Skip to contentSkip to conversation

Fixes for device probing on flaky connections

PATCH In review [email protected] 3 participants 7 messages Jun 15, 2022 – Jun 22, 2022 SCSI

lore
Share & tools
lore
#0 lore raw

cover Fixes for device probing on flaky connections

From: Martin Wilck <[email protected]>

We have seen device probing errors in FCoE uplink failover tests. After device loss and link recovery, spurious packet loss in the fabric may cause device rediscovery to fail. With the two following patches, the rediscovery failure could be avoided.

Reviews and comments welcome,
Martin

Hannes Reinecke (1):
scsi: add BLIST_RETRY_SCAN to ignore errors during scanning

Martin Wilck (1):
scsi: scan: retry INQUIRY after timeout

 drivers/scsi/scsi_scan.c    | 17 ++++++++++++++---
 include/scsi/scsi_devinfo.h |  4 +++-
 2 files changed, 17 insertions(+), 4 deletions(-)
Signature · 1 line
2.36.1
#1 lore raw

PATCH 1/2 scsi: add BLIST_RETRY_SCAN to ignore errors during scanning

From: Hannes Reinecke <[email protected]>

On flaky connections the INQUIRY command might run into a timeout, but as we already have performed a REPORT LUNS command we know that a LUN should be present. So ignore errors for the INQUIRY commands, and retry until we exhaust the number of retries.

---
 drivers/scsi/scsi_scan.c    | 12 +++++++++---
 include/scsi/scsi_devinfo.h |  4 +++-
 2 files changed, 12 insertions(+), 4 deletions(-)
drivers/scsi/scsi_scan.c +9 -3
index 91ac901a6682..b9b851ce1b72 100644
@@ -649,8 +649,6 @@ static int scsi_probe_lun(struct scsi_device *sdev, unsigned char *inq_result,
 	int pass, count, result;
 	struct scsi_sense_hdr sshdr;
 
-	*bflags = 0;
-
 	/* Perform up to 3 passes.  The first pass uses a conservative
 	 * transfer length of 36 unless sdev->inquiry_len specifies a
 	 * different value. */
@@ -697,6 +695,11 @@ static int scsi_probe_lun(struct scsi_device *sdev, unsigned char *inq_result,
 				    (sshdr.ascq == 0))
 					continue;
 			}
+			if (*bflags & BLIST_RETRY_SCAN) {
+				SCSI_LOG_SCAN_BUS(3, sdev_printk(KERN_INFO, sdev,
+						"scsi scan: retry inquiry after REPORT LUNs\n"));
+				continue;
+			}
 		} else if (result == 0) {
 			/*
 			 * if nothing was transferred, we try
@@ -709,6 +712,8 @@ static int scsi_probe_lun(struct scsi_device *sdev, unsigned char *inq_result,
 		break;
 	}
 
+	*bflags = 0;
+
 	if (result == 0) {
 		scsi_sanitize_inquiry_string(&inq_result[8], 8);
 		scsi_sanitize_inquiry_string(&inq_result[16], 16);
@@ -1531,9 +1536,10 @@ static int scsi_report_lun_scan(struct scsi_target *starget, blist_flags_t bflag
 				    " allowed by the host adapter\n", lun);
 		} else {
 			int res;
+			blist_flags_t bflags = BLIST_RETRY_SCAN;
 
 			res = scsi_probe_and_add_lun(starget,
-				lun, NULL, NULL, rescan, NULL);
+				lun, &bflags, NULL, rescan, NULL);
 			if (res == SCSI_SCAN_NO_RESPONSE) {
 				/*
 				 * Got some results, but now none, abort.
include/scsi/scsi_devinfo.h +3 -1
index 5d14adae21c7..e74a228d73d1 100644
@@ -68,8 +68,10 @@
 #define BLIST_RETRY_ITF		((__force blist_flags_t)(1ULL << 32))
 /* Always retry ABORTED_COMMAND with ASC 0xc1 */
 #define BLIST_RETRY_ASC_C1	((__force blist_flags_t)(1ULL << 33))
+/* Retry errors during scanning */
+#define BLIST_RETRY_SCAN	((__force blist_flags_t)(1ULL << 34))
 
-#define __BLIST_LAST_USED BLIST_RETRY_ASC_C1
+#define __BLIST_LAST_USED BLIST_RETRY_SCAN
 
 #define __BLIST_HIGH_UNUSED (~(__BLIST_LAST_USED | \
 			       (__force blist_flags_t) \
Signature · 1 line
2.36.1
#2 lore raw

Re: [PATCH 1/2] scsi: add BLIST_RETRY_SCAN to ignore errors during scanning

Martin,

drivers/scsi/scsi_scan.c static int scsi_report_lun_scan(struct scsi_target *starget, blist_flags_t bflag · line 1539

Quoted 5 lines
drivers/scsi/scsi_scan.c +1 -0
@@ -1531,9 +1536,10 @@ static int scsi_report_lun_scan(struct scsi_target *starget, blist_flags_t bflag
 				    " allowed by the host adapter\n", lun);
 		} else {
 			int res;
+			blist_flags_t bflags = BLIST_RETRY_SCAN;

I'm not a big fan of using the bflag as carrier of "I was reported and therefore must exist".

