net/mlx5e: Make function mlx5e_change_rep_mtu() static
[linux-2.6/btrfs-unstable.git] / drivers / scsi / scsi_error.c
blob946039117bf4f2c655db2bb1129ee75df1f1df21
1 /*
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
13 * minor cleanups.
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_common.h>
37 #include <scsi/scsi_transport.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_ioctl.h>
40 #include <scsi/scsi_dh.h>
41 #include <scsi/sg.h>
43 #include "scsi_priv.h"
44 #include "scsi_logging.h"
45 #include "scsi_transport_api.h"
47 #include <trace/events/scsi.h>
49 #include <asm/unaligned.h>
51 static void scsi_eh_done(struct scsi_cmnd *scmd);
54 * These should *probably* be handled by the host itself.
55 * Since it is allowed to sleep, it probably should.
57 #define BUS_RESET_SETTLE_TIME (10)
58 #define HOST_RESET_SETTLE_TIME (10)
60 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
61 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
62 struct scsi_cmnd *);
64 void scsi_eh_wakeup(struct Scsi_Host *shost)
66 lockdep_assert_held(shost->host_lock);
68 if (atomic_read(&shost->host_busy) == shost->host_failed) {
69 trace_scsi_eh_wakeup(shost);
70 wake_up_process(shost->ehandler);
71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
72 "Waking error handler thread\n"));
76 /**
77 * scsi_schedule_eh - schedule EH for SCSI host
78 * @shost: SCSI host to invoke error handling on.
80 * Schedule SCSI EH without scmd.
82 void scsi_schedule_eh(struct Scsi_Host *shost)
84 unsigned long flags;
86 spin_lock_irqsave(shost->host_lock, flags);
88 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
89 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
90 shost->host_eh_scheduled++;
91 scsi_eh_wakeup(shost);
94 spin_unlock_irqrestore(shost->host_lock, flags);
96 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
98 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
100 if (!shost->last_reset || shost->eh_deadline == -1)
101 return 0;
104 * 32bit accesses are guaranteed to be atomic
105 * (on all supported architectures), so instead
106 * of using a spinlock we can as well double check
107 * if eh_deadline has been set to 'off' during the
108 * time_before call.
110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
111 shost->eh_deadline > -1)
112 return 0;
114 return 1;
118 * scmd_eh_abort_handler - Handle command aborts
119 * @work: command to be aborted.
121 * Note: this function must be called only for a command that has timed out.
122 * Because the block layer marks a request as complete before it calls
123 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
124 * timed out do not have any effect. Hence it is safe to call
125 * scsi_finish_command() from this function.
127 void
128 scmd_eh_abort_handler(struct work_struct *work)
130 struct scsi_cmnd *scmd =
131 container_of(work, struct scsi_cmnd, abort_work.work);
132 struct scsi_device *sdev = scmd->device;
133 int rtn;
135 if (scsi_host_eh_past_deadline(sdev->host)) {
136 SCSI_LOG_ERROR_RECOVERY(3,
137 scmd_printk(KERN_INFO, scmd,
138 "eh timeout, not aborting\n"));
139 } else {
140 SCSI_LOG_ERROR_RECOVERY(3,
141 scmd_printk(KERN_INFO, scmd,
142 "aborting command\n"));
143 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
144 if (rtn == SUCCESS) {
145 set_host_byte(scmd, DID_TIME_OUT);
146 if (scsi_host_eh_past_deadline(sdev->host)) {
147 SCSI_LOG_ERROR_RECOVERY(3,
148 scmd_printk(KERN_INFO, scmd,
149 "eh timeout, not retrying "
150 "aborted command\n"));
151 } else if (!scsi_noretry_cmd(scmd) &&
152 (++scmd->retries <= scmd->allowed)) {
153 SCSI_LOG_ERROR_RECOVERY(3,
154 scmd_printk(KERN_WARNING, scmd,
155 "retry aborted command\n"));
156 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
157 return;
158 } else {
159 SCSI_LOG_ERROR_RECOVERY(3,
160 scmd_printk(KERN_WARNING, scmd,
161 "finish aborted command\n"));
162 scsi_finish_command(scmd);
163 return;
165 } else {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_INFO, scmd,
168 "cmd abort %s\n",
169 (rtn == FAST_IO_FAIL) ?
170 "not send" : "failed"));
174 scsi_eh_scmd_add(scmd);
178 * scsi_abort_command - schedule a command abort
179 * @scmd: scmd to abort.
181 * We only need to abort commands after a command timeout
183 static int
184 scsi_abort_command(struct scsi_cmnd *scmd)
186 struct scsi_device *sdev = scmd->device;
187 struct Scsi_Host *shost = sdev->host;
188 unsigned long flags;
190 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
192 * Retry after abort failed, escalate to next level.
194 SCSI_LOG_ERROR_RECOVERY(3,
195 scmd_printk(KERN_INFO, scmd,
196 "previous abort failed\n"));
197 BUG_ON(delayed_work_pending(&scmd->abort_work));
198 return FAILED;
201 spin_lock_irqsave(shost->host_lock, flags);
202 if (shost->eh_deadline != -1 && !shost->last_reset)
203 shost->last_reset = jiffies;
204 spin_unlock_irqrestore(shost->host_lock, flags);
206 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
207 SCSI_LOG_ERROR_RECOVERY(3,
208 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
209 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
210 return SUCCESS;
214 * scsi_eh_reset - call into ->eh_action to reset internal counters
215 * @scmd: scmd to run eh on.
217 * The scsi driver might be carrying internal state about the
218 * devices, so we need to call into the driver to reset the
219 * internal state once the error handler is started.
