Skip to contentSkip to conversation

autofs - fix fix symlinks arent checked for expiry

PATCH In review Ian Kent 2 participants 5 messages Dec 24, 2013 – Dec 27, 2013

Subsystems (2)
lore
Share & tools
lore

Ian Kent

author maintainer
#0 lore raw

PATCH 1/1 autofs - fix fix symlinks arent checked for expiry

When following a symlink the last_used counter is unconditionally updated causing the expire checks from user space to prevent expiry. Opps!

---
 fs/autofs4/symlink.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
fs/autofs4/symlink.c +2 -1
index 061a692..1e8ea19 100644
@@ -14,8 +14,9 @@
 
 static void *autofs4_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
+	struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
 	struct autofs_info *ino = autofs4_dentry_ino(dentry);
-	if (ino)
+	if (ino && !autofs4_oz_mode(sbi))
 		ino->last_used = jiffies;
 	nd_set_link(nd, dentry->d_inode->i_private);
 	return NULL;
#1 lore raw
Quoted 3 lines · On Tue, 24 Dec 2013 17:44:59 +0800 Ian Kent <[email protected]> wrote:

When following a symlink the last_used counter is unconditionally updated causing the expire checks from user space to prevent expiry. Opps!

A bit unclear. You're saying that userspace's act of checking expiry status will itself disrupt the expiry process?

Also, it's rather unclear what the userspace impact is here, and how severe it is. Please always carefully describe the user-visible impact so that others can decide which kernel version(s) need the patch.

fs/autofs4/symlink.c line 22

Quoted 13 lines
fs/autofs4/symlink.c +2 -1
@@ -14,8 +14,9 @@
 
 static void *autofs4_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
+	struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
 	struct autofs_info *ino = autofs4_dentry_ino(dentry);
-	if (ino)
+	if (ino && !autofs4_oz_mode(sbi))
 		ino->last_used = jiffies;
 	nd_set_link(nd, dentry->d_inode->i_private);
 	return NULL;

What kernel is this against? 3.13-rc5 is quite different:

static void *autofs4_follow_link(struct dentry *dentry, struct nameidata *nd)

{
	nd_set_link(nd, dentry->d_inode->i_private);
	return NULL;
}
author maintainer
#2 lore raw
Quoted 8 lines · Andrew Morton
Quoted 3 lines · On Tue, 24 Dec 2013 17:44:59 +0800 Ian Kent <[email protected]> wrote:

When following a symlink the last_used counter is unconditionally updated causing the expire checks from user space to prevent expiry. Opps!

A bit unclear. You're saying that userspace's act of checking expiry status will itself disrupt the expiry process?

If the user space expire code uses stat(2) instead of lstat(2), yes. It's quite possible this will be the case since it made no difference when not using symlinks in the autofs directory tree.

Quoted 4 lines

Also, it's rather unclear what the userspace impact is here, and how severe it is. Please always carefully describe the user-visible impact so that others can decide which kernel version(s) need the patch.

The impact of this is that symlinks within an an autofs directory tree don't cause a callback to the daemon so they can be expired (removed in this case).

autofs4_oz_mode() is the mechanism that's used to identify the user space process that's managing the automount tree. It's used in a number of places to prevent the process managing the tree from doing things like triggering mounts itself or updating the last_used counter.

It's a bit of a puzzle why it worked when I originally tested it. But later when I looked at it to work out why some symlinks weren't expiring it was obvious.

Do you want me to re-submit this with an updated description?

Quoted 5 lines
Quoted 5 lines
fs/autofs4/symlink.c +0 -0
@@ -14,8 +14,9 @@
 
Quoted 11 lines

fs/autofs4/symlink.c line 22

Quoted 9 lines
fs/autofs4/symlink.c +2 -1
 static void *autofs4_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
+	struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
 	struct autofs_info *ino = autofs4_dentry_ino(dentry);
-	if (ino)
+	if (ino && !autofs4_oz_mode(sbi))
 		ino->last_used = jiffies;
 	nd_set_link(nd, dentry->d_inode->i_private);
 	return NULL;

What kernel is this against? 3.13-rc5 is quite different:

That's a good question.
Which tree should I be basing patches on?

As it turns out it is against 3.13-rc5 which was the version the linus tree was at (when I pulled it) prior to mailing the patch.

Quoted 7 lines

static void *autofs4_follow_link(struct dentry *dentry, struct nameidata *nd)

{
	nd_set_link(nd, dentry->d_inode->i_private);
	return NULL;
}

Ian

#3 lore raw
Quoted 28 lines · On Fri, 27 Dec 2013 09:09:52 +0800 Ian Kent <[email protected]> wrote:
Quoted 8 lines · Andrew Morton
Quoted 3 lines · On Tue, 24 Dec 2013 17:44:59 +0800 Ian Kent <[email protected]> wrote:

When following a symlink the last_used counter is unconditionally updated causing the expire checks from user space to prevent expiry. Opps!

A bit unclear. You're saying that userspace's act of checking expiry status will itself disrupt the expiry process?

If the user space expire code uses stat(2) instead of lstat(2), yes. It's quite possible this will be the case since it made no difference when not using symlinks in the autofs directory tree.

Quoted 4 lines

Also, it's rather unclear what the userspace impact is here, and how severe it is. Please always carefully describe the user-visible impact so that others can decide which kernel version(s) need the patch.

