LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* [PATCH] f2fs: quota: fix potential deadlock
@ 2021-09-03  2:38 Chao Yu
  2021-12-31  1:06 ` [f2fs-dev] " Shachar Raindel
  0 siblings, 1 reply; 3+ messages in thread
From: Chao Yu @ 2021-09-03  2:38 UTC (permalink / raw)
  To: jaegeuk; +Cc: linux-f2fs-devel, linux-kernel, Chao Yu, Yi Zhuang

As Yi Zhuang reported in bugzilla:

https://bugzilla.kernel.org/show_bug.cgi?id=214299

There is potential deadlock during quota data flush as below:

Thread A:			Thread B:
f2fs_dquot_acquire
down_read(&sbi->quota_sem)
				f2fs_write_checkpoint
				block_operations
				f2fs_look_all
				down_write(&sbi->cp_rwsem)
f2fs_quota_write
f2fs_write_begin
__do_map_lock
f2fs_lock_op
down_read(&sbi->cp_rwsem)
				__need_flush_qutoa
				down_write(&sbi->quota_sem)

This patch changes block_operations() to use trylock, if it fails,
it means there is potential quota data updater, in this condition,
let's flush quota data first and then trylock again to check dirty
status of quota data.

The side effect is: in heavy race condition (e.g. multi quota data
upaters vs quota data flusher), it may decrease the probability of
synchronizing quota data successfully in checkpoint() due to limited
retry time of quota flush.

Reported-by: Yi Zhuang <zhuangyi1@huawei.com>
Signed-off-by: Chao Yu <chao@kernel.org>
---
 fs/f2fs/checkpoint.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
index 7d8803a4cbc2..6f6a7d812d60 100644
--- a/fs/f2fs/checkpoint.c
+++ b/fs/f2fs/checkpoint.c
@@ -1159,7 +1159,8 @@ static bool __need_flush_quota(struct f2fs_sb_info *sbi)
 	if (!is_journalled_quota(sbi))
 		return false;
 
-	down_write(&sbi->quota_sem);
+	if (!down_write_trylock(&sbi->quota_sem))
+		return true;
 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
 		ret = false;
 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
-- 
2.32.0


^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [f2fs-dev] [PATCH] f2fs: quota: fix potential deadlock
  2021-09-03  2:38 [PATCH] f2fs: quota: fix potential deadlock Chao Yu
@ 2021-12-31  1:06 ` Shachar Raindel
  0 siblings, 0 replies; 3+ messages in thread
From: Shachar Raindel @ 2021-12-31  1:06 UTC (permalink / raw)
  To: Chao Yu; +Cc: jaegeuk, linux-kernel, Yi Zhuang, linux-f2fs-devel

Somewhat late to the party (i.e. 3 months late), happy mailbox cleanup holidays!

On Thu, Sep 2, 2021 at 8:04 PM Chao Yu <chao@kernel.org> wrote:
>
> As Yi Zhuang reported in bugzilla:
>
> https://bugzilla.kernel.org/show_bug.cgi?id=214299
>
Bug report is for kernel 5.3. When I reported very similar deadlock in
google-msm 4.9 tree (
https://lore.kernel.org/linux-f2fs-devel/20201128174124.22397-1-shacharr@gmail.com/t/
), you pointed out that the code was missing commits which removed the
cp_rwsem grabbing (which are also missing from kernel 5.3):


commit 435cbab95e3966cd8310addd9e9b758dce0e8b84
Author: Jaegeuk Kim <jaegeuk@kernel.org>
Date:   Thu Apr 9 10:25:21 2020 -0700

     f2fs: fix quota_sync failure due to f2fs_lock_op

commit ca7f76e680745d3b8a386638045f85dac1c4b2f4
Author: Chao Yu <chao@kernel.org>
Date:   Fri May 29 17:29:47 2020 +0800

     f2fs: fix wrong discard space

commit 79963d967b492876fa17c8c2c2c17b7438683d9b
Author: Chao Yu <chao@kernel.org>
Date:   Thu Jun 18 14:36:23 2020 +0800

     f2fs: shrink node_write lock coverage

Is this patch needed with these commits applied?


>
> There is potential deadlock during quota data flush as below:
>
> Thread A:                       Thread B:
> f2fs_dquot_acquire
> down_read(&sbi->quota_sem)
>                                 f2fs_write_checkpoint
>                                 block_operations
>                                 f2fs_look_all
>                                 down_write(&sbi->cp_rwsem)
> f2fs_quota_write
> f2fs_write_begin
> __do_map_lock
> f2fs_lock_op
> down_read(&sbi->cp_rwsem)
>                                 __need_flush_qutoa
>                                 down_write(&sbi->quota_sem)
>
> This patch changes block_operations() to use trylock, if it fails,
> it means there is potential quota data updater, in this condition,
> let's flush quota data first and then trylock again to check dirty
> status of quota data.
>
> The side effect is: in heavy race condition (e.g. multi quota data
> upaters vs quota data flusher), it may decrease the probability of
> synchronizing quota data successfully in checkpoint() due to limited
> retry time of quota flush.
>
> Reported-by: Yi Zhuang <zhuangyi1@huawei.com>
> Signed-off-by: Chao Yu <chao@kernel.org>

As this patch is applied in the mainline kernel, can we CC -stable to
get this patch into the various Android kernels? Specifically,
https://android.googlesource.com/kernel/msm/+/refs/tags/android-12.0.0_r0.21/fs/f2fs/checkpoint.c#1147
needs this patch (alongside many other google-msm kernel branches).

