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>
28 #include <linux/jiffies.h>
30 #include <scsi/scsi.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_dbg.h>
33 #include <scsi/scsi_device.h>
34 #include <scsi/scsi_driver.h>
35 #include <scsi/scsi_eh.h>
36 #include <scsi/scsi_transport.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_ioctl.h>
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
42 #include "scsi_transport_api.h"
44 #include <trace/events/scsi.h>
46 static void scsi_eh_done(struct scsi_cmnd
*scmd
);
49 * These should *probably* be handled by the host itself.
50 * Since it is allowed to sleep, it probably should.
52 #define BUS_RESET_SETTLE_TIME (10)
53 #define HOST_RESET_SETTLE_TIME (10)
55 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
);
57 /* called with shost->host_lock held */
58 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
60 if (shost
->host_busy
== shost
->host_failed
) {
61 trace_scsi_eh_wakeup(shost
);
62 wake_up_process(shost
->ehandler
);
63 SCSI_LOG_ERROR_RECOVERY(5,
64 printk("Waking error handler thread\n"));
69 * scsi_schedule_eh - schedule EH for SCSI host
70 * @shost: SCSI host to invoke error handling on.
72 * Schedule SCSI EH without scmd.
74 void scsi_schedule_eh(struct Scsi_Host
*shost
)
78 spin_lock_irqsave(shost
->host_lock
, flags
);
80 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
81 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
82 shost
->host_eh_scheduled
++;
83 scsi_eh_wakeup(shost
);
86 spin_unlock_irqrestore(shost
->host_lock
, flags
);
88 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
91 * scsi_eh_scmd_add - add scsi cmd to error handling.
92 * @scmd: scmd to run eh on.
93 * @eh_flag: optional SCSI_EH flag.
98 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
100 struct Scsi_Host
*shost
= scmd
->device
->host
;
104 if (!shost
->ehandler
)
107 spin_lock_irqsave(shost
->host_lock
, flags
);
108 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
109 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
113 scmd
->eh_eflags
|= eh_flag
;
114 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
115 shost
->host_failed
++;
116 scsi_eh_wakeup(shost
);
118 spin_unlock_irqrestore(shost
->host_lock
, flags
);
123 * scsi_times_out - Timeout function for normal scsi commands.
124 * @req: request that is timing out.
127 * We do not need to lock this. There is the potential for a race
128 * only in that the normal completion handling might run, but if the
129 * normal completion function determines that the timer has already
130 * fired, then it mustn't do anything.
132 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
134 struct scsi_cmnd
*scmd
= req
->special
;
135 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
136 struct Scsi_Host
*host
= scmd
->device
->host
;
138 trace_scsi_dispatch_cmd_timeout(scmd
);
139 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
141 if (host
->transportt
->eh_timed_out
)
142 rtn
= host
->transportt
->eh_timed_out(scmd
);
143 else if (host
->hostt
->eh_timed_out
)
144 rtn
= host
->hostt
->eh_timed_out(scmd
);
146 scmd
->result
|= DID_TIME_OUT
<< 16;
148 if (unlikely(rtn
== BLK_EH_NOT_HANDLED
&&
149 !scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
)))
150 rtn
= BLK_EH_HANDLED
;
156 * scsi_block_when_processing_errors - Prevent cmds from being queued.
157 * @sdev: Device on which we are performing recovery.
160 * We block until the host is out of error recovery, and then check to
161 * see whether the host or the device is offline.
164 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
166 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
170 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
172 online
= scsi_device_online(sdev
);
174 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__
,
179 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
181 #ifdef CONFIG_SCSI_LOGGING
183 * scsi_eh_prt_fail_stats - Log info on failures.
184 * @shost: scsi host being recovered.
185 * @work_q: Queue of scsi cmds to process.
187 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
188 struct list_head
*work_q
)
190 struct scsi_cmnd
*scmd
;
191 struct scsi_device
*sdev
;
192 int total_failures
= 0;
195 int devices_failed
= 0;
197 shost_for_each_device(sdev
, shost
) {
198 list_for_each_entry(scmd
, work_q
, eh_entry
) {
199 if (scmd
->device
== sdev
) {
201 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
208 if (cmd_cancel
|| cmd_failed
) {
209 SCSI_LOG_ERROR_RECOVERY(3,
210 sdev_printk(KERN_INFO
, sdev
,
211 "%s: cmds failed: %d, cancel: %d\n",
212 __func__
, cmd_failed
,
220 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
221 " devices require eh work\n",
222 total_failures
, devices_failed
));
227 * scsi_check_sense - Examine scsi cmd sense
228 * @scmd: Cmd to have sense checked.
231 * SUCCESS or FAILED or NEEDS_RETRY or TARGET_ERROR
234 * When a deferred error is detected the current command has
235 * not been executed and needs retrying.
237 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
239 struct scsi_device
*sdev
= scmd
->device
;
240 struct scsi_sense_hdr sshdr
;
242 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
243 return FAILED
; /* no valid sense data */
245 if (scmd
->cmnd
[0] == TEST_UNIT_READY
&& scmd
->scsi_done
!= scsi_eh_done
)
247 * nasty: for mid-layer issued TURs, we need to return the
248 * actual sense data without any recovery attempt. For eh
249 * issued ones, we need to try to recover and interpret
253 if (scsi_sense_is_deferred(&sshdr
))
256 if (sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
&&
257 sdev
->scsi_dh_data
->scsi_dh
->check_sense
) {
260 rc
= sdev
->scsi_dh_data
->scsi_dh
->check_sense(sdev
, &sshdr
);
261 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
263 /* handler does not care. Drop down to default handling */
267 * Previous logic looked for FILEMARK, EOM or ILI which are
268 * mainly associated with tapes and returned SUCCESS.
270 if (sshdr
.response_code
== 0x70) {
272 if (scmd
->sense_buffer
[2] & 0xe0)
276 * descriptor format: look for "stream commands sense data
277 * descriptor" (see SSC-3). Assume single sense data
278 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
280 if ((sshdr
.additional_length
> 3) &&
281 (scmd
->sense_buffer
[8] == 0x4) &&
282 (scmd
->sense_buffer
[11] & 0xe0))
286 switch (sshdr
.sense_key
) {
289 case RECOVERED_ERROR
:
290 return /* soft_error */ SUCCESS
;
292 case ABORTED_COMMAND
:
293 if (sshdr
.asc
== 0x10) /* DIF */
300 * if we are expecting a cc/ua because of a bus reset that we
301 * performed, treat this just as a retry. otherwise this is
302 * information that we should pass up to the upper-level driver
303 * so that we can deal with it there.
