Re: [PATCH] libiscsi: add lock around task lists to fix list corruption regression

Hi again, Chris and others! We're hitting this again on 4.14.32. Looks like the path iscsi_queuecommand -> prepd_reject/prepd_fault results in iscsi_complete_task getting called with frwd_lock held instead of the requisite back_lock which might explain the task list corruption under heavy loads.

Re: [PATCH] libiscsi: add lock around task lists to fix list corruption regression

Hi! Was running pretty nicely for a week or so, until we got this one: [534630.679965] BUG: unable to handle kernel at 0078 [534630.684035] IP: [] iscsi_xmit_task+0x29/0xc0 [534630.685724] PGD a1fcd3067 [534630.687346] Oops: 0002 [#1] [534630.688846] Modules linked in:

Re: [PATCH] libiscsi: add lock around task lists to fix list corruption regression

On 02/23/2017 07:25 PM, Chris Leech wrote: > Yikes, my git-send-email settings suppressed the important CCs. Sorry! > > Guilherme and Ilkka, can you comment about your testing results or review > please? Hi Chris, thanks for looping me. Patch seems nice, I do have some points below, most

Re: [PATCH] libiscsi: add lock around task lists to fix list corruption regression

Yikes, my git-send-email settings suppressed the important CCs. Sorry! Guilherme and Ilkka, can you comment about your testing results or review please? - Chris Leech - Original Message - > There's a rather long standing regression from commit > 659743b [SCSI] libiscsi: Reduce locking

Re: [PATCH] libiscsi: add lock around task lists to fix list corruption regression

> "Chris" == Chris Leech writes: Chris> There's a rather long standing regression from commit 659743b Chris> [SCSI] libiscsi: Reduce locking contention in fast path Chris> Depending on iSCSI target behavior, it's possible to hit the case Chris> in iscsi_complete_task

[PATCH] libiscsi: add lock around task lists to fix list corruption regression

There's a rather long standing regression from commit 659743b [SCSI] libiscsi: Reduce locking contention in fast path Depending on iSCSI target behavior, it's possible to hit the case in iscsi_complete_task where the task is still on a pending list (!list_empty(>running)). When that happens the