LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* [PATCH] proc: extend /proc/<pid>/fdinfo/<fd>
@ 2008-02-09 12:01 Eugene Teo
  2008-02-11  9:08 ` Miklos Szeredi
  2008-02-11 10:20 ` KOSAKI Motohiro
  0 siblings, 2 replies; 5+ messages in thread
From: Eugene Teo @ 2008-02-09 12:01 UTC (permalink / raw)
  To: linux-kernel; +Cc: Miklos Szeredi

This patch extends /proc/<pid>/fdinfo/<fd> to report information about open
files, and pathname. This information can be useful to know when debugging an
application.

For each file descriptor, the information is provided in the following format:

kerndev: ~/code/kernel# cat /proc/`pgrep pickup`/fdinfo/6
mode:   0622
dev:    253,0
ino:    21463057
uid:    89
gid:    89
rdev:   0,0
pos:    0
flags:  04002 FD_CLOEXEC # if close-on-exec flag is set
path:   /var/spool/postfix/public/pickup

You can also use this to find out more information about locked open files:

kerndev: /proc# cat /proc/locks
1: POSIX  ADVISORY  WRITE 2663 fd:00:21398205 0 EOF
                          ^^^^       ^^^^^^^^
2: FLOCK  ADVISORY  WRITE 2654 fd:00:21398203 0 EOF
3: FLOCK  ADVISORY  WRITE 2564 fd:00:21463056 0 EOF
4: FLOCK  ADVISORY  WRITE 2266 fd:00:21398151 0 EOF
kerndev: /proc# cd 2663/fdinfo/
kerndev: /proc/2663/fdinfo# grep 21398205 * -B2 -A6
               ^^^^
3-mode:	0644
3-dev:	253,0
3:ino:	21398205
        ^^^^^^^^
3-uid:	0
3-gid:	0
3-rdev:	0,0
3-pos:	5
3-flags:	02 FD_CLOEXEC
3-path:	/var/run/atd.pid

Signed-off-by: Eugene Teo <eugeneteo@kernel.sg>
---
 fs/proc/base.c |   47 +++++++++++++++++++++++++++++++++++++++++------
 1 files changed, 41 insertions(+), 6 deletions(-)

diff --git a/fs/proc/base.c b/fs/proc/base.c
index c59852b..dde617f 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -1402,7 +1402,7 @@ out:
 	return ~0U;
 }
 
-#define PROC_FDINFO_MAX 64
+#define PROC_FDINFO_MAX 256
 
 static int proc_fd_info(struct inode *inode, struct dentry **dentry,
 			struct vfsmount **mnt, char *info)
@@ -1411,6 +1411,7 @@ static int proc_fd_info(struct inode *inode, struct dentry **dentry,
 	struct files_struct *files = NULL;
 	struct file *file;
 	int fd = proc_fd(inode);
+	int ret = -ENOENT;
 
 	if (task) {
 		files = get_files_struct(task);
@@ -1424,24 +1425,58 @@ static int proc_fd_info(struct inode *inode, struct dentry **dentry,
 		spin_lock(&files->file_lock);
 		file = fcheck_files(files, fd);
 		if (file) {
+			ret = 0;
+
 			if (mnt)
 				*mnt = mntget(file->f_path.mnt);
 			if (dentry)
 				*dentry = dget(file->f_path.dentry);
-			if (info)
+			/* called by proc_fdinfo_read() */
+			if (info) {
+				struct fdtable *fdt = files_fdtable(files);
+				struct dentry *d = dget(file->f_path.dentry);
+				struct vfsmount *v = mntget(file->f_path.mnt);
+				char *page = (char *)__get_free_page(GFP_TEMPORARY);
+				int gcoe = FD_ISSET(fd, fdt->close_on_exec);
+				int dev_nr = d->d_inode->i_sb->s_dev;
+				int rdev_nr = d->d_inode->i_rdev;
+
+				if (!page) {
+					ret = -ENOMEM;
+					goto out;
+				}
+
 				snprintf(info, PROC_FDINFO_MAX,
+					 "mode:\t0%o\n"
+					 "dev:\t%d,%d\n"
+					 "ino:\t%ld\n"
+					 "uid:\t%d\n"
+					 "gid:\t%d\n"
+					 "rdev:\t%d,%d\n"
 					 "pos:\t%lli\n"
-					 "flags:\t0%o\n",
+					 "flags:\t0%o %s\n"
+					 "path:\t%s\n",
+					 d->d_inode->i_mode & 0777,
+					 MAJOR(dev_nr), MINOR(dev_nr),
+					 d->d_inode->i_ino,
+					 file->f_uid, file->f_gid,
+					 MAJOR(rdev_nr), MINOR(rdev_nr),
 					 (long long) file->f_pos,
-					 file->f_flags);
+					 file->f_flags, gcoe ? "FD_CLOEXEC" : "",
+					 d_path(d, v, page, PAGE_SIZE));
+				free_page((unsigned long)page);
+out:
+				mntput(v);
+				dput(d);
+			}
 			spin_unlock(&files->file_lock);
 			put_files_struct(files);
-			return 0;
+			return ret;
 		}
 		spin_unlock(&files->file_lock);
 		put_files_struct(files);
 	}
-	return -ENOENT;
+	return ret;
 }
 
