			} else {
+				dlm_change_lockres_owner(dlm, res, mle->master);
+			}
 			spin_unlock(&res->spinlock);
 		}
-		/* master is known, detach if not already detached */
-		dlm_mle_detach_hb_events(dlm, mle);
-		dlm_put_mle(mle);
-		
+
+		/* master is known, detach if not already detached.
+		 * ensures that only one assert_master call will happen
+		 * on this mle. */
+		spin_lock(&dlm->spinlock);
+		spin_lock(&dlm->master_lock);
+
+		rr = atomic_read(&mle->mle_refs.refcount);
+		if (mle->inuse > 0) {
+			if (extra_ref && rr < 3)
+				err = 1;
+			else if (!extra_ref && rr < 2)
+				err = 1;
+		} else {
+			if (extra_ref && rr < 2)
+				err = 1;
+			else if (!extra_ref && rr < 1)
+				err = 1;
+		}
+		if (err) {
+			mlog(ML_ERROR, "%s:%.*s: got assert master from %u "
+			     "that will mess up this node, refs=%d, extra=%d, "
+			     "inuse=%d\n", dlm->name, namelen, name,
+			     assert->node_idx, rr, extra_ref, mle->inuse);
+			dlm_print_one_mle(mle);
+		}
+		list_del_init(&mle->list);
+		__dlm_mle_detach_hb_events(dlm, mle);
+		__dlm_put_mle(mle);
 		if (extra_ref) {
 			/* the assert master message now balances the extra
 		 	 * ref given by the master / migration request message.
 		 	 * if this is the last put, it will be removed
 		 	 * from the list. */
-			dlm_put_mle(mle);
+			__dlm_put_mle(mle);
+		}
+		spin_unlock(&dlm->master_lock);
+		spin_unlock(&dlm->spinlock);
+	} else if (res) {
+		if (res->owner != assert->node_idx) {
+			mlog(0, "assert_master from %u, but current "
+			     "owner is %u (%.*s), no mle\n", assert->node_idx,
+			     res->owner, namelen, name);
 		}
 	}
 
@@ -1788,12 +1924,12 @@ done:
 
 kill:
 	/* kill the caller! */
+	mlog(ML_ERROR, "Bad message received from another node.  Dumping state "
+	     "and killing the other node now!  This node is OK and can continue.\n");
+	__dlm_print_one_lock_resource(res);
 	spin_unlock(&res->spinlock);
 	spin_unlock(&dlm->spinlock);
 	dlm_lockres_put(res);
-	mlog(ML_ERROR, "Bad message received from another node.  Dumping state "
-	     "and killing the other node now!  This node is OK and can continue.\n");
-	dlm_dump_lock_resources(dlm);
 	dlm_put(dlm);
 	return -EINVAL;
 }
@@ -1803,7 +1939,7 @@ int dlm_dispatch_assert_master(struct dl
 			       int ignore_higher, u8 request_from, u32 flags)
 {
 	struct dlm_work_item *item;
-	item = kcalloc(1, sizeof(*item), GFP_KERNEL);
+	item = kcalloc(1, sizeof(*item), GFP_NOFS);
 	if (!item)
 		return -ENOMEM;
 
@@ -1825,7 +1961,7 @@ int dlm_dispatch_assert_master(struct dl
 	list_add_tail(&item->list, &dlm->work_list);
 	spin_unlock(&dlm->work_lock);
 
-	schedule_work(&dlm->dispatched_work);
+	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
 	return 0;
 }
 
@@ -1866,6 +2002,23 @@ static void dlm_assert_master_worker(str
 		}
 	}
 
+	/*
+	 * If we're migrating this lock to someone else, we are no
+	 * longer allowed to assert out own mastery.  OTOH, we need to
+	 * prevent migration from starting while we're still asserting
+	 * our dominance.  The reserved ast delays migration.
+	 */
+	spin_lock(&res->spinlock);
+	if (res->state & DLM_LOCK_RES_MIGRATING) {
+		mlog(0, "Someone asked us to assert mastery, but we're "
+		     "in the middle of migration.  Skipping assert, "
+		     "the new master will handle that.\n");
+		spin_unlock(&res->spinlock);
+		goto put;
+	} else
+		__dlm_lockres_reserve_ast(res);
+	spin_unlock(&res->spinlock);
+
 	/* this call now finishes out the nodemap
 	 * even if one or more nodes die */
 	mlog(0, "worker about to master %.*s here, this=%u\n",
@@ -1875,9 +2028,14 @@ static void dlm_assert_master_worker(str
 				   nodemap, flags);
 	if (ret < 0) {
 		/* no need to restart, we are done */
-		mlog_errno(ret);
+		if (!dlm_is_host_down(ret))
+			mlog_errno(ret);
 	}
 
+	/* Ok, we've asserted ourselves.  Let's let migration start. */
+	dlm_lockres_release_ast(dlm, res);
+
+put:
 	dlm_lockres_put(res);
 
 	mlog(0, "finished with dlm_assert_master_worker\n");
@@ -1916,6 +2074,7 @@ static int dlm_pre_master_reco_lockres(s
 				BUG();
 			/* host is down, so answer for that node would be
 			 * DLM_LOCK_RES_OWNER_UNKNOWN.  continue. */
+			ret = 0;
 		}
 
 		if (master != DLM_LOCK_RES_OWNER_UNKNOWN) {
@@ -2016,14 +2175,14 @@ int dlm_migrate_lockres(struct dlm_ctxt 
 	 */
 
 	ret = -ENOMEM;
-	mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_KERNEL);
+	mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_NOFS);
 	if (!mres) {
 		mlog_errno(ret);
 		goto leave;
 	}
 
 	mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache,
-								GFP_KERNEL);
+								GFP_NOFS);
 	if (!mle) {
 		mlog_errno(ret);
 		goto leave;
@@ -2117,7 +2276,7 @@ fail:
 	 * take both dlm->spinlock and dlm->master_lock */
 	spin_lock(&dlm->spinlock);
 	spin_lock(&dlm->master_lock);
-	dlm_get_mle(mle);
+	dlm_get_mle_inuse(mle);
 	spin_unlock(&dlm->master_lock);
 	spin_unlock(&dlm->spinlock);
 
@@ -2134,7 +2293,10 @@ fail:
 		/* migration failed, detach and clean up mle */
 		dlm_mle_detach_hb_events(dlm, mle);
 		dlm_put_mle(mle);
-		dlm_put_mle(mle);
+		dlm_put_mle_inuse(mle);
+		spin_lock(&res->spinlock);
+		res->state &= ~DLM_LOCK_RES_MIGRATING;
+		spin_unlock(&res->spinlock);
 		goto leave;
 	}
 
@@ -2164,8 +2326,8 @@ fail:
 			/* avoid hang during shutdown when migrating lockres 
 			 * to a node which also goes down */
 			if (dlm_is_node_dead(dlm, target)) {
-				mlog(0, "%s:%.*s: expected migration target %u "
-				     "is no longer up.  restarting.\n",
+				mlog(0, "%s:%.*s: expected migration "
+				     "target %u is no longer up, restarting\n",
 				     dlm->name, res->lockname.len,
 				     res->lockname.name, target);
 				ret = -ERESTARTSYS;
@@ -2175,7 +2337,10 @@ fail:
 			/* migration failed, detach and clean up mle */
 			dlm_mle_detach_hb_events(dlm, mle);
 			dlm_put_mle(mle);
-			dlm_put_mle(mle);
+			dlm_put_mle_inuse(mle);
+			spin_lock(&res->spinlock);
+			res->state &= ~DLM_LOCK_RES_MIGRATING;
+			spin_unlock(&res->spinlock);
 			goto leave;
 		}
 		/* TODO: if node died: stop, clean up, return error */
@@ -2191,7 +2356,7 @@ fail:
 
 	/* master is known, detach if not already detached */
 	dlm_mle_detach_hb_events(dlm, mle);
-	dlm_put_mle(mle);
+	dlm_put_mle_inuse(mle);
 	ret = 0;
 
 	dlm_lockres_calc_usage(dlm, res);
@@ -2462,7 +2627,7 @@ int dlm_migrate_request_handler(struct o
 	struct dlm_migrate_request *migrate = (struct dlm_migrate_request *) msg->buf;
 	struct dlm_master_list_entry *mle = NULL, *oldmle = NULL;
 	const char *name;
-	unsigned int namelen;
+	unsigned int namelen, hash;
 	int ret = 0;
 
 	if (!dlm_grab(dlm))
@@ -2470,10 +2635,11 @@ int dlm_migrate_request_handler(struct o
 
 	name = migrate->name;
 	namelen = migrate->namelen;
+	hash = dlm_lockid_hash(name, namelen);
 
 	/* preallocate.. if this fails, abort */
 	mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache,
-							 GFP_KERNEL);
+							 GFP_NOFS);
 
 	if (!mle) {
 		ret = -ENOMEM;
@@ -2482,7 +2648,7 @@ int dlm_migrate_request_handler(struct o
 
 	/* check for pre-existing lock */
 	spin_lock(&dlm->spinlock);
-	res = __dlm_lookup_lockres(dlm, name, namelen);
+	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
 	spin_lock(&dlm->master_lock);
 
 	if (res) {
@@ -2580,6 +2746,7 @@ static int dlm_add_migration_mle(struct 
 			/* remove it from the list so that only one
 			 * mle will be found */
 			list_del_init(&tmp->list);
+			__dlm_mle_detach_hb_events(dlm, mle);
 		}
 		spin_unlock(&tmp->spinlock);
 	}
@@ -2601,6 +2768,7 @@ void dlm_clean_master_list(struct dlm_ct
 	struct list_head *iter, *iter2;
 	struct dlm_master_list_entry *mle;
 	struct dlm_lock_resource *res;
+	unsigned int hash;
 
 	mlog_entry("dlm=%s, dead node=%u\n", dlm->name, dead_node);
 top:
@@ -2640,7 +2808,7 @@ top:
 				 * may result in the mle being unlinked and
 				 * freed, but there may still be a process
 				 * waiting in the dlmlock path which is fine. */
-				mlog(ML_ERROR, "node %u was expected master\n",
+				mlog(0, "node %u was expected master\n",
 				     dead_node);
 				atomic_set(&mle->woken, 1);
 				spin_unlock(&mle->spinlock);
@@ -2673,19 +2841,21 @@ top:
 
 		/* remove from the list early.  NOTE: unlinking
 		 * list_head while in list_for_each_safe */
+		__dlm_mle_detach_hb_events(dlm, mle);
 		spin_lock(&mle->spinlock);
 		list_del_init(&mle->list);
 		atomic_set(&mle->woken, 1);
 		spin_unlock(&mle->spinlock);
 		wake_up(&mle->wq);
 
-		mlog(0, "node %u died during migration from "
-		     "%u to %u!\n", dead_node,
+		mlog(0, "%s: node %u died during migration from "
+		     "%u to %u!\n", dlm->name, dead_node,
 		     mle->master, mle->new_master);
 		/* if there is a lockres associated with this
 	 	 * mle, find it and set its owner to UNKNOWN */
+		hash = dlm_lockid_hash(mle->u.name.name, mle->u.name.len);
 		res = __dlm_lookup_lockres(dlm, mle->u.name.name,
-					mle->u.name.len);
+					   mle->u.name.len, hash);
 		if (res) {
 			/* unfortunately if we hit this rare case, our
 		 	 * lock ordering is messed.  we need to drop
diff --git a/fs/ocfs2/dlm/dlmrecovery.c b/fs/ocfs2/dlm/dlmrecovery.c
index 805cbab..b5078fd 100644
--- a/fs/ocfs2/dlm/dlmrecovery.c
+++ b/fs/ocfs2/dlm/dlmrecovery.c
@@ -115,12 +115,37 @@ static u64 dlm_get_next_mig_cookie(void)
 	return c;
 }
 
+static inline void dlm_set_reco_dead_node(struct dlm_ctxt *dlm,
+					  u8 dead_node)
+{
+	assert_spin_locked(&dlm->spinlock);
+	if (dlm->reco.dead_node != dead_node)
+		mlog(0, "%s: changing dead_node from %u to %u\n",
+		     dlm->name, dlm->reco.dead_node, dead_node);
+	dlm->reco.dead_node = dead_node;
+}
+
+static inline void dlm_set_reco_master(struct dlm_ctxt *dlm,
+				       u8 master)
+{
+	assert_spin_locked(&dlm->spinlock);
+	mlog(0, "%s: changing new_master from %u to %u\n",
+	     dlm->name, dlm->reco.new_master, master);
+	dlm->reco.new_master = master;
+}
+
+static inline void __dlm_reset_recovery(struct dlm_ctxt *dlm)
+{
+	assert_spin_locked(&dlm->spinlock);
+	clear_bit(dlm->reco.dead_node, dlm->recovery_map);
+	dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
+	dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM);
+}
+
 static inline void dlm_reset_recovery(struct dlm_ctxt *dlm)
 {
 	spin_lock(&dlm->spinlock);
-	clear_bit(dlm->reco.dead_node, dlm->recovery_map);
-	dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
-	dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
+	__dlm_reset_recovery(dlm);
 	spin_unlock(&dlm->spinlock);
 }
 
@@ -132,12 +157,21 @@ void dlm_dispatch_work(void *data)
 	struct list_head *iter, *iter2;
 	struct dlm_work_item *item;
 	dlm_workfunc_t *workfunc;
+	int tot=0;
+
+	if (!dlm_joined(dlm))
+		return;
 
 	spin_lock(&dlm->work_lock);
 	list_splice_init(&dlm->work_list, &tmp_list);
 	spin_unlock(&dlm->work_lock);
 
 	list_for_each_safe(iter, iter2, &tmp_list) {
+		tot++;
+	}
+	mlog(0, "%s: work thread has %d work items\n", dlm->name, tot);
+
+	list_for_each_safe(iter, iter2, &tmp_list) {
 		item = list_entry(iter, struct dlm_work_item, list);
 		workfunc = item->func;
 		list_del_init(&item->list);
@@ -220,6 +254,52 @@ void dlm_complete_recovery_thread(struct
  *
  */
 
+static void dlm_print_reco_node_status(struct dlm_ctxt *dlm)
+{
+	struct dlm_reco_node_data *ndata;
+	struct dlm_lock_resource *res;
+
+	mlog(ML_NOTICE, "%s(%d): recovery info, state=%s, dead=%u, master=%u\n",
+	     dlm->name, dlm->dlm_reco_thread_task->pid,
+	     dlm->reco.state & DLM_RECO_STATE_ACTIVE ? "ACTIVE" : "inactive",
+	     dlm->reco.dead_node, dlm->reco.new_master);
+
+	list_for_each_entry(ndata, &dlm->reco.node_data, list) {
+		char *st = "unknown";
+		switch (ndata->state) {
+			case DLM_RECO_NODE_DATA_INIT:
+				st = "init";
+				break;
+			case DLM_RECO_NODE_DATA_REQUESTING:
+				st = "requesting";
+				break;
+			case DLM_RECO_NODE_DATA_DEAD:
+				st = "dead";
+				break;
+			case DLM_RECO_NODE_DATA_RECEIVING:
+				st = "receiving";
+				break;
+			case DLM_RECO_NODE_DATA_REQUESTED:
+				st = "requested";
+				break;
+			case DLM_RECO_NODE_DATA_DONE:
+				st = "done";
+				break;
+			case DLM_RECO_NODE_DATA_FINALIZE_SENT:
+				st = "finalize-sent";
+				break;
+			default:
+				st = "bad";
+				break;
+		}
+		mlog(ML_NOTICE, "%s: reco state, node %u, state=%s\n",
+		     dlm->name, ndata->node_num, st);
+	}
+	list_for_each_entry(res, &dlm->reco.resources, recovering) {
+		mlog(ML_NOTICE, "%s: lockres %.*s on recovering list\n",
+		     dlm->name, res->lockname.len, res->lockname.name);
+	}
+}
 
