From: Rob Evers <revers@redhat.com>
Date: Fri, 16 Jan 2009 14:36:00 -0500
Subject: [scsi] no-sense msgs, data corruption, but no i/o errors
Message-id: 20090116193503.12714.37477.sendpatchset@localhost.localdomain
O-Subject: [RHEL5.4 PATCH] scsi no-sense messages, data corruption, but no i/o errors
Bugzilla: 468088
RH-Acked-by: Mike Christie <mchristi@redhat.com>
RH-Acked-by: Doug Ledford <dledford@redhat.com>
RH-Acked-by: Tomas Henzl <thenzl@redhat.com>

bugzilla number:

    https://bugzilla.redhat.com/show_bug.cgi?id=468088

description:

    Quoted from the commit in Linus's tree:

    The current handling of NO_SENSE check condition is the same as
    RECOVERED_ERROR, and assumes that in both cases, the I/O was fully
    transferred.

    We have seen cases of arrays returning with NO_SENSE (no error), but
    the I/O was not completely transferred, thus residual set.  Thus,
    rather than return good_bytes as the entire transfer, set good_bytes
    to 0, so that the midlayer then applies the residual in calculating
    the transfer, and for sd, will fail the I/O and fall into a retry
    path.

Testing:

    The scsi_debug.c driver was modified to recreate the scenario.
    Periodically a fault was injected into the driver handling
    of scsi requests.  The fault scenario was that a check condtion
    status is returned with sense = 'no sense'.  Addtionally, during
    the fault injection, read data was corrupted.

    Before the patch was applied, dt was used to detect that data
    corruption was occurring, when the faults were being injected.

    After the patch was applied, the same test scenario resulted
    in no data corruption.

    Console messages are evident in both cases indicating that a
    'no sense' sense code is present during fault injection.

Upstream Status:

    This patch was accepted into Linus' kernel (2.6.28) in
    October 2008.

diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
index 32e5a09..4fd6e68 100644
--- a/drivers/scsi/sd.c
+++ b/drivers/scsi/sd.c
@@ -1015,7 +1015,6 @@ static void sd_rw_intr(struct scsi_cmnd * SCpnt)
 		good_bytes = (bad_lba - start_lba)*SCpnt->device->sector_size;
 		break;
 	case RECOVERED_ERROR:
-	case NO_SENSE:
 		/* Inform the user, but make sure that it's not treated
 		 * as a hard error.
 		 */
@@ -1024,6 +1023,15 @@ static void sd_rw_intr(struct scsi_cmnd * SCpnt)
 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
 		good_bytes = xfer_size;
 		break;
+	case NO_SENSE:
+		/* This indicates a false check condition, so ignore it.  An
+		 * unknown amount of data was transferred so treat it as an
+		 * error.
+		 */
+		scsi_print_sense("sd", SCpnt);
+		SCpnt->result = 0;
+		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
+		break;
 	case ILLEGAL_REQUEST:
 		if (SCpnt->device->use_10_for_rw &&
 		    (SCpnt->cmnd[0] == READ_10 ||