305 if (scmd
->device
->expecting_cc_ua
) {
307 * Because some device does not queue unit
308 * attentions correctly, we carefully check
309 * additional sense code and qualifier so as
310 * not to squash media change unit attention.
312 if (sshdr
.asc
!= 0x28 || sshdr
.ascq
!= 0x00) {
313 scmd
->device
->expecting_cc_ua
= 0;
318 * if the device is in the process of becoming ready, we
321 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
324 * if the device is not started, we need to wake
325 * the error handler to start the motor
327 if (scmd
->device
->allow_restart
&&
328 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
331 if (sshdr
.asc
== 0x3f && sshdr
.ascq
== 0x0e)
332 scmd_printk(KERN_WARNING
, scmd
,
333 "Warning! Received an indication that the "
334 "LUN assignments on this target have "
335 "changed. The Linux SCSI layer does not "
336 "automatically remap LUN assignments.\n");
337 else if (sshdr
.asc
== 0x3f)
338 scmd_printk(KERN_WARNING
, scmd
,
339 "Warning! Received an indication that the "
340 "operating parameters on this target have "
341 "changed. The Linux SCSI layer does not "
342 "automatically adjust these parameters.\n");
344 if (sshdr
.asc
== 0x38 && sshdr
.ascq
== 0x07)
345 scmd_printk(KERN_WARNING
, scmd
,
346 "Warning! Received an indication that the "
347 "LUN reached a thin provisioning soft "
351 * Pass the UA upwards for a determination in the completion
356 /* these are not supported */
358 case VOLUME_OVERFLOW
:
365 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
366 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
367 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
373 if (scmd
->device
->retry_hwerror
)
374 return ADD_TO_MLQUEUE
;
378 case ILLEGAL_REQUEST
:
379 if (sshdr
.asc
== 0x20 || /* Invalid command operation code */
380 sshdr
.asc
== 0x21 || /* Logical block address out of range */
381 sshdr
.asc
== 0x24 || /* Invalid field in cdb */
382 sshdr
.asc
== 0x26) { /* Parameter value invalid */
392 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
394 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
395 struct scsi_device
*tmp_sdev
;
397 if (!sht
->change_queue_depth
||
398 sdev
->queue_depth
>= sdev
->max_queue_depth
)
401 if (time_before(jiffies
,
402 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
405 if (time_before(jiffies
,
406 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
410 * Walk all devices of a target and do
413 shost_for_each_device(tmp_sdev
, sdev
->host
) {
414 if (tmp_sdev
->channel
!= sdev
->channel
||
415 tmp_sdev
->id
!= sdev
->id
||
416 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
419 * call back into LLD to increase queue_depth by one
420 * with ramp up reason code.
422 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1,
423 SCSI_QDEPTH_RAMP_UP
);
424 sdev
->last_queue_ramp_up
= jiffies
;
428 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
430 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
431 struct scsi_device
*tmp_sdev
;
433 if (!sht
->change_queue_depth
)
436 shost_for_each_device(tmp_sdev
, sdev
->host
) {
437 if (tmp_sdev
->channel
!= sdev
->channel
||
438 tmp_sdev
->id
!= sdev
->id
)
441 * We do not know the number of commands that were at
442 * the device when we got the queue full so we start
443 * from the highest possible value and work our way down.
445 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
- 1,
451 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
452 * @scmd: SCSI cmd to examine.
455 * This is *only* called when we are examining the status of commands
456 * queued during error recovery. the main difference here is that we
457 * don't allow for the possibility of retries here, and we are a lot
458 * more restrictive about what we consider acceptable.
460 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
463 * first check the host byte, to see if there is anything in there
464 * that would indicate what we need to do.
466 if (host_byte(scmd
->result
) == DID_RESET
) {
468 * rats. we are already in the error handler, so we now
469 * get to try and figure out what to do next. if the sense
470 * is valid, we have a pretty good idea of what to do.
471 * if not, we mark it as FAILED.
473 return scsi_check_sense(scmd
);
475 if (host_byte(scmd
->result
) != DID_OK
)
479 * next, check the message byte.
481 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
485 * now, check the status byte to see if this indicates
488 switch (status_byte(scmd
->result
)) {
490 scsi_handle_queue_ramp_up(scmd
->device
);
491 case COMMAND_TERMINATED
:
493 case CHECK_CONDITION
:
494 return scsi_check_sense(scmd
);
496 case INTERMEDIATE_GOOD
:
497 case INTERMEDIATE_C_GOOD
:
499 * who knows? FIXME(eric)
502 case RESERVATION_CONFLICT
:
503 if (scmd
->cmnd
[0] == TEST_UNIT_READY
)
504 /* it is a success, we probed the device and
507 /* otherwise, we failed to send the command */
510 scsi_handle_queue_full(scmd
->device
);
521 * scsi_eh_done - Completion function for error handling.
522 * @scmd: Cmd that is done.
524 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
526 struct completion
*eh_action
;
528 SCSI_LOG_ERROR_RECOVERY(3,
529 printk("%s scmd: %p result: %x\n",
530 __func__
, scmd
, scmd
->result
));
532 eh_action
= scmd
->device
->host
->eh_action
;
538 * scsi_try_host_reset - ask host adapter to reset itself
539 * @scmd: SCSI cmd to send host reset.
541 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
545 struct Scsi_Host
*host
= scmd
->device
->host
;
546 struct scsi_host_template
*hostt
= host
->hostt
;
548 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
551 if (!hostt
->eh_host_reset_handler
)
554 rtn
= hostt
->eh_host_reset_handler(scmd
);
556 if (rtn
== SUCCESS
) {
557 if (!hostt
->skip_settle_delay
)
558 ssleep(HOST_RESET_SETTLE_TIME
);
559 spin_lock_irqsave(host
->host_lock
, flags
);
560 scsi_report_bus_reset(host
, scmd_channel(scmd
));
561 spin_unlock_irqrestore(host
->host_lock
, flags
);
568 * scsi_try_bus_reset - ask host to perform a bus reset
569 * @scmd: SCSI cmd to send bus reset.