 #define DLM_RECO_THREAD_TIMEOUT_MS (5 * 1000)
 
@@ -267,11 +347,23 @@ int dlm_is_node_dead(struct dlm_ctxt *dl
 {
 	int dead;
 	spin_lock(&dlm->spinlock);
-	dead = test_bit(node, dlm->domain_map);
+	dead = !test_bit(node, dlm->domain_map);
 	spin_unlock(&dlm->spinlock);
 	return dead;
 }
 
+/* returns true if node is no longer in the domain
+ * could be dead or just not joined */
+static int dlm_is_node_recovered(struct dlm_ctxt *dlm, u8 node)
+{
+	int recovered;
+	spin_lock(&dlm->spinlock);
+	recovered = !test_bit(node, dlm->recovery_map);
+	spin_unlock(&dlm->spinlock);
+	return recovered;
+}
+
+
 int dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout)
 {
 	if (timeout) {
@@ -290,6 +382,24 @@ int dlm_wait_for_node_death(struct dlm_c
 	return 0;
 }
 
+int dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout)
+{
+	if (timeout) {
+		mlog(0, "%s: waiting %dms for notification of "
+		     "recovery of node %u\n", dlm->name, timeout, node);
+		wait_event_timeout(dlm->dlm_reco_thread_wq,
+			   dlm_is_node_recovered(dlm, node),
+			   msecs_to_jiffies(timeout));
+	} else {
+		mlog(0, "%s: waiting indefinitely for notification "
+		     "of recovery of node %u\n", dlm->name, node);
+		wait_event(dlm->dlm_reco_thread_wq,
+			   dlm_is_node_recovered(dlm, node));
+	}
+	/* for now, return 0 */
+	return 0;
+}
+
 /* callers of the top-level api calls (dlmlock/dlmunlock) should
  * block on the dlm->reco.event when recovery is in progress.
  * the dlm recovery thread will set this state when it begins
@@ -308,6 +418,13 @@ static int dlm_in_recovery(struct dlm_ct
 
 void dlm_wait_for_recovery(struct dlm_ctxt *dlm)
 {
+	if (dlm_in_recovery(dlm)) {
+		mlog(0, "%s: reco thread %d in recovery: "
+		     "state=%d, master=%u, dead=%u\n",
+		     dlm->name, dlm->dlm_reco_thread_task->pid,
+		     dlm->reco.state, dlm->reco.new_master,
+		     dlm->reco.dead_node);
+	}
 	wait_event(dlm->reco.event, !dlm_in_recovery(dlm));
 }
 
@@ -341,7 +458,7 @@ static int dlm_do_recovery(struct dlm_ct
 		mlog(0, "new master %u died while recovering %u!\n",
 		     dlm->reco.new_master, dlm->reco.dead_node);
 		/* unset the new_master, leave dead_node */
-		dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
+		dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM);
 	}
 
 	/* select a target to recover */
@@ -350,14 +467,14 @@ static int dlm_do_recovery(struct dlm_ct
 
 		bit = find_next_bit (dlm->recovery_map, O2NM_MAX_NODES+1, 0);
 		if (bit >= O2NM_MAX_NODES || bit < 0)
-			dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
+			dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
 		else
-			dlm->reco.dead_node = bit;
+			dlm_set_reco_dead_node(dlm, bit);
 	} else if (!test_bit(dlm->reco.dead_node, dlm->recovery_map)) {
 		/* BUG? */
 		mlog(ML_ERROR, "dead_node %u no longer in recovery map!\n",
 		     dlm->reco.dead_node);
-		dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
+		dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
 	}
 
 	if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) {
@@ -366,7 +483,8 @@ static int dlm_do_recovery(struct dlm_ct
 		/* return to main thread loop and sleep. */
 		return 0;
 	}
-	mlog(0, "recovery thread found node %u in the recovery map!\n",
+	mlog(0, "%s(%d):recovery thread found node %u in the recovery map!\n",
+	     dlm->name, dlm->dlm_reco_thread_task->pid,
 	     dlm->reco.dead_node);
 	spin_unlock(&dlm->spinlock);
 
@@ -389,8 +507,8 @@ static int dlm_do_recovery(struct dlm_ct
 		}
 		mlog(0, "another node will master this recovery session.\n");
 	}
-	mlog(0, "dlm=%s, new_master=%u, this node=%u, dead_node=%u\n",
-	     dlm->name, dlm->reco.new_master,
+	mlog(0, "dlm=%s (%d), new_master=%u, this node=%u, dead_node=%u\n",
+	     dlm->name, dlm->dlm_reco_thread_task->pid, dlm->reco.new_master,
 	     dlm->node_num, dlm->reco.dead_node);
 
 	/* it is safe to start everything back up here
@@ -402,11 +520,13 @@ static int dlm_do_recovery(struct dlm_ct
 	return 0;
 
 master_here:
-	mlog(0, "mastering recovery of %s:%u here(this=%u)!\n",
+	mlog(0, "(%d) mastering recovery of %s:%u here(this=%u)!\n",
+	     dlm->dlm_reco_thread_task->pid,
 	     dlm->name, dlm->reco.dead_node, dlm->node_num);
 
 	status = dlm_remaster_locks(dlm, dlm->reco.dead_node);
 	if (status < 0) {
+		/* we should never hit this anymore */
 		mlog(ML_ERROR, "error %d remastering locks for node %u, "
 		     "retrying.\n", status, dlm->reco.dead_node);
 		/* yield a bit to allow any final network messages
@@ -433,9 +553,16 @@ static int dlm_remaster_locks(struct dlm
 	int destroy = 0;
 	int pass = 0;
 
-	status = dlm_init_recovery_area(dlm, dead_node);
-	if (status < 0)
-		goto leave;
+	do {
+		/* we have become recovery master.  there is no escaping
+		 * this, so just keep trying until we get it. */
+		status = dlm_init_recovery_area(dlm, dead_node);
+		if (status < 0) {
+			mlog(ML_ERROR, "%s: failed to alloc recovery area, "
+			     "retrying\n", dlm->name);
+			msleep(1000);
+		}
+	} while (status != 0);
 
 	/* safe to access the node data list without a lock, since this
 	 * process is the only one to change the list */
@@ -452,16 +579,36 @@ static int dlm_remaster_locks(struct dlm
 			continue;
 		}
 
-		status = dlm_request_all_locks(dlm, ndata->node_num, dead_node);
-		if (status < 0) {
-			mlog_errno(status);
-			if (dlm_is_host_down(status))
-				ndata->state = DLM_RECO_NODE_DATA_DEAD;
-			else {
-				destroy = 1;
-				goto leave;
+		do {
+			status = dlm_request_all_locks(dlm, ndata->node_num,
+						       dead_node);
+			if (status < 0) {
+				mlog_errno(status);
+				if (dlm_is_host_down(status)) {
+					/* node died, ignore it for recovery */
+					status = 0;
+					ndata->state = DLM_RECO_NODE_DATA_DEAD;
+					/* wait for the domain map to catch up
+					 * with the network state. */
+					wait_event_timeout(dlm->dlm_reco_thread_wq,
+							   dlm_is_node_dead(dlm,
+								ndata->node_num),
+							   msecs_to_jiffies(1000));
+					mlog(0, "waited 1 sec for %u, "
+					     "dead? %s\n", ndata->node_num,
+					     dlm_is_node_dead(dlm, ndata->node_num) ?
+					     "yes" : "no");
+				} else {
+					/* -ENOMEM on the other node */
+					mlog(0, "%s: node %u returned "
+					     "%d during recovery, retrying "
+					     "after a short wait\n",
+					     dlm->name, ndata->node_num,
+					     status);
+					msleep(100);
+				}
 			}
-		}
+		} while (status != 0);
 
 		switch (ndata->state) {
 			case DLM_RECO_NODE_DATA_INIT:
@@ -473,10 +620,9 @@ static int dlm_remaster_locks(struct dlm
 				mlog(0, "node %u died after requesting "
 				     "recovery info for node %u\n",
 				     ndata->node_num, dead_node);
-				// start all over
-				destroy = 1;
-				status = -EAGAIN;
-				goto leave;
+				/* fine.  don't need this node's info.
+				 * continue without it. */
+				break;
 			case DLM_RECO_NODE_DATA_REQUESTING:
 				ndata->state = DLM_RECO_NODE_DATA_REQUESTED;
 				mlog(0, "now receiving recovery data from "
@@ -520,35 +666,26 @@ static int dlm_remaster_locks(struct dlm
 					BUG();
 					break;
 				case DLM_RECO_NODE_DATA_DEAD:
-					mlog(ML_NOTICE, "node %u died after "
+					mlog(0, "node %u died after "
 					     "requesting recovery info for "
 					     "node %u\n", ndata->node_num,
 					     dead_node);
-					spin_unlock(&dlm_reco_state_lock);
-					// start all over
-					destroy = 1;
-					status = -EAGAIN;
-					/* instead of spinning like crazy here,
-					 * wait for the domain map to catch up
-					 * with the network state.  otherwise this
-					 * can be hit hundreds of times before
-					 * the node is really seen as dead. */
-					wait_event_timeout(dlm->dlm_reco_thread_wq,
-							   dlm_is_node_dead(dlm,
-								ndata->node_num),
-							   msecs_to_jiffies(1000));
-					mlog(0, "waited 1 sec for %u, "
-					     "dead? %s\n", ndata->node_num,
-					     dlm_is_node_dead(dlm, ndata->node_num) ?
-					     "yes" : "no");
-					goto leave;
+					break;
 				case DLM_RECO_NODE_DATA_RECEIVING:
 				case DLM_RECO_NODE_DATA_REQUESTED:
+					mlog(0, "%s: node %u still in state %s\n",
+					     dlm->name, ndata->node_num,
+					     ndata->state==DLM_RECO_NODE_DATA_RECEIVING ?
+					     "receiving" : "requested");
 					all_nodes_done = 0;
 					break;
 				case DLM_RECO_NODE_DATA_DONE:
+					mlog(0, "%s: node %u state is done\n",
+					     dlm->name, ndata->node_num);
 					break;
 				case DLM_RECO_NODE_DATA_FINALIZE_SENT:
+					mlog(0, "%s: node %u state is finalize\n",
+					     dlm->name, ndata->node_num);
 					break;
 			}
 		}
@@ -578,7 +715,7 @@ static int dlm_remaster_locks(struct dlm
 			     jiffies, dlm->reco.dead_node,
 			     dlm->node_num, dlm->reco.new_master);
 			destroy = 1;
-			status = ret;
+			status = 0;
 			/* rescan everything marked dirty along the way */
 			dlm_kick_thread(dlm, NULL);
 			break;
@@ -591,7 +728,6 @@ static int dlm_remaster_locks(struct dlm
 
 	}
 
-leave:
 	if (destroy)
 		dlm_destroy_recovery_area(dlm, dead_node);
 
@@ -617,7 +753,7 @@ static int dlm_init_recovery_area(struct
 		}
 		BUG_ON(num == dead_node);
 
-		ndata = kcalloc(1, sizeof(*ndata), GFP_KERNEL);
+		ndata = kcalloc(1, sizeof(*ndata), GFP_NOFS);
 		if (!ndata) {
 			dlm_destroy_recovery_area(dlm, dead_node);
 			return -ENOMEM;
@@ -691,16 +827,25 @@ int dlm_request_all_locks_handler(struct
 	if (!dlm_grab(dlm))
 		return -EINVAL;
 
+	if (lr->dead_node != dlm->reco.dead_node) {
+		mlog(ML_ERROR, "%s: node %u sent dead_node=%u, but local "
+		     "dead_node is %u\n", dlm->name, lr->node_idx,
+		     lr->dead_node, dlm->reco.dead_node);
+		dlm_print_reco_node_status(dlm);
+		/* this is a hack */
+		dlm_put(dlm);
+		return -ENOMEM;
+	}
 	BUG_ON(lr->dead_node != dlm->reco.dead_node);
 
-	item = kcalloc(1, sizeof(*item), GFP_KERNEL);
+	item = kcalloc(1, sizeof(*item), GFP_NOFS);
 	if (!item) {
 		dlm_put(dlm);
 		return -ENOMEM;
 	}
 
 	/* this will get freed by dlm_request_all_locks_worker */
-	buf = (char *) __get_free_page(GFP_KERNEL);
+	buf = (char *) __get_free_page(GFP_NOFS);
 	if (!buf) {
 		kfree(item);
 		dlm_put(dlm);
@@ -715,7 +860,7 @@ int dlm_request_all_locks_handler(struct
 	spin_lock(&dlm->work_lock);
 	list_add_tail(&item->list, &dlm->work_list);
 	spin_unlock(&dlm->work_lock);
-	schedule_work(&dlm->dispatched_work);
+	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
 
 	dlm_put(dlm);
 	return 0;
@@ -730,32 +875,34 @@ static void dlm_request_all_locks_worker
 	struct list_head *iter;
 	int ret;
 	u8 dead_node, reco_master;
+	int skip_all_done = 0;
 
 	dlm = item->dlm;
 	dead_node = item->u.ral.dead_node;
 	reco_master = item->u.ral.reco_master;
 	mres = (struct dlm_migratable_lockres *)data;
 
+	mlog(0, "%s: recovery worker started, dead=%u, master=%u\n",
+	     dlm->name, dead_node, reco_master);
+
 	if (dead_node != dlm->reco.dead_node ||
 	    reco_master != dlm->reco.new_master) {
-		/* show extra debug info if the recovery state is messed */
-		mlog(ML_ERROR, "%s: bad reco state: reco(dead=%u, master=%u), "
-		     "request(dead=%u, master=%u)\n",
-		     dlm->name, dlm->reco.dead_node, dlm->reco.new_master,
-		     dead_node, reco_master);
-		mlog(ML_ERROR, "%s: name=%.*s master=%u locks=%u/%u flags=%u "
-		     "entry[0]={c=%u:%llu,l=%u,f=%u,t=%d,ct=%d,hb=%d,n=%u}\n",
-		     dlm->name, mres->lockname_len, mres->lockname, mres->master,
-		     mres->num_locks, mres->total_locks, mres->flags,
-		     dlm_get_lock_cookie_node(mres->ml[0].cookie),
-		     dlm_get_lock_cookie_seq(mres->ml[0].cookie),
-		     mres->ml[0].list, mres->ml[0].flags,
-		     mres->ml[0].type, mres->ml[0].convert_type,
-		     mres->ml[0].highest_blocked, mres->ml[0].node);
-		BUG();
+		/* worker could have been created before the recovery master
+		 * died.  if so, do not continue, but do not error. */
+		if (dlm->reco.new_master == O2NM_INVALID_NODE_NUM) {
+			mlog(ML_NOTICE, "%s: will not send recovery state, "
+			     "recovery master %u died, thread=(dead=%u,mas=%u)"
+			     " current=(dead=%u,mas=%u)\n", dlm->name,
+			     reco_master, dead_node, reco_master,
+			     dlm->reco.dead_node, dlm->reco.new_master);
+		} else {
+			mlog(ML_NOTICE, "%s: reco state invalid: reco(dead=%u, "
+			     "master=%u), request(dead=%u, master=%u)\n",
+			     dlm->name, dlm->reco.dead_node,
+			     dlm->reco.new_master, dead_node, reco_master);
+		}
+		goto leave;
 	}
-	BUG_ON(dead_node != dlm->reco.dead_node);
-	BUG_ON(reco_master != dlm->reco.new_master);
 
 	/* lock resources should have already been moved to the
  	 * dlm->reco.resources list.  now move items from that list
@@ -766,12 +913,20 @@ static void dlm_request_all_locks_worker
 	dlm_move_reco_locks_to_list(dlm, &resources, dead_node);
 
