[LTP] du01 with btrfs on systems with > 4k page size

[Stanislav Kholmanskikh <stanislav.kholmanskikh@oracle.com>]

On 11/09/2016 07:18 PM, Cyril Hrubis wrote:
> Hi!
>> On SPARC the default page size is 8k.
>>
>> And du01 fails with btrfs at check3:
>>
>> du01 3 TFAIL : 'du -a' failed
>> du01 4 TINFO : Looking for '[0-4][[:space:]]\.\/testdir\/testsymlink' in:
>> 10240  ./testfile
>> 8      ./testdir/testsymlink
>> 8      ./testdir
>> 10248  .
>> du01 4 TFAIL : 'du --all' failed
>> du01 5 TINFO : Looking for '[0-4][[:space:]]\.\/testdir\/testsymlink' in:
>> 10240  ./testfile
>> 8      ./testdir/testsymlink
>> 8      ./testdir
>> 10248  .
>>
>> i.e. the testsymlink is 8k whereas at most 4k is expected by the test case.
>>
>> In commit bdd09b1c6f2c8ad ("du01.sh: Fix failures on Btrfs on ppc64le")
>> a similar situation was fixed, but for check5 and check6. I'm curious
>> why check3 doesn't fail on ppc64. It seems it should fail with the
>> current code.
>>
>> Could, please, anybody with access to a ppc64 box run this test case
>> with btrfs and/or provide the output from commands:
>>
>> [root@skholman-m7 du]# cd /mnt
>> [root@skholman-m7 mnt]# mkdir basedir
>> [root@skholman-m7 mnt]# cd basedir/
>> [root@skholman-m7 basedir]# dd if=/dev/zero of=testfile bs=1M count=10
>> 10+0 records in
>> 10+0 records out
>> 10485760 bytes (10 MB) copied, 0.00825522 s, 1.3 GB/s
>> [root@skholman-m7 basedir]# mkdir -p testdir
>> [root@skholman-m7 basedir]# ln -s ../testfile testdir/testsymlink
>> [root@skholman-m7 basedir]# du -a
>> 10240	./testfile
>> 8	./testdir/testsymlink
>> 8	./testdir
>> 10248	.
>> [root@skholman-m7 basedir]#
> 
> It's really 64k for a symlink:
> 
> du -a
> 10240   ./testfile
> 64      ./testdir/testsymlink
> 64      ./testdir
> 10304   .
> 

Thank you.

> And the reason that it works is:
> 
> commit 9712f3122a46c43fccd694ee7204ec8c19cfacdc
> Author: Cyril Hrubis <chrubis@suse.cz>
> Date:   Wed Jan 13 17:33:51 2016 +0100
> 
>     commands: du01.sh; Btrfs fix.
> 
>     Btrfs reports symlinks to be 4 blocks in size.
> 
>     Also rename the test symlink so that it's clear it's
>     symlink.
> 
> 
> Since the grep ignores the 6 at the start of the pattern and succeeds.
> It's strange that logs I have in SUSE bugzilla says that it was 4k in
> size, but that may have been true for some other kernel, since the
> pagesize is configurable in kernel .config.
> 
> So I guess that we should fix the pattern to match anyhing that could be
> pagesize. For PPC64 that should be one of 4K/16K/64K/256K, it's 8K for
> sparc apparently so I guess that we should go for 2^n between 4 and
> 256.

Maybe determine the default page size at runtime (getconf PAGESIZE) and
check if testsymlink is either 0 or the page size?


> 
> And we may also specify that the pattern should start matching at the
> start of the line, otherwise the test will still pass if there was some
> garbage before the correct string.
> 

Agreed. For all the checks.