571 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
575 struct Scsi_Host
*host
= scmd
->device
->host
;
576 struct scsi_host_template
*hostt
= host
->hostt
;
578 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
581 if (!hostt
->eh_bus_reset_handler
)
584 rtn
= hostt
->eh_bus_reset_handler(scmd
);
586 if (rtn
== SUCCESS
) {
587 if (!hostt
->skip_settle_delay
)
588 ssleep(BUS_RESET_SETTLE_TIME
);
589 spin_lock_irqsave(host
->host_lock
, flags
);
590 scsi_report_bus_reset(host
, scmd_channel(scmd
));
591 spin_unlock_irqrestore(host
->host_lock
, flags
);
597 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
600 sdev
->expecting_cc_ua
= 1;
604 * scsi_try_target_reset - Ask host to perform a target reset
605 * @scmd: SCSI cmd used to send a target reset
608 * There is no timeout for this operation. if this operation is
609 * unreliable for a given host, then the host itself needs to put a
610 * timer on it, and set the host back to a consistent state prior to
613 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
617 struct Scsi_Host
*host
= scmd
->device
->host
;
618 struct scsi_host_template
*hostt
= host
->hostt
;
620 if (!hostt
->eh_target_reset_handler
)
623 rtn
= hostt
->eh_target_reset_handler(scmd
);
624 if (rtn
== SUCCESS
) {
625 spin_lock_irqsave(host
->host_lock
, flags
);
626 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
627 __scsi_report_device_reset
);
628 spin_unlock_irqrestore(host
->host_lock
, flags
);
635 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
636 * @scmd: SCSI cmd used to send BDR
639 * There is no timeout for this operation. if this operation is
640 * unreliable for a given host, then the host itself needs to put a
641 * timer on it, and set the host back to a consistent state prior to
644 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
647 struct scsi_host_template
*hostt
= scmd
->device
->host
->hostt
;
649 if (!hostt
->eh_device_reset_handler
)
652 rtn
= hostt
->eh_device_reset_handler(scmd
);
654 __scsi_report_device_reset(scmd
->device
, NULL
);
658 static int scsi_try_to_abort_cmd(struct scsi_host_template
*hostt
, struct scsi_cmnd
*scmd
)
660 if (!hostt
->eh_abort_handler
)
663 return hostt
->eh_abort_handler(scmd
);
666 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
668 if (scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
) != SUCCESS
)
669 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
670 if (scsi_try_target_reset(scmd
) != SUCCESS
)
671 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
672 scsi_try_host_reset(scmd
);
676 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
677 * @scmd: SCSI command structure to hijack
678 * @ses: structure to save restore information
679 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
680 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
681 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
683 * This function is used to save a scsi command information before re-execution
684 * as part of the error recovery process. If @sense_bytes is 0 the command
685 * sent must be one that does not transfer any data. If @sense_bytes != 0
686 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
687 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
689 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
690 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
692 struct scsi_device
*sdev
= scmd
->device
;
695 * We need saved copies of a number of fields - this is because
696 * error handling may need to overwrite these with different values
697 * to run different commands, and once error handling is complete,
698 * we will need to restore these values prior to running the actual
701 ses
->cmd_len
= scmd
->cmd_len
;
702 ses
->cmnd
= scmd
->cmnd
;
703 ses
->data_direction
= scmd
->sc_data_direction
;
704 ses
->sdb
= scmd
->sdb
;
705 ses
->next_rq
= scmd
->request
->next_rq
;
706 ses
->result
= scmd
->result
;
707 ses
->underflow
= scmd
->underflow
;
708 ses
->prot_op
= scmd
->prot_op
;
710 scmd
->prot_op
= SCSI_PROT_NORMAL
;
711 scmd
->cmnd
= ses
->eh_cmnd
;
712 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
713 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
714 scmd
->request
->next_rq
= NULL
;
717 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
719 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
721 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
722 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
723 scmd
->sdb
.table
.nents
= 1;
724 scmd
->cmnd
[0] = REQUEST_SENSE
;
725 scmd
->cmnd
[4] = scmd
->sdb
.length
;
726 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
728 scmd
->sc_data_direction
= DMA_NONE
;
730 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
731 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
732 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
738 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
739 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
740 (sdev
->lun
<< 5 & 0xe0);
743 * Zero the sense buffer. The scsi spec mandates that any
744 * untransferred sense data should be interpreted as being zero.
746 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
748 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
751 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
752 * @scmd: SCSI command structure to restore
753 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
755 * Undo any damage done by above scsi_eh_prep_cmnd().
757 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
760 * Restore original data
762 scmd
->cmd_len
= ses
->cmd_len
;
763 scmd
->cmnd
= ses
->cmnd
;
764 scmd
->sc_data_direction
= ses
->data_direction
;
765 scmd
->sdb
= ses
->sdb
;
766 scmd
->request
->next_rq
= ses
->next_rq
;
767 scmd
->result
= ses
->result
;
768 scmd
->underflow
= ses
->underflow
;
769 scmd
->prot_op
= ses
->prot_op
;
771 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
774 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
775 * @scmd: SCSI command structure to hijack
777 * @cmnd_size: size in bytes of @cmnd
778 * @timeout: timeout for this request
779 * @sense_bytes: size of sense data to copy or 0
781 * This function is used to send a scsi command down to a target device
782 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
785 * SUCCESS or FAILED or NEEDS_RETRY
787 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
788 int cmnd_size
, int timeout
, unsigned sense_bytes
)
790 struct scsi_device
*sdev
= scmd
->device
;
791 struct Scsi_Host
*shost
= sdev
->host
;
792 DECLARE_COMPLETION_ONSTACK(done
);
793 unsigned long timeleft
= timeout
;
794 struct scsi_eh_save ses
;
795 const unsigned long stall_for
= msecs_to_jiffies(100);
799 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
800 shost
->eh_action
= &done
;
803 scmd
->scsi_done
= scsi_eh_done
;
804 rtn
= shost
->hostt
->queuecommand(shost
, scmd
);
806 if (timeleft
> stall_for
) {
807 scsi_eh_restore_cmnd(scmd
, &ses
);
808 timeleft
-= stall_for
;
809 msleep(jiffies_to_msecs(stall_for
));
812 /* signal not to enter either branch of the if () below */
816 timeleft
= wait_for_completion_timeout(&done
, timeout
);
819 shost
->eh_action
= NULL
;
821 scsi_log_completion(scmd
, rtn
);
823 SCSI_LOG_ERROR_RECOVERY(3,
824 printk("%s: scmd: %p, timeleft: %ld\n",
825 __func__
, scmd
, timeleft
));
828 * If there is time left scsi_eh_done got called, and we will examine
829 * the actual status codes to see whether the command actually did
830 * complete normally, else if we have a zero return and no time left,
831 * the command must still be pending, so abort it and return FAILED.