 static int proc_fd_link(struct inode *inode, struct dentry **dentry,


^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] proc: extend /proc/<pid>/fdinfo/<fd>
  2008-02-09 12:01 [PATCH] proc: extend /proc/<pid>/fdinfo/<fd> Eugene Teo
@ 2008-02-11  9:08 ` Miklos Szeredi
  2008-02-11 14:23   ` Eugene Teo
  2008-02-11 10:20 ` KOSAKI Motohiro
  1 sibling, 1 reply; 5+ messages in thread
From: Miklos Szeredi @ 2008-02-11  9:08 UTC (permalink / raw)
  To: Eugene Teo; +Cc: linux-kernel


On Sat, 2008-02-09 at 20:01 +0800, Eugene Teo wrote:
> This patch extends /proc/<pid>/fdinfo/<fd> to report information about open
> files, and pathname. This information can be useful to know when debugging an
> application.
> 
> For each file descriptor, the information is provided in the following format:
> 
> kerndev: ~/code/kernel# cat /proc/`pgrep pickup`/fdinfo/6
> mode:   0622
> dev:    253,0
> ino:    21463057

These are already available via 'stat /proc/<pid>/fd/<fd>'

> uid:    89
> gid:    89

AFAICS file->f_[ug]id are only used by the netfilter code, but I guess
it wouldn't hurt to show them here.  But please add fields at the end of
the file, instead of the beggining.

> rdev:   0,0

This is in struct stat too.

> pos:    0
> flags:  04002 FD_CLOEXEC # if close-on-exec flag is set

FD_CLOEXEC should be on a separate line, because it's not an O_FOO flag.

> path:   /var/spool/postfix/public/pickup

readlink /proc/<pid>/fd/<fd>

Thanks,
Miklos

> 
> You can also use this to find out more information about locked open files:
> 
> kerndev: /proc# cat /proc/locks
> 1: POSIX  ADVISORY  WRITE 2663 fd:00:21398205 0 EOF
>                           ^^^^       ^^^^^^^^
> 2: FLOCK  ADVISORY  WRITE 2654 fd:00:21398203 0 EOF
> 3: FLOCK  ADVISORY  WRITE 2564 fd:00:21463056 0 EOF
> 4: FLOCK  ADVISORY  WRITE 2266 fd:00:21398151 0 EOF
> kerndev: /proc# cd 2663/fdinfo/
> kerndev: /proc/2663/fdinfo# grep 21398205 * -B2 -A6
>                ^^^^
> 3-mode:	0644
> 3-dev:	253,0
> 3:ino:	21398205
>         ^^^^^^^^
> 3-uid:	0
> 3-gid:	0
> 3-rdev:	0,0
> 3-pos:	5
> 3-flags:	02 FD_CLOEXEC
> 3-path:	/var/run/atd.pid
> 
> Signed-off-by: Eugene Teo <eugeneteo@kernel.sg>
> ---
>  fs/proc/base.c |   47 +++++++++++++++++++++++++++++++++++++++++------
>  1 files changed, 41 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index c59852b..dde617f 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -1402,7 +1402,7 @@ out:
>  	return ~0U;
>  }
>  
> -#define PROC_FDINFO_MAX 64
> +#define PROC_FDINFO_MAX 256
>  
>  static int proc_fd_info(struct inode *inode, struct dentry **dentry,
>  			struct vfsmount **mnt, char *info)
> @@ -1411,6 +1411,7 @@ static int proc_fd_info(struct inode *inode, struct dentry **dentry,
>  	struct files_struct *files = NULL;
>  	struct file *file;
>  	int fd = proc_fd(inode);
> +	int ret = -ENOENT;
>  
>  	if (task) {
>  		files = get_files_struct(task);
> @@ -1424,24 +1425,58 @@ static int proc_fd_info(struct inode *inode, struct dentry **dentry,
>  		spin_lock(&files->file_lock);
>  		file = fcheck_files(files, fd);
>  		if (file) {
> +			ret = 0;
> +
>  			if (mnt)
>  				*mnt = mntget(file->f_path.mnt);
>  			if (dentry)
>  				*dentry = dget(file->f_path.dentry);
> -			if (info)
> +			/* called by proc_fdinfo_read() */
> +			if (info) {
> +				struct fdtable *fdt = files_fdtable(files);
> +				struct dentry *d = dget(file->f_path.dentry);
> +				struct vfsmount *v = mntget(file->f_path.mnt);
> +				char *page = (char *)__get_free_page(GFP_TEMPORARY);
> +				int gcoe = FD_ISSET(fd, fdt->close_on_exec);
> +				int dev_nr = d->d_inode->i_sb->s_dev;
> +				int rdev_nr = d->d_inode->i_rdev;
> +
> +				if (!page) {
> +					ret = -ENOMEM;
> +					goto out;
> +				}
> +
>  				snprintf(info, PROC_FDINFO_MAX,
> +					 "mode:\t0%o\n"
> +					 "dev:\t%d,%d\n"
> +					 "ino:\t%ld\n"
> +					 "uid:\t%d\n"
> +					 "gid:\t%d\n"
> +					 "rdev:\t%d,%d\n"
>  					 "pos:\t%lli\n"
> -					 "flags:\t0%o\n",
> +					 "flags:\t0%o %s\n"
> +					 "path:\t%s\n",
> +					 d->d_inode->i_mode & 0777,
> +					 MAJOR(dev_nr), MINOR(dev_nr),
> +					 d->d_inode->i_ino,
> +					 file->f_uid, file->f_gid,
> +					 MAJOR(rdev_nr), MINOR(rdev_nr),
>  					 (long long) file->f_pos,
> -					 file->f_flags);
> +					 file->f_flags, gcoe ? "FD_CLOEXEC" : "",
> +					 d_path(d, v, page, PAGE_SIZE));
> +				free_page((unsigned long)page);
> +out:
> +				mntput(v);
> +				dput(d);
> +			}
>  			spin_unlock(&files->file_lock);
>  			put_files_struct(files);
> -			return 0;
> +			return ret;
>  		}
>  		spin_unlock(&files->file_lock);
>  		put_files_struct(files);
>  	}
> -	return -ENOENT;
> +	return ret;
>  }
>  
>  static int proc_fd_link(struct inode *inode, struct dentry **dentry,
> 


