[LU-7371] Wrong read length over isize

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.8.0
• Affects Version/s: Lustre 2.8.0
• Labels: patch
• Severity: 3
• Assignee: Alex Zhuravlev
• Reporter: Li Xi (Inactive)

Description

When running on osd-ldiskfs, if the isize is equal to 4095, a read length of 4096 will be returned because a wrong calculation of EOF.

Issue Links

• is related to: LU-12275 Client-side file data encryption

Comments

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/17060/
Subject: LU-7371 test: wrong read length over isize
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 7023698133970372031a16beac276e5e3e64cfbe

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/17020/
Subject: LU-7371 osd-ldiskfs: fix wrong read length over isize
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 97c4c162be77d2ee9bad5d800c9b5803f252caa0

[Gerrit Updater]

Li Xi ([e-mail]) uploaded a new patch: http://review.whamcloud.com/17060
Subject: LU-7371 test: wrong read length over isize
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: bc9893da37ae60918b61e2d5fd84b4cb87ad3b82

[Andreas Dilger]

I had originally thought dd if=/dev/zero of=$DIR/$tfile bs=4095 count=1 conv=sync would be enough to create the file at 4095 bytes, and then the with bs=4096 bytes would trigger the bug. If that doesn't work, then another option is to add an OBD_FAIL_OST_* check in the code to reproduce the original symptom.

[Li Xi (Inactive)]

I tried to write regression test. However, Lustre client has way to determine file size on client side. So, it seems hard to reproduce the issue on client side.

I collected following messages when doing following things on Lustre without patch (all Lustre client and servers runs on the same machine):

dd if=/dev/zero of=file bs=4095 count=1
sync
echo 3 > /proc/sys/vm/drop_caches
dd if=file of=/dev/null bs=1048576

[root@server1 lustre]# grep tgt_brw_read /tmp/lustre.log | grep leaving
00000020:00000001:0.0:1446642073.288612:0:10887:0:(tgt_handler.c:1915:tgt_brw_read()) Process leaving (rc=4096 : 4096 : 1000)
[root@server1 lustre]# grep 4095 /tmp/lustre.log
00020000:00000001:0.0:1446642073.282732:0:12771:0:(lov_offset.c:68:lov_stripe_size()) Process leaving (rc=4095 : 4095 : fff)
00020000:00000001:0.0:1446642073.282733:0:12771:0:(lov_offset.c:68:lov_stripe_size()) Process leaving (rc=4095 : 4095 : fff)
00000080:00000001:1.0:1446642073.288760:0:12771:0:(file.c:1302:ll_file_aio_read()) Process leaving (rc=4095 : 4095 : fff)
00000080:00000001:1.0:1446642073.288762:0:12771:0:(file.c:1332:ll_file_read()) Process leaving (rc=4095 : 4095 : fff)
00020000:00000001:1.0:1446642073.288856:0:12771:0:(lov_offset.c:68:lov_stripe_size()) Process leaving (rc=4095 : 4095 : fff)
00020000:00000001:1.0:1446642073.288857:0:12771:0:(lov_offset.c:68:lov_stripe_size()) Process leaving (rc=4095 : 4095 : fff)

[Joseph Gmitter (Inactive)]

Hi Alex,
Can you take a look at this issue?
Thanks.
Joe

[Gerrit Updater]

Li Xi ([e-mail]) uploaded a new patch: http://review.whamcloud.com/17020
Subject: LU-7371 osd-ldiskfs: fix wrong read length over isize
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 669fc00eb6ad6a6e9c94e3814326c50c65a7e3af