[SCSI] Allow FC LLD to fast-fail scsi eh by introducing new eh return
authorChristof Schmitt <christof.schmitt@de.ibm.com>
Wed, 24 Mar 2010 15:50:30 +0000 (16:50 +0100)
committerJames Bottomley <James.Bottomley@suse.de>
Sun, 11 Apr 2010 14:49:33 +0000 (09:49 -0500)
If the scsi eh is running and then a FC LLD calls
fc_remote_port_delete, the SCSI commands sent from the eh will fail.
To prevent this, a FC LLD can call fc_block_scsi_eh from the eh
callback, blocking the eh thread until the dev_loss_tmo fires or the
remote port is available again.

If (e.g. for a multipathing setup) the dev_loss_tmo is set to a very
large value, thus preventing the scsi device removal , the scsi eh can
block for a long time. For multipathing, the fast_io_fail_tmo is then
set to a low value to detect path problems sooner.

This patch introduces a new return code FAST_IO_FAIL. The function
fc_block_scsi_eh now returns FAST_IO_FAIL when the fast_io_fail_tmo
fires. This indicates that the LLD terminated all pending I/O requests
and there are no more pending SCSI commands for the scsi eh to wait
for. This return code can be passed back to the scsi eh to stop the
escalation and finish the recovery process for this device.

Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
drivers/scsi/scsi_error.c
drivers/scsi/scsi_transport_fc.c
include/scsi/scsi.h
include/scsi/scsi_transport_fc.h

index d45c69ca5737b335644fb4490bedf3fd6cf8779c..33175974b55af3202aa3964e2fecf882d8feddbd 100644 (file)
@@ -957,9 +957,10 @@ static int scsi_eh_abort_cmds(struct list_head *work_q,
                                                  "0x%p\n", current->comm,
                                                  scmd));
                rtn = scsi_try_to_abort_cmd(scmd);
-               if (rtn == SUCCESS) {
+               if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
                        scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
                        if (!scsi_device_online(scmd->device) ||
+                           rtn == FAST_IO_FAIL ||
                            !scsi_eh_tur(scmd)) {
                                scsi_eh_finish_cmd(scmd, done_q);
                        }
@@ -1086,8 +1087,9 @@ static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
                                                  " 0x%p\n", current->comm,
                                                  sdev));
                rtn = scsi_try_bus_device_reset(bdr_scmd);
-               if (rtn == SUCCESS) {
+               if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
                        if (!scsi_device_online(sdev) ||
+                           rtn == FAST_IO_FAIL ||
                            !scsi_eh_tur(bdr_scmd)) {
                                list_for_each_entry_safe(scmd, next,
                                                         work_q, eh_entry) {
@@ -1150,10 +1152,11 @@ static int scsi_eh_target_reset(struct Scsi_Host *shost,
                                                  "to target %d\n",
                                                  current->comm, id));
                rtn = scsi_try_target_reset(tgtr_scmd);
-               if (rtn == SUCCESS) {
+               if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
                        list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
                                if (id == scmd_id(scmd))
                                        if (!scsi_device_online(scmd->device) ||
+                                           rtn == FAST_IO_FAIL ||
                                            !scsi_eh_tur(tgtr_scmd))
                                                scsi_eh_finish_cmd(scmd,
                                                                   done_q);
@@ -1209,10 +1212,11 @@ static int scsi_eh_bus_reset(struct Scsi_Host *shost,
                                                  " %d\n", current->comm,
                                                  channel));
                rtn = scsi_try_bus_reset(chan_scmd);
-               if (rtn == SUCCESS) {
+               if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
                        list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
                                if (channel == scmd_channel(scmd))
                                        if (!scsi_device_online(scmd->device) ||
+                                           rtn == FAST_IO_FAIL ||
                                            !scsi_eh_tur(scmd))
                                                scsi_eh_finish_cmd(scmd,
                                                                   done_q);
@@ -1246,9 +1250,10 @@ static int scsi_eh_host_reset(struct list_head *work_q,
                                                  , current->comm));
 