 	/* now we can begin blasting lockreses without the dlm lock */
+
+	/* any errors returned will be due to the new_master dying,
+	 * the dlm_reco_thread should detect this */
 	list_for_each(iter, &resources) {
 		res = list_entry (iter, struct dlm_lock_resource, recovering);
 		ret = dlm_send_one_lockres(dlm, res, mres, reco_master,
 				   	DLM_MRES_RECOVERY);
-		if (ret < 0)
-			mlog_errno(ret);
+		if (ret < 0) {
+			mlog(ML_ERROR, "%s: node %u went down while sending "
+			     "recovery state for dead node %u, ret=%d\n", dlm->name,
+			     reco_master, dead_node, ret);
+			skip_all_done = 1;
+			break;
+		}
 	}
 
 	/* move the resources back to the list */
@@ -779,10 +934,15 @@ static void dlm_request_all_locks_worker
 	list_splice_init(&resources, &dlm->reco.resources);
 	spin_unlock(&dlm->spinlock);
 
-	ret = dlm_send_all_done_msg(dlm, dead_node, reco_master);
-	if (ret < 0)
-		mlog_errno(ret);
-
+	if (!skip_all_done) {
+		ret = dlm_send_all_done_msg(dlm, dead_node, reco_master);
+		if (ret < 0) {
+			mlog(ML_ERROR, "%s: node %u went down while sending "
+			     "recovery all-done for dead node %u, ret=%d\n",
+			     dlm->name, reco_master, dead_node, ret);
+		}
+	}
+leave:
 	free_page((unsigned long)data);
 }
 
@@ -801,8 +961,14 @@ static int dlm_send_all_done_msg(struct 
 
 	ret = o2net_send_message(DLM_RECO_DATA_DONE_MSG, dlm->key, &done_msg,
 				 sizeof(done_msg), send_to, &tmpret);
-	/* negative status is ignored by the caller */
-	if (ret >= 0)
+	if (ret < 0) {
+		if (!dlm_is_host_down(ret)) {
+			mlog_errno(ret);
+			mlog(ML_ERROR, "%s: unknown error sending data-done "
+			     "to %u\n", dlm->name, send_to);
+			BUG();
+		}
+	} else
 		ret = tmpret;
 	return ret;
 }
@@ -822,7 +988,11 @@ int dlm_reco_data_done_handler(struct o2
 	mlog(0, "got DATA DONE: dead_node=%u, reco.dead_node=%u, "
 	     "node_idx=%u, this node=%u\n", done->dead_node,
 	     dlm->reco.dead_node, done->node_idx, dlm->node_num);
-	BUG_ON(done->dead_node != dlm->reco.dead_node);
+
+	mlog_bug_on_msg((done->dead_node != dlm->reco.dead_node),
+			"Got DATA DONE: dead_node=%u, reco.dead_node=%u, "
+			"node_idx=%u, this node=%u\n", done->dead_node,
+			dlm->reco.dead_node, done->node_idx, dlm->node_num);
 
 	spin_lock(&dlm_reco_state_lock);
 	list_for_each(iter, &dlm->reco.node_data) {
@@ -1023,8 +1193,9 @@ static int dlm_add_lock_to_array(struct 
 		    ml->type == LKM_PRMODE) {
 			/* if it is already set, this had better be a PR
 			 * and it has to match */
-			if (mres->lvb[0] && (ml->type == LKM_EXMODE ||
-			    memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))) {
+			if (!dlm_lvb_is_empty(mres->lvb) &&
+			    (ml->type == LKM_EXMODE ||
+			     memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))) {
 				mlog(ML_ERROR, "mismatched lvbs!\n");
 				__dlm_print_one_lock_resource(lock->lockres);
 				BUG();
@@ -1083,22 +1254,25 @@ int dlm_send_one_lockres(struct dlm_ctxt
 			 * we must send it immediately. */
 			ret = dlm_send_mig_lockres_msg(dlm, mres, send_to,
 						       res, total_locks);
-			if (ret < 0) {
-				// TODO
-				mlog(ML_ERROR, "dlm_send_mig_lockres_msg "
-				     "returned %d, TODO\n", ret);
-				BUG();
-			}
+			if (ret < 0)
+				goto error;
 		}
 	}
 	/* flush any remaining locks */
 	ret = dlm_send_mig_lockres_msg(dlm, mres, send_to, res, total_locks);
-	if (ret < 0) {
-		// TODO
-		mlog(ML_ERROR, "dlm_send_mig_lockres_msg returned %d, "
-		     "TODO\n", ret);
+	if (ret < 0)
+		goto error;
+	return ret;
+
+error:
+	mlog(ML_ERROR, "%s: dlm_send_mig_lockres_msg returned %d\n",
+	     dlm->name, ret);
+	if (!dlm_is_host_down(ret))
 		BUG();
-	}
+	mlog(0, "%s: node %u went down while sending %s "
+	     "lockres %.*s\n", dlm->name, send_to,
+	     flags & DLM_MRES_RECOVERY ?  "recovery" : "migration",
+	     res->lockname.len, res->lockname.name);
 	return ret;
 }
 
@@ -1146,8 +1320,8 @@ int dlm_mig_lockres_handler(struct o2net
 		mlog(0, "all done flag.  all lockres data received!\n");
 
 	ret = -ENOMEM;
-	buf = kmalloc(be16_to_cpu(msg->data_len), GFP_KERNEL);
-	item = kcalloc(1, sizeof(*item), GFP_KERNEL);
+	buf = kmalloc(be16_to_cpu(msg->data_len), GFP_NOFS);
+	item = kcalloc(1, sizeof(*item), GFP_NOFS);
 	if (!buf || !item)
 		goto leave;
 
@@ -1238,7 +1412,7 @@ int dlm_mig_lockres_handler(struct o2net
 	spin_lock(&dlm->work_lock);
 	list_add_tail(&item->list, &dlm->work_list);
 	spin_unlock(&dlm->work_lock);
-	schedule_work(&dlm->dispatched_work);
+	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
 
 leave:
 	dlm_put(dlm);
@@ -1406,6 +1580,7 @@ int dlm_master_requery_handler(struct o2
 	struct dlm_ctxt *dlm = data;
 	struct dlm_master_requery *req = (struct dlm_master_requery *)msg->buf;
 	struct dlm_lock_resource *res = NULL;
+	unsigned int hash;
 	int master = DLM_LOCK_RES_OWNER_UNKNOWN;
 	u32 flags = DLM_ASSERT_MASTER_REQUERY;
 
@@ -1415,8 +1590,10 @@ int dlm_master_requery_handler(struct o2
 		return master;
 	}
 
+	hash = dlm_lockid_hash(req->name, req->namelen);
+
 	spin_lock(&dlm->spinlock);
-	res = __dlm_lookup_lockres(dlm, req->name, req->namelen);
+	res = __dlm_lookup_lockres(dlm, req->name, req->namelen, hash);
 	if (res) {
 		spin_lock(&res->spinlock);
 		master = res->owner;
@@ -1483,7 +1660,7 @@ static int dlm_process_recovery_data(str
 	struct dlm_lock *newlock = NULL;
 	struct dlm_lockstatus *lksb = NULL;
 	int ret = 0;
-	int i;
+	int i, bad;
 	struct list_head *iter;
 	struct dlm_lock *lock = NULL;
 
@@ -1553,28 +1730,48 @@ static int dlm_process_recovery_data(str
 		}
 		lksb->flags |= (ml->flags &
 				(DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB));
-			
-		if (mres->lvb[0]) {
+
+		if (ml->type == LKM_NLMODE)
+			goto skip_lvb;
+
+		if (!dlm_lvb_is_empty(mres->lvb)) {
 			if (lksb->flags & DLM_LKSB_PUT_LVB) {
 				/* other node was trying to update
 				 * lvb when node died.  recreate the
 				 * lksb with the updated lvb. */
 				memcpy(lksb->lvb, mres->lvb, DLM_LVB_LEN);
+				/* the lock resource lvb update must happen
+				 * NOW, before the spinlock is dropped.
+				 * we no longer wait for the AST to update
+				 * the lvb. */
+				memcpy(res->lvb, mres->lvb, DLM_LVB_LEN);
 			} else {
 				/* otherwise, the node is sending its 
 				 * most recent valid lvb info */
 				BUG_ON(ml->type != LKM_EXMODE &&
 				       ml->type != LKM_PRMODE);
-				if (res->lvb[0] && (ml->type == LKM_EXMODE ||
-				    memcmp(res->lvb, mres->lvb, DLM_LVB_LEN))) {
-					mlog(ML_ERROR, "received bad lvb!\n");
-					__dlm_print_one_lock_resource(res);
-					BUG();
+				if (!dlm_lvb_is_empty(res->lvb) &&
+ 				    (ml->type == LKM_EXMODE ||
+ 				     memcmp(res->lvb, mres->lvb, DLM_LVB_LEN))) {
+ 					int i;
+ 					mlog(ML_ERROR, "%s:%.*s: received bad "
+ 					     "lvb! type=%d\n", dlm->name,
+ 					     res->lockname.len,
+ 					     res->lockname.name, ml->type);
+ 					printk("lockres lvb=[");
+ 					for (i=0; i<DLM_LVB_LEN; i++)
+ 						printk("%02x", res->lvb[i]);
+ 					printk("]\nmigrated lvb=[");
+ 					for (i=0; i<DLM_LVB_LEN; i++)
+ 						printk("%02x", mres->lvb[i]);
+ 					printk("]\n");
+ 					dlm_print_one_lock_resource(res);
+ 					BUG();
 				}
 				memcpy(res->lvb, mres->lvb, DLM_LVB_LEN);
 			}
 		}
-
+skip_lvb:
 
 		/* NOTE:
 		 * wrt lock queue ordering and recovery:
@@ -1592,9 +1789,33 @@ static int dlm_process_recovery_data(str
 		 * relative to each other, but clearly *not*
 		 * preserved relative to locks from other nodes.
 		 */
+		bad = 0;
 		spin_lock(&res->spinlock);
-		dlm_lock_get(newlock);
-		list_add_tail(&newlock->list, queue);
+		list_for_each_entry(lock, queue, list) {
+			if (lock->ml.cookie == ml->cookie) {
+				u64 c = lock->ml.cookie;
+				mlog(ML_ERROR, "%s:%.*s: %u:%llu: lock already "
+				     "exists on this lockres!\n", dlm->name,
+				     res->lockname.len, res->lockname.name,
+				     dlm_get_lock_cookie_node(c),
+				     dlm_get_lock_cookie_seq(c));
+
+				mlog(ML_NOTICE, "sent lock: type=%d, conv=%d, "
+				     "node=%u, cookie=%u:%llu, queue=%d\n",
+	      			     ml->type, ml->convert_type, ml->node,
+				     dlm_get_lock_cookie_node(ml->cookie),
+				     dlm_get_lock_cookie_seq(ml->cookie),
+				     ml->list);
+
+				__dlm_print_one_lock_resource(res);
+				bad = 1;
+				break;
+			}
+		}
+		if (!bad) {
+			dlm_lock_get(newlock);
+			list_add_tail(&newlock->list, queue);
+		}
 		spin_unlock(&res->spinlock);
 	}
 	mlog(0, "done running all the locks\n");
@@ -1618,8 +1839,14 @@ void dlm_move_lockres_to_recovery_list(s
 	struct dlm_lock *lock;
 
 	res->state |= DLM_LOCK_RES_RECOVERING;
-	if (!list_empty(&res->recovering))
+	if (!list_empty(&res->recovering)) {
+		mlog(0,
+		     "Recovering res %s:%.*s, is already on recovery list!\n",
+		     dlm->name, res->lockname.len, res->lockname.name);
 		list_del_init(&res->recovering);
+	}
+	/* We need to hold a reference while on the recovery list */
+	dlm_lockres_get(res);
 	list_add_tail(&res->recovering, &dlm->reco.resources);
 
 	/* find any pending locks and put them back on proper list */
@@ -1708,9 +1935,11 @@ static void dlm_finish_local_lockres_rec
 			spin_lock(&res->spinlock);
 			dlm_change_lockres_owner(dlm, res, new_master);
 			res->state &= ~DLM_LOCK_RES_RECOVERING;
-			__dlm_dirty_lockres(dlm, res);
+			if (!__dlm_lockres_unused(res))
+				__dlm_dirty_lockres(dlm, res);
 			spin_unlock(&res->spinlock);
 			wake_up(&res->wq);
+			dlm_lockres_put(res);
 		}
 	}
 
@@ -1719,7 +1948,7 @@ static void dlm_finish_local_lockres_rec
 	 * the RECOVERING state and set the owner
 	 * if necessary */
 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
-		bucket = &(dlm->lockres_hash[i]);
+		bucket = dlm_lockres_hash(dlm, i);
 		hlist_for_each_entry(res, hash_iter, bucket, hash_node) {
 			if (res->state & DLM_LOCK_RES_RECOVERING) {
 				if (res->owner == dead_node) {
@@ -1743,11 +1972,13 @@ static void dlm_finish_local_lockres_rec
 					     dlm->name, res->lockname.len,
 					     res->lockname.name, res->owner);
 					list_del_init(&res->recovering);
+					dlm_lockres_put(res);
 				}
 				spin_lock(&res->spinlock);
 				dlm_change_lockres_owner(dlm, res, new_master);
 				res->state &= ~DLM_LOCK_RES_RECOVERING;
-				__dlm_dirty_lockres(dlm, res);
+				if (!__dlm_lockres_unused(res))
+					__dlm_dirty_lockres(dlm, res);
 				spin_unlock(&res->spinlock);
 				wake_up(&res->wq);
 			}
@@ -1884,7 +2115,7 @@ static void dlm_do_local_recovery_cleanu
 	 *    need to be fired as a result.
 	 */
 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
-		bucket = &(dlm->lockres_hash[i]);
+		bucket = dlm_lockres_hash(dlm, i);
 		hlist_for_each_entry(res, iter, bucket, hash_node) {
  			/* always prune any $RECOVERY entries for dead nodes,
  			 * otherwise hangs can occur during later recovery */
@@ -1924,6 +2155,20 @@ static void __dlm_hb_node_down(struct dl
 {
 	assert_spin_locked(&dlm->spinlock);
 
+	if (dlm->reco.new_master == idx) {
+		mlog(0, "%s: recovery master %d just died\n",
+		     dlm->name, idx);
+		if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
+			/* finalize1 was reached, so it is safe to clear
+			 * the new_master and dead_node.  that recovery
+			 * is complete. */
+			mlog(0, "%s: dead master %d had reached "
+			     "finalize1 state, clearing\n", dlm->name, idx);
+			dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE;
+			__dlm_reset_recovery(dlm);
+		}
+	}
+
 	/* check to see if the node is already considered dead */
 	if (!test_bit(idx, dlm->live_nodes_map)) {
 		mlog(0, "for domain %s, node %d is already dead. "
@@ -2087,7 +2332,7 @@ again:	
 
 			/* set the new_master to this node */
 			spin_lock(&dlm->spinlock);
-			dlm->reco.new_master = dlm->node_num;
+			dlm_set_reco_master(dlm, dlm->node_num);
 			spin_unlock(&dlm->spinlock);
 		}
 
@@ -2125,6 +2370,10 @@ again:	
 		mlog(0, "%s: reco master %u is ready to recover %u\n",
 		     dlm->name, dlm->reco.new_master, dlm->reco.dead_node);
 		status = -EEXIST;
+	} else if (ret == DLM_RECOVERING) {
+		mlog(0, "dlm=%s dlmlock says master node died (this=%u)\n",
+		     dlm->name, dlm->node_num);
+		goto again;
 	} else {
 		struct dlm_lock_resource *res;
 
@@ -2156,7 +2405,7 @@ static int dlm_send_begin_reco_message(s
 
 	mlog_entry("%u\n", dead_node);
 
