Bug 239625

Ok, I've been able to reproduce this. I'll see if I can track down the cause...

Also, somewhat related is the fact that you get different errors when you try to
open a file this way, depending on whether the inode already exists in kernel
(-EROFS if it doesn't, and -EEXIST if it does). The difference seems to be that
in the first case (dentry doesn't exist yet), we fall into this in open_namei:

        /* Negative dentry, just create the file */
        if (!dentry->d_inode) {

...in the other case, d_inode already exists for the dentry and we don't. I need
to test whether this is also an issue on other filesystems as well...

This seems to be NFS-specific. ext3 does not show the same behavior (it returns
-EEXIST every time), and doesn't call vfs_create (so it's not falling into the
same if condition above).

It looks like the issue is this in nfs_lookup:

        /* If we're doing an exclusive create, optimize away the lookup */
        if (nfs_is_exclusive_create(dir, nd))
                goto no_entry;

This seems to make the dentry returned be a negative dentry, presuming that the
create call will fix everything up. Commenting this out corrects the problem,
but it may actually be better to fix this on the backend so that if the create
fails we just remove the dentry.

Actually, I think we need this patch:

http://git.kernel.org/gitweb.cgi?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=fd6840714d9cf6e93f1d42b904860a94df316b85

I'll have a closer look.

(In reply to comment #5)
> Actually, I think we need this patch:
> 
> NFS: nfs_lookup - don't hash dentry when optimising away the lookup

That looks very promising.  Thanks for tracking it down, and I'll see if I can
verify that it fixes the problem on our systems.

Created attachment 154754 [details]
proposed patch

This patch is a backport of the one I linked to earlier. I've not tested it so
not sure if it will help this. It looks like it will though...

This request was evaluated by Red Hat Product Management for inclusion in a Red
Hat Enterprise Linux maintenance release.  Product Management has requested
further review of this request by Red Hat Engineering, for potential
inclusion in a Red Hat Enterprise Linux Update release for currently deployed
products.  This request is not yet committed for inclusion in an Update
release.

(In reply to comment #7)
> Created an attachment (id=154754) [edit]
> proposed patch
> 
> This patch is a backport of the one I linked to earlier. I've not tested it so
> not sure if it will help this. It looks like it will though...

Even without this patch, I cannot reproduce the failure on 2.6.9-55.EL.  (I used
the CentOS RPM because our IT staff hasn't fetched the RHEL4u5 updates yet.)

[root@flatline ~]# uname -a
Linux flatline 2.6.9-55.ELsmp #1 SMP Wed May 2 14:04:42 EDT 2007 x86_64 x86_64
x86_64 GNU/Linux
[root@flatline ~]# dmesg | head -2
Bootdata ok (command line is ro root=/dev/VolGroup00/LogVol00 rhgb quiet vga=0x317)
Linux version 2.6.9-55.ELsmp (mockbuild.org) (gcc version 3.4.6
20060404 (Red Hat 3.4.6-8)) #1 SMP Wed May 2 14:04:42 EDT 2007
[root@flatline ~]# mount -o ro plum:/home/tmp /mnt/tmp
[root@flatline ~]# (cd /mnt/tmp; tar xvf /tmp/issue.tar)
issue
tar: issue: Cannot open: Read-only file system
tar: Error exit delayed from previous errors
[root@flatline ~]# ls -l /mnt/tmp
total 4
-rw-r--r--  1 adi adi 76 May  9 17:58 issue

I've repeated the test 4 times just to be sure; this sequence reliably
reproduces the failure on 2.6.9-42.0.10, and reliably does not fail on 2.6.9-55.EL.

And yet, I cannot see any sign of a patch in -55 that would resolve it.

committed in stream U6 build 55.12. A test kernel with this patch is available
from http://people.redhat.com/~jbaron/rhel4/

An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.

http://rhn.redhat.com/errata/RHBA-2007-0791.html