221 static void scsi_eh_reset(struct scsi_cmnd *scmd)
223 if (!blk_rq_is_passthrough(scmd->request)) {
224 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
225 if (sdrv->eh_reset)
226 sdrv->eh_reset(scmd);
230 static void scsi_eh_inc_host_failed(struct rcu_head *head)
232 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
233 struct Scsi_Host *shost = scmd->device->host;
234 unsigned long flags;
236 spin_lock_irqsave(shost->host_lock, flags);
237 shost->host_failed++;
238 scsi_eh_wakeup(shost);
239 spin_unlock_irqrestore(shost->host_lock, flags);
243 * scsi_eh_scmd_add - add scsi cmd to error handling.
244 * @scmd: scmd to run eh on.
246 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
248 struct Scsi_Host *shost = scmd->device->host;
249 unsigned long flags;
250 int ret;
252 WARN_ON_ONCE(!shost->ehandler);
254 spin_lock_irqsave(shost->host_lock, flags);
255 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
256 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
257 WARN_ON_ONCE(ret);
259 if (shost->eh_deadline != -1 && !shost->last_reset)
260 shost->last_reset = jiffies;
262 scsi_eh_reset(scmd);
263 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
264 spin_unlock_irqrestore(shost->host_lock, flags);
266 * Ensure that all tasks observe the host state change before the
267 * host_failed change.
269 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
273 * scsi_times_out - Timeout function for normal scsi commands.
274 * @req: request that is timing out.
276 * Notes:
277 * We do not need to lock this. There is the potential for a race
278 * only in that the normal completion handling might run, but if the
279 * normal completion function determines that the timer has already
280 * fired, then it mustn't do anything.
282 enum blk_eh_timer_return scsi_times_out(struct request *req)
284 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
285 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
286 struct Scsi_Host *host = scmd->device->host;
288 trace_scsi_dispatch_cmd_timeout(scmd);
289 scsi_log_completion(scmd, TIMEOUT_ERROR);
291 if (host->eh_deadline != -1 && !host->last_reset)
292 host->last_reset = jiffies;
294 if (host->hostt->eh_timed_out)
295 rtn = host->hostt->eh_timed_out(scmd);
297 if (rtn == BLK_EH_NOT_HANDLED) {
298 if (scsi_abort_command(scmd) != SUCCESS) {
299 set_host_byte(scmd, DID_TIME_OUT);
300 scsi_eh_scmd_add(scmd);
304 return rtn;
308 * scsi_block_when_processing_errors - Prevent cmds from being queued.
309 * @sdev: Device on which we are performing recovery.
311 * Description:
312 * We block until the host is out of error recovery, and then check to
313 * see whether the host or the device is offline.
315 * Return value:
316 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
318 int scsi_block_when_processing_errors(struct scsi_device *sdev)
320 int online;
322 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
324 online = scsi_device_online(sdev);
326 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
327 "%s: rtn: %d\n", __func__, online));
329 return online;
331 EXPORT_SYMBOL(scsi_block_when_processing_errors);
333 #ifdef CONFIG_SCSI_LOGGING
335 * scsi_eh_prt_fail_stats - Log info on failures.
336 * @shost: scsi host being recovered.
337 * @work_q: Queue of scsi cmds to process.
339 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
340 struct list_head *work_q)
342 struct scsi_cmnd *scmd;
343 struct scsi_device *sdev;
344 int total_failures = 0;
345 int cmd_failed = 0;
346 int cmd_cancel = 0;
347 int devices_failed = 0;
349 shost_for_each_device(sdev, shost) {
350 list_for_each_entry(scmd, work_q, eh_entry) {
351 if (scmd->device == sdev) {
352 ++total_failures;
353 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
354 ++cmd_cancel;
355 else
356 ++cmd_failed;
360 if (cmd_cancel || cmd_failed) {
361 SCSI_LOG_ERROR_RECOVERY(3,
362 shost_printk(KERN_INFO, shost,
363 "%s: cmds failed: %d, cancel: %d\n",
364 __func__, cmd_failed,
365 cmd_cancel));
366 cmd_cancel = 0;
367 cmd_failed = 0;
368 ++devices_failed;
372 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
373 "Total of %d commands on %d"
374 " devices require eh work\n",
375 total_failures, devices_failed));
377 #endif
380 * scsi_report_lun_change - Set flag on all *other* devices on the same target
381 * to indicate that a UNIT ATTENTION is expected.
382 * @sdev: Device reporting the UNIT ATTENTION
384 static void scsi_report_lun_change(struct scsi_device *sdev)
386 sdev->sdev_target->expecting_lun_change = 1;
390 * scsi_report_sense - Examine scsi sense information and log messages for
391 * certain conditions, also issue uevents for some of them.
392 * @sdev: Device reporting the sense code
393 * @sshdr: sshdr to be examined
395 static void scsi_report_sense(struct scsi_device *sdev,
396 struct scsi_sense_hdr *sshdr)
398 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
400 if (sshdr->sense_key == UNIT_ATTENTION) {
401 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
402 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
403 sdev_printk(KERN_WARNING, sdev,
404 "Inquiry data has changed");
405 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
406 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
407 scsi_report_lun_change(sdev);
408 sdev_printk(KERN_WARNING, sdev,
409 "Warning! Received an indication that the "
410 "LUN assignments on this target have "
411 "changed. The Linux SCSI layer does not "
412 "automatically remap LUN assignments.\n");
413 } else if (sshdr->asc == 0x3f)
414 sdev_printk(KERN_WARNING, sdev,
415 "Warning! Received an indication that the "
416 "operating parameters on this target have "
417 "changed. The Linux SCSI layer does not "
418 "automatically adjust these parameters.\n");
420 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
421 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
422 sdev_printk(KERN_WARNING, sdev,
423 "Warning! Received an indication that the "
424 "LUN reached a thin provisioning soft "
425 "threshold.\n");
428 if (sshdr->asc == 0x29) {
429 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
430 sdev_printk(KERN_WARNING, sdev,
431 "Power-on or device reset occurred\n");
434 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
435 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
436 sdev_printk(KERN_WARNING, sdev,
437 "Mode parameters changed");
438 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
439 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
440 sdev_printk(KERN_WARNING, sdev,
441 "Asymmetric access state changed");
442 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
443 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
444 sdev_printk(KERN_WARNING, sdev,
445 "Capacity data has changed");
446 } else if (sshdr->asc == 0x2a)
447 sdev_printk(KERN_WARNING, sdev,
448 "Parameters changed");
451 if (evt_type != SDEV_EVT_MAXBITS) {
452 set_bit(evt_type, sdev->pending_events);
453 schedule_work(&sdev->event_work);
458 * scsi_check_sense - Examine scsi cmd sense
459 * @scmd: Cmd to have sense checked.