-	mlog(0, "dead node is %u\n", dead_node);
+	mlog(0, "%s: dead node is %u\n", dlm->name, dead_node);
 
 	spin_lock(&dlm->spinlock);
 	dlm_node_iter_init(dlm->domain_map, &iter);
@@ -2214,6 +2463,14 @@ retry:
 			 * another ENOMEM */
 			msleep(100);
 			goto retry;
+		} else if (ret == EAGAIN) {
+			mlog(0, "%s: trying to start recovery of node "
+			     "%u, but node %u is waiting for last recovery "
+			     "to complete, backoff for a bit\n", dlm->name,
+			     dead_node, nodenum);
+			/* TODO Look into replacing msleep with cond_resched() */
+			msleep(100);
+			goto retry;
 		}
 	}
 
@@ -2229,8 +2486,20 @@ int dlm_begin_reco_handler(struct o2net_
 	if (!dlm_grab(dlm))
 		return 0;
 
-	mlog(0, "node %u wants to recover node %u\n",
-		  br->node_idx, br->dead_node);
+	spin_lock(&dlm->spinlock);
+	if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
+		mlog(0, "%s: node %u wants to recover node %u (%u:%u) "
+		     "but this node is in finalize state, waiting on finalize2\n",
+		     dlm->name, br->node_idx, br->dead_node,
+		     dlm->reco.dead_node, dlm->reco.new_master);
+		spin_unlock(&dlm->spinlock);
+		return EAGAIN;
+	}
+	spin_unlock(&dlm->spinlock);
+
+	mlog(0, "%s: node %u wants to recover node %u (%u:%u)\n",
+	     dlm->name, br->node_idx, br->dead_node,
+	     dlm->reco.dead_node, dlm->reco.new_master);
 
 	dlm_fire_domain_eviction_callbacks(dlm, br->dead_node);
 
@@ -2252,8 +2521,8 @@ int dlm_begin_reco_handler(struct o2net_
 		     "node %u changing it to %u\n", dlm->name, 
 		     dlm->reco.dead_node, br->node_idx, br->dead_node);
 	}
-	dlm->reco.new_master = br->node_idx;
-	dlm->reco.dead_node = br->dead_node;
+	dlm_set_reco_master(dlm, br->node_idx);
+	dlm_set_reco_dead_node(dlm, br->dead_node);
 	if (!test_bit(br->dead_node, dlm->recovery_map)) {
 		mlog(0, "recovery master %u sees %u as dead, but this "
 		     "node has not yet.  marking %u as dead\n",
@@ -2272,10 +2541,16 @@ int dlm_begin_reco_handler(struct o2net_
 	spin_unlock(&dlm->spinlock);
 
 	dlm_kick_recovery_thread(dlm);
+
+	mlog(0, "%s: recovery started by node %u, for %u (%u:%u)\n",
+	     dlm->name, br->node_idx, br->dead_node,
+	     dlm->reco.dead_node, dlm->reco.new_master);
+
 	dlm_put(dlm);
 	return 0;
 }
 
+#define DLM_FINALIZE_STAGE2  0x01
 static int dlm_send_finalize_reco_message(struct dlm_ctxt *dlm)
 {
 	int ret = 0;
@@ -2283,25 +2558,31 @@ static int dlm_send_finalize_reco_messag
 	struct dlm_node_iter iter;
 	int nodenum;
 	int status;
+	int stage = 1;
 
-	mlog(0, "finishing recovery for node %s:%u\n",
-	     dlm->name, dlm->reco.dead_node);
+	mlog(0, "finishing recovery for node %s:%u, "
+	     "stage %d\n", dlm->name, dlm->reco.dead_node, stage);
 
 	spin_lock(&dlm->spinlock);
 	dlm_node_iter_init(dlm->domain_map, &iter);
 	spin_unlock(&dlm->spinlock);
 
+stage2:
 	memset(&fr, 0, sizeof(fr));
 	fr.node_idx = dlm->node_num;
 	fr.dead_node = dlm->reco.dead_node;
+	if (stage == 2)
+		fr.flags |= DLM_FINALIZE_STAGE2;
 
 	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
 		if (nodenum == dlm->node_num)
 			continue;
 		ret = o2net_send_message(DLM_FINALIZE_RECO_MSG, dlm->key,
 					 &fr, sizeof(fr), nodenum, &status);
-		if (ret >= 0) {
+		if (ret >= 0)
 			ret = status;
+		if (ret < 0) {
+			mlog_errno(ret);
 			if (dlm_is_host_down(ret)) {
 				/* this has no effect on this recovery 
 				 * session, so set the status to zero to 
@@ -2309,13 +2590,17 @@ static int dlm_send_finalize_reco_messag
 				mlog(ML_ERROR, "node %u went down after this "
 				     "node finished recovery.\n", nodenum);
 				ret = 0;
+				continue;
 			}
-		}
-		if (ret < 0) {
-			mlog_errno(ret);
 			break;
 		}
 	}
+	if (stage == 1) {
+		/* reset the node_iter back to the top and send finalize2 */
+		iter.curnode = -1;
+		stage = 2;
+		goto stage2;
+	}
 
 	return ret;
 }
@@ -2324,14 +2609,19 @@ int dlm_finalize_reco_handler(struct o2n
 {
 	struct dlm_ctxt *dlm = data;
 	struct dlm_finalize_reco *fr = (struct dlm_finalize_reco *)msg->buf;
+	int stage = 1;
 
 	/* ok to return 0, domain has gone away */
 	if (!dlm_grab(dlm))
 		return 0;
 
-	mlog(0, "node %u finalizing recovery of node %u\n",
-	     fr->node_idx, fr->dead_node);
+	if (fr->flags & DLM_FINALIZE_STAGE2)
+		stage = 2;
 
+	mlog(0, "%s: node %u finalizing recovery stage%d of "
+	     "node %u (%u:%u)\n", dlm->name, fr->node_idx, stage,
+	     fr->dead_node, dlm->reco.dead_node, dlm->reco.new_master);
+ 
 	spin_lock(&dlm->spinlock);
 
 	if (dlm->reco.new_master != fr->node_idx) {
@@ -2347,13 +2637,41 @@ int dlm_finalize_reco_handler(struct o2n
 		BUG();
 	}
 
-	dlm_finish_local_lockres_recovery(dlm, fr->dead_node, fr->node_idx);
-
-	spin_unlock(&dlm->spinlock);
+	switch (stage) {
+		case 1:
+			dlm_finish_local_lockres_recovery(dlm, fr->dead_node, fr->node_idx);
+			if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
+				mlog(ML_ERROR, "%s: received finalize1 from "
+				     "new master %u for dead node %u, but "
+				     "this node has already received it!\n",
+				     dlm->name, fr->node_idx, fr->dead_node);
+				dlm_print_reco_node_status(dlm);
+				BUG();
+			}
+			dlm->reco.state |= DLM_RECO_STATE_FINALIZE;
+			spin_unlock(&dlm->spinlock);
+			break;
+		case 2:
+			if (!(dlm->reco.state & DLM_RECO_STATE_FINALIZE)) {
+				mlog(ML_ERROR, "%s: received finalize2 from "
+				     "new master %u for dead node %u, but "
+				     "this node did not have finalize1!\n",
+				     dlm->name, fr->node_idx, fr->dead_node);
+				dlm_print_reco_node_status(dlm);
+				BUG();
+			}
+			dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE;
+			spin_unlock(&dlm->spinlock);
+			dlm_reset_recovery(dlm);
+			dlm_kick_recovery_thread(dlm);
+			break;
+		default:
+			BUG();
+	}
 
-	dlm_reset_recovery(dlm);
+	mlog(0, "%s: recovery done, reco master was %u, dead now %u, master now %u\n",
+	     dlm->name, fr->node_idx, dlm->reco.dead_node, dlm->reco.new_master);
 
-	dlm_kick_recovery_thread(dlm);
 	dlm_put(dlm);
 	return 0;
 }
diff --git a/fs/ocfs2/dlm/dlmthread.c b/fs/ocfs2/dlm/dlmthread.c
index 5be9d14..4641d06 100644
--- a/fs/ocfs2/dlm/dlmthread.c
+++ b/fs/ocfs2/dlm/dlmthread.c
@@ -39,6 +39,7 @@ #include <linux/socket.h>
 #include <linux/inet.h>
 #include <linux/timer.h>
 #include <linux/kthread.h>
+#include <linux/delay.h>
 
 
 #include "cluster/heartbeat.h"
@@ -53,6 +54,8 @@ #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_
 #include "cluster/masklog.h"
 
 static int dlm_thread(void *data);
+static void dlm_purge_lockres_now(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *lockres);
 
 static void dlm_flush_asts(struct dlm_ctxt *dlm);
 
@@ -80,7 +83,7 @@ repeat:
 }
 
 
-static int __dlm_lockres_unused(struct dlm_lock_resource *res)
+int __dlm_lockres_unused(struct dlm_lock_resource *res)
 {
 	if (list_empty(&res->granted) &&
 	    list_empty(&res->converting) &&
@@ -103,6 +106,20 @@ void __dlm_lockres_calc_usage(struct dlm
 	assert_spin_locked(&res->spinlock);
 
 	if (__dlm_lockres_unused(res)){
+		/* For now, just keep any resource we master */
+		if (res->owner == dlm->node_num)
+		{
+			if (!list_empty(&res->purge)) {
+				mlog(0, "we master %s:%.*s, but it is on "
+				     "the purge list.  Removing\n",
+				     dlm->name, res->lockname.len,
+				     res->lockname.name);
+				list_del_init(&res->purge);
+				dlm->purge_count--;
+			}
+			return;
+		}
+
 		if (list_empty(&res->purge)) {
 			mlog(0, "putting lockres %.*s from purge list\n",
 			     res->lockname.len, res->lockname.name);
@@ -110,10 +127,23 @@ void __dlm_lockres_calc_usage(struct dlm
 			res->last_used = jiffies;
 			list_add_tail(&res->purge, &dlm->purge_list);
 			dlm->purge_count++;
+
+			/* if this node is not the owner, there is
+			 * no way to keep track of who the owner could be.
+			 * unhash it to avoid serious problems. */
+			if (res->owner != dlm->node_num) {
+				mlog(0, "%s:%.*s: doing immediate "
+				     "purge of lockres owned by %u\n",
+				     dlm->name, res->lockname.len,
+				     res->lockname.name, res->owner);
+
+				dlm_purge_lockres_now(dlm, res);
+			}
 		}
 	} else if (!list_empty(&res->purge)) {
-		mlog(0, "removing lockres %.*s from purge list\n",
-		     res->lockname.len, res->lockname.name);
+		mlog(0, "removing lockres %.*s from purge list, "
+		     "owner=%u\n", res->lockname.len, res->lockname.name,
+		     res->owner);
 
 		list_del_init(&res->purge);
 		dlm->purge_count--;
@@ -165,6 +195,7 @@ again:
 	} else if (ret < 0) {
 		mlog(ML_NOTICE, "lockres %.*s: migrate failed, retrying\n",
 		     lockres->lockname.len, lockres->lockname.name);
+		msleep(100);
 		goto again;
 	}
 
@@ -178,6 +209,24 @@ finish:
 	__dlm_unhash_lockres(lockres);
 }
 
+/* make an unused lockres go away immediately.
+ * as soon as the dlm spinlock is dropped, this lockres
+ * will not be found. kfree still happens on last put. */
+static void dlm_purge_lockres_now(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *lockres)
+{
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&lockres->spinlock);
+
+	BUG_ON(!__dlm_lockres_unused(lockres));
+
+	if (!list_empty(&lockres->purge)) {
+		list_del_init(&lockres->purge);
+		dlm->purge_count--;
+	}
+	__dlm_unhash_lockres(lockres);
+}
+
 static void dlm_run_purge_list(struct dlm_ctxt *dlm,
 			       int purge_now)
 {
@@ -422,6 +471,8 @@ void __dlm_dirty_lockres(struct dlm_ctxt
 	/* don't shuffle secondary queues */
 	if ((res->owner == dlm->node_num) &&
 	    !(res->state & DLM_LOCK_RES_DIRTY)) {
+		/* ref for dirty_list */
+		dlm_lockres_get(res);
 		list_add_tail(&res->dirty, &dlm->dirty_list);
 		res->state |= DLM_LOCK_RES_DIRTY;
 	}
@@ -606,6 +657,8 @@ static int dlm_thread(void *data)
 			list_del_init(&res->dirty);
 			spin_unlock(&res->spinlock);
 			spin_unlock(&dlm->spinlock);
+			/* Drop dirty_list ref */
+			dlm_lockres_put(res);
 
 		 	/* lockres can be re-dirtied/re-added to the
 			 * dirty_list in this gap, but that is ok */
@@ -642,8 +695,9 @@ static int dlm_thread(void *data)
 			 * spinlock and do NOT have the dlm lock.
 			 * safe to reserve/queue asts and run the lists. */
 
-			mlog(0, "calling dlm_shuffle_lists with dlm=%p, "
-			     "res=%p\n", dlm, res);
+			mlog(0, "calling dlm_shuffle_lists with dlm=%s, "
+			     "res=%.*s\n", dlm->name,
+			     res->lockname.len, res->lockname.name);
 
 			/* called while holding lockres lock */
 			dlm_shuffle_lists(dlm, res);
@@ -657,6 +711,8 @@ in_progress:
 			/* if the lock was in-progress, stick
 			 * it on the back of the list */
 			if (delay) {
+				/* ref for dirty_list */
+				dlm_lockres_get(res);
 				spin_lock(&res->spinlock);
 				list_add_tail(&res->dirty, &dlm->dirty_list);
 				res->state |= DLM_LOCK_RES_DIRTY;
@@ -677,7 +733,7 @@ in_progress:
 
 		/* yield and continue right away if there is more work to do */
 		if (!n) {
-			yield();
+			cond_resched();
 			continue;
 		}
 
diff --git a/fs/ocfs2/dlm/dlmunlock.c b/fs/ocfs2/dlm/dlmunlock.c
index 7b1a275..1fe1112 100644
--- a/fs/ocfs2/dlm/dlmunlock.c
+++ b/fs/ocfs2/dlm/dlmunlock.c
@@ -319,6 +319,16 @@ static enum dlm_status dlm_send_remote_u
 
 	mlog_entry("%.*s\n", res->lockname.len, res->lockname.name);
 
+	if (owner == dlm->node_num) {
+		/* ended up trying to contact ourself.  this means
+		 * that the lockres had been remote but became local
+		 * via a migration.  just retry it, now as local */
+		mlog(0, "%s:%.*s: this node became the master due to a "
+		     "migration, re-evaluate now\n", dlm->name,
+		     res->lockname.len, res->lockname.name);
+		return DLM_FORWARD;
+	}
+
 	memset(&unlock, 0, sizeof(unlock));
 	unlock.node_idx = dlm->node_num;
 	unlock.flags = cpu_to_be32(flags);
diff --git a/fs/ocfs2/dlm/userdlm.c b/fs/ocfs2/dlm/userdlm.c
index 74ca4e5..e641b08 100644
--- a/fs/ocfs2/dlm/userdlm.c
+++ b/fs/ocfs2/dlm/userdlm.c
@@ -672,7 +672,7 @@ struct dlm_ctxt *user_dlm_register_conte
 	u32 dlm_key;
 	char *domain;
 
-	domain = kmalloc(name->len + 1, GFP_KERNEL);
+	domain = kmalloc(name->len + 1, GFP_NOFS);
 	if (!domain) {
 		mlog_errno(-ENOMEM);
 		return ERR_PTR(-ENOMEM);

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2006-05-17 22:21 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2006-05-17 22:21 UTC (permalink / raw)
  To: Andrew Morton, Linus Torvalds; +Cc: ocfs2-devel, linux-kernel

Ok, sending this patch set again. It seems to have fallen off the
stack, probably because my timing sucked, and I mailed it out just
after -rc4 was released :/

Some OCFS2 bug fixes, and a couple configfs ones. All of them are issues
that folks have actually hit.

