Ocfs2 devel - Sep 2015 - [patch 14/28] ocfs2: take inode lock in ocfs2_iop_set/get_acl()

From: Tariq Saeed <tariq.x.saeed at oracle.com>
Subject: ocfs2: take inode lock in ocfs2_iop_set/get_acl()

Orabug: 20189959

This bug in mainline code is pointed out by Mark Fasheh.  When
ocfs2_iop_set_acl() and ocfs2_iop_get_acl() are entered from VFS layer,
inode lock is not held.  This seems to be regression from older kernels. 
The patch is to fix that.

Signed-off-by: Tariq Saeed <tariq.x.saeed at oracle.com>
Cc: Mark Fasheh <mfasheh at suse.com>
Cc: Joel Becker <jlbec at evilplan.org>
Signed-off-by: Andrew Morton <akpm at linux-foundation.org>
---

 fs/ocfs2/acl.c |   26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)

diff -puN
fs/ocfs2/acl.c~resubmit-ocfs2_iop_set-get_acl-called-from-the-vfs-so-take-inode-lock-v2second-version
fs/ocfs2/acl.c
---
a/fs/ocfs2/acl.c~resubmit-ocfs2_iop_set-get_acl-called-from-the-vfs-so-take-inode-lock-v2second-version
+++ a/fs/ocfs2/acl.c
@@ -284,7 +284,19 @@ int ocfs2_set_acl(handle_t *handle,
 
 int ocfs2_iop_set_acl(struct inode *inode, struct posix_acl *acl, int type)
 {
-	return ocfs2_set_acl(NULL, inode, NULL, type, acl, NULL, NULL);
+	struct buffer_head *bh = NULL;
+	int status = 0;
+
+	status = ocfs2_inode_lock(inode, &bh, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		return status;
+	}
+	status = ocfs2_set_acl(NULL, inode, bh, type, acl, NULL, NULL);
+	ocfs2_inode_unlock(inode, 1);
+	brelse(bh);
+	return status;
 }
 
 struct posix_acl *ocfs2_iop_get_acl(struct inode *inode, int type)
@@ -292,19 +304,21 @@ struct posix_acl *ocfs2_iop_get_acl(stru
 	struct ocfs2_super *osb;
 	struct buffer_head *di_bh = NULL;
 	struct posix_acl *acl;
-	int ret = -EAGAIN;
+	int ret;
 
 	osb = OCFS2_SB(inode->i_sb);
 	if (!(osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL))
 		return NULL;
-
-	ret = ocfs2_read_inode_block(inode, &di_bh);
-	if (ret < 0)
+	ret = ocfs2_inode_lock(inode, &di_bh, 0);
+	if (ret < 0) {
+		if (ret != -ENOENT)
+			mlog_errno(ret);
 		return ERR_PTR(ret);
+	}
 
 	acl = ocfs2_get_acl_nolock(inode, type, di_bh);
 
+	ocfs2_inode_unlock(inode, 0);
 	brelse(di_bh);
-
 	return acl;
 }
_

On Wed, Aug 26, 2015 at 03:11:57PM -0700, Andrew Morton
wrote:
> From: Tariq Saeed <tariq.x.saeed at oracle.com>
> Subject: ocfs2: take inode lock in ocfs2_iop_set/get_acl()
> 
> Orabug: 20189959
> 
> This bug in mainline code is pointed out by Mark Fasheh.  When
> ocfs2_iop_set_acl() and ocfs2_iop_get_acl() are entered from VFS layer,
> inode lock is not held.  This seems to be regression from older kernels. 
> The patch is to fix that.
> 
> Signed-off-by: Tariq Saeed <tariq.x.saeed at oracle.com>
> Cc: Mark Fasheh <mfasheh at suse.com>
> Cc: Joel Becker <jlbec at evilplan.org>
> Signed-off-by: Andrew Morton <akpm at linux-foundation.org>
Thank you for fixing this Tariq,

Reviewed-by: Mark Fasheh <mfasheh at suse.de>

--
Mark Fasheh

On 08/31/2015 12:44 PM, Mark Fasheh wrote:
> On Wed, Aug 26, 2015 at 03:11:57PM -0700, Andrew Morton wrote:
>> From: Tariq Saeed <tariq.x.saeed at oracle.com>
>> Subject: ocfs2: take inode lock in ocfs2_iop_set/get_acl()
>>
>> Orabug: 20189959
>>
>> This bug in mainline code is pointed out by Mark Fasheh.  When
>> ocfs2_iop_set_acl() and ocfs2_iop_get_acl() are entered from VFS layer,
>> inode lock is not held.  This seems to be regression from older
kernels.
>> The patch is to fix that.
>>
>> Signed-off-by: Tariq Saeed <tariq.x.saeed at oracle.com>
>> Cc: Mark Fasheh <mfasheh at suse.com>
>> Cc: Joel Becker <jlbec at evilplan.org>
>> Signed-off-by: Andrew Morton <akpm at linux-foundation.org>
> Thank you for fixing this Tariq,
>
> Reviewed-by: Mark Fasheh <mfasheh at suse.de>
>
> --
> Mark Fasheh
Hi Mark,
I realized that taking inode lock at vfs entry points opens up a self 
deadlock window
if a remote conversion req  to EX is blocked. The reason is this code path.
fchmod|fchmodat
   -> chmod_common
     -> notify_change
       -> ocfs2_setattr (takes inode lock EX) <<===        ->
posix_acl_chmod
         -> get_acl
           -> ocfs2_iop_get_acl (inode lock PR blocks behind remote EX 
conv) *
           -> ocfs2_iop_set_acl (inode lock EX blocks behind remote EX 
conv) *

* - self deadlock

I think this can be solved by introducing a flag OCFS2_LOCK_RECURSIVE 
to  ocfs2_cluster_lock().
The meaning of this flag is this. If the requesting level is <= 
lockres->l_level, in that case ignore
OCFS2_LOCK_BLOCKED and just inc the holder count. This will work for all 
req levels if l_level is EX.

         if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
             !ocfs2_may_continue_on_blocked_lock(lockres, level) || 
!(arg_flags & (OCFS2_LOCK_RECURSIVE) ...

ocfs2_iop_get|set_acl() will pass OCFS2_LOCK_RECURSIVE to 
ocfs2_cluster_lock().

I am looking for suggestions.

Thanks
-Tariq Saeed