461 * Return value:
462 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
464 * Notes:
465 * When a deferred error is detected the current command has
466 * not been executed and needs retrying.
468 int scsi_check_sense(struct scsi_cmnd *scmd)
470 struct scsi_device *sdev = scmd->device;
471 struct scsi_sense_hdr sshdr;
473 if (! scsi_command_normalize_sense(scmd, &sshdr))
474 return FAILED; /* no valid sense data */
476 scsi_report_sense(sdev, &sshdr);
478 if (scsi_sense_is_deferred(&sshdr))
479 return NEEDS_RETRY;
481 if (sdev->handler && sdev->handler->check_sense) {
482 int rc;
484 rc = sdev->handler->check_sense(sdev, &sshdr);
485 if (rc != SCSI_RETURN_NOT_HANDLED)
486 return rc;
487 /* handler does not care. Drop down to default handling */
490 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
492 * nasty: for mid-layer issued TURs, we need to return the
493 * actual sense data without any recovery attempt. For eh
494 * issued ones, we need to try to recover and interpret
496 return SUCCESS;
499 * Previous logic looked for FILEMARK, EOM or ILI which are
500 * mainly associated with tapes and returned SUCCESS.
502 if (sshdr.response_code == 0x70) {
503 /* fixed format */
504 if (scmd->sense_buffer[2] & 0xe0)
505 return SUCCESS;
506 } else {
508 * descriptor format: look for "stream commands sense data
509 * descriptor" (see SSC-3). Assume single sense data
510 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
512 if ((sshdr.additional_length > 3) &&
513 (scmd->sense_buffer[8] == 0x4) &&
514 (scmd->sense_buffer[11] & 0xe0))
515 return SUCCESS;
518 switch (sshdr.sense_key) {
519 case NO_SENSE:
520 return SUCCESS;
521 case RECOVERED_ERROR:
522 return /* soft_error */ SUCCESS;
524 case ABORTED_COMMAND:
525 if (sshdr.asc == 0x10) /* DIF */
526 return SUCCESS;
528 return NEEDS_RETRY;
529 case NOT_READY:
530 case UNIT_ATTENTION:
532 * if we are expecting a cc/ua because of a bus reset that we
533 * performed, treat this just as a retry. otherwise this is
534 * information that we should pass up to the upper-level driver
535 * so that we can deal with it there.
537 if (scmd->device->expecting_cc_ua) {
539 * Because some device does not queue unit
540 * attentions correctly, we carefully check
541 * additional sense code and qualifier so as
542 * not to squash media change unit attention.
544 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
545 scmd->device->expecting_cc_ua = 0;
546 return NEEDS_RETRY;
550 * we might also expect a cc/ua if another LUN on the target
551 * reported a UA with an ASC/ASCQ of 3F 0E -
552 * REPORTED LUNS DATA HAS CHANGED.
554 if (scmd->device->sdev_target->expecting_lun_change &&
555 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
556 return NEEDS_RETRY;
558 * if the device is in the process of becoming ready, we
559 * should retry.
561 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
562 return NEEDS_RETRY;
564 * if the device is not started, we need to wake
565 * the error handler to start the motor
567 if (scmd->device->allow_restart &&
568 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
569 return FAILED;
571 * Pass the UA upwards for a determination in the completion
572 * functions.
574 return SUCCESS;
576 /* these are not supported */
577 case DATA_PROTECT:
578 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
579 /* Thin provisioning hard threshold reached */
580 set_host_byte(scmd, DID_ALLOC_FAILURE);
581 return SUCCESS;
583 /* FALLTHROUGH */
584 case COPY_ABORTED:
585 case VOLUME_OVERFLOW:
586 case MISCOMPARE:
587 case BLANK_CHECK:
588 set_host_byte(scmd, DID_TARGET_FAILURE);
589 return SUCCESS;
591 case MEDIUM_ERROR:
592 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
593 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
594 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
595 set_host_byte(scmd, DID_MEDIUM_ERROR);
596 return SUCCESS;
598 return NEEDS_RETRY;
600 case HARDWARE_ERROR:
601 if (scmd->device->retry_hwerror)
602 return ADD_TO_MLQUEUE;
603 else
604 set_host_byte(scmd, DID_TARGET_FAILURE);
605 /* FALLTHROUGH */
607 case ILLEGAL_REQUEST:
608 if (sshdr.asc == 0x20 || /* Invalid command operation code */
609 sshdr.asc == 0x21 || /* Logical block address out of range */
610 sshdr.asc == 0x22 || /* Invalid function */
611 sshdr.asc == 0x24 || /* Invalid field in cdb */
612 sshdr.asc == 0x26 || /* Parameter value invalid */
613 sshdr.asc == 0x27) { /* Write protected */
614 set_host_byte(scmd, DID_TARGET_FAILURE);
616 return SUCCESS;
618 default:
619 return SUCCESS;
622 EXPORT_SYMBOL_GPL(scsi_check_sense);
624 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
626 struct scsi_host_template *sht = sdev->host->hostt;
627 struct scsi_device *tmp_sdev;
629 if (!sht->track_queue_depth ||
630 sdev->queue_depth >= sdev->max_queue_depth)
631 return;
633 if (time_before(jiffies,
634 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
635 return;
637 if (time_before(jiffies,
638 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
639 return;
642 * Walk all devices of a target and do
643 * ramp up on them.
