Re: EXT4 ENOSPC Bug

[Theodore Tso]

On Mon, Feb 16, 2009 at 04:27:03PM +0100, Andres Freund wrote:
>
> So, yes, seems to be an inode allocation problem.
>

Andres, Alex, others,

I'm pretty sure the ENOSPC problem which you both found is an inode
allocation problem.  Some of you seem to have an easier time
reproducing it than others; could you try this patch, and periodically
scan your system logs for the message "ext4: find_group_flex failed,
fallback succeeded"?  If the problem goes away for you, and you find
the occasional aforemention message in your system log, that will
confirm what I suspect, which is the bug is in fs/ext4/inode.c's
find_group_flex() function.  (If I'm wrong, the fallback code will
activate only when the filesystem is genuinely out of inodes, which
should be very rare.)

More comments are in the patch header.  My current long-term plan for
dealing with this is to enhance find_group_orlov() to and
find_group_other() to understand about flex_bg's.

							- Ted

commit 1012e25b371b203164e4766a98f1e696df68b56d
Author: Theodore Ts'o <tytso@xxxxxxx>
Date:   Mon Feb 16 13:51:16 2009 -0500

    ext4: Add fallback for find_group_flex
    
    This is a workaround for find_group_flex() which badly needs to be
    replaced.  One of its problems (besides ignoring the Orlov algorithm)
    is that it is a bit hyperactive about returning failure under
    suspicious circumstances.  This can lead to spurious ENOSPC failures.
    Work around this for now by retrying the search using
    find_group_other() if find_group_flex() returns -1.  If
    find_group_other() succeeds when find_group_flex(), log a warning
    message.  I can't quite find the motivation to spend effort working on
    fixing find_group_flex() given that I want to replace it all anyway
    (and in fact work on the replacement code is underway), so we may
    leave the workaround for as long as find_group_flex() stays in the
    kernel...
    
    Signed-off-by: "Theodore Ts'o" <tytso@xxxxxxx>

diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
index a200059..21080ab 100644
--- a/fs/ext4/ialloc.c
+++ b/fs/ext4/ialloc.c
@@ -715,6 +715,13 @@ struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode)
 
 	if (sbi->s_log_groups_per_flex) {
 		ret2 = find_group_flex(sb, dir, &group);
+		if (ret2 == -1) {
+			ret2 = find_group_other(sb, dir, &group);
+			if (ret2 == 0)
+				printk(KERN_NOTICE "ext4: find_group_flex "
+				       "failed, fallback succeeded dir %lu\n",
+				       dir->i_ino);
+		}
 		goto got_group;
 	}
 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/