832 * If we never actually managed to issue the command, because
833 * ->queuecommand() kept returning non zero, use the rtn = FAILED
834 * value above (so don't execute either branch of the if)
837 rtn
= scsi_eh_completed_normally(scmd
);
838 SCSI_LOG_ERROR_RECOVERY(3,
839 printk("%s: scsi_eh_completed_normally %x\n",
856 scsi_abort_eh_cmnd(scmd
);
860 scsi_eh_restore_cmnd(scmd
, &ses
);
862 if (scmd
->request
->cmd_type
!= REQ_TYPE_BLOCK_PC
) {
863 struct scsi_driver
*sdrv
= scsi_cmd_to_driver(scmd
);
865 rtn
= sdrv
->eh_action(scmd
, cmnd
, cmnd_size
, rtn
);
872 * scsi_request_sense - Request sense data from a particular target.
873 * @scmd: SCSI cmd for request sense.
876 * Some hosts automatically obtain this information, others require
877 * that we obtain it on our own. This function will *not* return until
878 * the command either times out, or it completes.
880 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
882 return scsi_send_eh_cmnd(scmd
, NULL
, 0, scmd
->device
->eh_timeout
, ~0);
886 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
887 * @scmd: Original SCSI cmd that eh has finished.
888 * @done_q: Queue for processed commands.
891 * We don't want to use the normal command completion while we are are
892 * still handling errors - it may cause other commands to be queued,
893 * and that would disturb what we are doing. Thus we really want to
894 * keep a list of pending commands for final completion, and once we
895 * are ready to leave error handling we handle completion for real.
897 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
899 scmd
->device
->host
->host_failed
--;
901 list_move_tail(&scmd
->eh_entry
, done_q
);
903 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
906 * scsi_eh_get_sense - Get device sense data.
907 * @work_q: Queue of commands to process.
908 * @done_q: Queue of processed commands.
911 * See if we need to request sense information. if so, then get it
912 * now, so we have a better idea of what to do.
915 * This has the unfortunate side effect that if a shost adapter does
916 * not automatically request sense information, we end up shutting
917 * it down before we request it.
919 * All drivers should request sense information internally these days,
920 * so for now all I have to say is tough noogies if you end up in here.
922 * XXX: Long term this code should go away, but that needs an audit of
925 int scsi_eh_get_sense(struct list_head
*work_q
,
926 struct list_head
*done_q
)
928 struct scsi_cmnd
*scmd
, *next
;
931 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
932 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
933 SCSI_SENSE_VALID(scmd
))
936 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
937 "%s: requesting sense\n",
939 rtn
= scsi_request_sense(scmd
);
943 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
944 " result %x\n", scmd
,
946 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
948 rtn
= scsi_decide_disposition(scmd
);
951 * if the result was normal, then just pass it along to the
955 /* we don't want this command reissued, just
956 * finished with the sense data, so set
957 * retries to the max allowed to ensure it
958 * won't get reissued */
959 scmd
->retries
= scmd
->allowed
;
960 else if (rtn
!= NEEDS_RETRY
)
963 scsi_eh_finish_cmd(scmd
, done_q
);
966 return list_empty(work_q
);
968 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
971 * scsi_eh_tur - Send TUR to device.
972 * @scmd: &scsi_cmnd to send TUR
975 * 0 - Device is ready. 1 - Device NOT ready.
977 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
979 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
980 int retry_cnt
= 1, rtn
;
983 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6,
984 scmd
->device
->eh_timeout
, 0);
986 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
987 __func__
, scmd
, rtn
));
1002 * scsi_eh_test_devices - check if devices are responding from error recovery.
1003 * @cmd_list: scsi commands in error recovery.
1004 * @work_q: queue for commands which still need more error recovery
1005 * @done_q: queue for commands which are finished
1006 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1009 * Tests if devices are in a working state. Commands to devices now in
1010 * a working state are sent to the done_q while commands to devices which
1011 * are still failing to respond are returned to the work_q for more
1014 static int scsi_eh_test_devices(struct list_head
*cmd_list
,
1015 struct list_head
*work_q
,
1016 struct list_head
*done_q
, int try_stu
)
1018 struct scsi_cmnd
*scmd
, *next
;
1019 struct scsi_device
*sdev
;
1022 while (!list_empty(cmd_list
)) {
1023 scmd
= list_entry(cmd_list
->next
, struct scsi_cmnd
, eh_entry
);
1024 sdev
= scmd
->device
;
1026 finish_cmds
= !scsi_device_online(scmd
->device
) ||
1027 (try_stu
&& !scsi_eh_try_stu(scmd
) &&
1028 !scsi_eh_tur(scmd
)) ||
1031 list_for_each_entry_safe(scmd
, next
, cmd_list
, eh_entry
)
1032 if (scmd
->device
== sdev
) {
1034 scsi_eh_finish_cmd(scmd
, done_q
);
1036 list_move_tail(&scmd
->eh_entry
, work_q
);
1039 return list_empty(work_q
);
1044 * scsi_eh_abort_cmds - abort pending commands.
1045 * @work_q: &list_head for pending commands.
1046 * @done_q: &list_head for processed commands.
1049 * Try and see whether or not it makes sense to try and abort the
1050 * running command. This only works out to be the case if we have one
1051 * command that has timed out. If the command simply failed, it makes
1052 * no sense to try and abort the command, since as far as the shost
1053 * adapter is concerned, it isn't running.
1055 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
1056 struct list_head
*done_q
)
1058 struct scsi_cmnd
*scmd
, *next
;
1059 LIST_HEAD(check_list
);
1062 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1063 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
1065 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1066 "0x%p\n", current
->comm
,
1068 rtn
= scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
);
1069 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1070 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
1071 if (rtn
== FAST_IO_FAIL
)
1072 scsi_eh_finish_cmd(scmd
, done_q
);
1074 list_move_tail(&scmd
->eh_entry
, &check_list
);
1076 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1083 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1087 * scsi_eh_try_stu - Send START_UNIT to device.
1088 * @scmd: &scsi_cmnd to send START_UNIT
1091 * 0 - Device is ready. 1 - Device NOT ready.
1093 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1095 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1097 if (scmd
->device
->allow_restart
) {
1098 int i
, rtn
= NEEDS_RETRY
;
1100 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1101 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1111 * scsi_eh_stu - send START_UNIT if needed
1112 * @shost: &scsi host being recovered.