^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] proc: extend /proc/<pid>/fdinfo/<fd>
  2008-02-09 12:01 [PATCH] proc: extend /proc/<pid>/fdinfo/<fd> Eugene Teo
  2008-02-11  9:08 ` Miklos Szeredi
@ 2008-02-11 10:20 ` KOSAKI Motohiro
  2008-02-11 14:31   ` Eugene Teo
  1 sibling, 1 reply; 5+ messages in thread
From: KOSAKI Motohiro @ 2008-02-11 10:20 UTC (permalink / raw)
  To: Eugene Teo; +Cc: linux-kernel, Miklos Szeredi

Hi Eugene

In general, I think this patch isn't wrong idea.
but it shuld be brush up more, may be.

> kerndev: ~/code/kernel# cat /proc/`pgrep pickup`/fdinfo/6
> mode:   0622

I think this is inode attribute, but not fd attribute.

> dev:    253,0
> ino:    21463057

may be useful, agreed with you :)

> uid:    89
> gid:    89
> rdev:   0,0
> pos:    0
> flags:  04002 FD_CLOEXEC # if close-on-exec flag is set

agreed with Miklos's opinion.

> path:   /var/spool/postfix/public/pickup

may be, we shold think namespace..

> You can also use this to find out more information about locked open files:

is your requirement only fd -> ino pick up?

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] proc: extend /proc/<pid>/fdinfo/<fd>
  2008-02-11  9:08 ` Miklos Szeredi
@ 2008-02-11 14:23   ` Eugene Teo
  0 siblings, 0 replies; 5+ messages in thread
From: Eugene Teo @ 2008-02-11 14:23 UTC (permalink / raw)
  To: Miklos Szeredi; +Cc: linux-kernel

Hi Miklos,

<quote sender="Miklos Szeredi">
> On Sat, 2008-02-09 at 20:01 +0800, Eugene Teo wrote:
> > This patch extends /proc/<pid>/fdinfo/<fd> to report information about open
> > files, and pathname. This information can be useful to know when debugging an
> > application.
> > 
> > For each file descriptor, the information is provided in the following format:
> > 
> > kerndev: ~/code/kernel# cat /proc/`pgrep pickup`/fdinfo/6
> > mode:   0622
> > dev:    253,0
> > ino:    21463057
> 
> These are already available via 'stat /proc/<pid>/fd/<fd>'
> 
> > uid:    89
> > gid:    89
> 
> AFAICS file->f_[ug]id are only used by the netfilter code, but I guess
> it wouldn't hurt to show them here.  But please add fields at the end of
> the file, instead of the beggining.

Ok, will do.