Please pull from 'upstream-linus' branch of
git://oss.oracle.com/home/sourcebo/git/ocfs2.git

to receive the following updates:

 fs/Makefile           |    2 
 fs/configfs/dir.c     |  137 +++++++++++++++++++++++++++++++++++---------------
 fs/ocfs2/aops.c       |   46 ++++++++++++++--
 fs/ocfs2/aops.h       |    4 -
 fs/ocfs2/extent_map.c |    6 +-
 fs/ocfs2/file.c       |   86 ++++++++++++++++++++++---------
 fs/ocfs2/journal.c    |    8 +-
 fs/ocfs2/uptodate.c   |    4 -
 fs/ocfs2/vote.c       |    6 +-
 9 files changed, 213 insertions(+), 86 deletions(-)

Joel Becker:
      configfs: Fix a reference leak in configfs_mkdir().
      configfs: configfs_mkdir() failed to cleanup linkage.
      configfs: Make sure configfs_init() is called before consumers.

Mark Fasheh:
      ocfs2: take data locks around extend
      ocfs2: take meta data lock in ocfs2_file_aio_read()
      ocfs2: Don't populate uptodate cache in ocfs2_force_read_journal()

Sunil Mushran:
      ocfs2: fix gfp mask in some file system paths

diff --git a/fs/Makefile b/fs/Makefile
index 83bf478..078d3d1 100644
--- a/fs/Makefile
+++ b/fs/Makefile
@@ -45,6 +45,7 @@ obj-$(CONFIG_DNOTIFY)		+= dnotify.o
 obj-$(CONFIG_PROC_FS)		+= proc/
 obj-y				+= partitions/
 obj-$(CONFIG_SYSFS)		+= sysfs/
+obj-$(CONFIG_CONFIGFS_FS)	+= configfs/
 obj-y				+= devpts/
 
 obj-$(CONFIG_PROFILING)		+= dcookies.o
@@ -100,5 +101,4 @@ obj-$(CONFIG_BEFS_FS)		+= befs/
 obj-$(CONFIG_HOSTFS)		+= hostfs/
 obj-$(CONFIG_HPPFS)		+= hppfs/
 obj-$(CONFIG_DEBUG_FS)		+= debugfs/
-obj-$(CONFIG_CONFIGFS_FS)	+= configfs/
 obj-$(CONFIG_OCFS2_FS)		+= ocfs2/
diff --git a/fs/configfs/dir.c b/fs/configfs/dir.c
index 5638c8f..5f95218 100644
--- a/fs/configfs/dir.c
+++ b/fs/configfs/dir.c
@@ -505,13 +505,15 @@ static int populate_groups(struct config
 	int i;
 
 	if (group->default_groups) {
-		/* FYI, we're faking mkdir here
+		/*
+		 * FYI, we're faking mkdir here
 		 * I'm not sure we need this semaphore, as we're called
 		 * from our parent's mkdir.  That holds our parent's
 		 * i_mutex, so afaik lookup cannot continue through our
 		 * parent to find us, let alone mess with our tree.
 		 * That said, taking our i_mutex is closer to mkdir
-		 * emulation, and shouldn't hurt. */
+		 * emulation, and shouldn't hurt.
+		 */
 		mutex_lock(&dentry->d_inode->i_mutex);
 
 		for (i = 0; group->default_groups[i]; i++) {
@@ -546,20 +548,34 @@ static void unlink_obj(struct config_ite
 
 		item->ci_group = NULL;
 		item->ci_parent = NULL;
+
+		/* Drop the reference for ci_entry */
 		config_item_put(item);
 
+		/* Drop the reference for ci_parent */
 		config_group_put(group);
 	}
 }
 
 static void link_obj(struct config_item *parent_item, struct config_item *item)
 {
-	/* Parent seems redundant with group, but it makes certain
-	 * traversals much nicer. */
+	/*
+	 * Parent seems redundant with group, but it makes certain
+	 * traversals much nicer.
+	 */
 	item->ci_parent = parent_item;
+
+	/*
+	 * We hold a reference on the parent for the child's ci_parent
+	 * link.
+	 */
 	item->ci_group = config_group_get(to_config_group(parent_item));
 	list_add_tail(&item->ci_entry, &item->ci_group->cg_children);
 
+	/*
+	 * We hold a reference on the child for ci_entry on the parent's
+	 * cg_children
+	 */
 	config_item_get(item);
 }
 
@@ -684,6 +700,10 @@ static void client_drop_item(struct conf
 	type = parent_item->ci_type;
 	BUG_ON(!type);
 
+	/*
+	 * If ->drop_item() exists, it is responsible for the
+	 * config_item_put().
+	 */
 	if (type->ct_group_ops && type->ct_group_ops->drop_item)
 		type->ct_group_ops->drop_item(to_config_group(parent_item),
 						item);
@@ -694,23 +714,28 @@ static void client_drop_item(struct conf
 
 static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 {
-	int ret;
+	int ret, module_got = 0;
 	struct config_group *group;
 	struct config_item *item;
 	struct config_item *parent_item;
 	struct configfs_subsystem *subsys;
 	struct configfs_dirent *sd;
 	struct config_item_type *type;
-	struct module *owner;
+	struct module *owner = NULL;
 	char *name;
 
-	if (dentry->d_parent == configfs_sb->s_root)
-		return -EPERM;
+	if (dentry->d_parent == configfs_sb->s_root) {
+		ret = -EPERM;
+		goto out;
+	}
 
 	sd = dentry->d_parent->d_fsdata;
-	if (!(sd->s_type & CONFIGFS_USET_DIR))
-		return -EPERM;
+	if (!(sd->s_type & CONFIGFS_USET_DIR)) {
+		ret = -EPERM;
+		goto out;
+	}
 
+	/* Get a working ref for the duration of this function */
 	parent_item = configfs_get_config_item(dentry->d_parent);
 	type = parent_item->ci_type;
 	subsys = to_config_group(parent_item)->cg_subsys;
@@ -719,15 +744,16 @@ static int configfs_mkdir(struct inode *
 	if (!type || !type->ct_group_ops ||
 	    (!type->ct_group_ops->make_group &&
 	     !type->ct_group_ops->make_item)) {
-		config_item_put(parent_item);
-		return -EPERM;  /* What lack-of-mkdir returns */
+		ret = -EPERM;  /* Lack-of-mkdir returns -EPERM */
+		goto out_put;
 	}
 
 	name = kmalloc(dentry->d_name.len + 1, GFP_KERNEL);
 	if (!name) {
-		config_item_put(parent_item);
-		return -ENOMEM;
+		ret = -ENOMEM;
+		goto out_put;
 	}
+
 	snprintf(name, dentry->d_name.len + 1, "%s", dentry->d_name.name);
 
 	down(&subsys->su_sem);
@@ -748,40 +774,67 @@ static int configfs_mkdir(struct inode *
 
 	kfree(name);
 	if (!item) {
-		config_item_put(parent_item);
-		return -ENOMEM;
+		/*
+		 * If item == NULL, then link_obj() was never called.
+		 * There are no extra references to clean up.
+		 */
+		ret = -ENOMEM;
+		goto out_put;
 	}
 
-	ret = -EINVAL;
+	/*
+	 * link_obj() has been called (via link_group() for groups).
+	 * From here on out, errors must clean that up.
+	 */
+
 	type = item->ci_type;
-	if (type) {
-		owner = type->ct_owner;
-		if (try_module_get(owner)) {
-			if (group) {
-				ret = configfs_attach_group(parent_item,
-							    item,
-							    dentry);
-			} else {
-				ret = configfs_attach_item(parent_item,
-							   item,
-							   dentry);
-			}
+	if (!type) {
+		ret = -EINVAL;
+		goto out_unlink;
+	}
 
-			if (ret) {
-				down(&subsys->su_sem);
-				if (group)
-					unlink_group(group);
-				else
-					unlink_obj(item);
-				client_drop_item(parent_item, item);
-				up(&subsys->su_sem);
+	owner = type->ct_owner;
+	if (!try_module_get(owner)) {
+		ret = -EINVAL;
+		goto out_unlink;
+	}
 
-				config_item_put(parent_item);
-				module_put(owner);
-			}
-		}
+	/*
+	 * I hate doing it this way, but if there is
+	 * an error,  module_put() probably should
+	 * happen after any cleanup.
+	 */
+	module_got = 1;
+
+	if (group)
+		ret = configfs_attach_group(parent_item, item, dentry);
+	else
+		ret = configfs_attach_item(parent_item, item, dentry);
+
+out_unlink:
+	if (ret) {
+		/* Tear down everything we built up */
+		down(&subsys->su_sem);
+		if (group)
+			unlink_group(group);
+		else
+			unlink_obj(item);
+		client_drop_item(parent_item, item);
+		up(&subsys->su_sem);
+
+		if (module_got)
+			module_put(owner);
 	}
 
+out_put:
+	/*
+	 * link_obj()/link_group() took a reference from child->parent,
+	 * so the parent is safely pinned.  We can drop our working
+	 * reference.
+	 */
+	config_item_put(parent_item);
+
+out:
 	return ret;
 }
 
@@ -801,6 +854,7 @@ static int configfs_rmdir(struct inode *
 	if (sd->s_type & CONFIGFS_USET_DEFAULT)
 		return -EPERM;
 
+	/* Get a working ref until we have the child */
 	parent_item = configfs_get_config_item(dentry->d_parent);
 	subsys = to_config_group(parent_item)->cg_subsys;
 	BUG_ON(!subsys);
@@ -817,6 +871,7 @@ static int configfs_rmdir(struct inode *
 		return ret;
 	}
 
+	/* Get a working ref for the duration of this function */
 	item = configfs_get_config_item(dentry);
 
 	/* Drop reference from above, item already holds one. */
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 0d858d0..47152bf 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -276,13 +276,29 @@ static int ocfs2_writepage(struct page *
 	return ret;
 }
 
+/* This can also be called from ocfs2_write_zero_page() which has done
+ * it's own cluster locking. */
+int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
+			       unsigned from, unsigned to)
+{
+	int ret;
+
+	down_read(&OCFS2_I(inode)->ip_alloc_sem);
+
+	ret = block_prepare_write(page, from, to, ocfs2_get_block);
+
+	up_read(&OCFS2_I(inode)->ip_alloc_sem);
+
+	return ret;
+}
+
 /*
  * ocfs2_prepare_write() can be an outer-most ocfs2 call when it is called
  * from loopback.  It must be able to perform its own locking around
  * ocfs2_get_block().
  */
-int ocfs2_prepare_write(struct file *file, struct page *page,
-			unsigned from, unsigned to)
+static int ocfs2_prepare_write(struct file *file, struct page *page,
+			       unsigned from, unsigned to)
 {
 	struct inode *inode = page->mapping->host;
 	int ret;
@@ -295,11 +311,7 @@ int ocfs2_prepare_write(struct file *fil
 		goto out;
 	}
 
-	down_read(&OCFS2_I(inode)->ip_alloc_sem);
-
-	ret = block_prepare_write(page, from, to, ocfs2_get_block);
-
-	up_read(&OCFS2_I(inode)->ip_alloc_sem);
+	ret = ocfs2_prepare_write_nolock(inode, page, from, to);
 
 	ocfs2_meta_unlock(inode, 0);
 out:
@@ -625,11 +637,31 @@ static ssize_t ocfs2_direct_IO(int rw,
 	int ret;
 
 	mlog_entry_void();
+
+	/*
+	 * We get PR data locks even for O_DIRECT.  This allows
+	 * concurrent O_DIRECT I/O but doesn't let O_DIRECT with
+	 * extending and buffered zeroing writes race.  If they did
+	 * race then the buffered zeroing could be written back after
+	 * the O_DIRECT I/O.  It's one thing to tell people not to mix
+	 * buffered and O_DIRECT writes, but expecting them to
+	 * understand that file extension is also an implicit buffered
+	 * write is too much.  By getting the PR we force writeback of
+	 * the buffered zeroing before proceeding.
+	 */
+	ret = ocfs2_data_lock(inode, 0);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+	ocfs2_data_unlock(inode, 0);
+
 	ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
 					    inode->i_sb->s_bdev, iov, offset,
 					    nr_segs, 
 					    ocfs2_direct_IO_get_blocks,
 					    ocfs2_dio_end_io);
+out:
 	mlog_exit(ret);
 	return ret;
 }
diff --git a/fs/ocfs2/aops.h b/fs/ocfs2/aops.h
index d40456d..e88c3f0 100644
--- a/fs/ocfs2/aops.h
+++ b/fs/ocfs2/aops.h
@@ -22,8 +22,8 @@
 #ifndef OCFS2_AOPS_H
 #define OCFS2_AOPS_H
 
-int ocfs2_prepare_write(struct file *file, struct page *page,
-			unsigned from, unsigned to);
+int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
+			       unsigned from, unsigned to);
 
 struct ocfs2_journal_handle *ocfs2_start_walk_page_trans(struct inode *inode,
 							 struct page *page,
diff --git a/fs/ocfs2/extent_map.c b/fs/ocfs2/extent_map.c
index 4601fc2..1a5c690 100644
--- a/fs/ocfs2/extent_map.c
+++ b/fs/ocfs2/extent_map.c
@@ -569,7 +569,7 @@ static int ocfs2_extent_map_insert(struc
 
 	ret = -ENOMEM;
 	ctxt.new_ent = kmem_cache_alloc(ocfs2_em_ent_cachep,
-					GFP_KERNEL);
+					GFP_NOFS);
 	if (!ctxt.new_ent) {
 		mlog_errno(ret);
 		return ret;
@@ -583,14 +583,14 @@ static int ocfs2_extent_map_insert(struc
 		if (ctxt.need_left && !ctxt.left_ent) {
 			ctxt.left_ent =
 				kmem_cache_alloc(ocfs2_em_ent_cachep,
-						 GFP_KERNEL);
+						 GFP_NOFS);
 			if (!ctxt.left_ent)
 				break;
 		}
 		if (ctxt.need_right && !ctxt.right_ent) {
 			ctxt.right_ent =
 				kmem_cache_alloc(ocfs2_em_ent_cachep,
-						 GFP_KERNEL);
+						 GFP_NOFS);
 			if (!ctxt.right_ent)
 				break;
 		}
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 581eb45..a9559c8 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -613,7 +613,8 @@ leave:
 
 /* Some parts of this taken from generic_cont_expand, which turned out
  * to be too fragile to do exactly what we need without us having to
- * worry about recursive locking in ->commit_write(). */
+ * worry about recursive locking in ->prepare_write() and
+ * ->commit_write(). */
 static int ocfs2_write_zero_page(struct inode *inode,
 				 u64 size)
 {
@@ -641,7 +642,7 @@ static int ocfs2_write_zero_page(struct 
 		goto out;
 	}
 
-	ret = ocfs2_prepare_write(NULL, page, offset, offset);
+	ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
 	if (ret < 0) {
 		mlog_errno(ret);
 		goto out_unlock;
@@ -695,13 +696,26 @@ out:
 	return ret;
 }
 
+/* 
+ * A tail_to_skip value > 0 indicates that we're being called from
+ * ocfs2_file_aio_write(). This has the following implications:
+ *
+ * - we don't want to update i_size
+ * - di_bh will be NULL, which is fine because it's only used in the
+ *   case where we want to update i_size.
+ * - ocfs2_zero_extend() will then only be filling the hole created
+ *   between i_size and the start of the write.
+ */
 static int ocfs2_extend_file(struct inode *inode,
 			     struct buffer_head *di_bh,
-			     u64 new_i_size)
+			     u64 new_i_size,
+			     size_t tail_to_skip)
 {
 	int ret = 0;
 	u32 clusters_to_add;
 