1113 * @work_q: &list_head for pending commands.
1114 * @done_q: &list_head for processed commands.
1117 * If commands are failing due to not ready, initializing command required,
1118 * try revalidating the device, which will end up sending a start unit.
1120 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1121 struct list_head
*work_q
,
1122 struct list_head
*done_q
)
1124 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1125 struct scsi_device
*sdev
;
1127 shost_for_each_device(sdev
, shost
) {
1129 list_for_each_entry(scmd
, work_q
, eh_entry
)
1130 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1131 scsi_check_sense(scmd
) == FAILED
) {
1139 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1140 " 0x%p\n", current
->comm
, sdev
));
1142 if (!scsi_eh_try_stu(stu_scmd
)) {
1143 if (!scsi_device_online(sdev
) ||
1144 !scsi_eh_tur(stu_scmd
)) {
1145 list_for_each_entry_safe(scmd
, next
,
1147 if (scmd
->device
== sdev
)
1148 scsi_eh_finish_cmd(scmd
, done_q
);
1152 SCSI_LOG_ERROR_RECOVERY(3,
1153 printk("%s: START_UNIT failed to sdev:"
1154 " 0x%p\n", current
->comm
, sdev
));
1158 return list_empty(work_q
);
1163 * scsi_eh_bus_device_reset - send bdr if needed
1164 * @shost: scsi host being recovered.
1165 * @work_q: &list_head for pending commands.
1166 * @done_q: &list_head for processed commands.
1169 * Try a bus device reset. Still, look to see whether we have multiple
1170 * devices that are jammed or not - if we have multiple devices, it
1171 * makes no sense to try bus_device_reset - we really would need to try
1172 * a bus_reset instead.
1174 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1175 struct list_head
*work_q
,
1176 struct list_head
*done_q
)
1178 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1179 struct scsi_device
*sdev
;
1182 shost_for_each_device(sdev
, shost
) {
1184 list_for_each_entry(scmd
, work_q
, eh_entry
)
1185 if (scmd
->device
== sdev
) {
1193 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1194 " 0x%p\n", current
->comm
,
1196 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1197 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1198 if (!scsi_device_online(sdev
) ||
1199 rtn
== FAST_IO_FAIL
||
1200 !scsi_eh_tur(bdr_scmd
)) {
1201 list_for_each_entry_safe(scmd
, next
,
1203 if (scmd
->device
== sdev
)
1204 scsi_eh_finish_cmd(scmd
,
1209 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1217 return list_empty(work_q
);
1221 * scsi_eh_target_reset - send target reset if needed
1222 * @shost: scsi host being recovered.
1223 * @work_q: &list_head for pending commands.
1224 * @done_q: &list_head for processed commands.
1227 * Try a target reset.
1229 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1230 struct list_head
*work_q
,
1231 struct list_head
*done_q
)
1233 LIST_HEAD(tmp_list
);
1234 LIST_HEAD(check_list
);
1236 list_splice_init(work_q
, &tmp_list
);
1238 while (!list_empty(&tmp_list
)) {
1239 struct scsi_cmnd
*next
, *scmd
;
1243 scmd
= list_entry(tmp_list
.next
, struct scsi_cmnd
, eh_entry
);
1246 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1248 current
->comm
, id
));
1249 rtn
= scsi_try_target_reset(scmd
);
1250 if (rtn
!= SUCCESS
&& rtn
!= FAST_IO_FAIL
)
1251 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1254 current
->comm
, id
));
1255 list_for_each_entry_safe(scmd
, next
, &tmp_list
, eh_entry
) {
1256 if (scmd_id(scmd
) != id
)
1260 list_move_tail(&scmd
->eh_entry
, &check_list
);
1261 else if (rtn
== FAST_IO_FAIL
)
1262 scsi_eh_finish_cmd(scmd
, done_q
);
1264 /* push back on work queue for further processing */
1265 list_move(&scmd
->eh_entry
, work_q
);
1269 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1273 * scsi_eh_bus_reset - send a bus reset
1274 * @shost: &scsi host being recovered.
1275 * @work_q: &list_head for pending commands.
1276 * @done_q: &list_head for processed commands.
1278 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1279 struct list_head
*work_q
,
1280 struct list_head
*done_q
)
1282 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1283 LIST_HEAD(check_list
);
1284 unsigned int channel
;
1288 * we really want to loop over the various channels, and do this on
1289 * a channel by channel basis. we should also check to see if any
1290 * of the failed commands are on soft_reset devices, and if so, skip
1294 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1296 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1297 if (channel
== scmd_channel(scmd
)) {
1301 * FIXME add back in some support for
1302 * soft_reset devices.
1309 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1310 " %d\n", current
->comm
,
1312 rtn
= scsi_try_bus_reset(chan_scmd
);
1313 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1314 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1315 if (channel
== scmd_channel(scmd
)) {
1316 if (rtn
== FAST_IO_FAIL
)
1317 scsi_eh_finish_cmd(scmd
,
1320 list_move_tail(&scmd
->eh_entry
,
1325 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1326 " failed chan: %d\n",
1331 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1335 * scsi_eh_host_reset - send a host reset
1336 * @work_q: list_head for processed commands.
1337 * @done_q: list_head for processed commands.
1339 static int scsi_eh_host_reset(struct list_head
*work_q
,
1340 struct list_head
*done_q
)
1342 struct scsi_cmnd
*scmd
, *next
;
1343 LIST_HEAD(check_list
);
1346 if (!list_empty(work_q
)) {
1347 scmd
= list_entry(work_q
->next
,
1348 struct scsi_cmnd
, eh_entry
);
1350 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1353 rtn
= scsi_try_host_reset(scmd
);
1354 if (rtn
== SUCCESS
) {
1355 list_splice_init(work_q
, &check_list
);
1356 } else if (rtn
== FAST_IO_FAIL
) {
1357 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1358 scsi_eh_finish_cmd(scmd
, done_q
);
1361 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1366 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 1);
1370 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1371 * @work_q: list_head for processed commands.
1372 * @done_q: list_head for processed commands.
1374 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1375 struct list_head
*done_q
)
1377 struct scsi_cmnd
*scmd
, *next
;
1379 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1380 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1381 "not ready after error recovery\n");
1382 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1383 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1385 * FIXME: Handle lost cmds.
