[LU-16160] take ldlm lock when queue sync pages

• Type: Bug
• Resolution: Fixed
• Priority: Major
• Fix Version/s: Lustre 2.16.0, Lustre 2.15.3
• Severity: 3
• Assignee: Zhenyu Xu
• Reporter: Zhenyu Xu

Description

osc_queue_sync_pages() add osc_extent to osc_object's IO extent list without taking ldlm locks, and then it calls osc_io_unplug_async() to queue the IO work for the client.

I think the IO extent should take ldlm locks while waiting in the IO work queue.

Issue Links

• is related to: LU-16401 various crashes with cl_page_discard vs readhead race
• is related to: LU-16224 rw_seq_cst_vs_drop_caches dies with SIGBUS
• is related to: LU-16156 stale read during IOR test due LU-14541
• is related to: LU-14541 Memory reclaim caused a stale data read
• is related to: LU-15815 fast_read/stale data/reclaim workround causes SIGBUS

Comments

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/49647/
Subject: LU-16160 llite: SIGBUS is possible on a race with page reclaim
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: b4da788a819f82d35b685d6ee7f02809c05ca005

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/49541/
Subject: LU-16160 revert: "llite: clear stale page's uptodate bit"
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 84c9618190f9e3a526ce51dc4995fcfa3a9ed265

[Peter Jones]

panda Patrick ported your attached patch to master and pushed it into gerrit, so we can compare and contrast both similar approaches. To that end, please can you confirm that nothing was altered during the porting? Thanks!

[Gerrit Updater]

"Patrick Farrell <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/49647
Subject: LU-16160 llite: SIGBUS is possible on a race with page reclaim
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 371d94fe71b103c1712bf8f2b1bb1c026cf31de4

[Andrew Perepechko]

Peter, sure, I'll do that. Thank you

[Peter Jones]

Andrew

Bobijam feels that your attached patch takes a similar approach to his latest LU-16160 patch. Could you please review the latter in gerrit to flag any issues that should deter us from landing this to master?

Thanks

Peter

[Peter Jones]

Thanks very much Andrew. Is it possible to push the patch into gerrit so it is easier for us to provide testing/review feedback?

[Andrew Perepechko]

I've uploaded the workaround for stale data+SIGBUS issue that we are currently using, as an attachment. We're going to replace it with a better (design-wise/performance-wise) solution if/when we have it. Not sure if it's useful but we'd like to share.

[Gerrit Updater]

"Zhenyu Xu <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/49553
Subject: LU-16160 llite: handle filemap_fault() returns SIGBUS error
Project: fs/lustre-release
Branch: b2_15
Current Patch Set: 1
Commit: 7b9342563fc08f800fa4379b189f3b5df6ebd570

[Gerrit Updater]

"Zhenyu Xu <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/49549
Subject: LU-16160 llite: handle filemap_fault() returns SIGBUS error
Project: fs/lustre-release
Branch: b2_12
Current Patch Set: 1
Commit: 2e20813f4f355d6c8d0cf0ac861fefbf35d7f6f3

[Cory Spitz]

> Please reopen if more work is to be tracked under this ticket
Reopening since bobijam has posted a revert and more work against this ticket.