645 shost_for_each_device(tmp_sdev, sdev->host) {
646 if (tmp_sdev->channel != sdev->channel ||
647 tmp_sdev->id != sdev->id ||
648 tmp_sdev->queue_depth == sdev->max_queue_depth)
649 continue;
651 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
652 sdev->last_queue_ramp_up = jiffies;
656 static void scsi_handle_queue_full(struct scsi_device *sdev)
658 struct scsi_host_template *sht = sdev->host->hostt;
659 struct scsi_device *tmp_sdev;
661 if (!sht->track_queue_depth)
662 return;
664 shost_for_each_device(tmp_sdev, sdev->host) {
665 if (tmp_sdev->channel != sdev->channel ||
666 tmp_sdev->id != sdev->id)
667 continue;
669 * We do not know the number of commands that were at
670 * the device when we got the queue full so we start
671 * from the highest possible value and work our way down.
673 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
678 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
679 * @scmd: SCSI cmd to examine.
681 * Notes:
682 * This is *only* called when we are examining the status of commands
683 * queued during error recovery. the main difference here is that we
684 * don't allow for the possibility of retries here, and we are a lot
685 * more restrictive about what we consider acceptable.
687 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
690 * first check the host byte, to see if there is anything in there
691 * that would indicate what we need to do.
693 if (host_byte(scmd->result) == DID_RESET) {
695 * rats. we are already in the error handler, so we now
696 * get to try and figure out what to do next. if the sense
697 * is valid, we have a pretty good idea of what to do.
698 * if not, we mark it as FAILED.
700 return scsi_check_sense(scmd);
702 if (host_byte(scmd->result) != DID_OK)
703 return FAILED;
706 * next, check the message byte.
708 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
709 return FAILED;
712 * now, check the status byte to see if this indicates
713 * anything special.
715 switch (status_byte(scmd->result)) {
716 case GOOD:
717 scsi_handle_queue_ramp_up(scmd->device);
718 /* FALLTHROUGH */
719 case COMMAND_TERMINATED:
720 return SUCCESS;
721 case CHECK_CONDITION:
722 return scsi_check_sense(scmd);
723 case CONDITION_GOOD:
724 case INTERMEDIATE_GOOD:
725 case INTERMEDIATE_C_GOOD:
727 * who knows? FIXME(eric)
729 return SUCCESS;
730 case RESERVATION_CONFLICT:
731 if (scmd->cmnd[0] == TEST_UNIT_READY)
732 /* it is a success, we probed the device and
733 * found it */
734 return SUCCESS;
735 /* otherwise, we failed to send the command */
736 return FAILED;
737 case QUEUE_FULL:
738 scsi_handle_queue_full(scmd->device);
739 /* fall through */
740 case BUSY:
741 return NEEDS_RETRY;
742 default:
743 return FAILED;
745 return FAILED;
749 * scsi_eh_done - Completion function for error handling.
750 * @scmd: Cmd that is done.
752 static void scsi_eh_done(struct scsi_cmnd *scmd)
754 struct completion *eh_action;
756 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
757 "%s result: %x\n", __func__, scmd->result));
759 eh_action = scmd->device->host->eh_action;
760 if (eh_action)
761 complete(eh_action);
765 * scsi_try_host_reset - ask host adapter to reset itself
766 * @scmd: SCSI cmd to send host reset.
768 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
770 unsigned long flags;
771 int rtn;
772 struct Scsi_Host *host = scmd->device->host;
773 struct scsi_host_template *hostt = host->hostt;
775 SCSI_LOG_ERROR_RECOVERY(3,
776 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
778 if (!hostt->eh_host_reset_handler)
779 return FAILED;
781 rtn = hostt->eh_host_reset_handler(scmd);
783 if (rtn == SUCCESS) {
784 if (!hostt->skip_settle_delay)
785 ssleep(HOST_RESET_SETTLE_TIME);
786 spin_lock_irqsave(host->host_lock, flags);
787 scsi_report_bus_reset(host, scmd_channel(scmd));
788 spin_unlock_irqrestore(host->host_lock, flags);
791 return rtn;
795 * scsi_try_bus_reset - ask host to perform a bus reset
796 * @scmd: SCSI cmd to send bus reset.
798 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
800 unsigned long flags;
801 int rtn;
802 struct Scsi_Host *host = scmd->device->host;
803 struct scsi_host_template *hostt = host->hostt;
805 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
806 "%s: Snd Bus RST\n", __func__));
808 if (!hostt->eh_bus_reset_handler)
809 return FAILED;
811 rtn = hostt->eh_bus_reset_handler(scmd);
813 if (rtn == SUCCESS) {
814 if (!hostt->skip_settle_delay)
815 ssleep(BUS_RESET_SETTLE_TIME);
816 spin_lock_irqsave(host->host_lock, flags);
817 scsi_report_bus_reset(host, scmd_channel(scmd));
818 spin_unlock_irqrestore(host->host_lock, flags);
821 return rtn;
824 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
826 sdev->was_reset = 1;
827 sdev->expecting_cc_ua = 1;
831 * scsi_try_target_reset - Ask host to perform a target reset
832 * @scmd: SCSI cmd used to send a target reset
834 * Notes:
835 * There is no timeout for this operation. if this operation is
836 * unreliable for a given host, then the host itself needs to put a
837 * timer on it, and set the host back to a consistent state prior to
838 * returning.
