2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/gfp.h>
20 #include <linux/timer.h>
21 #include <linux/string.h>
22 #include <linux/kernel.h>
23 #include <linux/freezer.h>
24 #include <linux/kthread.h>
25 #include <linux/interrupt.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
29 #include <scsi/scsi.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_transport.h>
35 #include <scsi/scsi_host.h>
36 #include <scsi/scsi_ioctl.h>
38 #include "scsi_priv.h"
39 #include "scsi_logging.h"
40 #include "scsi_transport_api.h"
42 #include <trace/events/scsi.h>
44 #define SENSE_TIMEOUT (10*HZ)
47 * These should *probably* be handled by the host itself.
48 * Since it is allowed to sleep, it probably should.
50 #define BUS_RESET_SETTLE_TIME (10)
51 #define HOST_RESET_SETTLE_TIME (10)
53 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
);
55 /* called with shost->host_lock held */
56 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
58 if (shost
->host_busy
== shost
->host_failed
) {
59 trace_scsi_eh_wakeup(shost
);
60 wake_up_process(shost
->ehandler
);
61 SCSI_LOG_ERROR_RECOVERY(5,
62 printk("Waking error handler thread\n"));
67 * scsi_schedule_eh - schedule EH for SCSI host
68 * @shost: SCSI host to invoke error handling on.
70 * Schedule SCSI EH without scmd.
72 void scsi_schedule_eh(struct Scsi_Host
*shost
)
76 spin_lock_irqsave(shost
->host_lock
, flags
);
78 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
79 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
80 shost
->host_eh_scheduled
++;
81 scsi_eh_wakeup(shost
);
84 spin_unlock_irqrestore(shost
->host_lock
, flags
);
86 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
89 * scsi_eh_scmd_add - add scsi cmd to error handling.
90 * @scmd: scmd to run eh on.
91 * @eh_flag: optional SCSI_EH flag.
96 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
98 struct Scsi_Host
*shost
= scmd
->device
->host
;
102 if (!shost
->ehandler
)
105 spin_lock_irqsave(shost
->host_lock
, flags
);
106 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
107 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
111 scmd
->eh_eflags
|= eh_flag
;
112 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
113 shost
->host_failed
++;
114 scsi_eh_wakeup(shost
);
116 spin_unlock_irqrestore(shost
->host_lock
, flags
);
121 * scsi_times_out - Timeout function for normal scsi commands.
122 * @req: request that is timing out.
125 * We do not need to lock this. There is the potential for a race
126 * only in that the normal completion handling might run, but if the
127 * normal completion function determines that the timer has already
128 * fired, then it mustn't do anything.
130 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
132 struct scsi_cmnd
*scmd
= req
->special
;
133 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
134 struct Scsi_Host
*host
= scmd
->device
->host
;
136 trace_scsi_dispatch_cmd_timeout(scmd
);
137 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
139 if (host
->transportt
->eh_timed_out
)
140 rtn
= host
->transportt
->eh_timed_out(scmd
);
141 else if (host
->hostt
->eh_timed_out
)
142 rtn
= host
->hostt
->eh_timed_out(scmd
);
144 if (unlikely(rtn
== BLK_EH_NOT_HANDLED
&&
145 !scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
146 scmd
->result
|= DID_TIME_OUT
<< 16;
147 rtn
= BLK_EH_HANDLED
;
154 * scsi_block_when_processing_errors - Prevent cmds from being queued.
155 * @sdev: Device on which we are performing recovery.
158 * We block until the host is out of error recovery, and then check to
159 * see whether the host or the device is offline.
162 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
164 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
168 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
170 online
= scsi_device_online(sdev
);
172 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__
,
177 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
179 #ifdef CONFIG_SCSI_LOGGING
181 * scsi_eh_prt_fail_stats - Log info on failures.
182 * @shost: scsi host being recovered.
183 * @work_q: Queue of scsi cmds to process.
185 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
186 struct list_head
*work_q
)
188 struct scsi_cmnd
*scmd
;
189 struct scsi_device
*sdev
;
190 int total_failures
= 0;
193 int devices_failed
= 0;
195 shost_for_each_device(sdev
, shost
) {
196 list_for_each_entry(scmd
, work_q
, eh_entry
) {
197 if (scmd
->device
== sdev
) {
199 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
206 if (cmd_cancel
|| cmd_failed
) {
207 SCSI_LOG_ERROR_RECOVERY(3,
208 sdev_printk(KERN_INFO
, sdev
,
209 "%s: cmds failed: %d, cancel: %d\n",
210 __func__
, cmd_failed
,
218 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
219 " devices require eh work\n",
220 total_failures
, devices_failed
));
225 * scsi_check_sense - Examine scsi cmd sense
226 * @scmd: Cmd to have sense checked.
229 * SUCCESS or FAILED or NEEDS_RETRY or TARGET_ERROR
232 * When a deferred error is detected the current command has
233 * not been executed and needs retrying.
235 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
237 struct scsi_device
*sdev
= scmd
->device
;
238 struct scsi_sense_hdr sshdr
;
240 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
241 return FAILED
; /* no valid sense data */
243 if (scsi_sense_is_deferred(&sshdr
))
246 if (sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
&&
247 sdev
->scsi_dh_data
->scsi_dh
->check_sense
) {
250 rc
= sdev
->scsi_dh_data
->scsi_dh
->check_sense(sdev
, &sshdr
);
251 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
253 /* handler does not care. Drop down to default handling */
257 * Previous logic looked for FILEMARK, EOM or ILI which are
258 * mainly associated with tapes and returned SUCCESS.
260 if (sshdr
.response_code
== 0x70) {
262 if (scmd
->sense_buffer
[2] & 0xe0)
266 * descriptor format: look for "stream commands sense data
267 * descriptor" (see SSC-3). Assume single sense data
268 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
270 if ((sshdr
.additional_length
> 3) &&
271 (scmd
->sense_buffer
[8] == 0x4) &&
272 (scmd
->sense_buffer
[11] & 0xe0))
276 switch (sshdr
.sense_key
) {
279 case RECOVERED_ERROR
:
280 return /* soft_error */ SUCCESS
;
282 case ABORTED_COMMAND
:
283 if (sshdr
.asc
== 0x10) /* DIF */
290 * if we are expecting a cc/ua because of a bus reset that we
291 * performed, treat this just as a retry. otherwise this is
292 * information that we should pass up to the upper-level driver
293 * so that we can deal with it there.