1388 scsi_eh_finish_cmd(scmd
, done_q
);
1394 * scsi_noretry_cmd - determinte if command should be failed fast
1395 * @scmd: SCSI cmd to examine.
1397 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1399 switch (host_byte(scmd
->result
)) {
1403 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1405 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
);
1407 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1408 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1411 case DID_SOFT_ERROR
:
1412 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1415 switch (status_byte(scmd
->result
)) {
1416 case CHECK_CONDITION
:
1418 * assume caller has checked sense and determinted
1419 * the check condition was retryable.
1421 if (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
||
1422 scmd
->request
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1430 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1431 * @scmd: SCSI cmd to examine.
1434 * This is *only* called when we are examining the status after sending
1435 * out the actual data command. any commands that are queued for error
1436 * recovery (e.g. test_unit_ready) do *not* come through here.
1438 * When this routine returns failed, it means the error handler thread
1439 * is woken. In cases where the error code indicates an error that
1440 * doesn't require the error handler read (i.e. we don't need to
1441 * abort/reset), this function should return SUCCESS.
1443 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1448 * if the device is offline, then we clearly just pass the result back
1449 * up to the top level.
1451 if (!scsi_device_online(scmd
->device
)) {
1452 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1459 * first check the host byte, to see if there is anything in there
1460 * that would indicate what we need to do.
1462 switch (host_byte(scmd
->result
)) {
1463 case DID_PASSTHROUGH
:
1465 * no matter what, pass this through to the upper layer.
1466 * nuke this special code so that it looks like we are saying
1469 scmd
->result
&= 0xff00ffff;
1473 * looks good. drop through, and check the next byte.
1476 case DID_NO_CONNECT
:
1477 case DID_BAD_TARGET
:
1480 * note - this means that we just report the status back
1481 * to the top level driver, not that we actually think
1482 * that it indicates SUCCESS.
1486 * when the low level driver returns did_soft_error,
1487 * it is responsible for keeping an internal retry counter
1488 * in order to avoid endless loops (db)
1490 * actually this is a bug in this function here. we should
1491 * be mindful of the maximum number of retries specified
1492 * and not get stuck in a loop.
1494 case DID_SOFT_ERROR
:
1500 return ADD_TO_MLQUEUE
;
1501 case DID_TRANSPORT_DISRUPTED
:
1503 * LLD/transport was disrupted during processing of the IO.
1504 * The transport class is now blocked/blocking,
1505 * and the transport will decide what to do with the IO
1506 * based on its timers and recovery capablilities if
1507 * there are enough retries.
1510 case DID_TRANSPORT_FAILFAST
:
1512 * The transport decided to failfast the IO (most likely
1513 * the fast io fail tmo fired), so send IO directly upwards.
1517 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1518 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1520 * execute reservation conflict processing code
1530 * when we scan the bus, we get timeout messages for
1531 * these commands if there is no device available.
1532 * other hosts report did_no_connect for the same thing.
1534 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1535 scmd
->cmnd
[0] == INQUIRY
)) {
1547 * next, check the message byte.
1549 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1553 * check the status byte to see if this indicates anything special.
1555 switch (status_byte(scmd
->result
)) {
1557 scsi_handle_queue_full(scmd
->device
);
1559 * the case of trying to send too many commands to a
1560 * tagged queueing device.
1564 * device can't talk to us at the moment. Should only
1565 * occur (SAM-3) when the task queue is empty, so will cause
1566 * the empty queue handling to trigger a stall in the
1569 return ADD_TO_MLQUEUE
;
1571 scsi_handle_queue_ramp_up(scmd
->device
);
1572 case COMMAND_TERMINATED
:
1576 case CHECK_CONDITION
:
1577 rtn
= scsi_check_sense(scmd
);
1578 if (rtn
== NEEDS_RETRY
)
1580 else if (rtn
== TARGET_ERROR
) {
1582 * Need to modify host byte to signal a
1583 * permanent target failure
1585 set_host_byte(scmd
, DID_TARGET_FAILURE
);
1588 /* if rtn == FAILED, we have no sense information;
1589 * returning FAILED will wake the error handler thread
1590 * to collect the sense and redo the decide
1593 case CONDITION_GOOD
:
1594 case INTERMEDIATE_GOOD
:
1595 case INTERMEDIATE_C_GOOD
:
1598 * who knows? FIXME(eric)
1602 case RESERVATION_CONFLICT
:
1603 sdev_printk(KERN_INFO
, scmd
->device
,
1604 "reservation conflict\n");
1605 set_host_byte(scmd
, DID_NEXUS_FAILURE
);
1606 return SUCCESS
; /* causes immediate i/o error */
1614 /* we requeue for retry because the error was retryable, and
1615 * the request was not marked fast fail. Note that above,
1616 * even if the request is marked fast fail, we still requeue
1617 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1618 if ((++scmd
->retries
) <= scmd
->allowed
1619 && !scsi_noretry_cmd(scmd
)) {
1623 * no more retries - report this one back to upper level.
1629 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1631 __blk_put_request(req
->q
, req
);
1635 * scsi_eh_lock_door - Prevent medium removal for the specified device
1636 * @sdev: SCSI device to prevent medium removal
1639 * We must be called from process context.
1642 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1643 * head of the devices request queue, and continue.
1645 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1647 struct request
*req
;
1650 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1651 * request becomes available
1653 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1655 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1659 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1662 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1664 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1665 req
->cmd_flags
|= REQ_QUIET
;
1666 req
->timeout
= 10 * HZ
;
1669 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1673 * scsi_restart_operations - restart io operations to the specified host.
1674 * @shost: Host we are restarting.
1677 * When we entered the error handler, we blocked all further i/o to
1678 * this device. we need to 'reverse' this process.
1680 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1682 struct scsi_device
*sdev
;
1683 unsigned long flags
;
1686 * If the door was locked, we need to insert a door lock request
1687 * onto the head of the SCSI request queue for the device. There
1688 * is no point trying to lock the door of an off-line device.
1690 shost_for_each_device(sdev
, shost
) {
1691 if (scsi_device_online(sdev
) && sdev
->locked
)
1692 scsi_eh_lock_door(sdev
);
1696 * next free up anything directly waiting upon the host. this
1697 * will be requests for character device operations, and also for
1698 * ioctls to queued block devices.
1700 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1703 spin_lock_irqsave(shost
->host_lock
, flags
);
1704 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1705 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1706 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1707 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1709 wake_up(&shost
->host_wait
);
1712 * finally we need to re-initiate requests that may be pending. we will
1713 * have had everything blocked while error handling is taking place, and
1714 * now that error recovery is done, we will need to ensure that these
1715 * requests are started.
