xfs
[Top] [All Lists]

Re: stress test failures

To: marcelo.magallon@xxxxxxxxxxx (Marcelo E. Magallon)
Subject: Re: stress test failures
From: Timothy Shimmin <tes@xxxxxxxxxxxxxxxxxxxxxxx>
Date: Tue, 6 Feb 2001 13:00:42 +1100 (EDT)
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <20010205104330.A1108@xxxxxxxxxxxxxxx> from "Marcelo E. Magallon" at Feb 05, 2001 10:43:30 AM
Sender: owner-linux-xfs@xxxxxxxxxxx
Hi Marcelo,

You wrote:
>  I got a failure on test 020, short version:
> 
>  020 - output mismatch (see 020.out.bad)
>  65a66,68
>  > 0001520 e5 79 0d 00 03 00 00 00 c0 79 0d 00 00 00 00 00
>  > 0001540 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>  > *
>  
Looking at 020, the test is testing setting a long value
for the extended attribute.
It does this by trying to set a value of 100K worth of zeroes.
The attr(1) code truncates this to ATTR_MAX_VALUELEN (64K).
(I will have to update the man page for this - it says 256K). 
So it is quite concerning that when we extract back the 64K
value we get some non-zero values.
I will have to look at the code and see where we can trace
what is going wrong.

>  051 also fails:
> 
>  051 - output mismatch (see 051.out.bad)
>  24c24
>  < sh: ./file1: Permission denied
>  ---
>  > Test was executed
...
It seems that ALL the tests which verified that access
would be denied did not happen - access was granted in
each case.
Even the first permissions test case failed which actually had the DAC
with no access permissions other than for the owner.

I use a small program, xfstests/src/runas, to set up the 
effective uid, e-gid, and supplementary groups for a process. 
Could you do the following as root to verify it is working:
    [root@sagan src]# runas -u 10001 -g 10002 -s 10003 -s 10004 id
    uid=0(root) gid=0(root) euid=10001 egid=10002 groups=10003,10004


>  > Test was executed
>  56c56
>  < file1 [u::---,g::---,o::---,u:1002:r-x,m::rwx]
>  ---
>  > file1 [u::---,g::---,o::---,u:simvis:r-x,m::rwx]
>  65,66c65,66
>  < file1 [u::---,g::---,o::---,u:1002:r-x,m::rwx]
>  < file1 [u::---,g::---,o::---,g:1002:r-x,m::rwx]
>  ---
>  > file1 [u::---,g::---,o::---,u:simvis:r-x,m::rwx]
>  > file1 [u::---,g::---,o::---,g:xxxxxx:r-x,m::rwx]
>  
>  "xxxxxxx" is the name of the group with gid 1002 (someone arround here
>  could get uneasy if I disclose it).  I guess chacl needs a "numeric"
>  flag.
Ok, this is a problem with the qa test - oops.
I can't really change the text output of the acl cmd
because it uses acl_to_short_text() which is an acl posix function.
I will fix up the qa test to do some filtering/mapping.

BTW, test 051, will not run (and will output a msg to that effect)
if the chacl(1) command isn't there or if the ACL syscall is not
turned on in the kernel.

Thanks,
Tim.

<Prev in Thread] Current Thread [Next in Thread>