295 if (scmd
->device
->expecting_cc_ua
) {
297 * Because some device does not queue unit
298 * attentions correctly, we carefully check
299 * additional sense code and qualifier so as
300 * not to squash media change unit attention.
302 if (sshdr
.asc
!= 0x28 || sshdr
.ascq
!= 0x00) {
303 scmd
->device
->expecting_cc_ua
= 0;
308 * if the device is in the process of becoming ready, we
311 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
314 * if the device is not started, we need to wake
315 * the error handler to start the motor
317 if (scmd
->device
->allow_restart
&&
318 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
321 if (sshdr
.asc
== 0x3f && sshdr
.ascq
== 0x0e)
322 scmd_printk(KERN_WARNING
, scmd
,
323 "Warning! Received an indication that the "
324 "LUN assignments on this target have "
325 "changed. The Linux SCSI layer does not "
326 "automatically remap LUN assignments.\n");
327 else if (sshdr
.asc
== 0x3f)
328 scmd_printk(KERN_WARNING
, scmd
,
329 "Warning! Received an indication that the "
330 "operating parameters on this target have "
331 "changed. The Linux SCSI layer does not "
332 "automatically adjust these parameters.\n");
334 if (sshdr
.asc
== 0x38 && sshdr
.ascq
== 0x07)
335 scmd_printk(KERN_WARNING
, scmd
,
336 "Warning! Received an indication that the "
337 "LUN reached a thin provisioning soft "
341 * Pass the UA upwards for a determination in the completion
346 /* these are not supported */
348 case VOLUME_OVERFLOW
:
355 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
356 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
357 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
363 if (scmd
->device
->retry_hwerror
)
364 return ADD_TO_MLQUEUE
;
368 case ILLEGAL_REQUEST
:
374 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
376 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
377 struct scsi_device
*tmp_sdev
;
379 if (!sht
->change_queue_depth
||
380 sdev
->queue_depth
>= sdev
->max_queue_depth
)
383 if (time_before(jiffies
,
384 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
387 if (time_before(jiffies
,
388 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
392 * Walk all devices of a target and do
395 shost_for_each_device(tmp_sdev
, sdev
->host
) {
396 if (tmp_sdev
->channel
!= sdev
->channel
||
397 tmp_sdev
->id
!= sdev
->id
||
398 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
401 * call back into LLD to increase queue_depth by one
402 * with ramp up reason code.
404 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1,
405 SCSI_QDEPTH_RAMP_UP
);
406 sdev
->last_queue_ramp_up
= jiffies
;
410 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
412 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
413 struct scsi_device
*tmp_sdev
;
415 if (!sht
->change_queue_depth
)
418 shost_for_each_device(tmp_sdev
, sdev
->host
) {
419 if (tmp_sdev
->channel
!= sdev
->channel
||
420 tmp_sdev
->id
!= sdev
->id
)
423 * We do not know the number of commands that were at
424 * the device when we got the queue full so we start
425 * from the highest possible value and work our way down.
427 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
- 1,
433 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
434 * @scmd: SCSI cmd to examine.
437 * This is *only* called when we are examining the status of commands
438 * queued during error recovery. the main difference here is that we
439 * don't allow for the possibility of retries here, and we are a lot
440 * more restrictive about what we consider acceptable.
442 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
445 * first check the host byte, to see if there is anything in there
446 * that would indicate what we need to do.
448 if (host_byte(scmd
->result
) == DID_RESET
) {
450 * rats. we are already in the error handler, so we now
451 * get to try and figure out what to do next. if the sense
452 * is valid, we have a pretty good idea of what to do.
453 * if not, we mark it as FAILED.
455 return scsi_check_sense(scmd
);
457 if (host_byte(scmd
->result
) != DID_OK
)
461 * next, check the message byte.
463 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
467 * now, check the status byte to see if this indicates
470 switch (status_byte(scmd
->result
)) {
472 scsi_handle_queue_ramp_up(scmd
->device
);
473 case COMMAND_TERMINATED
:
475 case CHECK_CONDITION
:
476 return scsi_check_sense(scmd
);
478 case INTERMEDIATE_GOOD
:
479 case INTERMEDIATE_C_GOOD
:
481 * who knows? FIXME(eric)
484 case RESERVATION_CONFLICT
:
485 if (scmd
->cmnd
[0] == TEST_UNIT_READY
)
486 /* it is a success, we probed the device and
489 /* otherwise, we failed to send the command */
492 scsi_handle_queue_full(scmd
->device
);
503 * scsi_eh_done - Completion function for error handling.
504 * @scmd: Cmd that is done.
506 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
508 struct completion
*eh_action
;
510 SCSI_LOG_ERROR_RECOVERY(3,
511 printk("%s scmd: %p result: %x\n",
512 __func__
, scmd
, scmd
->result
));
514 eh_action
= scmd
->device
->host
->eh_action
;
520 * scsi_try_host_reset - ask host adapter to reset itself
521 * @scmd: SCSI cmd to send hsot reset.
523 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
527 struct Scsi_Host
*host
= scmd
->device
->host
;
528 struct scsi_host_template
*hostt
= host
->hostt
;
530 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
533 if (!hostt
->eh_host_reset_handler
)
536 rtn
= hostt
->eh_host_reset_handler(scmd
);
538 if (rtn
== SUCCESS
) {
539 if (!hostt
->skip_settle_delay
)
540 ssleep(HOST_RESET_SETTLE_TIME
);
541 spin_lock_irqsave(host
->host_lock
, flags
);
542 scsi_report_bus_reset(host
, scmd_channel(scmd
));
543 spin_unlock_irqrestore(host
->host_lock
, flags
);
550 * scsi_try_bus_reset - ask host to perform a bus reset
551 * @scmd: SCSI cmd to send bus reset.