840 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
842 unsigned long flags;
843 int rtn;
844 struct Scsi_Host *host = scmd->device->host;
845 struct scsi_host_template *hostt = host->hostt;
847 if (!hostt->eh_target_reset_handler)
848 return FAILED;
850 rtn = hostt->eh_target_reset_handler(scmd);
851 if (rtn == SUCCESS) {
852 spin_lock_irqsave(host->host_lock, flags);
853 __starget_for_each_device(scsi_target(scmd->device), NULL,
854 __scsi_report_device_reset);
855 spin_unlock_irqrestore(host->host_lock, flags);
858 return rtn;
862 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
863 * @scmd: SCSI cmd used to send BDR
865 * Notes:
866 * There is no timeout for this operation. if this operation is
867 * unreliable for a given host, then the host itself needs to put a
868 * timer on it, and set the host back to a consistent state prior to
869 * returning.
871 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
873 int rtn;
874 struct scsi_host_template *hostt = scmd->device->host->hostt;
876 if (!hostt->eh_device_reset_handler)
877 return FAILED;
879 rtn = hostt->eh_device_reset_handler(scmd);
880 if (rtn == SUCCESS)
881 __scsi_report_device_reset(scmd->device, NULL);
882 return rtn;
886 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
887 * @hostt: SCSI driver host template
888 * @scmd: SCSI cmd used to send a target reset
890 * Return value:
891 * SUCCESS, FAILED, or FAST_IO_FAIL
893 * Notes:
894 * SUCCESS does not necessarily indicate that the command
895 * has been aborted; it only indicates that the LLDDs
896 * has cleared all references to that command.
897 * LLDDs should return FAILED only if an abort was required
898 * but could not be executed. LLDDs should return FAST_IO_FAIL
899 * if the device is temporarily unavailable (eg due to a
900 * link down on FibreChannel)
902 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
903 struct scsi_cmnd *scmd)
905 if (!hostt->eh_abort_handler)
906 return FAILED;
908 return hostt->eh_abort_handler(scmd);
911 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
913 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
914 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
915 if (scsi_try_target_reset(scmd) != SUCCESS)
916 if (scsi_try_bus_reset(scmd) != SUCCESS)
917 scsi_try_host_reset(scmd);
921 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
922 * @scmd: SCSI command structure to hijack
923 * @ses: structure to save restore information
924 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
925 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
926 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
928 * This function is used to save a scsi command information before re-execution
929 * as part of the error recovery process. If @sense_bytes is 0 the command
930 * sent must be one that does not transfer any data. If @sense_bytes != 0
931 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
932 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
934 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
935 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
937 struct scsi_device *sdev = scmd->device;
940 * We need saved copies of a number of fields - this is because
941 * error handling may need to overwrite these with different values
942 * to run different commands, and once error handling is complete,
943 * we will need to restore these values prior to running the actual
944 * command.
946 ses->cmd_len = scmd->cmd_len;
947 ses->cmnd = scmd->cmnd;
948 ses->data_direction = scmd->sc_data_direction;
949 ses->sdb = scmd->sdb;
950 ses->next_rq = scmd->request->next_rq;
951 ses->result = scmd->result;
952 ses->underflow = scmd->underflow;
953 ses->prot_op = scmd->prot_op;
954 ses->eh_eflags = scmd->eh_eflags;
956 scmd->prot_op = SCSI_PROT_NORMAL;
957 scmd->eh_eflags = 0;
958 scmd->cmnd = ses->eh_cmnd;
959 memset(scmd->cmnd, 0, BLK_MAX_CDB);
960 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
961 scmd->request->next_rq = NULL;
962 scmd->result = 0;
964 if (sense_bytes) {
965 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
966 sense_bytes);
967 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
968 scmd->sdb.length);
969 scmd->sdb.table.sgl = &ses->sense_sgl;
970 scmd->sc_data_direction = DMA_FROM_DEVICE;
971 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
972 scmd->cmnd[0] = REQUEST_SENSE;
973 scmd->cmnd[4] = scmd->sdb.length;
974 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
975 } else {
976 scmd->sc_data_direction = DMA_NONE;
977 if (cmnd) {
978 BUG_ON(cmnd_size > BLK_MAX_CDB);
979 memcpy(scmd->cmnd, cmnd, cmnd_size);
980 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
984 scmd->underflow = 0;
986 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
987 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
988 (sdev->lun << 5 & 0xe0);
991 * Zero the sense buffer. The scsi spec mandates that any
992 * untransferred sense data should be interpreted as being zero.
994 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
996 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
999 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1000 * @scmd: SCSI command structure to restore
1001 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1003 * Undo any damage done by above scsi_eh_prep_cmnd().
1005 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1008 * Restore original data
1010 scmd->cmd_len = ses->cmd_len;
1011 scmd->cmnd = ses->cmnd;
1012 scmd->sc_data_direction = ses->data_direction;
1013 scmd->sdb = ses->sdb;
1014 scmd->request->next_rq = ses->next_rq;
1015 scmd->result = ses->result;
1016 scmd->underflow = ses->underflow;
1017 scmd->prot_op = ses->prot_op;
1018 scmd->eh_eflags = ses->eh_eflags;
1020 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1023 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1024 * @scmd: SCSI command structure to hijack
1025 * @cmnd: CDB to send
1026 * @cmnd_size: size in bytes of @cmnd
1027 * @timeout: timeout for this request
1028 * @sense_bytes: size of sense data to copy or 0
1030 * This function is used to send a scsi command down to a target device
1031 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1033 * Return value:
1034 * SUCCESS or FAILED or NEEDS_RETRY
1036 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1037 int cmnd_size, int timeout, unsigned sense_bytes)
1039 struct scsi_device *sdev = scmd->device;
1040 struct Scsi_Host *shost = sdev->host;
1041 DECLARE_COMPLETION_ONSTACK(done);
1042 unsigned long timeleft = timeout;
1043 struct scsi_eh_save ses;
1044 const unsigned long stall_for = msecs_to_jiffies(100);
1045 int rtn;
1047 retry:
1048 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1049 shost->eh_action = &done;
1051 scsi_log_send(scmd);
1052 scmd->scsi_done = scsi_eh_done;
1053 rtn = shost->hostt->queuecommand(shost, scmd);
1054 if (rtn) {
1055 if (timeleft > stall_for) {
1056 scsi_eh_restore_cmnd(scmd, &ses);
1057 timeleft -= stall_for;
1058 msleep(jiffies_to_msecs(stall_for));
1059 goto retry;
1061 /* signal not to enter either branch of the if () below */
1062 timeleft = 0;
1063 rtn = FAILED;
1064 } else {
1065 timeleft = wait_for_completion_timeout(&done, timeout);
1066 rtn = SUCCESS;
1069 shost->eh_action = NULL;
1071 scsi_log_completion(scmd, rtn);
1073 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1074 "%s timeleft: %ld\n",
1075 __func__, timeleft));
1078 * If there is time left scsi_eh_done got called, and we will examine
1079 * the actual status codes to see whether the command actually did
1080 * complete normally, else if we have a zero return and no time left,
1081 * the command must still be pending, so abort it and return FAILED.
