[Lustre-discuss] LustreError: 26019:0:(file.c:3143:ll_inode_revalidate_fini()) failure -2 inode

Erich Focht efocht at gmail.com
Mon Jan 23 05:20:33 PST 2012


we are seeing this error a lot since we updated NFS exporting clients to 1.8.7 (oracle

> Jan 16 15:19:16 xxfs1 kernel: LustreError: 7312:0:(file.c:3329:ll_inode_revalidate_fini()) failure -2 inode 87425037
> Jan 16 15:19:16 xxfs1 kernel: LustreError: 7312:0:(file.c:3329:ll_inode_revalidate_fini()) Skipped 729 previous similar messages
I'm not sure whether the similar messages mean we see the same message for the same inode or for another one. In any case: the inode (file) is there, it was not deleted (as suggested in the last email by Oleg).

Looking at the code we seem to be in the "obscure races" case here:

static int ll_inode_revalidate_fini(struct inode *inode, int rc) {
        if (rc == -ENOENT) { /* Already unlinked. Just update nlink
                              * and return success */
                inode->i_nlink = 0;
                /* This path cannot be hit for regular files unless in
                 * case of obscure races, so no need to to validate
                 * size. */
                if (!S_ISREG(inode->i_mode) &&
                        return 0;

Could this, by any chance, be triggered by the patch that avoids dentry rehashing:
https://bugzilla.lustre.org/attachment.cgi?id=33119&action=diff  ?
This patch is included in 1.8.7 (oracle version).


On 06/17/2011 05:00 AM, Oleg Drokin wrote:
> Hello!
> On Jun 16, 2011, at 11:05 AM, fenix.serega at gmail.com wrote:
>> Lustre 1.8
>> A lot of LustreErrors on client:
>> ustreError: 8747:0:(file.c:3143:ll_inode_revalidate_fini()) Skipped 6 previous similar messages
>> LustreError: 8747:0:(file.c:3143:ll_inode_revalidate_fini()) failure -2 inode 63486047
>> What does it mean !? failure -2 !? It seems all working correctly, no errors on OSS's or MDS.
> Likely that's just a race you see when a lot of apps work in the same dir doing e.g. ls and rm.
> See https://bugzilla.lustre.org/show_bug.cgi?id=16377 in particular comment #40
> The message just tells us that a file we believe exists (we got it from lookup) does not
> exist anymore when we are trying to fetch its attributes.
> Bye,
>     Oleg
> --
> Oleg Drokin
> Senior Software Engineer
> Whamcloud, Inc.

More information about the lustre-discuss mailing list