553 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
557 struct Scsi_Host
*host
= scmd
->device
->host
;
558 struct scsi_host_template
*hostt
= host
->hostt
;
560 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
563 if (!hostt
->eh_bus_reset_handler
)
566 rtn
= hostt
->eh_bus_reset_handler(scmd
);
568 if (rtn
== SUCCESS
) {
569 if (!hostt
->skip_settle_delay
)
570 ssleep(BUS_RESET_SETTLE_TIME
);
571 spin_lock_irqsave(host
->host_lock
, flags
);
572 scsi_report_bus_reset(host
, scmd_channel(scmd
));
573 spin_unlock_irqrestore(host
->host_lock
, flags
);
579 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
582 sdev
->expecting_cc_ua
= 1;
586 * scsi_try_target_reset - Ask host to perform a target reset
587 * @scmd: SCSI cmd used to send a target reset
590 * There is no timeout for this operation. if this operation is
591 * unreliable for a given host, then the host itself needs to put a
592 * timer on it, and set the host back to a consistent state prior to
595 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
599 struct Scsi_Host
*host
= scmd
->device
->host
;
600 struct scsi_host_template
*hostt
= host
->hostt
;
602 if (!hostt
->eh_target_reset_handler
)
605 rtn
= hostt
->eh_target_reset_handler(scmd
);
606 if (rtn
== SUCCESS
) {
607 spin_lock_irqsave(host
->host_lock
, flags
);
608 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
609 __scsi_report_device_reset
);
610 spin_unlock_irqrestore(host
->host_lock
, flags
);
617 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
618 * @scmd: SCSI cmd used to send BDR
621 * There is no timeout for this operation. if this operation is
622 * unreliable for a given host, then the host itself needs to put a
623 * timer on it, and set the host back to a consistent state prior to
626 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
629 struct scsi_host_template
*hostt
= scmd
->device
->host
->hostt
;
631 if (!hostt
->eh_device_reset_handler
)
634 rtn
= hostt
->eh_device_reset_handler(scmd
);
636 __scsi_report_device_reset(scmd
->device
, NULL
);
640 static int scsi_try_to_abort_cmd(struct scsi_host_template
*hostt
, struct scsi_cmnd
*scmd
)
642 if (!hostt
->eh_abort_handler
)
645 return hostt
->eh_abort_handler(scmd
);
648 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
650 if (scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
) != SUCCESS
)
651 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
652 if (scsi_try_target_reset(scmd
) != SUCCESS
)
653 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
654 scsi_try_host_reset(scmd
);
658 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
659 * @scmd: SCSI command structure to hijack
660 * @ses: structure to save restore information
661 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
662 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
663 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
665 * This function is used to save a scsi command information before re-execution
666 * as part of the error recovery process. If @sense_bytes is 0 the command
667 * sent must be one that does not transfer any data. If @sense_bytes != 0
668 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
669 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
671 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
672 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
674 struct scsi_device
*sdev
= scmd
->device
;
677 * We need saved copies of a number of fields - this is because
678 * error handling may need to overwrite these with different values
679 * to run different commands, and once error handling is complete,
680 * we will need to restore these values prior to running the actual
683 ses
->cmd_len
= scmd
->cmd_len
;
684 ses
->cmnd
= scmd
->cmnd
;
685 ses
->data_direction
= scmd
->sc_data_direction
;
686 ses
->sdb
= scmd
->sdb
;
687 ses
->next_rq
= scmd
->request
->next_rq
;
688 ses
->result
= scmd
->result
;
689 ses
->underflow
= scmd
->underflow
;
690 ses
->prot_op
= scmd
->prot_op
;
692 scmd
->prot_op
= SCSI_PROT_NORMAL
;
693 scmd
->cmnd
= ses
->eh_cmnd
;
694 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
695 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
696 scmd
->request
->next_rq
= NULL
;
699 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
701 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
703 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
704 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
705 scmd
->sdb
.table
.nents
= 1;
706 scmd
->cmnd
[0] = REQUEST_SENSE
;
707 scmd
->cmnd
[4] = scmd
->sdb
.length
;
708 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
710 scmd
->sc_data_direction
= DMA_NONE
;
712 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
713 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
714 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
720 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
721 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
722 (sdev
->lun
<< 5 & 0xe0);
725 * Zero the sense buffer. The scsi spec mandates that any
726 * untransferred sense data should be interpreted as being zero.
728 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
730 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
733 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
734 * @scmd: SCSI command structure to restore
735 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
737 * Undo any damage done by above scsi_eh_prep_cmnd().
739 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
742 * Restore original data
744 scmd
->cmd_len
= ses
->cmd_len
;
745 scmd
->cmnd
= ses
->cmnd
;
746 scmd
->sc_data_direction
= ses
->data_direction
;
747 scmd
->sdb
= ses
->sdb
;
748 scmd
->request
->next_rq
= ses
->next_rq
;
749 scmd
->result
= ses
->result
;
750 scmd
->underflow
= ses
->underflow
;
751 scmd
->prot_op
= ses
->prot_op
;
753 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
756 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
757 * @scmd: SCSI command structure to hijack
759 * @cmnd_size: size in bytes of @cmnd
760 * @timeout: timeout for this request
761 * @sense_bytes: size of sense data to copy or 0
763 * This function is used to send a scsi command down to a target device
764 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
767 * SUCCESS or FAILED or NEEDS_RETRY
769 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
770 int cmnd_size
, int timeout
, unsigned sense_bytes
)
772 struct scsi_device
*sdev
= scmd
->device
;
773 struct Scsi_Host
*shost
= sdev
->host
;
774 DECLARE_COMPLETION_ONSTACK(done
);
775 unsigned long timeleft
;
776 struct scsi_eh_save ses
;
779 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
780 shost
->eh_action
= &done
;
783 scmd
->scsi_done
= scsi_eh_done
;
784 shost
->hostt
->queuecommand(shost
, scmd
);
786 timeleft
= wait_for_completion_timeout(&done
, timeout
);
788 shost
->eh_action
= NULL
;
790 scsi_log_completion(scmd
, SUCCESS
);
792 SCSI_LOG_ERROR_RECOVERY(3,
793 printk("%s: scmd: %p, timeleft: %ld\n",
794 __func__
, scmd
, timeleft
));
797 * If there is time left scsi_eh_done got called, and we will
798 * examine the actual status codes to see whether the command
799 * actually did complete normally, else tell the host to forget
800 * about this command.