1082 * If we never actually managed to issue the command, because
1083 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1084 * value above (so don't execute either branch of the if)
1086 if (timeleft) {
1087 rtn = scsi_eh_completed_normally(scmd);
1088 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1089 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1091 switch (rtn) {
1092 case SUCCESS:
1093 case NEEDS_RETRY:
1094 case FAILED:
1095 break;
1096 case ADD_TO_MLQUEUE:
1097 rtn = NEEDS_RETRY;
1098 break;
1099 default:
1100 rtn = FAILED;
1101 break;
1103 } else if (rtn != FAILED) {
1104 scsi_abort_eh_cmnd(scmd);
1105 rtn = FAILED;
1108 scsi_eh_restore_cmnd(scmd, &ses);
1110 return rtn;
1114 * scsi_request_sense - Request sense data from a particular target.
1115 * @scmd: SCSI cmd for request sense.
1117 * Notes:
1118 * Some hosts automatically obtain this information, others require
1119 * that we obtain it on our own. This function will *not* return until
1120 * the command either times out, or it completes.
1122 static int scsi_request_sense(struct scsi_cmnd *scmd)
1124 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1127 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1129 if (!blk_rq_is_passthrough(scmd->request)) {
1130 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1131 if (sdrv->eh_action)
1132 rtn = sdrv->eh_action(scmd, rtn);
1134 return rtn;
1138 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1139 * @scmd: Original SCSI cmd that eh has finished.
1140 * @done_q: Queue for processed commands.
1142 * Notes:
1143 * We don't want to use the normal command completion while we are are
1144 * still handling errors - it may cause other commands to be queued,
1145 * and that would disturb what we are doing. Thus we really want to
1146 * keep a list of pending commands for final completion, and once we
1147 * are ready to leave error handling we handle completion for real.
1149 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1151 list_move_tail(&scmd->eh_entry, done_q);
1153 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1156 * scsi_eh_get_sense - Get device sense data.
1157 * @work_q: Queue of commands to process.
1158 * @done_q: Queue of processed commands.
1160 * Description:
1161 * See if we need to request sense information. if so, then get it
1162 * now, so we have a better idea of what to do.
1164 * Notes:
1165 * This has the unfortunate side effect that if a shost adapter does
1166 * not automatically request sense information, we end up shutting
1167 * it down before we request it.
1169 * All drivers should request sense information internally these days,
1170 * so for now all I have to say is tough noogies if you end up in here.
1172 * XXX: Long term this code should go away, but that needs an audit of
1173 * all LLDDs first.
1175 int scsi_eh_get_sense(struct list_head *work_q,
1176 struct list_head *done_q)
1178 struct scsi_cmnd *scmd, *next;
1179 struct Scsi_Host *shost;
1180 int rtn;
1183 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1184 * should not get sense.
1186 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1187 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1188 SCSI_SENSE_VALID(scmd))
1189 continue;
1191 shost = scmd->device->host;
1192 if (scsi_host_eh_past_deadline(shost)) {
1193 SCSI_LOG_ERROR_RECOVERY(3,
1194 scmd_printk(KERN_INFO, scmd,
1195 "%s: skip request sense, past eh deadline\n",
1196 current->comm));
1197 break;
1199 if (status_byte(scmd->result) != CHECK_CONDITION)
1201 * don't request sense if there's no check condition
1202 * status because the error we're processing isn't one
1203 * that has a sense code (and some devices get
1204 * confused by sense requests out of the blue)
1206 continue;
1208 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1209 "%s: requesting sense\n",
1210 current->comm));
1211 rtn = scsi_request_sense(scmd);
1212 if (rtn != SUCCESS)
1213 continue;
1215 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1216 "sense requested, result %x\n", scmd->result));
1217 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1219 rtn = scsi_decide_disposition(scmd);
1222 * if the result was normal, then just pass it along to the
1223 * upper level.
1225 if (rtn == SUCCESS)
1226 /* we don't want this command reissued, just
1227 * finished with the sense data, so set
1228 * retries to the max allowed to ensure it
1229 * won't get reissued */
1230 scmd->retries = scmd->allowed;
1231 else if (rtn != NEEDS_RETRY)
1232 continue;
1234 scsi_eh_finish_cmd(scmd, done_q);
1237 return list_empty(work_q);
1239 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1242 * scsi_eh_tur - Send TUR to device.
1243 * @scmd: &scsi_cmnd to send TUR
1245 * Return value:
1246 * 0 - Device is ready. 1 - Device NOT ready.
1248 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1250 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1251 int retry_cnt = 1, rtn;
1253 retry_tur:
1254 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1255 scmd->device->eh_timeout, 0);
1257 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1258 "%s return: %x\n", __func__, rtn));
1260 switch (rtn) {
1261 case NEEDS_RETRY:
1262 if (retry_cnt--)
1263 goto retry_tur;
1264 /*FALLTHRU*/
1265 case SUCCESS:
1266 return 0;
1267 default:
1268 return 1;
1273 * scsi_eh_test_devices - check if devices are responding from error recovery.