+	BUG_ON(!tail_to_skip && !di_bh);
+
 	/* setattr sometimes calls us like this. */
 	if (new_i_size == 0)
 		goto out;
@@ -714,27 +728,44 @@ static int ocfs2_extend_file(struct inod
 		OCFS2_I(inode)->ip_clusters;
 
 	if (clusters_to_add) {
-		ret = ocfs2_extend_allocation(inode, clusters_to_add);
+		/* 
+		 * protect the pages that ocfs2_zero_extend is going to
+		 * be pulling into the page cache.. we do this before the
+		 * metadata extend so that we don't get into the situation
+		 * where we've extended the metadata but can't get the data
+		 * lock to zero.
+		 */
+		ret = ocfs2_data_lock(inode, 1);
 		if (ret < 0) {
 			mlog_errno(ret);
 			goto out;
 		}
 
-		ret = ocfs2_zero_extend(inode, new_i_size);
+		ret = ocfs2_extend_allocation(inode, clusters_to_add);
 		if (ret < 0) {
 			mlog_errno(ret);
-			goto out;
+			goto out_unlock;
 		}
-	} 
 
-	/* No allocation required, we just use this helper to
-	 * do a trivial update of i_size. */
-	ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out;
+		ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out_unlock;
+		}
+	}
+
+	if (!tail_to_skip) {
+		/* We're being called from ocfs2_setattr() which wants
+		 * us to update i_size */
+		ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
+		if (ret < 0)
+			mlog_errno(ret);
 	}
 
+out_unlock:
+	if (clusters_to_add) /* this is the only case in which we lock */
+		ocfs2_data_unlock(inode, 1);
+
 out:
 	return ret;
 }
@@ -793,7 +824,7 @@ #define OCFS2_VALID_ATTRS (ATTR_ATIME | 
 		if (i_size_read(inode) > attr->ia_size)
 			status = ocfs2_truncate_file(inode, bh, attr->ia_size);
 		else
-			status = ocfs2_extend_file(inode, bh, attr->ia_size);
+			status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
 		if (status < 0) {
 			if (status != -ENOSPC)
 				mlog_errno(status);
@@ -1049,21 +1080,12 @@ static ssize_t ocfs2_file_aio_write(stru
 		if (!clusters)
 			break;
 
-		ret = ocfs2_extend_allocation(inode, clusters);
+		ret = ocfs2_extend_file(inode, NULL, newsize, count);
 		if (ret < 0) {
 			if (ret != -ENOSPC)
 				mlog_errno(ret);
 			goto out;
 		}
-
-		/* Fill any holes which would've been created by this
-		 * write. If we're O_APPEND, this will wind up
-		 * (correctly) being a noop. */
-		ret = ocfs2_zero_extend(inode, (u64) newsize - count);
-		if (ret < 0) {
-			mlog_errno(ret);
-			goto out;
-		}
 		break;
 	}
 
@@ -1146,6 +1168,22 @@ static ssize_t ocfs2_file_aio_read(struc
 		ocfs2_iocb_set_rw_locked(iocb);
 	}
 
+	/*
+	 * We're fine letting folks race truncates and extending
+	 * writes with read across the cluster, just like they can
+	 * locally. Hence no rw_lock during read.
+	 * 
+	 * Take and drop the meta data lock to update inode fields
+	 * like i_size. This allows the checks down below
+	 * generic_file_aio_read() a chance of actually working. 
+	 */
+	ret = ocfs2_meta_lock(inode, NULL, NULL, 0);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto bail;
+	}
+	ocfs2_meta_unlock(inode, 0);
+
 	ret = generic_file_aio_read(iocb, buf, count, iocb->ki_pos);
 	if (ret == -EINVAL)
 		mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
index 6a610ae..eebc3cf 100644
--- a/fs/ocfs2/journal.c
+++ b/fs/ocfs2/journal.c
@@ -117,7 +117,7 @@ struct ocfs2_journal_handle *ocfs2_alloc
 {
 	struct ocfs2_journal_handle *retval = NULL;
 
-	retval = kcalloc(1, sizeof(*retval), GFP_KERNEL);
+	retval = kcalloc(1, sizeof(*retval), GFP_NOFS);
 	if (!retval) {
 		mlog(ML_ERROR, "Failed to allocate memory for journal "
 		     "handle!\n");
@@ -870,9 +870,11 @@ #define CONCURRENT_JOURNAL_FILL 32
 		if (p_blocks > CONCURRENT_JOURNAL_FILL)
 			p_blocks = CONCURRENT_JOURNAL_FILL;
 
+		/* We are reading journal data which should not
+		 * be put in the uptodate cache */
 		status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
 					   p_blkno, p_blocks, bhs, 0,
-					   inode);
+					   NULL);
 		if (status < 0) {
 			mlog_errno(status);
 			goto bail;
@@ -982,7 +984,7 @@ static void ocfs2_queue_recovery_complet
 {
 	struct ocfs2_la_recovery_item *item;
 
-	item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_KERNEL);
+	item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
 	if (!item) {
 		/* Though we wish to avoid it, we are in fact safe in
 		 * skipping local alloc cleanup as fsck.ocfs2 is more
diff --git a/fs/ocfs2/uptodate.c b/fs/ocfs2/uptodate.c
index 04a684d..b8a00a7 100644
--- a/fs/ocfs2/uptodate.c
+++ b/fs/ocfs2/uptodate.c
@@ -337,7 +337,7 @@ static void __ocfs2_set_buffer_uptodate(
 	     (unsigned long long)oi->ip_blkno,
 	     (unsigned long long)block, expand_tree);
 
-	new = kmem_cache_alloc(ocfs2_uptodate_cachep, GFP_KERNEL);
+	new = kmem_cache_alloc(ocfs2_uptodate_cachep, GFP_NOFS);
 	if (!new) {
 		mlog_errno(-ENOMEM);
 		return;
@@ -349,7 +349,7 @@ static void __ocfs2_set_buffer_uptodate(
 		 * has no way of tracking that. */
 		for(i = 0; i < OCFS2_INODE_MAX_CACHE_ARRAY; i++) {
 			tree[i] = kmem_cache_alloc(ocfs2_uptodate_cachep,
-						   GFP_KERNEL);
+						   GFP_NOFS);
 			if (!tree[i]) {
 				mlog_errno(-ENOMEM);
 				goto out_free;
diff --git a/fs/ocfs2/vote.c b/fs/ocfs2/vote.c
index 53049a2..ee42765 100644
--- a/fs/ocfs2/vote.c
+++ b/fs/ocfs2/vote.c
@@ -586,7 +586,7 @@ static struct ocfs2_net_wait_ctxt *ocfs2
 {
 	struct ocfs2_net_wait_ctxt *w;
 
-	w = kcalloc(1, sizeof(*w), GFP_KERNEL);
+	w = kcalloc(1, sizeof(*w), GFP_NOFS);
 	if (!w) {
 		mlog_errno(-ENOMEM);
 		goto bail;
@@ -749,7 +749,7 @@ static struct ocfs2_vote_msg * ocfs2_new
 
 	BUG_ON(!ocfs2_is_valid_vote_request(type));
 
-	request = kcalloc(1, sizeof(*request), GFP_KERNEL);
+	request = kcalloc(1, sizeof(*request), GFP_NOFS);
 	if (!request) {
 		mlog_errno(-ENOMEM);
 	} else {
@@ -1129,7 +1129,7 @@ static int ocfs2_handle_vote_message(str
 	struct ocfs2_super *osb = data;
 	struct ocfs2_vote_work *work;
 
-	work = kmalloc(sizeof(struct ocfs2_vote_work), GFP_KERNEL);
+	work = kmalloc(sizeof(struct ocfs2_vote_work), GFP_NOFS);
 	if (!work) {
 		status = -ENOMEM;
 		mlog_errno(status);

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2006-05-12  0:21 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2006-05-12  0:21 UTC (permalink / raw)
  To: Andrew Morton, Linus Torvalds; +Cc: ocfs2-devel, linux-kernel

Some OCFS2 bug fixes, and a couple configfs ones. All of them are issues
that folks have actually hit.

Please pull from 'upstream-linus' branch of
git://oss.oracle.com/home/sourcebo/git/ocfs2.git

to receive the following bug fixes:

 fs/Makefile           |    2 
 fs/configfs/dir.c     |  137 +++++++++++++++++++++++++++++++++++---------------
 fs/ocfs2/aops.c       |   46 ++++++++++++++--
 fs/ocfs2/aops.h       |    4 -
 fs/ocfs2/extent_map.c |    6 +-
 fs/ocfs2/file.c       |   86 ++++++++++++++++++++++---------
 fs/ocfs2/journal.c    |    8 +-
 fs/ocfs2/uptodate.c   |    4 -
 fs/ocfs2/vote.c       |    6 +-
 9 files changed, 213 insertions(+), 86 deletions(-)

Joel Becker:
      configfs: Fix a reference leak in configfs_mkdir().
      configfs: configfs_mkdir() failed to cleanup linkage.
      configfs: Make sure configfs_init() is called before consumers.

Mark Fasheh:
      ocfs2: take data locks around extend
      ocfs2: take meta data lock in ocfs2_file_aio_read()
      ocfs2: Don't populate uptodate cache in ocfs2_force_read_journal()

Sunil Mushran:
      ocfs2: fix gfp mask in some file system paths

diff --git a/fs/Makefile b/fs/Makefile
index 83bf478..078d3d1 100644
--- a/fs/Makefile
+++ b/fs/Makefile
@@ -45,6 +45,7 @@ obj-$(CONFIG_DNOTIFY)		+= dnotify.o
 obj-$(CONFIG_PROC_FS)		+= proc/
 obj-y				+= partitions/
 obj-$(CONFIG_SYSFS)		+= sysfs/
+obj-$(CONFIG_CONFIGFS_FS)	+= configfs/
 obj-y				+= devpts/
 
 obj-$(CONFIG_PROFILING)		+= dcookies.o
@@ -100,5 +101,4 @@ obj-$(CONFIG_BEFS_FS)		+= befs/
 obj-$(CONFIG_HOSTFS)		+= hostfs/
 obj-$(CONFIG_HPPFS)		+= hppfs/
 obj-$(CONFIG_DEBUG_FS)		+= debugfs/
-obj-$(CONFIG_CONFIGFS_FS)	+= configfs/
 obj-$(CONFIG_OCFS2_FS)		+= ocfs2/
diff --git a/fs/configfs/dir.c b/fs/configfs/dir.c
index 5638c8f..5f95218 100644
--- a/fs/configfs/dir.c
+++ b/fs/configfs/dir.c
@@ -505,13 +505,15 @@ static int populate_groups(struct config
 	int i;
 
 	if (group->default_groups) {
-		/* FYI, we're faking mkdir here
+		/*
+		 * FYI, we're faking mkdir here
 		 * I'm not sure we need this semaphore, as we're called
 		 * from our parent's mkdir.  That holds our parent's
 		 * i_mutex, so afaik lookup cannot continue through our
 		 * parent to find us, let alone mess with our tree.
 		 * That said, taking our i_mutex is closer to mkdir
-		 * emulation, and shouldn't hurt. */
+		 * emulation, and shouldn't hurt.
+		 */
 		mutex_lock(&dentry->d_inode->i_mutex);
 
 		for (i = 0; group->default_groups[i]; i++) {
@@ -546,20 +548,34 @@ static void unlink_obj(struct config_ite
 
 		item->ci_group = NULL;
 		item->ci_parent = NULL;
+
+		/* Drop the reference for ci_entry */
 		config_item_put(item);
 
+		/* Drop the reference for ci_parent */
 		config_group_put(group);
 	}
 }
 
 static void link_obj(struct config_item *parent_item, struct config_item *item)
 {
-	/* Parent seems redundant with group, but it makes certain
-	 * traversals much nicer. */
+	/*
+	 * Parent seems redundant with group, but it makes certain
+	 * traversals much nicer.
+	 */
 	item->ci_parent = parent_item;
+
+	/*
+	 * We hold a reference on the parent for the child's ci_parent
+	 * link.
+	 */
 	item->ci_group = config_group_get(to_config_group(parent_item));
 	list_add_tail(&item->ci_entry, &item->ci_group->cg_children);
 
+	/*
+	 * We hold a reference on the child for ci_entry on the parent's
+	 * cg_children
+	 */
 	config_item_get(item);
 }
 
@@ -684,6 +700,10 @@ static void client_drop_item(struct conf
 	type = parent_item->ci_type;
 	BUG_ON(!type);
 
+	/*
+	 * If ->drop_item() exists, it is responsible for the
+	 * config_item_put().
+	 */
 	if (type->ct_group_ops && type->ct_group_ops->drop_item)
 		type->ct_group_ops->drop_item(to_config_group(parent_item),
 						item);
@@ -694,23 +714,28 @@ static void client_drop_item(struct conf
 
 static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 {
-	int ret;
+	int ret, module_got = 0;
 	struct config_group *group;
 	struct config_item *item;
 	struct config_item *parent_item;
 	struct configfs_subsystem *subsys;
 	struct configfs_dirent *sd;
 	struct config_item_type *type;
-	struct module *owner;
+	struct module *owner = NULL;
 	char *name;
 
-	if (dentry->d_parent == configfs_sb->s_root)
-		return -EPERM;
+	if (dentry->d_parent == configfs_sb->s_root) {
+		ret = -EPERM;
+		goto out;
+	}
 
 	sd = dentry->d_parent->d_fsdata;
-	if (!(sd->s_type & CONFIGFS_USET_DIR))
-		return -EPERM;
+	if (!(sd->s_type & CONFIGFS_USET_DIR)) {
+		ret = -EPERM;
+		goto out;
+	}
 
+	/* Get a working ref for the duration of this function */
 	parent_item = configfs_get_config_item(dentry->d_parent);
 	type = parent_item->ci_type;
 	subsys = to_config_group(parent_item)->cg_subsys;
@@ -719,15 +744,16 @@ static int configfs_mkdir(struct inode *
 	if (!type || !type->ct_group_ops ||
 	    (!type->ct_group_ops->make_group &&
 	     !type->ct_group_ops->make_item)) {
-		config_item_put(parent_item);
-		return -EPERM;  /* What lack-of-mkdir returns */
+		ret = -EPERM;  /* Lack-of-mkdir returns -EPERM */
+		goto out_put;
 	}
 
 	name = kmalloc(dentry->d_name.len + 1, GFP_KERNEL);
 	if (!name) {
-		config_item_put(parent_item);
-		return -ENOMEM;
+		ret = -ENOMEM;
+		goto out_put;
 	}
+
 	snprintf(name, dentry->d_name.len + 1, "%s", dentry->d_name.name);
 
 	down(&subsys->su_sem);
@@ -748,40 +774,67 @@ static int configfs_mkdir(struct inode *
 
 	kfree(name);
 	if (!item) {
-		config_item_put(parent_item);
-		return -ENOMEM;
+		/*
+		 * If item == NULL, then link_obj() was never called.
+		 * There are no extra references to clean up.
+		 */
+		ret = -ENOMEM;
+		goto out_put;
 	}
 
-	ret = -EINVAL;
+	/*
+	 * link_obj() has been called (via link_group() for groups).
+	 * From here on out, errors must clean that up.
+	 */
+
 	type = item->ci_type;
-	if (type) {
-		owner = type->ct_owner;
-		if (try_module_get(owner)) {
-			if (group) {
-				ret = configfs_attach_group(parent_item,
-							    item,
-							    dentry);
-			} else {
-				ret = configfs_attach_item(parent_item,
-							   item,
-							   dentry);
-			}
+	if (!type) {
+		ret = -EINVAL;
+		goto out_unlink;
+	}
 