The impact of this is that symlinks within an an autofs directory tree don't cause a callback to the daemon so they can be expired (removed in this case).

autofs4_oz_mode() is the mechanism that's used to identify the user space process that's managing the automount tree. It's used in a number of places to prevent the process managing the tree from doing things like triggering mounts itself or updating the last_used counter.

Quoted 5 lines

It's a bit of a puzzle why it worked when I originally tested it. But later when I looked at it to work out why some symlinks weren't expiring it was obvious.

Do you want me to re-submit this with an updated description?

Yes please. The questions which a bugfix changelog should answer are "should this be backported to -stable and if so, why". It's also helpful if it answers "if not, why not".

Quoted 17 lines
Quoted 5 lines
Quoted 5 lines
fs/autofs4/symlink.c +0 -0
@@ -14,8 +14,9 @@
 
Quoted 11 lines

fs/autofs4/symlink.c line 22

Quoted 9 lines
fs/autofs4/symlink.c +2 -1
 static void *autofs4_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
+	struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
 	struct autofs_info *ino = autofs4_dentry_ino(dentry);
-	if (ino)
+	if (ino && !autofs4_oz_mode(sbi))
 		ino->last_used = jiffies;
 	nd_set_link(nd, dentry->d_inode->i_private);
 	return NULL;

What kernel is this against? 3.13-rc5 is quite different:

Quoted 2 lines

That's a good question.
Which tree should I be basing patches on?

Current Linus git is almost always the one to target.

Quoted 2 lines

As it turns out it is against 3.13-rc5 which was the version the linus tree was at (when I pulled it) prior to mailing the patch.

Nope, autofs4_follow_link() looks like
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/fs/autofs4/symlink.c
all the way back to 3.12.

author maintainer
#4 lore raw
Quoted 34 lines · Andrew Morton
Quoted 32 lines · On Fri, 27 Dec 2013 09:09:52 +0800 Ian Kent <[email protected]> wrote:
Quoted 8 lines · Andrew Morton
Quoted 3 lines · On Tue, 24 Dec 2013 17:44:59 +0800 Ian Kent <[email protected]> wrote:

When following a symlink the last_used counter is unconditionally updated causing the expire checks from user space to prevent expiry. Opps!

A bit unclear. You're saying that userspace's act of checking expiry status will itself disrupt the expiry process?

If the user space expire code uses stat(2) instead of lstat(2), yes. It's quite possible this will be the case since it made no difference when not using symlinks in the autofs directory tree.

Quoted 4 lines

Also, it's rather unclear what the userspace impact is here, and how severe it is. Please always carefully describe the user-visible impact so that others can decide which kernel version(s) need the patch.

The impact of this is that symlinks within an an autofs directory tree don't cause a callback to the daemon so they can be expired (removed in this case).

autofs4_oz_mode() is the mechanism that's used to identify the user space process that's managing the automount tree. It's used in a number of places to prevent the process managing the tree from doing things like triggering mounts itself or updating the last_used counter.

It's a bit of a puzzle why it worked when I originally tested it. But later when I looked at it to work out why some symlinks weren't expiring it was obvious.

Quoted 5 lines
Quoted 1 line

Do you want me to re-submit this with an updated description?

Yes please. The questions which a bugfix changelog should answer are "should this be backported to -stable and if so, why". It's also helpful if it answers "if not, why not".

Yes, that's not clear.
I'd say it's not necessary to back port.

But I'm really keen for this to be merged in the current kernel because I'm making upstream changes to autofs (adding a parser for amd style mount maps) that will be able to make use of it (I can give details if you want).

It's probably a month or two before I commit them to git, and longer before they get into a source release.

And I can include the kernel patch in the autofs git tree for those that need it.

OTOH this affects am-utils (the amd package, http://www.am-utils.org/)
now, so since it's a fairly small change, maybe it should be back ported.

Unfortunately it's even worse than it sounds because the am-utils package isn't receiving much attention, both upstream (for about six years now) and downstream (certainly in Fedora anyway).

am-utils in Fedora hasn't worked properly for quite some time and I was unable to get any sensible assistance from upstream.

I'll put together a more detailed description based on above and merge autofs-fix-symlinks-arent-checked-for-expiry.patch and autofs-fix-fix-symlinks-arent-checked-for-expiry.patch
and re-submit.

Quoted 30 lines
Quoted 18 lines
Quoted 6 lines
Quoted 6 lines
fs/autofs4/symlink.c +0 -0
@@ -14,8 +14,9 @@
 
Quoted 11 lines

fs/autofs4/symlink.c line 22

Quoted 9 lines
fs/autofs4/symlink.c +2 -1
 static void *autofs4_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
+	struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
 	struct autofs_info *ino = autofs4_dentry_ino(dentry);
-	if (ino)
+	if (ino && !autofs4_oz_mode(sbi))
 		ino->last_used = jiffies;
 	nd_set_link(nd, dentry->d_inode->i_private);
 	return NULL;

What kernel is this against? 3.13-rc5 is quite different:

Quoted 2 lines

That's a good question.
Which tree should I be basing patches on?

Current Linus git is almost always the one to target.

Quoted 2 lines

As it turns out it is against 3.13-rc5 which was the version the linus tree was at (when I pulled it) prior to mailing the patch.

Nope, autofs4_follow_link() looks like
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/fs/autofs4/symlink.c
all the way back to 3.12.