[RFC] Make scsi error recovery play nice with devices blocked by transport

[Michael Reed <mdr@sgi.com>]

[-- Attachment #1: Type: text/plain, Size: 969 bytes --]

Error recovery doesn't interact very well with fc targets which
have been blocked by the fc transport.  Error recovery continues
to attempt to recover the target and ends up marking the fc target
offline.  Once offline, if the target returns before the remote port is
removed, commands which could have been successfully reissued instead
are completed with an error status due to the offline status
of the target.

This patch makes a couple of hopefully minor tweaks to the error
recovery logic to work better with targets which have been blocked
by the transport.

First, if the target is blocked and error recovery gives up, don't
put the device offline.  Either the transport will delete the target
thus disposing of any queued requests or it will unblock the target and
requests will be reissued.

Second, if a device is blocked, queue up commands being flushed from
the done queue for retry instead of completing them with an error.

Comments?

Thanks,
 Mike Reed

[-- Attachment #2: scsi_fc_recovery.patch --]
[-- Type: text/x-patch, Size: 2165 bytes --]


===========================================================================
linux/drivers/scsi/scsi_error.c
===========================================================================

--- a/drivers/scsi/scsi_error.c	2005-12-08 20:55:52.000000000 -0800
+++ b/drivers/scsi/scsi_error.c	2005-12-08 20:43:25.773184520 -0800
@@ -1125,10 +1125,14 @@
 	struct scsi_cmnd *scmd, *next;
 
 	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
-		sdev_printk(KERN_INFO, scmd->device,
-			    "scsi: Device offlined - not"
-			    " ready after error recovery\n");
-		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
+		/* if blocked, transport will provide final device disposition */
+		if (!scsi_device_blocked(scmd->device)) {
+			sdev_printk(KERN_INFO, scmd->device,
+				    "scsi: Device offlined - not"
+				    " ready after error recovery\n");
+			scsi_device_set_state(scmd->device, SDEV_OFFLINE);
+		}
+
 		if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
 			/*
 			 * FIXME: Handle lost cmds.
@@ -1455,9 +1459,10 @@
 
 	list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
 		list_del_init(&scmd->eh_entry);
-		if (scsi_device_online(scmd->device) &&
+		if (scsi_device_blocked(scmd->device) ||
+		    (scsi_device_online(scmd->device) &&
 		    !blk_noretry_request(scmd->request) &&
-		    (++scmd->retries < scmd->allowed)) {
+		    (++scmd->retries < scmd->allowed))) {
 			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
 							  " retry cmd: %p\n",
 							  current->comm,

===========================================================================
linux/include/scsi/scsi_device.h
===========================================================================

--- a/include/scsi/scsi_device.h	2005-12-08 20:55:52.000000000 -0800
+++ b/include/scsi/scsi_device.h	2005-11-17 12:17:38.235614397 -0800
@@ -275,6 +275,11 @@
 			    int data_direction, void *buffer, unsigned bufflen,
 			    struct scsi_sense_hdr *, int timeout, int retries);
 
+static inline int scsi_device_blocked(struct scsi_device *sdev)
+{
+	return sdev->sdev_state == SDEV_BLOCK;
+}
+
 static inline unsigned int sdev_channel(struct scsi_device *sdev)
 {
 	return sdev->channel;