ext4: fix ext4_free_inode() vs. ext4_claim_inode() race
authorEric Sandeen <sandeen@redhat.com>
Wed, 4 Mar 2009 23:38:18 +0000 (18:38 -0500)
committerTheodore Ts'o <tytso@mit.edu>
Wed, 4 Mar 2009 23:38:18 +0000 (18:38 -0500)
I was seeing fsck errors on inode bitmaps after a 4 thread
dbench run on a 4 cpu machine:

Inode bitmap differences: -50736 -(50752--50753) etc...

I believe that this is because ext4_free_inode() uses atomic
bitops, and although ext4_new_inode() *used* to also use atomic
bitops for synchronization, commit
393418676a7602e1d7d3f6e560159c65c8cbd50e changed this to use
the sb_bgl_lock, so that we could also synchronize against
read_inode_bitmap and initialization of uninit inode tables.

However, that change left ext4_free_inode using atomic bitops,
which I think leaves no synchronization between setting &
unsetting bits in the inode table.

The below patch fixes it for me, although I wonder if we're
getting at all heavy-handed with this spinlock...

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
fs/ext4/ialloc.c

index f18a919be70becf286c74364345a464e4d6e75e6..627f8c3337a3a24314624683ceac71ed30c72e95 100644 (file)
@@ -188,7 +188,7 @@ void ext4_free_inode(handle_t *handle, struct inode *inode)
        struct ext4_group_desc *gdp;
        struct ext4_super_block *es;
        struct ext4_sb_info *sbi;
-       int fatal = 0, err, count;
+       int fatal = 0, err, count, cleared;
        ext4_group_t flex_group;
 
        if (atomic_read(&inode->i_count) > 1) {
@@ -248,8 +248,10 @@ void ext4_free_inode(handle_t *handle, struct inode *inode)
                goto error_return;
 
        /* Ok, now we can actually update the inode bitmaps.. */
-       if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
-                                       bit, bitmap_bh->b_data))
+       spin_lock(sb_bgl_lock(sbi, block_group));
+       cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
+       spin_unlock(sb_bgl_lock(sbi, block_group));
+       if (!cleared)
                ext4_error(sb, "ext4_free_inode",
                           "bit already cleared for inode %lu", ino);
        else {