[LU-2479] sanity.sh test_129: max dir size limit work incorrectly.

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.2.0, Lustre 2.1.1, Lustre 2.1.2, Lustre 2.1.3, Lustre 1.8.x (1.8.0 - 1.8.5)
• Fix Version/s: Lustre 2.6.0, Lustre 2.5.2
• Type: Bug
• Priority: Major
• Reporter: Alexey Lyashkov
• Assignee: WC Triage
• Resolution: Fixed
• Labels: patch
• Severity: 3

Description

easy replicate by hand.
[root@rhel6-64 test1]# for i in `seq 10000000`; do touch $i; done
...
[root@rhel6-64 test1]# echo 16384 >/sys/fs/ldiskfs/loop0/max_dir_size
root@rhel6-64 lustre]# ls -ls
total 256
256 drwxr-xr-x 2 root root 258048 Dec 12 12:28 test1
[root@rhel6-64 lustre]# mkdir test1/proba
mkdir: cannot create directory `test1/proba': File too large
[root@rhel6-64 test1]# echo 0 >/sys/fs/ldiskfs/loop0/max_dir_size
[root@rhel6-64 lustre]# cd test1
[root@rhel6-64 test1]# rm -rf *
[root@rhel6-64 test1]# ls -l
total 0
[root@rhel6-64 test1]# echo 16384 >/sys/fs/ldiskfs/loop0/max_dir_size
[root@rhel6-64 test1]# cd ..
[root@rhel6-64 lustre]# ls -ls
total 264
264 drwxr-xr-x 2 root root 266240 Dec 12 12:28 test1
[root@rhel6-64 lustre]# mkdir test1/proba
mkdir: cannot create directory `test1/proba': File too large
[root@rhel6-64 lustre]#

mkdir failed due wrong assumption - if we allocate an inode that is always allocate a space in directory, but ext4 isn't truncate a directory after removing an objects from it.

that is caused random fails in sanity 129.
+ do_node mft51.xyua 'echo 16384 >/sys/fs/ldiskfs/loop0/max_dir_size'
+ local verbose=false
+ '[' xmft51.xyua = x--verbose ']'
+ local HOST=mft51.xyua
+ shift
+ local 'myPDSH=/usr/bin/pdsh -R ssh -S -w '
+ '[' mft51.xyua = mft51.xyua ']'
+ myPDSH=no_dsh
+ false
+ '[' no_dsh = rsh ']'
+ false
+ sed 's/^mft51.xyua: //'
+ no_dsh mft51.xyua '(PATH=$PATH:/usr/lib64/lustre/utils:/usr/lib64/lustre/tests:/sbin:/usr/sbin; cd /root; LUSTRE="/usr/lib64/lustre" sh -c "echo 16384 >/sys/fs/ldiskfs/loop0/max_dir_size")'
+ shift
+ eval '(PATH=$PATH:/usr/lib64/lustre/utils:/usr/lib64/lustre/tests:/sbin:/usr/sbin;' cd '/root;' 'LUSTRE="/usr/lib64/lustre"' sh -c '"echo' 16384 '>/sys/fs/ldiskfs/loop0/max_dir_size")'
++ PATH=/usr/lib64/lustre/tests/mpi:/usr/lib64/lustre/tests/racer:/usr/lib64/lustre/../lustre-iokit/sgpdd-survey:/usr/lib64/lustre/tests:/usr/lib64/lustre/utils/gss:/usr/lib64/lustre/utils:/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin:/sbin::/usr/lib64/lustre/utils:/usr/lib64/lustre/tests:/sbin:/usr/sbin
++ cd /root
++ LUSTRE=/usr/lib64/lustre
++ sh -c 'echo 16384 >/sys/fs/ldiskfs/loop0/max_dir_size'
+ return 0
+ mkdir -p /mnt/lustre/d0.sanity/d129
mkdir: cannot create directory `/mnt/lustre/d0.sanity': File too large
+ ls -l /mnt/lustre
total 0

Issue Links

• is related to: LU-4654 sanity test_129: current dir size 122...

Comments

[Alexey Lyashkov]

correct implementation landed into ext4 upstream as

commit df981d03eeff7971ac7e6ff37000bfa702327ef1
Author: Theodore Ts'o <[e-mail]>
Date: Fri Aug 17 09:48:17 2012 -0400

ext4: add max_dir_size_kb mount option

by checking a directory limit inside ext4_append function.

[Keith Mannthey (Inactive)]

This same issue has been seen in our test rig.

https://maloo.whamcloud.com/test_sessions/00d6be9e-5fbf-11e2-84d4-52540035b04c

[Keith Mannthey (Inactive)]

Alexey Lyashkov, Have you tested the above change and does it fix the error?

[Alexey Lyashkov]

Keith,

That is correct.
Alexander Boyko <[e-mail]> work on that issue from Xyratex side and will be ready to submit fix shortly (first inspection round passed).

[Alexander Boyko]

http://review.whamcloud.com/#change,5140

[Alexander Boyko]

http://review.whamcloud.com/#change,5140
Maloo found the issue, it is relate to LU-2640

[Alexander Boyko]

Status update:
I don`t understand why this patch should not be landed if it`s broke SLES11.
1) this patch fix serios bug with simple reproducer, and result is - file creation would failed in empty directory.
2) SLES11 patches are not landed yet.
3) this fix is port from upstream ext4 branch and SLES need to include it also.

[James A Simmons]

New patch address the SLES11 issues. Will work with RHEL6.3, SLES11 SP1 and SP2. Haven't tried it on RHEL6.4 yet.

[Keith Mannthey (Inactive)]

Patch has been merged.

[Vladimir Saveliev]

The directory max size check in ldiskfs_append() should not be applied when an inode does not refer to a directory.
Please take a look
http://review.whamcloud.com/8137

[Emoly Liu]

The directory max size check in ldiskfs_append() should not be applied when an inode does not refer to a directory.
Please take a look
http://review.whamcloud.com/8137

backport to b2_5: http://review.whamcloud.com/10043