803 rtn
= scsi_eh_completed_normally(scmd
);
804 SCSI_LOG_ERROR_RECOVERY(3,
805 printk("%s: scsi_eh_completed_normally %x\n",
822 scsi_abort_eh_cmnd(scmd
);
826 scsi_eh_restore_cmnd(scmd
, &ses
);
831 * scsi_request_sense - Request sense data from a particular target.
832 * @scmd: SCSI cmd for request sense.
835 * Some hosts automatically obtain this information, others require
836 * that we obtain it on our own. This function will *not* return until
837 * the command either times out, or it completes.
839 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
841 return scsi_send_eh_cmnd(scmd
, NULL
, 0, SENSE_TIMEOUT
, ~0);
845 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
846 * @scmd: Original SCSI cmd that eh has finished.
847 * @done_q: Queue for processed commands.
850 * We don't want to use the normal command completion while we are are
851 * still handling errors - it may cause other commands to be queued,
852 * and that would disturb what we are doing. Thus we really want to
853 * keep a list of pending commands for final completion, and once we
854 * are ready to leave error handling we handle completion for real.
856 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
858 scmd
->device
->host
->host_failed
--;
860 list_move_tail(&scmd
->eh_entry
, done_q
);
862 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
865 * scsi_eh_get_sense - Get device sense data.
866 * @work_q: Queue of commands to process.
867 * @done_q: Queue of processed commands.
870 * See if we need to request sense information. if so, then get it
871 * now, so we have a better idea of what to do.
874 * This has the unfortunate side effect that if a shost adapter does
875 * not automatically request sense information, we end up shutting
876 * it down before we request it.
878 * All drivers should request sense information internally these days,
879 * so for now all I have to say is tough noogies if you end up in here.
881 * XXX: Long term this code should go away, but that needs an audit of
884 int scsi_eh_get_sense(struct list_head
*work_q
,
885 struct list_head
*done_q
)
887 struct scsi_cmnd
*scmd
, *next
;
890 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
891 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
892 SCSI_SENSE_VALID(scmd
))
895 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
896 "%s: requesting sense\n",
898 rtn
= scsi_request_sense(scmd
);
902 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
903 " result %x\n", scmd
,
905 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
907 rtn
= scsi_decide_disposition(scmd
);
910 * if the result was normal, then just pass it along to the
914 /* we don't want this command reissued, just
915 * finished with the sense data, so set
916 * retries to the max allowed to ensure it
917 * won't get reissued */
918 scmd
->retries
= scmd
->allowed
;
919 else if (rtn
!= NEEDS_RETRY
)
922 scsi_eh_finish_cmd(scmd
, done_q
);
925 return list_empty(work_q
);
927 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
930 * scsi_eh_tur - Send TUR to device.
931 * @scmd: &scsi_cmnd to send TUR
934 * 0 - Device is ready. 1 - Device NOT ready.
936 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
938 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
939 int retry_cnt
= 1, rtn
;
942 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6, SENSE_TIMEOUT
, 0);
944 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
945 __func__
, scmd
, rtn
));
960 * scsi_eh_test_devices - check if devices are responding from error recovery.
961 * @cmd_list: scsi commands in error recovery.
962 * @work_q: queue for commands which still need more error recovery
963 * @done_q: queue for commands which are finished
964 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
967 * Tests if devices are in a working state. Commands to devices now in
968 * a working state are sent to the done_q while commands to devices which
969 * are still failing to respond are returned to the work_q for more
972 static int scsi_eh_test_devices(struct list_head
*cmd_list
,
973 struct list_head
*work_q
,
974 struct list_head
*done_q
, int try_stu
)
976 struct scsi_cmnd
*scmd
, *next
;
977 struct scsi_device
*sdev
;
980 while (!list_empty(cmd_list
)) {
981 scmd
= list_entry(cmd_list
->next
, struct scsi_cmnd
, eh_entry
);
984 finish_cmds
= !scsi_device_online(scmd
->device
) ||
985 (try_stu
&& !scsi_eh_try_stu(scmd
) &&
986 !scsi_eh_tur(scmd
)) ||
989 list_for_each_entry_safe(scmd
, next
, cmd_list
, eh_entry
)
990 if (scmd
->device
== sdev
) {
992 scsi_eh_finish_cmd(scmd
, done_q
);
994 list_move_tail(&scmd
->eh_entry
, work_q
);
997 return list_empty(work_q
);
1002 * scsi_eh_abort_cmds - abort pending commands.
1003 * @work_q: &list_head for pending commands.
1004 * @done_q: &list_head for processed commands.
1007 * Try and see whether or not it makes sense to try and abort the
1008 * running command. This only works out to be the case if we have one
1009 * command that has timed out. If the command simply failed, it makes
1010 * no sense to try and abort the command, since as far as the shost
1011 * adapter is concerned, it isn't running.
1013 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
1014 struct list_head
*done_q
)
1016 struct scsi_cmnd
*scmd
, *next
;
1017 LIST_HEAD(check_list
);
1020 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1021 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
1023 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1024 "0x%p\n", current
->comm
,
1026 rtn
= scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
);
1027 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1028 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
1029 if (rtn
== FAST_IO_FAIL
)
1030 scsi_eh_finish_cmd(scmd
, done_q
);
1032 list_move_tail(&scmd
->eh_entry
, &check_list
);
1034 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1041 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1045 * scsi_eh_try_stu - Send START_UNIT to device.
1046 * @scmd: &scsi_cmnd to send START_UNIT
1049 * 0 - Device is ready. 1 - Device NOT ready.
1051 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1053 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1055 if (scmd
->device
->allow_restart
) {
1056 int i
, rtn
= NEEDS_RETRY
;
1058 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1059 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1069 * scsi_eh_stu - send START_UNIT if needed
1070 * @shost: &scsi host being recovered.
1071 * @work_q: &list_head for pending commands.
1072 * @done_q: &list_head for processed commands.
1075 * If commands are failing due to not ready, initializing command required,
1076 * try revalidating the device, which will end up sending a start unit.
