[LU-13693] lfs getstripe should avoid opening regular files

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.14.0
• Severity: 3
• Assignee: John Hammond
• Reporter: John Hammond

Description

lfs getstripe uses an ioctl on the parent directory to get the file striping but still opens the file to call the OBD_IOC_GETNAME ioctl in get_mds_md_size(). Rather than doing this we should just use a large enough size for the lum buffer. This avoids interfering with existing file locks and leases on the file.

Issue Links

• is related to: LU-14623 sanity test_210: FAIL: multiop failed / still running
• is related to: LU-14322 Interop: sanityn test 51e fails with 'multiop failed'
• is related to: LU-11264 llapi_* routines demonstrate poor performance
• is related to: LU-13668 open-for-read should not conflict with mirror creation

Comments

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/39159/
Subject: LU-13693 lfs: check early for MDS_OPEN_DIRECTORY
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 3ea729fe822d3c39df6d8a5fb530dd8a6901c91c

[Gerrit Updater]

John L. Hammond ([e-mail]) uploaded a new patch: https://review.whamcloud.com/39159
Subject: LU-13693 lfs: check early for MDS_OPEN_DIRECTORY
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 9f9de2072416a3f66583c4cbfb3d405abe2f3f13

[Peter Jones]

Landed for 2.14

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/38979/
Subject: LU-13693 lfs: avoid opening regular files for getstripe
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 3eff6bada2bc789fe6089e792e4043ef7ec8396f

[Andreas Dilger]

Does it make sense to check in ll_lov_getstripe_ea_info() if the filename is already in the client dcache and get the xattr directly from the client cache, something like:

        qstr = QSTR_INIT(filename, strlen(filename));
        dchild = d_lookup(dparent, &qstr);

(dparent can be passed from the caller). The benefit of this would depend on whether the file is typically already present on the client when "lfs getstripe" is called or not, but the dcache lookup would be much lower overhead than an RPC (an O(1) lockless hash table lookup for the dcache these days).

[Gerrit Updater]

John L. Hammond ([e-mail]) uploaded a new patch: https://review.whamcloud.com/38979
Subject: LU-13693 lfs: avoid opening regular files for getstripe
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 9f8dce9f48ec3078fc798393e75ccd276bc30948