> > rdev:   0,0
> 
> This is in struct stat too.
> 
> > pos:    0
> > flags:  04002 FD_CLOEXEC # if close-on-exec flag is set
> 
> FD_CLOEXEC should be on a separate line, because it's not an O_FOO flag.
> > path:   /var/spool/postfix/public/pickup
> 
> readlink /proc/<pid>/fd/<fd>

Thanks for the review. I agree that some of the information can be omitted
since it can be accessed using stat and readlink. I still think it is useful
to have ino in the output though.

Here's the updated patch:

[PATCH] proc: extend /proc/<pid>/fdinfo/<fd>

This patch extends /proc/<pid>/fdinfo/<fd> to report information about open
files, and pathname. This information can be useful to know when debugging an
application.

For each file descriptor, the information is provided in the following format:

kerndev: ~/code/kernel$ cat /proc/`pgrep firefox-bin`/fdinfo/29
pos:    49152
flags:  02
uid:    500
gid:    500
ino:    13205793
fd_flags:       FD_CLOEXEC # if close-on-exec flag is set

Signed-off-by: Eugene Teo <eugeneteo@kernel.sg>
---
 fs/proc/base.c |   22 ++++++++++++++++++----
 1 file changed, 18 insertions(+), 4 deletions(-)

diff --git a/fs/proc/base.c b/fs/proc/base.c
index c59852b..5ea5dee 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -1402,7 +1402,7 @@ out:
 	return ~0U;
 }
 
-#define PROC_FDINFO_MAX 64
+#define PROC_FDINFO_MAX 128
 
 static int proc_fd_info(struct inode *inode, struct dentry **dentry,
 			struct vfsmount **mnt, char *info)
@@ -1428,12 +1428,26 @@ static int proc_fd_info(struct inode *inode, struct dentry **dentry,
 				*mnt = mntget(file->f_path.mnt);
 			if (dentry)
 				*dentry = dget(file->f_path.dentry);
-			if (info)
+			/* called by proc_fdinfo_read() */
+			if (info) {
+				struct fdtable *fdt = files_fdtable(files);
+				struct dentry *d = dget(file->f_path.dentry);
+				int gcoe = FD_ISSET(fd, fdt->close_on_exec);
+
 				snprintf(info, PROC_FDINFO_MAX,
 					 "pos:\t%lli\n"
-					 "flags:\t0%o\n",
+					 "flags:\t0%o\n"
+					 "uid:\t%d\n"
+					 "gid:\t%d\n"
+					 "ino:\t%ld\n"
+					 "fd_flags:\t%s\n",
 					 (long long) file->f_pos,
-					 file->f_flags);
+					 file->f_flags,
+					 file->f_uid, file->f_gid,
+					 d->d_inode->i_ino,
+					 gcoe ? "FD_CLOEXEC" : "0");
+				dput(d);
+			}
 			spin_unlock(&files->file_lock);
 			put_files_struct(files);
 			return 0;


^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] proc: extend /proc/<pid>/fdinfo/<fd>
  2008-02-11 10:20 ` KOSAKI Motohiro
@ 2008-02-11 14:31   ` Eugene Teo
  0 siblings, 0 replies; 5+ messages in thread
From: Eugene Teo @ 2008-02-11 14:31 UTC (permalink / raw)
  To: KOSAKI Motohiro; +Cc: linux-kernel, Miklos Szeredi

Hi Motohiro-san,

<quote sender="KOSAKI Motohiro">
> In general, I think this patch isn't wrong idea.
> but it shuld be brush up more, may be.

Thanks.

> > kerndev: ~/code/kernel# cat /proc/`pgrep pickup`/fdinfo/6
> > mode:   0622
> 
> I think this is inode attribute, but not fd attribute.

Yes, it is an inode attribute.

> > dev:    253,0
> > ino:    21463057
> 
> may be useful, agreed with you :)

Thanks.

> > uid:    89
> > gid:    89
> > rdev:   0,0
> > pos:    0
> > flags:  04002 FD_CLOEXEC # if close-on-exec flag is set
> 
> agreed with Miklos's opinion.
> 
> > path:   /var/spool/postfix/public/pickup
> 
> may be, we shold think namespace..

Besides readlink, this info can be gathered by running:
ls -laF /proc/<pid>/fd/<fd>

So, I agree with Miklo's opinion that we can omit this in the output.

> > You can also use this to find out more information about locked open files:
> 
> is your requirement only fd -> ino pick up?

Not only this, but it is useful to know quickly if the open file descriptor
has a close-on-exec flag set.

Thanks,
Eugene

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2008-02-11 14:35 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2008-02-09 12:01 [PATCH] proc: extend /proc/<pid>/fdinfo/<fd> Eugene Teo
2008-02-11  9:08 ` Miklos Szeredi
2008-02-11 14:23   ` Eugene Teo
2008-02-11 10:20 ` KOSAKI Motohiro
2008-02-11 14:31   ` Eugene Teo

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).