1078 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1079 struct list_head
*work_q
,
1080 struct list_head
*done_q
)
1082 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1083 struct scsi_device
*sdev
;
1085 shost_for_each_device(sdev
, shost
) {
1087 list_for_each_entry(scmd
, work_q
, eh_entry
)
1088 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1089 scsi_check_sense(scmd
) == FAILED
) {
1097 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1098 " 0x%p\n", current
->comm
, sdev
));
1100 if (!scsi_eh_try_stu(stu_scmd
)) {
1101 if (!scsi_device_online(sdev
) ||
1102 !scsi_eh_tur(stu_scmd
)) {
1103 list_for_each_entry_safe(scmd
, next
,
1105 if (scmd
->device
== sdev
)
1106 scsi_eh_finish_cmd(scmd
, done_q
);
1110 SCSI_LOG_ERROR_RECOVERY(3,
1111 printk("%s: START_UNIT failed to sdev:"
1112 " 0x%p\n", current
->comm
, sdev
));
1116 return list_empty(work_q
);
1121 * scsi_eh_bus_device_reset - send bdr if needed
1122 * @shost: scsi host being recovered.
1123 * @work_q: &list_head for pending commands.
1124 * @done_q: &list_head for processed commands.
1127 * Try a bus device reset. Still, look to see whether we have multiple
1128 * devices that are jammed or not - if we have multiple devices, it
1129 * makes no sense to try bus_device_reset - we really would need to try
1130 * a bus_reset instead.
1132 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1133 struct list_head
*work_q
,
1134 struct list_head
*done_q
)
1136 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1137 struct scsi_device
*sdev
;
1140 shost_for_each_device(sdev
, shost
) {
1142 list_for_each_entry(scmd
, work_q
, eh_entry
)
1143 if (scmd
->device
== sdev
) {
1151 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1152 " 0x%p\n", current
->comm
,
1154 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1155 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1156 if (!scsi_device_online(sdev
) ||
1157 rtn
== FAST_IO_FAIL
||
1158 !scsi_eh_tur(bdr_scmd
)) {
1159 list_for_each_entry_safe(scmd
, next
,
1161 if (scmd
->device
== sdev
)
1162 scsi_eh_finish_cmd(scmd
,
1167 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1175 return list_empty(work_q
);
1179 * scsi_eh_target_reset - send target reset if needed
1180 * @shost: scsi host being recovered.
1181 * @work_q: &list_head for pending commands.
1182 * @done_q: &list_head for processed commands.
1185 * Try a target reset.
1187 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1188 struct list_head
*work_q
,
1189 struct list_head
*done_q
)
1191 LIST_HEAD(tmp_list
);
1192 LIST_HEAD(check_list
);
1194 list_splice_init(work_q
, &tmp_list
);
1196 while (!list_empty(&tmp_list
)) {
1197 struct scsi_cmnd
*next
, *scmd
;
1201 scmd
= list_entry(tmp_list
.next
, struct scsi_cmnd
, eh_entry
);
1204 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1206 current
->comm
, id
));
1207 rtn
= scsi_try_target_reset(scmd
);
1208 if (rtn
!= SUCCESS
&& rtn
!= FAST_IO_FAIL
)
1209 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1212 current
->comm
, id
));
1213 list_for_each_entry_safe(scmd
, next
, &tmp_list
, eh_entry
) {
1214 if (scmd_id(scmd
) != id
)
1218 list_move_tail(&scmd
->eh_entry
, &check_list
);
1219 else if (rtn
== FAST_IO_FAIL
)
1220 scsi_eh_finish_cmd(scmd
, done_q
);
1222 /* push back on work queue for further processing */
1223 list_move(&scmd
->eh_entry
, work_q
);
1227 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1231 * scsi_eh_bus_reset - send a bus reset
1232 * @shost: &scsi host being recovered.
1233 * @work_q: &list_head for pending commands.
1234 * @done_q: &list_head for processed commands.
1236 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1237 struct list_head
*work_q
,
1238 struct list_head
*done_q
)
1240 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1241 LIST_HEAD(check_list
);
1242 unsigned int channel
;
1246 * we really want to loop over the various channels, and do this on
1247 * a channel by channel basis. we should also check to see if any
1248 * of the failed commands are on soft_reset devices, and if so, skip
1252 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1254 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1255 if (channel
== scmd_channel(scmd
)) {
1259 * FIXME add back in some support for
1260 * soft_reset devices.
1267 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1268 " %d\n", current
->comm
,
1270 rtn
= scsi_try_bus_reset(chan_scmd
);
1271 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1272 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1273 if (channel
== scmd_channel(scmd
)) {
1274 if (rtn
== FAST_IO_FAIL
)
1275 scsi_eh_finish_cmd(scmd
,
1278 list_move_tail(&scmd
->eh_entry
,
1283 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1284 " failed chan: %d\n",
1289 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1293 * scsi_eh_host_reset - send a host reset
1294 * @work_q: list_head for processed commands.
1295 * @done_q: list_head for processed commands.
1297 static int scsi_eh_host_reset(struct list_head
*work_q
,
1298 struct list_head
*done_q
)
1300 struct scsi_cmnd
*scmd
, *next
;
1301 LIST_HEAD(check_list
);
1304 if (!list_empty(work_q
)) {
1305 scmd
= list_entry(work_q
->next
,
1306 struct scsi_cmnd
, eh_entry
);
1308 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1311 rtn
= scsi_try_host_reset(scmd
);
1312 if (rtn
== SUCCESS
) {
1313 list_splice_init(work_q
, &check_list
);
1314 } else if (rtn
== FAST_IO_FAIL
) {
1315 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1316 scsi_eh_finish_cmd(scmd
, done_q
);
1319 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1324 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 1);
1328 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1329 * @work_q: list_head for processed commands.
1330 * @done_q: list_head for processed commands.
1332 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1333 struct list_head
*done_q
)
1335 struct scsi_cmnd
*scmd
, *next
;
1337 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1338 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1339 "not ready after error recovery\n");
1340 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1341 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1343 * FIXME: Handle lost cmds.
1346 scsi_eh_finish_cmd(scmd
, done_q
);
1352 * scsi_noretry_cmd - determinte if command should be failed fast
1353 * @scmd: SCSI cmd to examine.