1274 * @cmd_list: scsi commands in error recovery.
1275 * @work_q: queue for commands which still need more error recovery
1276 * @done_q: queue for commands which are finished
1277 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1279 * Decription:
1280 * Tests if devices are in a working state. Commands to devices now in
1281 * a working state are sent to the done_q while commands to devices which
1282 * are still failing to respond are returned to the work_q for more
1283 * processing.
1285 static int scsi_eh_test_devices(struct list_head *cmd_list,
1286 struct list_head *work_q,
1287 struct list_head *done_q, int try_stu)
1289 struct scsi_cmnd *scmd, *next;
1290 struct scsi_device *sdev;
1291 int finish_cmds;
1293 while (!list_empty(cmd_list)) {
1294 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1295 sdev = scmd->device;
1297 if (!try_stu) {
1298 if (scsi_host_eh_past_deadline(sdev->host)) {
1299 /* Push items back onto work_q */
1300 list_splice_init(cmd_list, work_q);
1301 SCSI_LOG_ERROR_RECOVERY(3,
1302 sdev_printk(KERN_INFO, sdev,
1303 "%s: skip test device, past eh deadline",
1304 current->comm));
1305 break;
1309 finish_cmds = !scsi_device_online(scmd->device) ||
1310 (try_stu && !scsi_eh_try_stu(scmd) &&
1311 !scsi_eh_tur(scmd)) ||
1312 !scsi_eh_tur(scmd);
1314 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1315 if (scmd->device == sdev) {
1316 if (finish_cmds &&
1317 (try_stu ||
1318 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1319 scsi_eh_finish_cmd(scmd, done_q);
1320 else
1321 list_move_tail(&scmd->eh_entry, work_q);
1324 return list_empty(work_q);
1328 * scsi_eh_try_stu - Send START_UNIT to device.
1329 * @scmd: &scsi_cmnd to send START_UNIT
1331 * Return value:
1332 * 0 - Device is ready. 1 - Device NOT ready.
1334 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1336 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1338 if (scmd->device->allow_restart) {
1339 int i, rtn = NEEDS_RETRY;
1341 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1342 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1344 if (rtn == SUCCESS)
1345 return 0;
1348 return 1;
1352 * scsi_eh_stu - send START_UNIT if needed
1353 * @shost: &scsi host being recovered.
1354 * @work_q: &list_head for pending commands.
1355 * @done_q: &list_head for processed commands.
1357 * Notes:
1358 * If commands are failing due to not ready, initializing command required,
1359 * try revalidating the device, which will end up sending a start unit.
1361 static int scsi_eh_stu(struct Scsi_Host *shost,
1362 struct list_head *work_q,
1363 struct list_head *done_q)
1365 struct scsi_cmnd *scmd, *stu_scmd, *next;
1366 struct scsi_device *sdev;
1368 shost_for_each_device(sdev, shost) {
1369 if (scsi_host_eh_past_deadline(shost)) {
1370 SCSI_LOG_ERROR_RECOVERY(3,
1371 sdev_printk(KERN_INFO, sdev,
1372 "%s: skip START_UNIT, past eh deadline\n",
1373 current->comm));
1374 break;
1376 stu_scmd = NULL;
1377 list_for_each_entry(scmd, work_q, eh_entry)
1378 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1379 scsi_check_sense(scmd) == FAILED ) {
1380 stu_scmd = scmd;
1381 break;
1384 if (!stu_scmd)
1385 continue;
1387 SCSI_LOG_ERROR_RECOVERY(3,
1388 sdev_printk(KERN_INFO, sdev,
1389 "%s: Sending START_UNIT\n",
1390 current->comm));
1392 if (!scsi_eh_try_stu(stu_scmd)) {
1393 if (!scsi_device_online(sdev) ||
1394 !scsi_eh_tur(stu_scmd)) {
1395 list_for_each_entry_safe(scmd, next,
1396 work_q, eh_entry) {
1397 if (scmd->device == sdev &&
1398 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1399 scsi_eh_finish_cmd(scmd, done_q);
1402 } else {
1403 SCSI_LOG_ERROR_RECOVERY(3,
1404 sdev_printk(KERN_INFO, sdev,
1405 "%s: START_UNIT failed\n",
1406 current->comm));
1410 return list_empty(work_q);
1415 * scsi_eh_bus_device_reset - send bdr if needed
1416 * @shost: scsi host being recovered.
1417 * @work_q: &list_head for pending commands.
1418 * @done_q: &list_head for processed commands.
1420 * Notes:
1421 * Try a bus device reset. Still, look to see whether we have multiple
1422 * devices that are jammed or not - if we have multiple devices, it
1423 * makes no sense to try bus_device_reset - we really would need to try
1424 * a bus_reset instead.
1426 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1427 struct list_head *work_q,
1428 struct list_head *done_q)
1430 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1431 struct scsi_device *sdev;
1432 int rtn;
1434 shost_for_each_device(sdev, shost) {
1435 if (scsi_host_eh_past_deadline(shost)) {
1436 SCSI_LOG_ERROR_RECOVERY(3,
1437 sdev_printk(KERN_INFO, sdev,
1438 "%s: skip BDR, past eh deadline\n",
1439 current->comm));
1440 break;
1442 bdr_scmd = NULL;
1443 list_for_each_entry(scmd, work_q, eh_entry)
1444 if (scmd->device == sdev) {
1445 bdr_scmd = scmd;
1446 break;
1449 if (!bdr_scmd)
1450 continue;
1452 SCSI_LOG_ERROR_RECOVERY(3,
1453 sdev_printk(KERN_INFO, sdev,
1454 "%s: Sending BDR\n", current->comm));
1455 rtn = scsi_try_bus_device_reset(bdr_scmd);
1456 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1457 if (!scsi_device_online(sdev) ||
1458 rtn == FAST_IO_FAIL ||
1459 !scsi_eh_tur(bdr_scmd)) {
1460 list_for_each_entry_safe(scmd, next,
1461 work_q, eh_entry) {
1462 if (scmd->device == sdev &&
1463 scsi_eh_action(scmd, rtn) != FAILED)
1464 scsi_eh_finish_cmd(scmd,
1465 done_q);
1468 } else {
1469 SCSI_LOG_ERROR_RECOVERY(3,
1470 sdev_printk(KERN_INFO, sdev,
1471 "%s: BDR failed\n", current->comm));
1475 return list_empty(work_q);
1479 * scsi_eh_target_reset - send target reset if needed
1480 * @shost: scsi host being recovered.
