ocfs2: fix deadlock caused by recursive locking in xattr
authorEric Ren <zren@suse.com>
Fri, 23 Jun 2017 22:08:55 +0000 (15:08 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 23 Jun 2017 23:15:55 +0000 (16:15 -0700)
Another deadlock path caused by recursive locking is reported.  This
kind of issue was introduced since commit 743b5f1434f5 ("ocfs2: take
inode lock in ocfs2_iop_set/get_acl()").  Two deadlock paths have been
fixed by commit b891fa5024a9 ("ocfs2: fix deadlock issue when taking
inode lock at vfs entry points").  Yes, we intend to fix this kind of
case in incremental way, because it's hard to find out all possible
paths at once.

This one can be reproduced like this.  On node1, cp a large file from
home directory to ocfs2 mountpoint.  While on node2, run
setfacl/getfacl.  Both nodes will hang up there.  The backtraces:

On node1:
  __ocfs2_cluster_lock.isra.39+0x357/0x740 [ocfs2]
  ocfs2_inode_lock_full_nested+0x17d/0x840 [ocfs2]
  ocfs2_write_begin+0x43/0x1a0 [ocfs2]
  generic_perform_write+0xa9/0x180
  __generic_file_write_iter+0x1aa/0x1d0
  ocfs2_file_write_iter+0x4f4/0xb40 [ocfs2]
  __vfs_write+0xc3/0x130
  vfs_write+0xb1/0x1a0
  SyS_write+0x46/0xa0

On node2:
  __ocfs2_cluster_lock.isra.39+0x357/0x740 [ocfs2]
  ocfs2_inode_lock_full_nested+0x17d/0x840 [ocfs2]
  ocfs2_xattr_set+0x12e/0xe80 [ocfs2]
  ocfs2_set_acl+0x22d/0x260 [ocfs2]
  ocfs2_iop_set_acl+0x65/0xb0 [ocfs2]
  set_posix_acl+0x75/0xb0
  posix_acl_xattr_set+0x49/0xa0
  __vfs_setxattr+0x69/0x80
  __vfs_setxattr_noperm+0x72/0x1a0
  vfs_setxattr+0xa7/0xb0
  setxattr+0x12d/0x190
  path_setxattr+0x9f/0xb0
  SyS_setxattr+0x14/0x20

Fix this one by using ocfs2_inode_{lock|unlock}_tracker, which is
exported by commit 439a36b8ef38 ("ocfs2/dlmglue: prepare tracking logic
to avoid recursive cluster lock").

Link: http://lkml.kernel.org/r/20170622014746.5815-1-zren@suse.com
Fixes: 743b5f1434f5 ("ocfs2: take inode lock in ocfs2_iop_set/get_acl()")
Signed-off-by: Eric Ren <zren@suse.com>
Reported-by: Thomas Voegtle <tv@lio96.de>
Tested-by: Thomas Voegtle <tv@lio96.de>
Reviewed-by: Joseph Qi <jiangqi903@gmail.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/ocfs2/dlmglue.c
fs/ocfs2/xattr.c

index 3b7c937a36b528e67511a23b136215ffaab6d8e4..4689940a953c2f7fc3b3a0e07a7d01c8c09e7d9a 100644 (file)
@@ -2591,6 +2591,10 @@ void ocfs2_inode_unlock_tracker(struct inode *inode,
        struct ocfs2_lock_res *lockres;
 
        lockres = &OCFS2_I(inode)->ip_inode_lockres;
+       /* had_lock means that the currect process already takes the cluster
+        * lock previously. If had_lock is 1, we have nothing to do here, and
+        * it will get unlocked where we got the lock.
+        */
        if (!had_lock) {
                ocfs2_remove_holder(lockres, oh);
                ocfs2_inode_unlock(inode, ex);
index 3c5384d9b3a549f319b114a782c7daab966a2c28..f70c3778d600c6be63996572bee6fd46ac03440c 100644 (file)
@@ -1328,20 +1328,21 @@ static int ocfs2_xattr_get(struct inode *inode,
                           void *buffer,
                           size_t buffer_size)
 {
-       int ret;
+       int ret, had_lock;
        struct buffer_head *di_bh = NULL;
+       struct ocfs2_lock_holder oh;
 
-       ret = ocfs2_inode_lock(inode, &di_bh, 0);
-       if (ret < 0) {
-               mlog_errno(ret);
-               return ret;
+       had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
+       if (had_lock < 0) {
+               mlog_errno(had_lock);
+               return had_lock;
        }
        down_read(&OCFS2_I(inode)->ip_xattr_sem);
        ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
                                     name, buffer, buffer_size);
        up_read(&OCFS2_I(inode)->ip_xattr_sem);
 
-       ocfs2_inode_unlock(inode, 0);
+       ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
 
        brelse(di_bh);
 
@@ -3537,11 +3538,12 @@ int ocfs2_xattr_set(struct inode *inode,
 {
        struct buffer_head *di_bh = NULL;
        struct ocfs2_dinode *di;
-       int ret, credits, ref_meta = 0, ref_credits = 0;
+       int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
        struct inode *tl_inode = osb->osb_tl_inode;
        struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
        struct ocfs2_refcount_tree *ref_tree = NULL;
+       struct ocfs2_lock_holder oh;
 
        struct ocfs2_xattr_info xi = {
                .xi_name_index = name_index,
@@ -3572,8 +3574,9 @@ int ocfs2_xattr_set(struct inode *inode,
                return -ENOMEM;
        }
 
-       ret = ocfs2_inode_lock(inode, &di_bh, 1);
-       if (ret < 0) {
+       had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
+       if (had_lock < 0) {
+               ret = had_lock;
                mlog_errno(ret);
                goto cleanup_nolock;
        }
@@ -3670,7 +3673,7 @@ cleanup:
                if (ret)
                        mlog_errno(ret);
        }
-       ocfs2_inode_unlock(inode, 1);
+       ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
 cleanup_nolock:
        brelse(di_bh);
        brelse(xbs.xattr_bh);