1355 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1357 switch (host_byte(scmd
->result
)) {
1361 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1363 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
);
1365 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1366 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1369 case DID_SOFT_ERROR
:
1370 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1373 switch (status_byte(scmd
->result
)) {
1374 case CHECK_CONDITION
:
1376 * assume caller has checked sense and determinted
1377 * the check condition was retryable.
1379 if (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
||
1380 scmd
->request
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1388 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1389 * @scmd: SCSI cmd to examine.
1392 * This is *only* called when we are examining the status after sending
1393 * out the actual data command. any commands that are queued for error
1394 * recovery (e.g. test_unit_ready) do *not* come through here.
1396 * When this routine returns failed, it means the error handler thread
1397 * is woken. In cases where the error code indicates an error that
1398 * doesn't require the error handler read (i.e. we don't need to
1399 * abort/reset), this function should return SUCCESS.
1401 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1406 * if the device is offline, then we clearly just pass the result back
1407 * up to the top level.
1409 if (!scsi_device_online(scmd
->device
)) {
1410 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1417 * first check the host byte, to see if there is anything in there
1418 * that would indicate what we need to do.
1420 switch (host_byte(scmd
->result
)) {
1421 case DID_PASSTHROUGH
:
1423 * no matter what, pass this through to the upper layer.
1424 * nuke this special code so that it looks like we are saying
1427 scmd
->result
&= 0xff00ffff;
1431 * looks good. drop through, and check the next byte.
1434 case DID_NO_CONNECT
:
1435 case DID_BAD_TARGET
:
1438 * note - this means that we just report the status back
1439 * to the top level driver, not that we actually think
1440 * that it indicates SUCCESS.
1444 * when the low level driver returns did_soft_error,
1445 * it is responsible for keeping an internal retry counter
1446 * in order to avoid endless loops (db)
1448 * actually this is a bug in this function here. we should
1449 * be mindful of the maximum number of retries specified
1450 * and not get stuck in a loop.
1452 case DID_SOFT_ERROR
:
1458 return ADD_TO_MLQUEUE
;
1459 case DID_TRANSPORT_DISRUPTED
:
1461 * LLD/transport was disrupted during processing of the IO.
1462 * The transport class is now blocked/blocking,
1463 * and the transport will decide what to do with the IO
1464 * based on its timers and recovery capablilities if
1465 * there are enough retries.
1468 case DID_TRANSPORT_FAILFAST
:
1470 * The transport decided to failfast the IO (most likely
1471 * the fast io fail tmo fired), so send IO directly upwards.
1475 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1476 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1478 * execute reservation conflict processing code
1488 * when we scan the bus, we get timeout messages for
1489 * these commands if there is no device available.
1490 * other hosts report did_no_connect for the same thing.
1492 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1493 scmd
->cmnd
[0] == INQUIRY
)) {
1505 * next, check the message byte.
1507 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1511 * check the status byte to see if this indicates anything special.
1513 switch (status_byte(scmd
->result
)) {
1515 scsi_handle_queue_full(scmd
->device
);
1517 * the case of trying to send too many commands to a
1518 * tagged queueing device.
1522 * device can't talk to us at the moment. Should only
1523 * occur (SAM-3) when the task queue is empty, so will cause
1524 * the empty queue handling to trigger a stall in the
1527 return ADD_TO_MLQUEUE
;
1529 scsi_handle_queue_ramp_up(scmd
->device
);
1530 case COMMAND_TERMINATED
:
1534 case CHECK_CONDITION
:
1535 rtn
= scsi_check_sense(scmd
);
1536 if (rtn
== NEEDS_RETRY
)
1538 else if (rtn
== TARGET_ERROR
) {
1540 * Need to modify host byte to signal a
1541 * permanent target failure
1543 scmd
->result
|= (DID_TARGET_FAILURE
<< 16);
1546 /* if rtn == FAILED, we have no sense information;
1547 * returning FAILED will wake the error handler thread
1548 * to collect the sense and redo the decide
1551 case CONDITION_GOOD
:
1552 case INTERMEDIATE_GOOD
:
1553 case INTERMEDIATE_C_GOOD
:
1556 * who knows? FIXME(eric)
1560 case RESERVATION_CONFLICT
:
1561 sdev_printk(KERN_INFO
, scmd
->device
,
1562 "reservation conflict\n");
1563 scmd
->result
|= (DID_NEXUS_FAILURE
<< 16);
1564 return SUCCESS
; /* causes immediate i/o error */
1572 /* we requeue for retry because the error was retryable, and
1573 * the request was not marked fast fail. Note that above,
1574 * even if the request is marked fast fail, we still requeue
1575 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1576 if ((++scmd
->retries
) <= scmd
->allowed
1577 && !scsi_noretry_cmd(scmd
)) {
1581 * no more retries - report this one back to upper level.
1587 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1589 __blk_put_request(req
->q
, req
);
1593 * scsi_eh_lock_door - Prevent medium removal for the specified device
1594 * @sdev: SCSI device to prevent medium removal
1597 * We must be called from process context.
1600 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1601 * head of the devices request queue, and continue.
1603 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1605 struct request
*req
;
1608 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1609 * request becomes available
1611 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1613 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1617 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1620 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1622 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1623 req
->cmd_flags
|= REQ_QUIET
;
1624 req
->timeout
= 10 * HZ
;
1627 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1631 * scsi_restart_operations - restart io operations to the specified host.
1632 * @shost: Host we are restarting.
1635 * When we entered the error handler, we blocked all further i/o to
1636 * this device. we need to 'reverse' this process.
1638 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1640 struct scsi_device
*sdev
;
1641 unsigned long flags
;
1644 * If the door was locked, we need to insert a door lock request
1645 * onto the head of the SCSI request queue for the device. There
1646 * is no point trying to lock the door of an off-line device.
1648 shost_for_each_device(sdev
, shost
) {
1649 if (scsi_device_online(sdev
) && sdev
->locked
)
1650 scsi_eh_lock_door(sdev
);
1654 * next free up anything directly waiting upon the host. this
1655 * will be requests for character device operations, and also for
1656 * ioctls to queued block devices.
