[LU-9921] LNet peer discovery list handling

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.11.0
• Severity: 3
• Assignee: Amir Shehata (Inactive)
• Reporter: John Hammond

Description

In lnet_destroy_peer_locked() we should add an assertion that the lp_dc_pendq list is empty. Similarly for lnet_destroy_peer_ni_locked() and lpni_peer_nis, lpni_on_remote_peer_ni_list, .... The list heads added and modified by the dynamic discovery series should be audited.

In lnet_peer_discovery_complete() the list handing appears unsafe. lnet_send() and lnet_finalize() may/will modify msg_list. So list_for_each_entry_safe() must be used. Based on how other code used msg_list, it looks like we need to call list_del() or list_del_init() before calling either function. At the end of lnet_peer_discovery_complete() we should assert that pending_msgs is empty. The dynamic discovery series should be audited for similarly unsafe list handling.

List heads in the_lnet should be statically initialized.

Issue Links

• is duplicated by: LU-9928 conf-sanity: test_32d IP: [<ffffffffc0bc8083>] lnet_cpt_of_md+0x13/0x260 [lnet]

Comments

[Joseph Gmitter (Inactive)]

Patch has landed to master for 2.11.0

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/28723/
Subject: LU-9921 lnet: resolve unsafe list access
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 62c3c8d148563781198abe11b2be213708b165cb

[Gerrit Updater]

Amir Shehata ([e-mail]) uploaded a new patch: https://review.whamcloud.com/28723
Subject: LU-9921 lnet: resolve unsafe list access
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 9dc66d8b1bf5a26ab32bc333039ea4059a516938