ufs: more deadlock prevention on tail unpacking
authorAl Viro <viro@zeniv.linux.org.uk>
Thu, 15 Jun 2017 04:42:56 +0000 (00:42 -0400)
committerAl Viro <viro@zeniv.linux.org.uk>
Thu, 15 Jun 2017 04:42:56 +0000 (00:42 -0400)
->s_lock is not needed for ufs_change_blocknr()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/ufs/balloc.c

index d56d9bc..0315fea 100644 (file)
@@ -478,6 +478,7 @@ u64 ufs_new_fragments(struct inode *inode, void *p, u64 fragment,
        if (result) {
                ufs_clear_frags(inode, result + oldcount, newcount - oldcount,
                                locked_page != NULL);
+               mutex_unlock(&UFS_SB(sb)->s_lock);
                ufs_change_blocknr(inode, fragment - oldcount, oldcount,
                                   uspi->s_sbbase + tmp,
                                   uspi->s_sbbase + result, locked_page);
@@ -487,7 +488,6 @@ u64 ufs_new_fragments(struct inode *inode, void *p, u64 fragment,
                UFS_I(inode)->i_lastfrag = max(UFS_I(inode)->i_lastfrag,
                                                fragment + count);
                write_sequnlock(&UFS_I(inode)->meta_lock);
-               mutex_unlock(&UFS_SB(sb)->s_lock);
                if (newcount < request)
                        ufs_free_fragments (inode, result + newcount, request - newcount);
                ufs_free_fragments (inode, tmp, oldcount);