-			if (ret) {
-				down(&subsys->su_sem);
-				if (group)
-					unlink_group(group);
-				else
-					unlink_obj(item);
-				client_drop_item(parent_item, item);
-				up(&subsys->su_sem);
+	owner = type->ct_owner;
+	if (!try_module_get(owner)) {
+		ret = -EINVAL;
+		goto out_unlink;
+	}
 
-				config_item_put(parent_item);
-				module_put(owner);
-			}
-		}
+	/*
+	 * I hate doing it this way, but if there is
+	 * an error,  module_put() probably should
+	 * happen after any cleanup.
+	 */
+	module_got = 1;
+
+	if (group)
+		ret = configfs_attach_group(parent_item, item, dentry);
+	else
+		ret = configfs_attach_item(parent_item, item, dentry);
+
+out_unlink:
+	if (ret) {
+		/* Tear down everything we built up */
+		down(&subsys->su_sem);
+		if (group)
+			unlink_group(group);
+		else
+			unlink_obj(item);
+		client_drop_item(parent_item, item);
+		up(&subsys->su_sem);
+
+		if (module_got)
+			module_put(owner);
 	}
 
+out_put:
+	/*
+	 * link_obj()/link_group() took a reference from child->parent,
+	 * so the parent is safely pinned.  We can drop our working
+	 * reference.
+	 */
+	config_item_put(parent_item);
+
+out:
 	return ret;
 }
 
@@ -801,6 +854,7 @@ static int configfs_rmdir(struct inode *
 	if (sd->s_type & CONFIGFS_USET_DEFAULT)
 		return -EPERM;
 
+	/* Get a working ref until we have the child */
 	parent_item = configfs_get_config_item(dentry->d_parent);
 	subsys = to_config_group(parent_item)->cg_subsys;
 	BUG_ON(!subsys);
@@ -817,6 +871,7 @@ static int configfs_rmdir(struct inode *
 		return ret;
 	}
 
+	/* Get a working ref for the duration of this function */
 	item = configfs_get_config_item(dentry);
 
 	/* Drop reference from above, item already holds one. */
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 0d858d0..47152bf 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -276,13 +276,29 @@ static int ocfs2_writepage(struct page *
 	return ret;
 }
 
+/* This can also be called from ocfs2_write_zero_page() which has done
+ * it's own cluster locking. */
+int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
+			       unsigned from, unsigned to)
+{
+	int ret;
+
+	down_read(&OCFS2_I(inode)->ip_alloc_sem);
+
+	ret = block_prepare_write(page, from, to, ocfs2_get_block);
+
+	up_read(&OCFS2_I(inode)->ip_alloc_sem);
+
+	return ret;
+}
+
 /*
  * ocfs2_prepare_write() can be an outer-most ocfs2 call when it is called
  * from loopback.  It must be able to perform its own locking around
  * ocfs2_get_block().
  */
-int ocfs2_prepare_write(struct file *file, struct page *page,
-			unsigned from, unsigned to)
+static int ocfs2_prepare_write(struct file *file, struct page *page,
+			       unsigned from, unsigned to)
 {
 	struct inode *inode = page->mapping->host;
 	int ret;
@@ -295,11 +311,7 @@ int ocfs2_prepare_write(struct file *fil
 		goto out;
 	}
 
-	down_read(&OCFS2_I(inode)->ip_alloc_sem);
-
-	ret = block_prepare_write(page, from, to, ocfs2_get_block);
-
-	up_read(&OCFS2_I(inode)->ip_alloc_sem);
+	ret = ocfs2_prepare_write_nolock(inode, page, from, to);
 
 	ocfs2_meta_unlock(inode, 0);
 out:
@@ -625,11 +637,31 @@ static ssize_t ocfs2_direct_IO(int rw,
 	int ret;
 
 	mlog_entry_void();
+
+	/*
+	 * We get PR data locks even for O_DIRECT.  This allows
+	 * concurrent O_DIRECT I/O but doesn't let O_DIRECT with
+	 * extending and buffered zeroing writes race.  If they did
+	 * race then the buffered zeroing could be written back after
+	 * the O_DIRECT I/O.  It's one thing to tell people not to mix
+	 * buffered and O_DIRECT writes, but expecting them to
+	 * understand that file extension is also an implicit buffered
+	 * write is too much.  By getting the PR we force writeback of
+	 * the buffered zeroing before proceeding.
+	 */
+	ret = ocfs2_data_lock(inode, 0);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+	ocfs2_data_unlock(inode, 0);
+
 	ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
 					    inode->i_sb->s_bdev, iov, offset,
 					    nr_segs, 
 					    ocfs2_direct_IO_get_blocks,
 					    ocfs2_dio_end_io);
+out:
 	mlog_exit(ret);
 	return ret;
 }
diff --git a/fs/ocfs2/aops.h b/fs/ocfs2/aops.h
index d40456d..e88c3f0 100644
--- a/fs/ocfs2/aops.h
+++ b/fs/ocfs2/aops.h
@@ -22,8 +22,8 @@
 #ifndef OCFS2_AOPS_H
 #define OCFS2_AOPS_H
 
-int ocfs2_prepare_write(struct file *file, struct page *page,
-			unsigned from, unsigned to);
+int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
+			       unsigned from, unsigned to);
 
 struct ocfs2_journal_handle *ocfs2_start_walk_page_trans(struct inode *inode,
 							 struct page *page,
diff --git a/fs/ocfs2/extent_map.c b/fs/ocfs2/extent_map.c
index 4601fc2..1a5c690 100644
--- a/fs/ocfs2/extent_map.c
+++ b/fs/ocfs2/extent_map.c
@@ -569,7 +569,7 @@ static int ocfs2_extent_map_insert(struc
 
 	ret = -ENOMEM;
 	ctxt.new_ent = kmem_cache_alloc(ocfs2_em_ent_cachep,
-					GFP_KERNEL);
+					GFP_NOFS);
 	if (!ctxt.new_ent) {
 		mlog_errno(ret);
 		return ret;
@@ -583,14 +583,14 @@ static int ocfs2_extent_map_insert(struc
 		if (ctxt.need_left && !ctxt.left_ent) {
 			ctxt.left_ent =
 				kmem_cache_alloc(ocfs2_em_ent_cachep,
-						 GFP_KERNEL);
+						 GFP_NOFS);
 			if (!ctxt.left_ent)
 				break;
 		}
 		if (ctxt.need_right && !ctxt.right_ent) {
 			ctxt.right_ent =
 				kmem_cache_alloc(ocfs2_em_ent_cachep,
-						 GFP_KERNEL);
+						 GFP_NOFS);
 			if (!ctxt.right_ent)
 				break;
 		}
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 581eb45..a9559c8 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -613,7 +613,8 @@ leave:
 
 /* Some parts of this taken from generic_cont_expand, which turned out
  * to be too fragile to do exactly what we need without us having to
- * worry about recursive locking in ->commit_write(). */
+ * worry about recursive locking in ->prepare_write() and
+ * ->commit_write(). */
 static int ocfs2_write_zero_page(struct inode *inode,
 				 u64 size)
 {
@@ -641,7 +642,7 @@ static int ocfs2_write_zero_page(struct 
 		goto out;
 	}
 
-	ret = ocfs2_prepare_write(NULL, page, offset, offset);
+	ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
 	if (ret < 0) {
 		mlog_errno(ret);
 		goto out_unlock;
@@ -695,13 +696,26 @@ out:
 	return ret;
 }
 
+/* 
+ * A tail_to_skip value > 0 indicates that we're being called from
+ * ocfs2_file_aio_write(). This has the following implications:
+ *
+ * - we don't want to update i_size
+ * - di_bh will be NULL, which is fine because it's only used in the
+ *   case where we want to update i_size.
+ * - ocfs2_zero_extend() will then only be filling the hole created
+ *   between i_size and the start of the write.
+ */
 static int ocfs2_extend_file(struct inode *inode,
 			     struct buffer_head *di_bh,
-			     u64 new_i_size)
+			     u64 new_i_size,
+			     size_t tail_to_skip)
 {
 	int ret = 0;
 	u32 clusters_to_add;
 
+	BUG_ON(!tail_to_skip && !di_bh);
+
 	/* setattr sometimes calls us like this. */
 	if (new_i_size == 0)
 		goto out;
@@ -714,27 +728,44 @@ static int ocfs2_extend_file(struct inod
 		OCFS2_I(inode)->ip_clusters;
 
 	if (clusters_to_add) {
-		ret = ocfs2_extend_allocation(inode, clusters_to_add);
+		/* 
+		 * protect the pages that ocfs2_zero_extend is going to
+		 * be pulling into the page cache.. we do this before the
+		 * metadata extend so that we don't get into the situation
+		 * where we've extended the metadata but can't get the data
+		 * lock to zero.
+		 */
+		ret = ocfs2_data_lock(inode, 1);
 		if (ret < 0) {
 			mlog_errno(ret);
 			goto out;
 		}
 
-		ret = ocfs2_zero_extend(inode, new_i_size);
+		ret = ocfs2_extend_allocation(inode, clusters_to_add);
 		if (ret < 0) {
 			mlog_errno(ret);
-			goto out;
+			goto out_unlock;
 		}
-	} 
 
-	/* No allocation required, we just use this helper to
-	 * do a trivial update of i_size. */
-	ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out;
+		ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out_unlock;
+		}
+	}
+
+	if (!tail_to_skip) {
+		/* We're being called from ocfs2_setattr() which wants
+		 * us to update i_size */
+		ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
+		if (ret < 0)
+			mlog_errno(ret);
 	}
 
+out_unlock:
+	if (clusters_to_add) /* this is the only case in which we lock */
+		ocfs2_data_unlock(inode, 1);
+
 out:
 	return ret;
 }
@@ -793,7 +824,7 @@ #define OCFS2_VALID_ATTRS (ATTR_ATIME | 
 		if (i_size_read(inode) > attr->ia_size)
 			status = ocfs2_truncate_file(inode, bh, attr->ia_size);
 		else
-			status = ocfs2_extend_file(inode, bh, attr->ia_size);
+			status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
 		if (status < 0) {
 			if (status != -ENOSPC)
 				mlog_errno(status);
@@ -1049,21 +1080,12 @@ static ssize_t ocfs2_file_aio_write(stru
 		if (!clusters)
 			break;
 
-		ret = ocfs2_extend_allocation(inode, clusters);
+		ret = ocfs2_extend_file(inode, NULL, newsize, count);
 		if (ret < 0) {
 			if (ret != -ENOSPC)
 				mlog_errno(ret);
 			goto out;
 		}
-
-		/* Fill any holes which would've been created by this
-		 * write. If we're O_APPEND, this will wind up
-		 * (correctly) being a noop. */
-		ret = ocfs2_zero_extend(inode, (u64) newsize - count);
-		if (ret < 0) {
-			mlog_errno(ret);
-			goto out;
-		}
 		break;
 	}
 
@@ -1146,6 +1168,22 @@ static ssize_t ocfs2_file_aio_read(struc
 		ocfs2_iocb_set_rw_locked(iocb);
 	}
 
+	/*
+	 * We're fine letting folks race truncates and extending
+	 * writes with read across the cluster, just like they can
+	 * locally. Hence no rw_lock during read.
+	 * 
+	 * Take and drop the meta data lock to update inode fields
+	 * like i_size. This allows the checks down below
+	 * generic_file_aio_read() a chance of actually working. 
+	 */
+	ret = ocfs2_meta_lock(inode, NULL, NULL, 0);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto bail;
+	}
+	ocfs2_meta_unlock(inode, 0);
+
 	ret = generic_file_aio_read(iocb, buf, count, iocb->ki_pos);
 	if (ret == -EINVAL)
 		mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
index 6a610ae..eebc3cf 100644
--- a/fs/ocfs2/journal.c
+++ b/fs/ocfs2/journal.c
@@ -117,7 +117,7 @@ struct ocfs2_journal_handle *ocfs2_alloc
 {
 	struct ocfs2_journal_handle *retval = NULL;
 
-	retval = kcalloc(1, sizeof(*retval), GFP_KERNEL);
+	retval = kcalloc(1, sizeof(*retval), GFP_NOFS);
 	if (!retval) {
 		mlog(ML_ERROR, "Failed to allocate memory for journal "
 		     "handle!\n");
@@ -870,9 +870,11 @@ #define CONCURRENT_JOURNAL_FILL 32
 		if (p_blocks > CONCURRENT_JOURNAL_FILL)
 			p_blocks = CONCURRENT_JOURNAL_FILL;
 
+		/* We are reading journal data which should not
+		 * be put in the uptodate cache */
 		status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
 					   p_blkno, p_blocks, bhs, 0,
-					   inode);
+					   NULL);
 		if (status < 0) {
 			mlog_errno(status);
 			goto bail;
@@ -982,7 +984,7 @@ static void ocfs2_queue_recovery_complet
 {
 	struct ocfs2_la_recovery_item *item;
 
-	item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_KERNEL);
+	item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
 	if (!item) {
 		/* Though we wish to avoid it, we are in fact safe in
 		 * skipping local alloc cleanup as fsck.ocfs2 is more
diff --git a/fs/ocfs2/uptodate.c b/fs/ocfs2/uptodate.c
index 04a684d..b8a00a7 100644
--- a/fs/ocfs2/uptodate.c
+++ b/fs/ocfs2/uptodate.c
@@ -337,7 +337,7 @@ static void __ocfs2_set_buffer_uptodate(
 	     (unsigned long long)oi->ip_blkno,
 	     (unsigned long long)block, expand_tree);
 
-	new = kmem_cache_alloc(ocfs2_uptodate_cachep, GFP_KERNEL);
+	new = kmem_cache_alloc(ocfs2_uptodate_cachep, GFP_NOFS);
 	if (!new) {
 		mlog_errno(-ENOMEM);
 		return;
@@ -349,7 +349,7 @@ static void __ocfs2_set_buffer_uptodate(
 		 * has no way of tracking that. */
 		for(i = 0; i < OCFS2_INODE_MAX_CACHE_ARRAY; i++) {
 			tree[i] = kmem_cache_alloc(ocfs2_uptodate_cachep,
-						   GFP_KERNEL);
+						   GFP_NOFS);
 			if (!tree[i]) {
 				mlog_errno(-ENOMEM);
 				goto out_free;
diff --git a/fs/ocfs2/vote.c b/fs/ocfs2/vote.c
index 53049a2..ee42765 100644
--- a/fs/ocfs2/vote.c
+++ b/fs/ocfs2/vote.c
@@ -586,7 +586,7 @@ static struct ocfs2_net_wait_ctxt *ocfs2
 {
 	struct ocfs2_net_wait_ctxt *w;
 
-	w = kcalloc(1, sizeof(*w), GFP_KERNEL);
+	w = kcalloc(1, sizeof(*w), GFP_NOFS);
 	if (!w) {
 		mlog_errno(-ENOMEM);
 		goto bail;
@@ -749,7 +749,7 @@ static struct ocfs2_vote_msg * ocfs2_new
 
 	BUG_ON(!ocfs2_is_valid_vote_request(type));
 
-	request = kcalloc(1, sizeof(*request), GFP_KERNEL);
+	request = kcalloc(1, sizeof(*request), GFP_NOFS);
 	if (!request) {
 		mlog_errno(-ENOMEM);
 	} else {
@@ -1129,7 +1129,7 @@ static int ocfs2_handle_vote_message(str
 	struct ocfs2_super *osb = data;
 	struct ocfs2_vote_work *work;
 
-	work = kmalloc(sizeof(struct ocfs2_vote_work), GFP_KERNEL);
+	work = kmalloc(sizeof(struct ocfs2_vote_work), GFP_NOFS);
 	if (!work) {
 		status = -ENOMEM;
 		mlog_errno(status);

^ permalink raw reply	[flat|nested] 28+ messages in thread

* [git patches] ocfs2 fixes
@ 2006-04-10 23:03 Mark Fasheh
  0 siblings, 0 replies; 28+ messages in thread
From: Mark Fasheh @ 2006-04-10 23:03 UTC (permalink / raw)
  To: Andrew Morton, Linus Torvalds; +Cc: ocfs2-devel, linux-kernel

We're post rc1 so this is fixes only. The configfs ones are minor, but
small, trivial, and obvious - so I included those as well.