Also: Why isn't patch #2 sufficient?

Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering
#3 lore raw

Re: [PATCH 1/2] scsi: add BLIST_RETRY_SCAN to ignore errors during scanning

Quoted 15 lines · Martin K. Petersen

Martin,

Quoted 7 lines
(quoted diff) +0 -0
@@ -1531,9 +1536,10 @@ static int scsi_report_lun_scan(struct

scsi_target *starget, blist_flags_t bflag
                                    " allowed by the host adapter\n", lun);
                } else {
                        int res;
+                       blist_flags_t bflags = BLIST_RETRY_SCAN;

I'm not a big fan of using the bflag as carrier of "I was reported and
therefore must exist".

Quoted 1 line

Also: Why isn't patch #2 sufficient?

I think it is. I can resubmit just #2 if you prefer and Hannes agrees.

Regards,
Martin

#4 lore raw

Re: [PATCH 1/2] scsi: add BLIST_RETRY_SCAN to ignore errors during scanning

Quoted 16 lines · Martin Wilck
Quoted 15 lines · Martin K. Petersen

Martin,

Quoted 7 lines
(quoted diff) +0 -0
@@ -1531,9 +1536,10 @@ static int scsi_report_lun_scan(struct

scsi_target *starget, blist_flags_t bflag
                                    " allowed by the host
adapter\n", lun);
                } else {
                        int res;
+                       blist_flags_t bflags = BLIST_RETRY_SCAN;

I'm not a big fan of using the bflag as carrier of "I was reported and
therefore must exist".

Quoted 4 lines
Quoted 1 line

Also: Why isn't patch #2 sufficient?

I think it is. I can resubmit just #2 if you prefer and Hannes agrees.

I'm fine with just adding #2; #1 is really just there to provide the original behaviour. Device probing is one of the most arcane areas in the SCSI stack due to all the various quirks etc and I didn't want
to change anything here.

But if it's okay, it's okay :-)

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke		           Kernel Storage Architect
[email protected]			                  +49 911 74053 688
SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), GF: Felix Imendörffer
#5 lore raw

Re: [PATCH 1/2] scsi: add BLIST_RETRY_SCAN to ignore errors during scanning

Quoted 24 lines · Hannes Reinecke
Quoted 18 lines · Martin Wilck
Quoted 17 lines · Martin K. Petersen

Martin,

Quoted 8 lines
(quoted diff) +0 -0
@@ -1531,9 +1536,10 @@ static int scsi_report_lun_scan(struct

scsi_target *starget, blist_flags_t bflag
                                     " allowed by the host adapter\n", lun);
                 } else {
                         int res;
+                       blist_flags_t bflags = BLIST_RETRY_SCAN;

I'm not a big fan of using the bflag as carrier of "I was reported
and
therefore must exist".

Quoted 5 lines
Quoted 1 line

Also: Why isn't patch #2 sufficient?

I think it is. I can resubmit just #2 if you prefer and Hannes agrees.

Quoted 6 lines

I'm fine with just adding #2; #1 is really just there to provide the original behaviour. Device probing is one of the most arcane areas in the SCSI stack due to all the various quirks etc and I didn't want to change anything here.

But if it's okay, it's okay :-)

Alright. To be certain, I'll ask our partner for another test.

Martin

#6 lore raw

PATCH 2/2 scsi: scan: retry INQUIRY after timeout

In reply to [email protected]

From: Martin Wilck <[email protected]>

The SCSI mid layer doesn't retry commands after DID_TIME_OUT (see scsi_noretry_cmd()). Packet loss in the fabric can cause spurious timeouts during SCSI device probing, causing device probing to fail. This has been observed in FCoE uplink failover tests, for example.

This patch fixes the issue by retrying the INQUIRY up to 3 times (in practice, we never observed more than a single retry),

---
 drivers/scsi/scsi_scan.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/scsi/scsi_scan.c +5 -0
index b9b851ce1b72..f0a248bd1cd3 100644
@@ -700,6 +700,11 @@ static int scsi_probe_lun(struct scsi_device *sdev, unsigned char *inq_result,
 						"scsi scan: retry inquiry after REPORT LUNs\n"));
 				continue;
 			}
+			if (host_byte(result) == DID_TIME_OUT) {
+				SCSI_LOG_SCAN_BUS(3, sdev_printk(KERN_INFO, sdev,
+						"scsi scan: retry inquiry after timeout\n"));
+				continue;
+			}
 		} else if (result == 0) {
 			/*
 			 * if nothing was transferred, we try
Signature · 1 line
2.36.1