1658 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1661 spin_lock_irqsave(shost
->host_lock
, flags
);
1662 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1663 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1664 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1665 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1667 wake_up(&shost
->host_wait
);
1670 * finally we need to re-initiate requests that may be pending. we will
1671 * have had everything blocked while error handling is taking place, and
1672 * now that error recovery is done, we will need to ensure that these
1673 * requests are started.
1675 scsi_run_host_queues(shost
);
1679 * scsi_eh_ready_devs - check device ready state and recover if not.
1680 * @shost: host to be recovered.
1681 * @work_q: &list_head for pending commands.
1682 * @done_q: &list_head for processed commands.
1684 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1685 struct list_head
*work_q
,
1686 struct list_head
*done_q
)
1688 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1689 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1690 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1691 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1692 if (!scsi_eh_host_reset(work_q
, done_q
))
1693 scsi_eh_offline_sdevs(work_q
,
1696 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1699 * scsi_eh_flush_done_q - finish processed commands or retry them.
1700 * @done_q: list_head of processed commands.
1702 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1704 struct scsi_cmnd
*scmd
, *next
;
1706 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1707 list_del_init(&scmd
->eh_entry
);
1708 if (scsi_device_online(scmd
->device
) &&
1709 !scsi_noretry_cmd(scmd
) &&
1710 (++scmd
->retries
<= scmd
->allowed
)) {
1711 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1715 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1718 * If just we got sense for the device (called
1719 * scsi_eh_get_sense), scmd->result is already
1720 * set, do not set DRIVER_TIMEOUT.
1723 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1724 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1726 current
->comm
, scmd
));
1727 scsi_finish_command(scmd
);
1731 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1734 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1735 * @shost: Host to unjam.
1738 * When we come in here, we *know* that all commands on the bus have
1739 * either completed, failed or timed out. we also know that no further
1740 * commands are being sent to the host, so things are relatively quiet
1741 * and we have freedom to fiddle with things as we wish.
1743 * This is only the *default* implementation. it is possible for
1744 * individual drivers to supply their own version of this function, and
1745 * if the maintainer wishes to do this, it is strongly suggested that
1746 * this function be taken as a template and modified. this function
1747 * was designed to correctly handle problems for about 95% of the
1748 * different cases out there, and it should always provide at least a
1749 * reasonable amount of error recovery.
1751 * Any command marked 'failed' or 'timeout' must eventually have
1752 * scsi_finish_cmd() called for it. we do all of the retry stuff
1753 * here, so when we restart the host after we return it should have an
1756 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1758 unsigned long flags
;
1759 LIST_HEAD(eh_work_q
);
1760 LIST_HEAD(eh_done_q
);
1762 spin_lock_irqsave(shost
->host_lock
, flags
);
1763 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1764 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1766 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1768 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1769 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1770 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1772 scsi_eh_flush_done_q(&eh_done_q
);
1776 * scsi_error_handler - SCSI error handler thread
1777 * @data: Host for which we are running.
1780 * This is the main error handling loop. This is run as a kernel thread
1781 * for every SCSI host and handles all error handling activity.
1783 int scsi_error_handler(void *data
)
1785 struct Scsi_Host
*shost
= data
;
1788 * We use TASK_INTERRUPTIBLE so that the thread is not
1789 * counted against the load average as a running process.
1790 * We never actually get interrupted because kthread_run
1791 * disables signal delivery for the created thread.
1793 set_current_state(TASK_INTERRUPTIBLE
);
1794 while (!kthread_should_stop()) {
1795 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1796 shost
->host_failed
!= shost
->host_busy
) {
1797 SCSI_LOG_ERROR_RECOVERY(1,
1798 printk("Error handler scsi_eh_%d sleeping\n",
1801 set_current_state(TASK_INTERRUPTIBLE
);
1805 __set_current_state(TASK_RUNNING
);
1806 SCSI_LOG_ERROR_RECOVERY(1,
1807 printk("Error handler scsi_eh_%d waking up\n",
1811 * We have a host that is failing for some reason. Figure out
1812 * what we need to do to get it up and online again (if we can).
1813 * If we fail, we end up taking the thing offline.
1815 if (scsi_autopm_get_host(shost
) != 0) {
1816 SCSI_LOG_ERROR_RECOVERY(1,
1817 printk(KERN_ERR
"Error handler scsi_eh_%d "
1818 "unable to autoresume\n",
1823 if (shost
->transportt
->eh_strategy_handler
)
1824 shost
->transportt
->eh_strategy_handler(shost
);
1826 scsi_unjam_host(shost
);
1829 * Note - if the above fails completely, the action is to take
1830 * individual devices offline and flush the queue of any
1831 * outstanding requests that may have been pending. When we
1832 * restart, we restart any I/O to any other devices on the bus
1833 * which are still online.
1835 scsi_restart_operations(shost
);
1836 scsi_autopm_put_host(shost
);
1837 set_current_state(TASK_INTERRUPTIBLE
);
1839 __set_current_state(TASK_RUNNING
);
1841 SCSI_LOG_ERROR_RECOVERY(1,
1842 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1843 shost
->ehandler
= NULL
;
1848 * Function: scsi_report_bus_reset()
1850 * Purpose: Utility function used by low-level drivers to report that
1851 * they have observed a bus reset on the bus being handled.
1853 * Arguments: shost - Host in question
1854 * channel - channel on which reset was observed.
1858 * Lock status: Host lock must be held.
1860 * Notes: This only needs to be called if the reset is one which
1861 * originates from an unknown location. Resets originated
1862 * by the mid-level itself don't need to call this, but there
1863 * should be no harm.
1865 * The main purpose of this is to make sure that a CHECK_CONDITION
1866 * is properly treated.
1868 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1870 struct scsi_device
*sdev
;
1872 __shost_for_each_device(sdev
, shost
) {
1873 if (channel
== sdev_channel(sdev
))
1874 __scsi_report_device_reset(sdev
, NULL
);
1877 EXPORT_SYMBOL(scsi_report_bus_reset
);
1880 * Function: scsi_report_device_reset()
1882 * Purpose: Utility function used by low-level drivers to report that
1883 * they have observed a device reset on the device being handled.
1885 * Arguments: shost - Host in question
1886 * channel - channel on which reset was observed
1887 * target - target on which reset was observed
1891 * Lock status: Host lock must be held
1893 * Notes: This only needs to be called if the reset is one which
1894 * originates from an unknown location. Resets originated
1895 * by the mid-level itself don't need to call this, but there
1896 * should be no harm.