Thanks,
--Shachar

^ permalink raw reply	[flat|nested] 3+ messages in thread

* [PATCH] f2fs: quota: fix potential deadlock
@ 2021-07-19  8:46 Chao Yu
  0 siblings, 0 replies; 3+ messages in thread
From: Chao Yu @ 2021-07-19  8:46 UTC (permalink / raw)
  To: jaegeuk; +Cc: linux-f2fs-devel, linux-kernel, Chao Yu, Chao Yu

xfstest generic/587 reports a deadlock issue as below:

======================================================
WARNING: possible circular locking dependency detected
5.14.0-rc1 #69 Not tainted
------------------------------------------------------
repquota/8606 is trying to acquire lock:
ffff888022ac9320 (&sb->s_type->i_mutex_key#18){+.+.}-{3:3}, at: f2fs_quota_sync+0x207/0x300 [f2fs]

but task is already holding lock:
ffff8880084bcde8 (&sbi->quota_sem){.+.+}-{3:3}, at: f2fs_quota_sync+0x59/0x300 [f2fs]

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #2 (&sbi->quota_sem){.+.+}-{3:3}:
       __lock_acquire+0x648/0x10b0
       lock_acquire+0x128/0x470
       down_read+0x3b/0x2a0
       f2fs_quota_sync+0x59/0x300 [f2fs]
       f2fs_quota_on+0x48/0x100 [f2fs]
       do_quotactl+0x5e3/0xb30
       __x64_sys_quotactl+0x23a/0x4e0
       do_syscall_64+0x3b/0x90
       entry_SYSCALL_64_after_hwframe+0x44/0xae

-> #1 (&sbi->cp_rwsem){++++}-{3:3}:
       __lock_acquire+0x648/0x10b0
       lock_acquire+0x128/0x470
       down_read+0x3b/0x2a0
       f2fs_unlink+0x353/0x670 [f2fs]
       vfs_unlink+0x1c7/0x380
       do_unlinkat+0x413/0x4b0
       __x64_sys_unlinkat+0x50/0xb0
       do_syscall_64+0x3b/0x90
       entry_SYSCALL_64_after_hwframe+0x44/0xae

-> #0 (&sb->s_type->i_mutex_key#18){+.+.}-{3:3}:
       check_prev_add+0xdc/0xb30
       validate_chain+0xa67/0xb20
       __lock_acquire+0x648/0x10b0
       lock_acquire+0x128/0x470
       down_write+0x39/0xc0
       f2fs_quota_sync+0x207/0x300 [f2fs]
       do_quotactl+0xaff/0xb30
       __x64_sys_quotactl+0x23a/0x4e0
       do_syscall_64+0x3b/0x90
       entry_SYSCALL_64_after_hwframe+0x44/0xae

other info that might help us debug this:

Chain exists of:
  &sb->s_type->i_mutex_key#18 --> &sbi->cp_rwsem --> &sbi->quota_sem

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&sbi->quota_sem);
                               lock(&sbi->cp_rwsem);
                               lock(&sbi->quota_sem);
  lock(&sb->s_type->i_mutex_key#18);

 *** DEADLOCK ***

3 locks held by repquota/8606:
 #0: ffff88801efac0e0 (&type->s_umount_key#53){++++}-{3:3}, at: user_get_super+0xd9/0x190
 #1: ffff8880084bc380 (&sbi->cp_rwsem){++++}-{3:3}, at: f2fs_quota_sync+0x3e/0x300 [f2fs]
 #2: ffff8880084bcde8 (&sbi->quota_sem){.+.+}-{3:3}, at: f2fs_quota_sync+0x59/0x300 [f2fs]

stack backtrace:
CPU: 6 PID: 8606 Comm: repquota Not tainted 5.14.0-rc1 #69
Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
Call Trace:
 dump_stack_lvl+0xce/0x134
 dump_stack+0x17/0x20
 print_circular_bug.isra.0.cold+0x239/0x253
 check_noncircular+0x1be/0x1f0
 check_prev_add+0xdc/0xb30
 validate_chain+0xa67/0xb20
 __lock_acquire+0x648/0x10b0
 lock_acquire+0x128/0x470
 down_write+0x39/0xc0
 f2fs_quota_sync+0x207/0x300 [f2fs]
 do_quotactl+0xaff/0xb30
 __x64_sys_quotactl+0x23a/0x4e0
 do_syscall_64+0x3b/0x90
 entry_SYSCALL_64_after_hwframe+0x44/0xae
RIP: 0033:0x7f883b0b4efe

