[LU-11686] sanity test 807 fails with mounted subdirectory with error '/mnt/lustre/d807.sanity/single_dd expected blocks: 1, got: 0'

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Affects Version/s: Lustre 2.12.0
• Labels: LSOM always_except
• Environment: subdirectory mount
• Severity: 3
• Assignee: WC Triage
• Reporter: James Nunez (Inactive)

Description

sanity test_807 verifies the LSOM syncing tool. When a Lustre client mounts a subdirectory on the Lustre file system, sanity test 807 fails and it looks like the LSOM sync tool, llsom_sync, does not return the correct information. From the failure at https://testing.whamcloud.com/test_sets/37edc450-e915-11e8-b67f-52540065bddc, the client test log contains

== sanity test 807: verify LSOM syncing tool ========================================================= 20:25:23 (1542313523)
…
CMD: onyx-43vm12 /usr/sbin/lctl --device lustre-MDT0000 changelog_register -n
Registered 1 changelog users: 'cl1'
…
llite.lustre-ffff93c57c1b9000.xattr_cache=0
1+0 records in
1+0 records out
1048576 bytes (1.0 MB) copied, 0.00356313 s, 294 MB/s
Test SOM for muti-client (2) writes
CMD: onyx-43vm10 multiop /mnt/lustre/f807.sanity Oz0w1048576c
CMD: onyx-43vm9.onyx.whamcloud.com multiop /mnt/lustre/f807.sanity Oz1048576w1048576c
Start to sync 3 records.
 sanity test_807: @@@@@@ FAIL: /mnt/lustre/d807.sanity/single_dd expected blocks: 1, got: 0 

To reproduce this error, set the environment variable FILESET or, in autotest, use the test-parameter “envdefinitions=FILESET=/subdir”.

Logs for other failures are at
https://testing.whamcloud.com/test_sets/c994a85c-d3b1-11e8-9238-52540065bddc
https://testing.whamcloud.com/test_sets/71f98f92-d0a8-11e8-9238-52540065bddc

Issue Links

• is related to: LU-8585 All Lustre test suites should pass with subdirectory mount

Comments

[Andreas Dilger]

Need to re-enable this subtest and see if it is passing.

[Andreas Dilger]

Currently, sanity.sh test_807 is skipped when FILESET is set, but I suspect this is now fixed by LU-8585.

[Andreas Dilger]

This is failing because llapi_open_by_fid() used by llsom_sync does not work in a subdirectory mount to prevent FID-guessing attacks from within a subdir. I suspect that this will not be an issue for real deployments, since the llsom_sync tool would not be running inside the subdir, so it is IMHO a not a critical issue. As a workaround, I think it would be possible to use llapi_fid2path() within a subdirectory mount to get the proper relative pathname to open the file. Whether that would be useful in production is a separate question.

Note minor typo in test_807 - "muti" -> "multi" in a few places.

Also, in llsom_sync.c it defines PATH_MAX but doesn't use it anywhere.

[Peter Jones]

Qian

Can you please advise?

Thanks

Peter