1898 * The main purpose of this is to make sure that a CHECK_CONDITION
1899 * is properly treated.
1901 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1903 struct scsi_device
*sdev
;
1905 __shost_for_each_device(sdev
, shost
) {
1906 if (channel
== sdev_channel(sdev
) &&
1907 target
== sdev_id(sdev
))
1908 __scsi_report_device_reset(sdev
, NULL
);
1911 EXPORT_SYMBOL(scsi_report_device_reset
);
1914 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1919 * Function: scsi_reset_provider
1921 * Purpose: Send requested reset to a bus or device at any phase.
1923 * Arguments: device - device to send reset to
1924 * flag - reset type (see scsi.h)
1926 * Returns: SUCCESS/FAILURE.
1928 * Notes: This is used by the SCSI Generic driver to provide
1929 * Bus/Device reset capability.
1932 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1934 struct scsi_cmnd
*scmd
;
1935 struct Scsi_Host
*shost
= dev
->host
;
1937 unsigned long flags
;
1940 if (scsi_autopm_get_host(shost
) < 0)
1943 scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1944 blk_rq_init(NULL
, &req
);
1945 scmd
->request
= &req
;
1947 scmd
->cmnd
= req
.cmd
;
1949 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1950 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1954 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1956 spin_lock_irqsave(shost
->host_lock
, flags
);
1957 shost
->tmf_in_progress
= 1;
1958 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1961 case SCSI_TRY_RESET_DEVICE
:
1962 rtn
= scsi_try_bus_device_reset(scmd
);
1966 case SCSI_TRY_RESET_TARGET
:
1967 rtn
= scsi_try_target_reset(scmd
);
1971 case SCSI_TRY_RESET_BUS
:
1972 rtn
= scsi_try_bus_reset(scmd
);
1976 case SCSI_TRY_RESET_HOST
:
1977 rtn
= scsi_try_host_reset(scmd
);
1983 spin_lock_irqsave(shost
->host_lock
, flags
);
1984 shost
->tmf_in_progress
= 0;
1985 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1988 * be sure to wake up anyone who was sleeping or had their queue
1989 * suspended while we performed the TMF.
1991 SCSI_LOG_ERROR_RECOVERY(3,
1992 printk("%s: waking up host to restart after TMF\n",
1995 wake_up(&shost
->host_wait
);
1997 scsi_run_host_queues(shost
);
1999 scsi_next_command(scmd
);
2000 scsi_autopm_put_host(shost
);
2003 EXPORT_SYMBOL(scsi_reset_provider
);
2006 * scsi_normalize_sense - normalize main elements from either fixed or
2007 * descriptor sense data format into a common format.
2009 * @sense_buffer: byte array containing sense data returned by device
2010 * @sb_len: number of valid bytes in sense_buffer
2011 * @sshdr: pointer to instance of structure that common
2012 * elements are written to.
2015 * The "main elements" from sense data are: response_code, sense_key,
2016 * asc, ascq and additional_length (only for descriptor format).
2018 * Typically this function can be called after a device has
2019 * responded to a SCSI command with the CHECK_CONDITION status.
2022 * 1 if valid sense data information found, else 0;
2024 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
2025 struct scsi_sense_hdr
*sshdr
)
2027 if (!sense_buffer
|| !sb_len
)
2030 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
2032 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
2034 if (!scsi_sense_valid(sshdr
))
2037 if (sshdr
->response_code
>= 0x72) {
2042 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
2044 sshdr
->asc
= sense_buffer
[2];
2046 sshdr
->ascq
= sense_buffer
[3];
2048 sshdr
->additional_length
= sense_buffer
[7];
2054 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
2056 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
2057 sb_len
: (sense_buffer
[7] + 8);
2059 sshdr
->asc
= sense_buffer
[12];
2061 sshdr
->ascq
= sense_buffer
[13];
2067 EXPORT_SYMBOL(scsi_normalize_sense
);
2069 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
2070 struct scsi_sense_hdr
*sshdr
)
2072 return scsi_normalize_sense(cmd
->sense_buffer
,
2073 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2075 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2078 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2079 * @sense_buffer: byte array of descriptor format sense data
2080 * @sb_len: number of valid bytes in sense_buffer
2081 * @desc_type: value of descriptor type to find
2082 * (e.g. 0 -> information)
2085 * only valid when sense data is in descriptor format
2088 * pointer to start of (first) descriptor if found else NULL
2090 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2093 int add_sen_len
, add_len
, desc_len
, k
;
2096 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2098 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2100 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2101 add_sen_len
: (sb_len
- 8);
2102 descp
= &sense_buffer
[8];
2103 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2105 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2106 desc_len
= add_len
+ 2;
2107 if (descp
[0] == desc_type
)
2109 if (add_len
< 0) // short descriptor ??
2114 EXPORT_SYMBOL(scsi_sense_desc_find
);
2117 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2118 * @sense_buffer: byte array of sense data
2119 * @sb_len: number of valid bytes in sense_buffer
2120 * @info_out: pointer to 64 integer where 8 or 4 byte information
2121 * field will be placed if found.
2124 * 1 if information field found, 0 if not found.
2126 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2135 switch (sense_buffer
[0] & 0x7f) {
2138 if (sense_buffer
[0] & 0x80) {
2139 *info_out
= (sense_buffer
[3] << 24) +
2140 (sense_buffer
[4] << 16) +
2141 (sense_buffer
[5] << 8) + sense_buffer
[6];
2147 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2149 if (ucp
&& (0xa == ucp
[1])) {
2151 for (j
= 0; j
< 8; ++j
) {
2164 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2167 * scsi_build_sense_buffer - build sense data in a buffer
2168 * @desc: Sense format (non zero == descriptor format,
2169 * 0 == fixed format)
2170 * @buf: Where to build sense data
2172 * @asc: Additional sense code
2173 * @ascq: Additional sense code qualifier
2176 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2179 buf
[0] = 0x72; /* descriptor, current */
2185 buf
[0] = 0x70; /* fixed, current */
2192 EXPORT_SYMBOL(scsi_build_sense_buffer
);