The root cause is ABBA deadlock of inode lock and cp_rwsem,
reorder locks in f2fs_quota_sync() as below to fix this issue:
- lock inode
- lock cp_rwsem
- lock quota_sem

Fixes: db6ec53b7e03 ("f2fs: add a rw_sem to cover quota flag changes")
Signed-off-by: Chao Yu <chao@kernel.org>
---
 fs/f2fs/super.c | 84 ++++++++++++++++++++++++++++---------------------
 1 file changed, 48 insertions(+), 36 deletions(-)

diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index 8fecd3050ccd..72eb9d70969f 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -2518,6 +2518,33 @@ static int f2fs_enable_quotas(struct super_block *sb)
 	return 0;
 }
 
+static int f2fs_quota_sync_file(struct f2fs_sb_info *sbi, int type)
+{
+	struct quota_info *dqopt = sb_dqopt(sbi->sb);
+	struct address_space *mapping = dqopt->files[type]->i_mapping;
+	int ret = 0;
+
+	ret = dquot_writeback_dquots(sbi->sb, type);
+	if (ret)
+		goto out;
+
+	ret = filemap_fdatawrite(mapping);
+	if (ret)
+		goto out;
+
+	/* if we are using journalled quota */
+	if (is_journalled_quota(sbi))
+		goto out;
+
+	ret = filemap_fdatawait(mapping);
+
+	truncate_inode_pages(&dqopt->files[type]->i_data, 0);
+out:
+	if (ret)
+		set_sbi_flag(sbi, SBI_QUOTA_NEED_REPAIR);
+	return ret;
+}
+
 int f2fs_quota_sync(struct super_block *sb, int type)
 {
 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
@@ -2525,57 +2552,42 @@ int f2fs_quota_sync(struct super_block *sb, int type)
 	int cnt;
 	int ret;
 
-	/*
-	 * do_quotactl
-	 *  f2fs_quota_sync
-	 *  down_read(quota_sem)
-	 *  dquot_writeback_dquots()
-	 *  f2fs_dquot_commit
-	 *                            block_operation
-	 *                            down_read(quota_sem)
-	 */
-	f2fs_lock_op(sbi);
-
-	down_read(&sbi->quota_sem);
-	ret = dquot_writeback_dquots(sb, type);
-	if (ret)
-		goto out;
-
 	/*
 	 * Now when everything is written we can discard the pagecache so
 	 * that userspace sees the changes.
 	 */
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
-		struct address_space *mapping;
 
 		if (type != -1 && cnt != type)
 			continue;
-		if (!sb_has_quota_active(sb, cnt))
-			continue;
 
-		mapping = dqopt->files[cnt]->i_mapping;
+		if (!sb_has_quota_active(sb, type))
+			return 0;
 
-		ret = filemap_fdatawrite(mapping);
-		if (ret)
-			goto out;
+		inode_lock(dqopt->files[cnt]);
 
-		/* if we are using journalled quota */
-		if (is_journalled_quota(sbi))
-			continue;
+		/*
+		 * do_quotactl
+		 *  f2fs_quota_sync
+		 *  down_read(quota_sem)
+		 *  dquot_writeback_dquots()
+		 *  f2fs_dquot_commit
+		 *			      block_operation
+		 *			      down_read(quota_sem)
+		 */
+		f2fs_lock_op(sbi);
+		down_read(&sbi->quota_sem);
 
-		ret = filemap_fdatawait(mapping);
-		if (ret)
-			set_sbi_flag(F2FS_SB(sb), SBI_QUOTA_NEED_REPAIR);
+		ret = f2fs_quota_sync_file(sbi, cnt);
+
+		up_read(&sbi->quota_sem);
+		f2fs_unlock_op(sbi);
 
-		inode_lock(dqopt->files[cnt]);
-		truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
 		inode_unlock(dqopt->files[cnt]);
+
+		if (ret)
+			break;
 	}
-out:
-	if (ret)
-		set_sbi_flag(F2FS_SB(sb), SBI_QUOTA_NEED_REPAIR);
-	up_read(&sbi->quota_sem);
-	f2fs_unlock_op(sbi);
 	return ret;
 }
 
-- 
2.22.1


^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2021-12-31  1:06 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-09-03  2:38 [PATCH] f2fs: quota: fix potential deadlock Chao Yu
2021-12-31  1:06 ` [f2fs-dev] " Shachar Raindel
  -- strict thread matches above, loose matches on Subject: below --
2021-07-19  8:46 Chao Yu

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).