git: 726d8e185206 - stable/13 - mps(4): Fix unmatched devq release.

From: Alexander Motin <mav_at_FreeBSD.org>
Date: Thu, 06 Jan 2022 19:04:38 UTC
The branch stable/13 has been updated by mav:

URL: https://cgit.FreeBSD.org/src/commit/?id=726d8e185206b25c9c894c66f28b405c5b204dec

commit 726d8e185206b25c9c894c66f28b405c5b204dec
Author:     Warner Losh <imp@FreeBSD.org>
AuthorDate: 2021-12-02 20:53:44 +0000
Commit:     Alexander Motin <mav@FreeBSD.org>
CommitDate: 2022-01-06 19:03:38 +0000

    mps(4): Fix unmatched devq release.
    
    Port 9781c28c6d63 and a8837c77efd0 to the mps driver.  Before this
    change devq was frozen only if some command was sent to the target after
    reset started, but release was called always.  This change freezes the
    devq immediately, leaving mprsas_action_scsiio() check only to cover
    race condition due to different lock devq use.
    
    This should also avoid unnecessary requeue of the commands, creating
    additional log noise and confusing some broken apps. It also avoids a
    'busy' requeue of I/Os failing when we're doing recovery that takes
    longer than the normal busy timeout. These I/Os failing can lead to
    filesystems being unmounted in the force unmount case for I/O errors.
    
    Sponsored by:           Netflix
    Reviewed by:            mav
    Differential Revision:  https://reviews.freebsd.org/D33228
    
    (cherry picked from commit a10253cffea84c0c980a36ba6776b00ed96c3e3b)
---
 sys/dev/mps/mps_sas.c | 38 ++++++++++++++++++++++----------------
 1 file changed, 22 insertions(+), 16 deletions(-)

diff --git a/sys/dev/mps/mps_sas.c b/sys/dev/mps/mps_sas.c
index 2d580edca9eb..d21cf1e2204e 100644
--- a/sys/dev/mps/mps_sas.c
+++ b/sys/dev/mps/mps_sas.c
@@ -235,8 +235,6 @@ mpssas_alloc_tm(struct mps_softc *sc)
 void
 mpssas_free_tm(struct mps_softc *sc, struct mps_command *tm)
 {
-	int target_id = 0xFFFFFFFF;
-
 	if (tm == NULL)
 		return;
 
@@ -245,13 +243,11 @@ mpssas_free_tm(struct mps_softc *sc, struct mps_command *tm)
 	 * free the resources used for freezing the devq.  Must clear the
 	 * INRESET flag as well or scsi I/O will not work.
 	 */
-	if (tm->cm_targ != NULL) {
-		tm->cm_targ->flags &= ~MPSSAS_TARGET_INRESET;
-		target_id = tm->cm_targ->tid;
-	}
 	if (tm->cm_ccb) {
-		mps_dprint(sc, MPS_INFO, "Unfreezing devq for target ID %d\n",
-		    target_id);
+		mps_dprint(sc, MPS_XINFO | MPS_RECOVERY,
+		    "Unfreezing devq for target ID %d\n",
+		    tm->cm_targ->tid);
+		tm->cm_targ->flags &= ~MPSSAS_TARGET_INRESET;
 		xpt_release_devq(tm->cm_ccb->ccb_h.path, 1, TRUE);
 		xpt_free_path(tm->cm_ccb->ccb_h.path);
 		xpt_free_ccb(tm->cm_ccb);
@@ -1699,11 +1695,11 @@ mpssas_action_scsiio(struct mpssas_softc *sassc, union ccb *ccb)
 	}
 
 	/*
-	 * If target has a reset in progress, freeze the devq and return.  The
-	 * devq will be released when the TM reset is finished.
+	 * If target has a reset in progress, the devq should be frozen.
+	 * Geting here we likely hit a race, so just requeue.
 	 */
 	if (targ->flags & MPSSAS_TARGET_INRESET) {
-		ccb->ccb_h.status = CAM_BUSY | CAM_DEV_QFRZN;
+		ccb->ccb_h.status = CAM_REQUEUE_REQ | CAM_DEV_QFRZN;
 		mps_dprint(sc, MPS_INFO, "%s: Freezing devq for target ID %d\n",
 		    __func__, targ->tid);
 		xpt_freeze_devq(ccb->ccb_h.path, 1);
@@ -3243,11 +3239,15 @@ mpssas_async(void *callback_arg, uint32_t code, struct cam_path *path,
 }
 
 /*
- * Set the INRESET flag for this target so that no I/O will be sent to
- * the target until the reset has completed.  If an I/O request does
- * happen, the devq will be frozen.  The CCB holds the path which is
- * used to release the devq.  The devq is released and the CCB is freed
+ * Freeze the devq and set the INRESET flag so that no I/O will be sent to
+ * the target until the reset has completed.  The CCB holds the path which
+ * is used to release the devq.  The devq is released and the CCB is freed
  * when the TM completes.
+ * We only need to do this when we're entering reset, not at each time we
+ * need to send an abort (which will happen if multiple commands timeout
+ * while we're sending the abort). We do not release the queue for each
+ * command we complete (just at the end when we free the tm), so freezing
+ * it each time doesn't make sense.
  */
 void
 mpssas_prepare_for_tm(struct mps_softc *sc, struct mps_command *tm,
@@ -3265,7 +3265,13 @@ mpssas_prepare_for_tm(struct mps_softc *sc, struct mps_command *tm,
 		} else {
 			tm->cm_ccb = ccb;
 			tm->cm_targ = target;
-			target->flags |= MPSSAS_TARGET_INRESET;
+			if ((target->flags & MPSSAS_TARGET_INRESET) == 0) {
+				mps_dprint(sc, MPS_XINFO | MPS_RECOVERY,
+				    "%s: Freezing devq for target ID %d\n",
+				    __func__, target->tid);
+				xpt_freeze_devq(ccb->ccb_h.path, 1);
+				target->flags |= MPSSAS_TARGET_INRESET;
+			}
 		}
 	}
 }