[LU-5639] Message is hashed to invalid match-table of LNet request portal

• Type: Bug
• Resolution: Fixed
• Priority: Blocker
• Fix Version/s: Lustre 2.7.0, Lustre 2.5.4
• Affects Version/s: Lustre 2.5.3
• Severity: 3
• Assignee: Liang Zhen (Inactive)
• Reporter: Liang Zhen (Inactive)

Description

When new message arrived at LNet request portal, if there is no receiving buffer (ME/MD) in match-table of the partition that message arrived in, either because service does not post any buffer in that partition, or because server is under memory pressure, LNet will increase a integer (rotor) then hash this message to another match-table within a different partition which has receiving buffers. This rotor should be unsigned when hashing, otherwise we will refer to illegal address when the integer is overflowed.

This issue should be rare because Lustre services will always post enough request buffers

Issue Links

• is duplicated by: LU-5962 OSS crash on lnet_ptl_match_md() due to a null pointer because ptl->ptl_rotor is negatif

Comments

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/11999/
Subject: LU-5639 lnet: portal spreading rotor should be unsigned
Project: fs/lustre-release
Branch: b2_5
Current Patch Set:
Commit: 14e2be44141bcec05202a83493eb34fe602dadc9

[Peter Jones]

Landed for 2.7

[Liang Zhen (Inactive)]

patch has landed to 2.7, but we also need this for 2.5:

http://review.whamcloud.com/#/c/11999/

[Liang Zhen (Inactive)]

patch is here: http://review.whamcloud.com/#/c/11936/