[LU-9971] MR: ABA problem in lnet_discover_peer_locked

• Status: Resolved
• Project: Lustre
• Fix Version/s: Lustre 2.13.0, Lustre 2.12.6
• Type: Bug
• Priority: Minor
• Reporter: Amir Shehata (Inactive)
• Assignee: Amir Shehata (Inactive)
• Resolution: Fixed
• Severity: 3

Description

In lnet_discover_peer_locked() when we drop and re-acquire the net lock there is a possibility that the lpni might've been relinked, opening the window for an ABA problem.

Issue Links

• is related to: LU-10281 conf-sanity: test_54a hung at lnet_di...
• is related to: LU-13652 [1575337.260035] LNetError: 8719:0:(p...
• is related to: LU-12519 sanity-sec test 31 crashes with ASSER...

Comments

[Olaf Weber]

Not so much an ABA problem as a use-after-free: the lp pointer might point to a different lnet_peer at the same address. Reshuffling the reference count addref and decref on lp a bit we can easily ensure that even if lpni is now linked to a different peer, at least that peer must have a different address.

[Gerrit Updater]

Olaf Weber ([e-mail]) uploaded a new patch: https://review.whamcloud.com/28944
Subject: LU-9971 lnet: use after free in lnet_discover_peer_locked()
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 0f1814c07e9e9107dc6b92584ab9ed999621fee6

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/28944/
Subject: LU-9971 lnet: use after free in lnet_discover_peer_locked()
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 2b5b551b15d96588f8f309b5a08c11cab203efeb

[Andriy Skulysh]

The patch has a defect caused by landing of LU-11299.

[Gerrit Updater]

Amir Shehata ([e-mail]) uploaded a new patch: https://review.whamcloud.com/35446
Subject: LU-9971 lnet: fix peer ref counting
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 086739cf3d0424da664cd11c7798f7c93153b95b

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/35446/
Subject: LU-9971 lnet: fix peer ref counting
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: dbcddb4824f08153f145327e2bfe1bf4a2becc4f

[Peter Jones]

Both patches landed for 2.13

[Gerrit Updater]

Amir Shehata ([e-mail]) uploaded a new patch: https://review.whamcloud.com/38891
Subject: LU-9971 lnet: use after free in lnet_discover_peer_locked()
Project: fs/lustre-release
Branch: b2_12
Current Patch Set: 1
Commit: da45d5fe3b64f38391f25511eccbacf250bac28b

[Gerrit Updater]

Amir Shehata ([e-mail]) uploaded a new patch: https://review.whamcloud.com/38892
Subject: LU-9971 lnet: fix peer ref counting
Project: fs/lustre-release
Branch: b2_12
Current Patch Set: 1
Commit: 9312614ca535a6b7c4d4c81756237a9192322460

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/38891/
Subject: LU-9971 lnet: use after free in lnet_discover_peer_locked()
Project: fs/lustre-release
Branch: b2_12
Current Patch Set:
Commit: 3f3c839f97298f6e65e2e053fc2ece59c39931dc