1717 scsi_run_host_queues(shost
);
1720 * if eh is active and host_eh_scheduled is pending we need to re-run
1721 * recovery. we do this check after scsi_run_host_queues() to allow
1722 * everything pent up since the last eh run a chance to make forward
1723 * progress before we sync again. Either we'll immediately re-run
1724 * recovery or scsi_device_unbusy() will wake us again when these
1725 * pending commands complete.
1727 spin_lock_irqsave(shost
->host_lock
, flags
);
1728 if (shost
->host_eh_scheduled
)
1729 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
1730 WARN_ON(scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
));
1731 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1735 * scsi_eh_ready_devs - check device ready state and recover if not.
1736 * @shost: host to be recovered.
1737 * @work_q: &list_head for pending commands.
1738 * @done_q: &list_head for processed commands.
1740 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1741 struct list_head
*work_q
,
1742 struct list_head
*done_q
)
1744 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1745 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1746 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1747 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1748 if (!scsi_eh_host_reset(work_q
, done_q
))
1749 scsi_eh_offline_sdevs(work_q
,
1752 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1755 * scsi_eh_flush_done_q - finish processed commands or retry them.
1756 * @done_q: list_head of processed commands.
1758 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1760 struct scsi_cmnd
*scmd
, *next
;
1762 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1763 list_del_init(&scmd
->eh_entry
);
1764 if (scsi_device_online(scmd
->device
) &&
1765 !scsi_noretry_cmd(scmd
) &&
1766 (++scmd
->retries
<= scmd
->allowed
)) {
1767 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1771 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1774 * If just we got sense for the device (called
1775 * scsi_eh_get_sense), scmd->result is already
1776 * set, do not set DRIVER_TIMEOUT.
1779 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1780 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1782 current
->comm
, scmd
));
1783 scsi_finish_command(scmd
);
1787 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1790 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1791 * @shost: Host to unjam.
1794 * When we come in here, we *know* that all commands on the bus have
1795 * either completed, failed or timed out. we also know that no further
1796 * commands are being sent to the host, so things are relatively quiet
1797 * and we have freedom to fiddle with things as we wish.
1799 * This is only the *default* implementation. it is possible for
1800 * individual drivers to supply their own version of this function, and
1801 * if the maintainer wishes to do this, it is strongly suggested that
1802 * this function be taken as a template and modified. this function
1803 * was designed to correctly handle problems for about 95% of the
1804 * different cases out there, and it should always provide at least a
1805 * reasonable amount of error recovery.
1807 * Any command marked 'failed' or 'timeout' must eventually have
1808 * scsi_finish_cmd() called for it. we do all of the retry stuff
1809 * here, so when we restart the host after we return it should have an
1812 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1814 unsigned long flags
;
1815 LIST_HEAD(eh_work_q
);
1816 LIST_HEAD(eh_done_q
);
1818 spin_lock_irqsave(shost
->host_lock
, flags
);
1819 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1820 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1822 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1824 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1825 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1826 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1828 scsi_eh_flush_done_q(&eh_done_q
);
1832 * scsi_error_handler - SCSI error handler thread
1833 * @data: Host for which we are running.
1836 * This is the main error handling loop. This is run as a kernel thread
1837 * for every SCSI host and handles all error handling activity.
1839 int scsi_error_handler(void *data
)
1841 struct Scsi_Host
*shost
= data
;
1844 * We use TASK_INTERRUPTIBLE so that the thread is not
1845 * counted against the load average as a running process.
1846 * We never actually get interrupted because kthread_run
1847 * disables signal delivery for the created thread.
1849 while (!kthread_should_stop()) {
1850 set_current_state(TASK_INTERRUPTIBLE
);
1851 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1852 shost
->host_failed
!= shost
->host_busy
) {
1853 SCSI_LOG_ERROR_RECOVERY(1,
1854 printk("Error handler scsi_eh_%d sleeping\n",
1860 __set_current_state(TASK_RUNNING
);
1861 SCSI_LOG_ERROR_RECOVERY(1,
1862 printk("Error handler scsi_eh_%d waking up\n",
1866 * We have a host that is failing for some reason. Figure out
1867 * what we need to do to get it up and online again (if we can).
1868 * If we fail, we end up taking the thing offline.
1870 if (!shost
->eh_noresume
&& scsi_autopm_get_host(shost
) != 0) {
1871 SCSI_LOG_ERROR_RECOVERY(1,
1872 printk(KERN_ERR
"Error handler scsi_eh_%d "
1873 "unable to autoresume\n",
1878 if (shost
->transportt
->eh_strategy_handler
)
1879 shost
->transportt
->eh_strategy_handler(shost
);
1881 scsi_unjam_host(shost
);
1884 * Note - if the above fails completely, the action is to take
1885 * individual devices offline and flush the queue of any
1886 * outstanding requests that may have been pending. When we
1887 * restart, we restart any I/O to any other devices on the bus
1888 * which are still online.
1890 scsi_restart_operations(shost
);
1891 if (!shost
->eh_noresume
)
1892 scsi_autopm_put_host(shost
);
1894 __set_current_state(TASK_RUNNING
);
1896 SCSI_LOG_ERROR_RECOVERY(1,
1897 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1898 shost
->ehandler
= NULL
;
1903 * Function: scsi_report_bus_reset()
1905 * Purpose: Utility function used by low-level drivers to report that
1906 * they have observed a bus reset on the bus being handled.
1908 * Arguments: shost - Host in question
1909 * channel - channel on which reset was observed.
1913 * Lock status: Host lock must be held.
1915 * Notes: This only needs to be called if the reset is one which
1916 * originates from an unknown location. Resets originated
1917 * by the mid-level itself don't need to call this, but there
1918 * should be no harm.
1920 * The main purpose of this is to make sure that a CHECK_CONDITION
1921 * is properly treated.
1923 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1925 struct scsi_device
*sdev
;
1927 __shost_for_each_device(sdev
, shost
) {
1928 if (channel
== sdev_channel(sdev
))
1929 __scsi_report_device_reset(sdev
, NULL
);
1932 EXPORT_SYMBOL(scsi_report_bus_reset
);
1935 * Function: scsi_report_device_reset()
1937 * Purpose: Utility function used by low-level drivers to report that
1938 * they have observed a device reset on the device being handled.
