ocfs2: fix deadlock issue when taking inode lock at vfs entry points
authorEric Ren <zren@suse.com>
Wed, 22 Feb 2017 23:40:44 +0000 (15:40 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 23 Feb 2017 00:41:27 +0000 (16:41 -0800)
Commit 743b5f1434f5 ("ocfs2: take inode lock in ocfs2_iop_set/get_acl()")
results in a deadlock, as the author "Tariq Saeed" realized shortly
after the patch was merged.  The discussion happened here

  https://oss.oracle.com/pipermail/ocfs2-devel/2015-September/011085.html

The reason why taking cluster inode lock at vfs entry points opens up a
self deadlock window, is explained in the previous patch of this series.

So far, we have seen two different code paths that have this issue.

1. do_sys_open
     may_open
      inode_permission
       ocfs2_permission
        ocfs2_inode_lock() <=== take PR
         generic_permission
          get_acl
           ocfs2_iop_get_acl
            ocfs2_inode_lock() <=== take PR

2. fchmod|fchmodat
    chmod_common
     notify_change
      ocfs2_setattr <=== take EX
       posix_acl_chmod
        get_acl
         ocfs2_iop_get_acl <=== take PR
        ocfs2_iop_set_acl <=== take EX

Fixes them by adding the tracking logic (in the previous patch) for these
funcs above, ocfs2_permission(), ocfs2_iop_[set|get]_acl(),
ocfs2_setattr().

Link: http://lkml.kernel.org/r/20170117100948.11657-3-zren@suse.com
Signed-off-by: Eric Ren <zren@suse.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Reviewed-by: Joseph Qi <jiangqi903@gmail.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/ocfs2/acl.c
fs/ocfs2/file.c

index bed1fcb630888746d28ba17e6ed3c0fdb6b2aa31..dc22ba8c710ff10de13526d52712ec4ef16cbb96 100644 (file)
@@ -283,16 +283,14 @@ int ocfs2_set_acl(handle_t *handle,
 int ocfs2_iop_set_acl(struct inode *inode, struct posix_acl *acl, int type)
 {
        struct buffer_head *bh = NULL;
-       int status = 0;
+       int status, had_lock;
+       struct ocfs2_lock_holder oh;
 
-       status = ocfs2_inode_lock(inode, &bh, 1);
-       if (status < 0) {
-               if (status != -ENOENT)
-                       mlog_errno(status);
-               return status;
-       }
+       had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
+       if (had_lock < 0)
+               return had_lock;
        status = ocfs2_set_acl(NULL, inode, bh, type, acl, NULL, NULL);
-       ocfs2_inode_unlock(inode, 1);
+       ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
        brelse(bh);
        return status;
 }
@@ -302,21 +300,20 @@ struct posix_acl *ocfs2_iop_get_acl(struct inode *inode, int type)
        struct ocfs2_super *osb;
        struct buffer_head *di_bh = NULL;
        struct posix_acl *acl;
-       int ret;
+       int had_lock;
+       struct ocfs2_lock_holder oh;
 
        osb = OCFS2_SB(inode->i_sb);
        if (!(osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL))
                return NULL;
-       ret = ocfs2_inode_lock(inode, &di_bh, 0);
-       if (ret < 0) {
-               if (ret != -ENOENT)
-                       mlog_errno(ret);
-               return ERR_PTR(ret);
-       }
+
+       had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
+       if (had_lock < 0)
+               return ERR_PTR(had_lock);
 
        acl = ocfs2_get_acl_nolock(inode, type, di_bh);
 
-       ocfs2_inode_unlock(inode, 0);
+       ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
        brelse(di_bh);
        return acl;
 }
index c4889655d32b5360228bb0c3deb7e729de855163..7b6a146327d7166f7505af53c1c50e07ae1310a1 100644 (file)
@@ -1138,6 +1138,8 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
        handle_t *handle = NULL;
        struct dquot *transfer_to[MAXQUOTAS] = { };
        int qtype;
+       int had_lock;
+       struct ocfs2_lock_holder oh;
 
        trace_ocfs2_setattr(inode, dentry,
                            (unsigned long long)OCFS2_I(inode)->ip_blkno,
@@ -1173,11 +1175,30 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
                }
        }
 
-       status = ocfs2_inode_lock(inode, &bh, 1);
-       if (status < 0) {
-               if (status != -ENOENT)
-                       mlog_errno(status);
+       had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
+       if (had_lock < 0) {
+               status = had_lock;
                goto bail_unlock_rw;
+       } else if (had_lock) {
+               /*
+                * As far as we know, ocfs2_setattr() could only be the first
+                * VFS entry point in the call chain of recursive cluster
+                * locking issue.
+                *
+                * For instance:
+                * chmod_common()
+                *  notify_change()
+                *   ocfs2_setattr()
+                *    posix_acl_chmod()
+                *     ocfs2_iop_get_acl()
+                *
+                * But, we're not 100% sure if it's always true, because the
+                * ordering of the VFS entry points in the call chain is out
+                * of our control. So, we'd better dump the stack here to
+                * catch the other cases of recursive locking.
+                */
+               mlog(ML_ERROR, "Another case of recursive locking:\n");
+               dump_stack();
        }
        inode_locked = 1;
 
@@ -1260,8 +1281,8 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
 bail_commit:
        ocfs2_commit_trans(osb, handle);
 bail_unlock:
-       if (status) {
-               ocfs2_inode_unlock(inode, 1);
+       if (status && inode_locked) {
+               ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
                inode_locked = 0;
        }
 bail_unlock_rw:
@@ -1279,7 +1300,7 @@ bail:
                        mlog_errno(status);
        }
        if (inode_locked)
-               ocfs2_inode_unlock(inode, 1);
+               ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
 
        brelse(bh);
        return status;
@@ -1320,21 +1341,32 @@ bail:
 
 int ocfs2_permission(struct inode *inode, int mask)
 {
-       int ret;
+       int ret, had_lock;
+       struct ocfs2_lock_holder oh;
 
        if (mask & MAY_NOT_BLOCK)
                return -ECHILD;
 
-       ret = ocfs2_inode_lock(inode, NULL, 0);
-       if (ret) {
-               if (ret != -ENOENT)
-                       mlog_errno(ret);
+       had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
+       if (had_lock < 0) {
+               ret = had_lock;
                goto out;
+       } else if (had_lock) {
+               /* See comments in ocfs2_setattr() for details.
+                * The call chain of this case could be:
+                * do_sys_open()
+                *  may_open()
+                *   inode_permission()
+                *    ocfs2_permission()
+                *     ocfs2_iop_get_acl()
+                */
+               mlog(ML_ERROR, "Another case of recursive locking:\n");
+               dump_stack();
        }
 
        ret = generic_permission(inode, mask);
 
-       ocfs2_inode_unlock(inode, 0);
+       ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
 out:
        return ret;
 }