Bug 426138

Description of problem:


Version-Release number of selected component (if applicable):

kernel-2.6.18-53.1.4.el5

(tested on both x86_64 and i686).

How reproducible:

100%

Steps to Reproduce:
1.  On an EL3 machine running 2.4.21-53.EL mount an NFS fs from server, do an ls -l
2.  e.g. mount  zex:/local/scratch/ /mnt/testing/
3.  ls -al /mnt/testing/test
  
Actual results:

$ ls -al /mnt/testing/test           
ls: /mnt/testing/test: Input/output error
-rw-r--r--    1 jp107    other         367 Dec 18 16:24 /mnt/testing/test

Expected results:

$ ls -al /mnt/testing/test           
-rw-r--r--    1 jp107    other         367 Dec 18 16:24 /mnt/testing/test

Additional info:

This seems to be caused by a bug in the nfs acl support code and mounting
without acls makes the error go away, e.g.

umount /mnt/testing
mount -o noacl zex:/local/scratch/ /mnt/testing/
ls -al /mnt/testing/test

doesn't show the problem.  The problem also doesn't seem to happen when using
newer (2.6) kernels so EL4/EL5 clients arn't affected.  This is using v3 mounts btw.

Looking at the logs on the server and tracing the traffic suggested the acl
issue and a few web searches showed up a discussion on the kernel list for a
similar sounding issue in (plain) 2.6.19.

  http://linux.derkeiler.com/Mailing-Lists/Kernel/2007-01/msg03478.html

(for example) describes a fix which ought to help, and adding something
equivalent to the kernel-2.6.18-53.1.4.el5 srpm does seem to fix it for my
simple tests.

Without that patch it seems to not fill in the nfsd_acl_versions[vers] fields so
probably doesn't handle ACLs at all as far as I can follow, but you probably
understand this stuff much better than I do!

I'll attach the patch I used to test, I re-did the patch from his post but
should just have the two changes he suggests.  I don't know if that made it into
the plain kernel tree or not...