Details

Reviewers

mav
trasz

Commits

rG9811763b417a: iscsi: Validate DataSN values in Data-In PDUs in the initiator.
rG4f0f5bf99591: iscsi: Validate DataSN values in Data-In PDUs in the initiator.

Summary

As is done in the target, require that DataSN values are consecutive
and in-order. If an out of order Data-In PDU is received, force a
session reconnect. In addition, when a SCSI Response PDU is received,
verify that the ExpDataSN field matches the count of Data-In PDUs
received for this command. If not, fail the I/O request.

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

jhb created this revision.Aug 17 2021, 9:28 PM

Herald added a subscriber: imp. · View Herald TranscriptAug 17 2021, 9:28 PM

jhb requested review of this revision.Aug 17 2021, 9:28 PM

Harbormaster completed remote builds in B41108: Diff 93847.Aug 17 2021, 9:28 PM

jhb added a parent revision: D31577: iscsi: Teach the iSCSI stack about "large" PDUs..Aug 17 2021, 9:28 PM

jhb added a reviewer: trasz.

Generally it looks good to me. I am just thinking whether in case of bhssr_expdatasn mismatch we should also kill the connection, since it should not happen.

In D31594#712266, @mav wrote:

Generally it looks good to me. I am just thinking whether in case of bhssr_expdatasn mismatch we should also kill the connection, since it should not happen.

Mmmm, quite possibly, I just wasn't sure how to do it here.

jhb added inline comments.Aug 18 2021, 5:52 PM

sys/dev/iscsi/iscsi.c
944	If I force a reconnect and skip the call to xpt_done() at the end of the function, will the ccb get resubmitted after the connection has reconnected?

mav added inline comments.Aug 18 2021, 6:24 PM

sys/dev/iscsi/iscsi.c
944	As I see, iscsi_session_terminate_tasks() completes with CAM_REQUEUE_REQ only I/Os in is_outstanding queue. So for that to happen this check and connection termination should be done before iscsi_outstanding_remove() above.