1940 * Arguments: shost - Host in question
1941 * channel - channel on which reset was observed
1942 * target - target on which reset was observed
1946 * Lock status: Host lock must be held
1948 * Notes: This only needs to be called if the reset is one which
1949 * originates from an unknown location. Resets originated
1950 * by the mid-level itself don't need to call this, but there
1951 * should be no harm.
1953 * The main purpose of this is to make sure that a CHECK_CONDITION
1954 * is properly treated.
1956 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1958 struct scsi_device
*sdev
;
1960 __shost_for_each_device(sdev
, shost
) {
1961 if (channel
== sdev_channel(sdev
) &&
1962 target
== sdev_id(sdev
))
1963 __scsi_report_device_reset(sdev
, NULL
);
1966 EXPORT_SYMBOL(scsi_report_device_reset
);
1969 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1974 * Function: scsi_reset_provider
1976 * Purpose: Send requested reset to a bus or device at any phase.
1978 * Arguments: device - device to send reset to
1979 * flag - reset type (see scsi.h)
1981 * Returns: SUCCESS/FAILURE.
1983 * Notes: This is used by the SCSI Generic driver to provide
1984 * Bus/Device reset capability.
1987 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1989 struct scsi_cmnd
*scmd
;
1990 struct Scsi_Host
*shost
= dev
->host
;
1992 unsigned long flags
;
1995 if (scsi_autopm_get_host(shost
) < 0)
1998 scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1999 blk_rq_init(NULL
, &req
);
2000 scmd
->request
= &req
;
2002 scmd
->cmnd
= req
.cmd
;
2004 scmd
->scsi_done
= scsi_reset_provider_done_command
;
2005 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
2009 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
2011 spin_lock_irqsave(shost
->host_lock
, flags
);
2012 shost
->tmf_in_progress
= 1;
2013 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2016 case SCSI_TRY_RESET_DEVICE
:
2017 rtn
= scsi_try_bus_device_reset(scmd
);
2021 case SCSI_TRY_RESET_TARGET
:
2022 rtn
= scsi_try_target_reset(scmd
);
2026 case SCSI_TRY_RESET_BUS
:
2027 rtn
= scsi_try_bus_reset(scmd
);
2031 case SCSI_TRY_RESET_HOST
:
2032 rtn
= scsi_try_host_reset(scmd
);
2038 spin_lock_irqsave(shost
->host_lock
, flags
);
2039 shost
->tmf_in_progress
= 0;
2040 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2043 * be sure to wake up anyone who was sleeping or had their queue
2044 * suspended while we performed the TMF.
2046 SCSI_LOG_ERROR_RECOVERY(3,
2047 printk("%s: waking up host to restart after TMF\n",
2050 wake_up(&shost
->host_wait
);
2052 scsi_run_host_queues(shost
);
2054 scsi_next_command(scmd
);
2055 scsi_autopm_put_host(shost
);
2058 EXPORT_SYMBOL(scsi_reset_provider
);
2061 * scsi_normalize_sense - normalize main elements from either fixed or
2062 * descriptor sense data format into a common format.
2064 * @sense_buffer: byte array containing sense data returned by device
2065 * @sb_len: number of valid bytes in sense_buffer
2066 * @sshdr: pointer to instance of structure that common
2067 * elements are written to.
2070 * The "main elements" from sense data are: response_code, sense_key,
2071 * asc, ascq and additional_length (only for descriptor format).
2073 * Typically this function can be called after a device has
2074 * responded to a SCSI command with the CHECK_CONDITION status.
2077 * 1 if valid sense data information found, else 0;
2079 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
2080 struct scsi_sense_hdr
*sshdr
)
2082 if (!sense_buffer
|| !sb_len
)
2085 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
2087 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
2089 if (!scsi_sense_valid(sshdr
))
2092 if (sshdr
->response_code
>= 0x72) {
2097 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
2099 sshdr
->asc
= sense_buffer
[2];
2101 sshdr
->ascq
= sense_buffer
[3];
2103 sshdr
->additional_length
= sense_buffer
[7];
2109 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
2111 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
2112 sb_len
: (sense_buffer
[7] + 8);
2114 sshdr
->asc
= sense_buffer
[12];
2116 sshdr
->ascq
= sense_buffer
[13];
2122 EXPORT_SYMBOL(scsi_normalize_sense
);
2124 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
2125 struct scsi_sense_hdr
*sshdr
)
2127 return scsi_normalize_sense(cmd
->sense_buffer
,
2128 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2130 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2133 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2134 * @sense_buffer: byte array of descriptor format sense data
2135 * @sb_len: number of valid bytes in sense_buffer
2136 * @desc_type: value of descriptor type to find
2137 * (e.g. 0 -> information)
2140 * only valid when sense data is in descriptor format
2143 * pointer to start of (first) descriptor if found else NULL
2145 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2148 int add_sen_len
, add_len
, desc_len
, k
;
2151 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2153 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2155 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2156 add_sen_len
: (sb_len
- 8);
2157 descp
= &sense_buffer
[8];
2158 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2160 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2161 desc_len
= add_len
+ 2;
2162 if (descp
[0] == desc_type
)
2164 if (add_len
< 0) // short descriptor ??
2169 EXPORT_SYMBOL(scsi_sense_desc_find
);
2172 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2173 * @sense_buffer: byte array of sense data
2174 * @sb_len: number of valid bytes in sense_buffer
2175 * @info_out: pointer to 64 integer where 8 or 4 byte information
2176 * field will be placed if found.
2179 * 1 if information field found, 0 if not found.
2181 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2190 switch (sense_buffer
[0] & 0x7f) {
2193 if (sense_buffer
[0] & 0x80) {
2194 *info_out
= (sense_buffer
[3] << 24) +
2195 (sense_buffer
[4] << 16) +
2196 (sense_buffer
[5] << 8) + sense_buffer
[6];
2202 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2204 if (ucp
&& (0xa == ucp
[1])) {
2206 for (j
= 0; j
< 8; ++j
) {
2219 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2222 * scsi_build_sense_buffer - build sense data in a buffer
2223 * @desc: Sense format (non zero == descriptor format,
2224 * 0 == fixed format)
2225 * @buf: Where to build sense data
2227 * @asc: Additional sense code
2228 * @ascq: Additional sense code qualifier
2231 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2234 buf
[0] = 0x72; /* descriptor, current */
2240 buf
[0] = 0x70; /* fixed, current */
2247 EXPORT_SYMBOL(scsi_build_sense_buffer
);