                rtn = scsi_try_host_reset(scmd);
-               if (rtn == SUCCESS) {
+               if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
                        list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
                                if (!scsi_device_online(scmd->device) ||
+                                   rtn == FAST_IO_FAIL ||
                                    (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
                                    !scsi_eh_tur(scmd))
                                        scsi_eh_finish_cmd(scmd, done_q);
index 55fe730a860633c46b190bd9a8aca89933a1db1e..06813789145ce09f65e67f3deee8a8bbc0eeb0c4 100644 (file)
@@ -3197,23 +3197,33 @@ fc_scsi_scan_rport(struct work_struct *work)
  *
  * This routine can be called from a FC LLD scsi_eh callback. It
  * blocks the scsi_eh thread until the fc_rport leaves the
- * FC_PORTSTATE_BLOCKED. This is necessary to avoid the scsi_eh
- * failing recovery actions for blocked rports which would lead to
- * offlined SCSI devices.
+ * FC_PORTSTATE_BLOCKED, or the fast_io_fail_tmo fires. This is
+ * necessary to avoid the scsi_eh failing recovery actions for blocked
+ * rports which would lead to offlined SCSI devices.
+ *
+ * Returns: 0 if the fc_rport left the state FC_PORTSTATE_BLOCKED.
+ *         FAST_IO_FAIL if the fast_io_fail_tmo fired, this should be
+ *         passed back to scsi_eh.
  */
-void fc_block_scsi_eh(struct scsi_cmnd *cmnd)
+int fc_block_scsi_eh(struct scsi_cmnd *cmnd)
 {
        struct Scsi_Host *shost = cmnd->device->host;
        struct fc_rport *rport = starget_to_rport(scsi_target(cmnd->device));
        unsigned long flags;
 
        spin_lock_irqsave(shost->host_lock, flags);
-       while (rport->port_state == FC_PORTSTATE_BLOCKED) {
+       while (rport->port_state == FC_PORTSTATE_BLOCKED &&
+              !(rport->flags & FC_RPORT_FAST_FAIL_TIMEDOUT)) {
                spin_unlock_irqrestore(shost->host_lock, flags);
                msleep(1000);
                spin_lock_irqsave(shost->host_lock, flags);
        }
        spin_unlock_irqrestore(shost->host_lock, flags);
+
+       if (rport->flags & FC_RPORT_FAST_FAIL_TIMEDOUT)
+               return FAST_IO_FAIL;
+
+       return 0;
 }
 EXPORT_SYMBOL(fc_block_scsi_eh);
 
index 8b4deca996ad93444f9eb2583928af28cf078ff5..832f41f37385c19d7567fec2b6908d2d83abb18b 100644 (file)
@@ -423,6 +423,7 @@ static inline int scsi_is_wlun(unsigned int lun)
 #define ADD_TO_MLQUEUE  0x2006
 #define TIMEOUT_ERROR   0x2007
 #define SCSI_RETURN_NOT_HANDLED   0x2008
+#define FAST_IO_FAIL   0x2009
 
 /*
  * Midlevel queue return values.
index 8e86a94faf06c08ba7e1cd5d12d7418ec5f5d6b0..87d81b3ce5649bf73f22ab6dfd46ea102a064dcb 100644 (file)
@@ -807,6 +807,6 @@ void fc_host_post_vendor_event(struct Scsi_Host *shost, u32 event_number,
 struct fc_vport *fc_vport_create(struct Scsi_Host *shost, int channel,
                struct fc_vport_identifiers *);
 int fc_vport_terminate(struct fc_vport *vport);
-void fc_block_scsi_eh(struct scsi_cmnd *cmnd);
+int fc_block_scsi_eh(struct scsi_cmnd *cmnd);
 
 #endif /* SCSI_TRANSPORT_FC_H */