Please pull from 'upstream-linus' branch of
git://oss.oracle.com/home/sourcebo/git/ocfs2.git

to receive the following updates:

 fs/Kconfig                   |    2 -
 fs/configfs/dir.c            |    2 -
 fs/ocfs2/cluster/heartbeat.c |   40 ++++++++++++++++++-----
 fs/ocfs2/dlm/userdlm.c       |   74 ++++++++++++++++++++++++++++++++-----------
 fs/ocfs2/file.c              |   19 ++++++-----
 5 files changed, 101 insertions(+), 36 deletions(-)

Adrian Bunk:
      CONFIGFS_FS must depend on SYSFS

Eric Sesterhenn:
      Bogus NULL pointer check in fs/configfs/dir.c

Mark Fasheh:
      ocfs2: multi node truncate fix
      ocfs2: remove an overly aggressive BUG() in dlmfs
      ocfs2: catch an invalid ast case in dlmfs
      ocfs2: Handle the DLM_CANCELGRANT case in user_unlock_ast()
      ocfs2: test and set teardown flag early in user_dlm_destroy_lock()
      ocfs2: Better I/O error handling in heartbeat

diff --git a/fs/Kconfig b/fs/Kconfig
index e207be6..97f3174 100644
--- a/fs/Kconfig
+++ b/fs/Kconfig
@@ -861,7 +861,7 @@ config RAMFS
 
 config CONFIGFS_FS
 	tristate "Userspace-driven configuration filesystem (EXPERIMENTAL)"
-	depends on EXPERIMENTAL
+	depends on SYSFS && EXPERIMENTAL
 	help
 	  configfs is a ram-based filesystem that provides the converse
 	  of sysfs's functionality. Where sysfs is a filesystem-based
diff --git a/fs/configfs/dir.c b/fs/configfs/dir.c
index 8ed9b06..5638c8f 100644
--- a/fs/configfs/dir.c
+++ b/fs/configfs/dir.c
@@ -504,7 +504,7 @@ static int populate_groups(struct config
 	int ret = 0;
 	int i;
 
-	if (group && group->default_groups) {
+	if (group->default_groups) {
 		/* FYI, we're faking mkdir here
 		 * I'm not sure we need this semaphore, as we're called
 		 * from our parent's mkdir.  That holds our parent's
diff --git a/fs/ocfs2/cluster/heartbeat.c b/fs/ocfs2/cluster/heartbeat.c
index bff0f0d..21f38ac 100644
--- a/fs/ocfs2/cluster/heartbeat.c
+++ b/fs/ocfs2/cluster/heartbeat.c
@@ -153,6 +153,7 @@ struct o2hb_region {
 struct o2hb_bio_wait_ctxt {
 	atomic_t          wc_num_reqs;
 	struct completion wc_io_complete;
+	int               wc_error;
 };
 
 static void o2hb_write_timeout(void *arg)
@@ -186,6 +187,7 @@ static inline void o2hb_bio_wait_init(st
 {
 	atomic_set(&wc->wc_num_reqs, num_ios);
 	init_completion(&wc->wc_io_complete);
+	wc->wc_error = 0;
 }
 
 /* Used in error paths too */
@@ -218,8 +220,10 @@ static int o2hb_bio_end_io(struct bio *b
 {
 	struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
 
-	if (error)
+	if (error) {
 		mlog(ML_ERROR, "IO Error %d\n", error);
+		wc->wc_error = error;
+	}
 
 	if (bio->bi_size)
 		return 1;
@@ -390,6 +394,8 @@ static int o2hb_read_slots(struct o2hb_r
 
 bail_and_wait:
 	o2hb_wait_on_io(reg, &wc);
+	if (wc.wc_error && !status)
+		status = wc.wc_error;
 
 	if (bios) {
 		for(i = 0; i < num_bios; i++)
@@ -790,20 +796,24 @@ static int o2hb_highest_node(unsigned lo
 	return highest;
 }
 
-static void o2hb_do_disk_heartbeat(struct o2hb_region *reg)
+static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
 {
 	int i, ret, highest_node, change = 0;
 	unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
 	struct bio *write_bio;
 	struct o2hb_bio_wait_ctxt write_wc;
 
-	if (o2nm_configured_node_map(configured_nodes, sizeof(configured_nodes)))
-		return;
+	ret = o2nm_configured_node_map(configured_nodes,
+				       sizeof(configured_nodes));
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
 
 	highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
 	if (highest_node >= O2NM_MAX_NODES) {
 		mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n");
-		return;
+		return -EINVAL;
 	}
 
 	/* No sense in reading the slots of nodes that don't exist
@@ -813,7 +823,7 @@ static void o2hb_do_disk_heartbeat(struc
 	ret = o2hb_read_slots(reg, highest_node + 1);
 	if (ret < 0) {
 		mlog_errno(ret);
-		return;
+		return ret;
 	}
 
 	/* With an up to date view of the slots, we can check that no
@@ -831,7 +841,7 @@ static void o2hb_do_disk_heartbeat(struc
 	ret = o2hb_issue_node_write(reg, &write_bio, &write_wc);
 	if (ret < 0) {
 		mlog_errno(ret);
-		return;
+		return ret;
 	}
 
 	i = -1;
@@ -847,6 +857,15 @@ static void o2hb_do_disk_heartbeat(struc
 	 */
 	o2hb_wait_on_io(reg, &write_wc);
 	bio_put(write_bio);
+	if (write_wc.wc_error) {
+		/* Do not re-arm the write timeout on I/O error - we
+		 * can't be sure that the new block ever made it to
+		 * disk */
+		mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
+		     write_wc.wc_error, reg->hr_dev_name);
+		return write_wc.wc_error;
+	}
+
 	o2hb_arm_write_timeout(reg);
 
 	/* let the person who launched us know when things are steady */
@@ -854,6 +873,8 @@ static void o2hb_do_disk_heartbeat(struc
 		if (atomic_dec_and_test(&reg->hr_steady_iterations))
 			wake_up(&o2hb_steady_queue);
 	}
+
+	return 0;
 }
 
 /* Subtract b from a, storing the result in a. a *must* have a larger
@@ -913,7 +934,10 @@ static int o2hb_thread(void *data)
 		 * likely to time itself out. */
 		do_gettimeofday(&before_hb);
 
-		o2hb_do_disk_heartbeat(reg);
+		i = 0;
+		do {
+			ret = o2hb_do_disk_heartbeat(reg);
+		} while (ret && ++i < 2);
 
 		do_gettimeofday(&after_hb);
 		elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
diff --git a/fs/ocfs2/dlm/userdlm.c b/fs/ocfs2/dlm/userdlm.c
index c3764f4..74ca4e5 100644
--- a/fs/ocfs2/dlm/userdlm.c
+++ b/fs/ocfs2/dlm/userdlm.c
@@ -139,6 +139,10 @@ static void user_ast(void *opaque)
 		return;
 	}
 
+	mlog_bug_on_msg(lockres->l_requested == LKM_IVMODE,
+			"Lockres %s, requested ivmode. flags 0x%x\n",
+			lockres->l_name, lockres->l_flags);
+
 	/* we're downconverting. */
 	if (lockres->l_requested < lockres->l_level) {
 		if (lockres->l_requested <=
@@ -229,23 +233,42 @@ static void user_unlock_ast(void *opaque
 
 	mlog(0, "UNLOCK AST called on lock %s\n", lockres->l_name);
 
-	if (status != DLM_NORMAL)
+	if (status != DLM_NORMAL && status != DLM_CANCELGRANT)
 		mlog(ML_ERROR, "Dlm returns status %d\n", status);
 
 	spin_lock(&lockres->l_lock);
-	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN)
+	/* The teardown flag gets set early during the unlock process,
+	 * so test the cancel flag to make sure that this ast isn't
+	 * for a concurrent cancel. */
+	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN
+	    && !(lockres->l_flags & USER_LOCK_IN_CANCEL)) {
 		lockres->l_level = LKM_IVMODE;
-	else {
+	} else if (status == DLM_CANCELGRANT) {
+		mlog(0, "Lock %s, cancel fails, flags 0x%x\n",
+		     lockres->l_name, lockres->l_flags);
+		/* We tried to cancel a convert request, but it was
+		 * already granted. Don't clear the busy flag - the
+		 * ast should've done this already. */
+		BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
+		lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
+		goto out_noclear;
+	} else {
+		BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
+		/* Cancel succeeded, we want to re-queue */
+		mlog(0, "Lock %s, cancel succeeds, flags 0x%x\n",
+		     lockres->l_name, lockres->l_flags);
 		lockres->l_requested = LKM_IVMODE; /* cancel an
 						    * upconvert
 						    * request. */
 		lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
 		/* we want the unblock thread to look at it again
 		 * now. */
-		__user_dlm_queue_lockres(lockres);
+		if (lockres->l_flags & USER_LOCK_BLOCKED)
+			__user_dlm_queue_lockres(lockres);
 	}
 
 	lockres->l_flags &= ~USER_LOCK_BUSY;
+out_noclear:
 	spin_unlock(&lockres->l_lock);
 
 	wake_up(&lockres->l_event);
@@ -268,13 +291,26 @@ static void user_dlm_unblock_lock(void *
 
 	spin_lock(&lockres->l_lock);
 
-	BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
-	BUG_ON(!(lockres->l_flags & USER_LOCK_QUEUED));
+	mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
+			"Lockres %s, flags 0x%x\n",
+			lockres->l_name, lockres->l_flags);
 
-	/* notice that we don't clear USER_LOCK_BLOCKED here. That's
-	 * for user_ast to do. */
+	/* notice that we don't clear USER_LOCK_BLOCKED here. If it's
+	 * set, we want user_ast clear it. */
 	lockres->l_flags &= ~USER_LOCK_QUEUED;
 
+	/* It's valid to get here and no longer be blocked - if we get
+	 * several basts in a row, we might be queued by the first
+	 * one, the unblock thread might run and clear the queued
+	 * flag, and finally we might get another bast which re-queues
+	 * us before our ast for the downconvert is called. */
+	if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
+		mlog(0, "Lockres %s, flags 0x%x: queued but not blocking\n",
+			lockres->l_name, lockres->l_flags);
+		spin_unlock(&lockres->l_lock);
+		goto drop_ref;
+	}
+
 	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
 		mlog(0, "lock is in teardown so we do nothing\n");
 		spin_unlock(&lockres->l_lock);
@@ -282,7 +318,9 @@ static void user_dlm_unblock_lock(void *
 	}
 
 	if (lockres->l_flags & USER_LOCK_BUSY) {
-		mlog(0, "BUSY flag detected...\n");
+		mlog(0, "Cancel lock %s, flags 0x%x\n",
+		     lockres->l_name, lockres->l_flags);
+
 		if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
 			spin_unlock(&lockres->l_lock);
 			goto drop_ref;
@@ -296,14 +334,7 @@ static void user_dlm_unblock_lock(void *
 				   LKM_CANCEL,
 				   user_unlock_ast,
 				   lockres);
-		if (status == DLM_CANCELGRANT) {
-			/* If we got this, then the ast was fired
-			 * before we could cancel. We cleanup our
-			 * state, and restart the function. */
-			spin_lock(&lockres->l_lock);
-			lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
-			spin_unlock(&lockres->l_lock);
-		} else if (status != DLM_NORMAL)
+		if (status != DLM_NORMAL)
 			user_log_dlm_error("dlmunlock", status, lockres);
 		goto drop_ref;
 	}
@@ -581,6 +612,14 @@ int user_dlm_destroy_lock(struct user_lo
 	mlog(0, "asked to destroy %s\n", lockres->l_name);
 
 	spin_lock(&lockres->l_lock);
+	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
+		mlog(0, "Lock is already torn down\n");
+		spin_unlock(&lockres->l_lock);
+		return 0;
+	}
+
+	lockres->l_flags |= USER_LOCK_IN_TEARDOWN;
+
 	while (lockres->l_flags & USER_LOCK_BUSY) {
 		spin_unlock(&lockres->l_lock);
 
@@ -606,7 +645,6 @@ int user_dlm_destroy_lock(struct user_lo
 
 	lockres->l_flags &= ~USER_LOCK_ATTACHED;
 	lockres->l_flags |= USER_LOCK_BUSY;
-	lockres->l_flags |= USER_LOCK_IN_TEARDOWN;
 	spin_unlock(&lockres->l_lock);
 
 	mlog(0, "unlocking lockres %s\n", lockres->l_name);
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 34e903a..581eb45 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -260,6 +260,17 @@ static int ocfs2_truncate_file(struct in
 	if (new_i_size == le64_to_cpu(fe->i_size))
 		goto bail;
 
+	/* This forces other nodes to sync and drop their pages. Do
+	 * this even if we have a truncate without allocation change -
+	 * ocfs2 cluster sizes can be much greater than page size, so
+	 * we have to truncate them anyway.  */
+	status = ocfs2_data_lock(inode, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	ocfs2_data_unlock(inode, 1);
+
 	if (le32_to_cpu(fe->i_clusters) ==
 	    ocfs2_clusters_for_bytes(osb->sb, new_i_size)) {
 		mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
@@ -272,14 +283,6 @@ static int ocfs2_truncate_file(struct in
 		goto bail;
 	}
 
-	/* This forces other nodes to sync and drop their pages */
-	status = ocfs2_data_lock(inode, 1);
-	if (status < 0) {
-		mlog_errno(status);
-		goto bail;
-	}
-	ocfs2_data_unlock(inode, 1);
-
 	/* alright, we're going to need to do a full blown alloc size
 	 * change. Orphan the inode so that recovery can complete the
 	 * truncate if necessary. This does the task of marking

^ permalink raw reply	[flat|nested] 28+ messages in thread

end of thread, other threads:[~2009-03-13 17:14 UTC | newest]

Thread overview: 28+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2007-01-22 19:02 [git patches] ocfs2 fixes Mark Fasheh
  -- strict thread matches above, loose matches on Subject: below --
2009-03-13 17:14 [git patches] Ocfs2 fixes Mark Fasheh
2009-02-26 19:57 Mark Fasheh
2008-12-17 19:34 Mark Fasheh
2008-12-01 22:54 Mark Fasheh
2008-11-04 18:32 Mark Fasheh
2008-06-16 20:12 Mark Fasheh
2008-06-01 16:16 Mark Fasheh
2008-05-02 18:35 Mark Fasheh
2008-03-10 22:44 [git patches] ocfs2 fixes Mark Fasheh
2008-03-04 18:46 Mark Fasheh
2007-12-17 18:58 Mark Fasheh
2007-11-28  1:15 Mark Fasheh
2007-11-07 17:41 Mark Fasheh
2007-09-21 16:36 Mark Fasheh
2007-09-11 22:32 Mark Fasheh
2007-08-10 18:43 Mark Fasheh
2007-06-08 17:48 Mark Fasheh
2007-05-03 18:57 Mark Fasheh
2007-03-28 20:29 Mark Fasheh
2007-02-02 17:28 Mark Fasheh
2007-01-29 20:04 Mark Fasheh
2006-12-29 23:18 Mark Fasheh
2006-06-21 18:43 Mark Fasheh
2006-06-19 18:46 Mark Fasheh
2006-05-17 22:21 Mark Fasheh
2006-05-12  0:21 Mark Fasheh
2006-04-10 23:03 Mark Fasheh

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).