1481 * @work_q: &list_head for pending commands.
1482 * @done_q: &list_head for processed commands.
1484 * Notes:
1485 * Try a target reset.
1487 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1488 struct list_head *work_q,
1489 struct list_head *done_q)
1491 LIST_HEAD(tmp_list);
1492 LIST_HEAD(check_list);
1494 list_splice_init(work_q, &tmp_list);
1496 while (!list_empty(&tmp_list)) {
1497 struct scsi_cmnd *next, *scmd;
1498 int rtn;
1499 unsigned int id;
1501 if (scsi_host_eh_past_deadline(shost)) {
1502 /* push back on work queue for further processing */
1503 list_splice_init(&check_list, work_q);
1504 list_splice_init(&tmp_list, work_q);
1505 SCSI_LOG_ERROR_RECOVERY(3,
1506 shost_printk(KERN_INFO, shost,
1507 "%s: Skip target reset, past eh deadline\n",
1508 current->comm));
1509 return list_empty(work_q);
1512 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1513 id = scmd_id(scmd);
1515 SCSI_LOG_ERROR_RECOVERY(3,
1516 shost_printk(KERN_INFO, shost,
1517 "%s: Sending target reset to target %d\n",
1518 current->comm, id));
1519 rtn = scsi_try_target_reset(scmd);
1520 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1521 SCSI_LOG_ERROR_RECOVERY(3,
1522 shost_printk(KERN_INFO, shost,
1523 "%s: Target reset failed"
1524 " target: %d\n",
1525 current->comm, id));
1526 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1527 if (scmd_id(scmd) != id)
1528 continue;
1530 if (rtn == SUCCESS)
1531 list_move_tail(&scmd->eh_entry, &check_list);
1532 else if (rtn == FAST_IO_FAIL)
1533 scsi_eh_finish_cmd(scmd, done_q);
1534 else
1535 /* push back on work queue for further processing */
1536 list_move(&scmd->eh_entry, work_q);
1540 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1544 * scsi_eh_bus_reset - send a bus reset
1545 * @shost: &scsi host being recovered.
1546 * @work_q: &list_head for pending commands.
1547 * @done_q: &list_head for processed commands.
1549 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1550 struct list_head *work_q,
1551 struct list_head *done_q)
1553 struct scsi_cmnd *scmd, *chan_scmd, *next;
1554 LIST_HEAD(check_list);
1555 unsigned int channel;
1556 int rtn;
1559 * we really want to loop over the various channels, and do this on
1560 * a channel by channel basis. we should also check to see if any
1561 * of the failed commands are on soft_reset devices, and if so, skip
1562 * the reset.
1565 for (channel = 0; channel <= shost->max_channel; channel++) {
1566 if (scsi_host_eh_past_deadline(shost)) {
1567 list_splice_init(&check_list, work_q);
1568 SCSI_LOG_ERROR_RECOVERY(3,
1569 shost_printk(KERN_INFO, shost,
1570 "%s: skip BRST, past eh deadline\n",
1571 current->comm));
1572 return list_empty(work_q);
1575 chan_scmd = NULL;
1576 list_for_each_entry(scmd, work_q, eh_entry) {
1577 if (channel == scmd_channel(scmd)) {
1578 chan_scmd = scmd;
1579 break;
1581 * FIXME add back in some support for
1582 * soft_reset devices.
1587 if (!chan_scmd)
1588 continue;
1589 SCSI_LOG_ERROR_RECOVERY(3,
1590 shost_printk(KERN_INFO, shost,
1591 "%s: Sending BRST chan: %d\n",
1592 current->comm, channel));
1593 rtn = scsi_try_bus_reset(chan_scmd);
1594 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1595 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1596 if (channel == scmd_channel(scmd)) {
1597 if (rtn == FAST_IO_FAIL)
1598 scsi_eh_finish_cmd(scmd,
1599 done_q);
1600 else
1601 list_move_tail(&scmd->eh_entry,
1602 &check_list);
1605 } else {
1606 SCSI_LOG_ERROR_RECOVERY(3,
1607 shost_printk(KERN_INFO, shost,
1608 "%s: BRST failed chan: %d\n",
1609 current->comm, channel));
1612 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1616 * scsi_eh_host_reset - send a host reset
1617 * @shost: host to be reset.
1618 * @work_q: &list_head for pending commands.
1619 * @done_q: &list_head for processed commands.
1621 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1622 struct list_head *work_q,
1623 struct list_head *done_q)
1625 struct scsi_cmnd *scmd, *next;
1626 LIST_HEAD(check_list);
1627 int rtn;
1629 if (!list_empty(work_q)) {
1630 scmd = list_entry(work_q->next,
1631 struct scsi_cmnd, eh_entry);
1633 SCSI_LOG_ERROR_RECOVERY(3,
1634 shost_printk(KERN_INFO, shost,
1635 "%s: Sending HRST\n",
1636 current->comm));
1638 rtn = scsi_try_host_reset(scmd);
1639 if (rtn == SUCCESS) {
1640 list_splice_init(work_q, &check_list);
1641 } else if (rtn == FAST_IO_FAIL) {
1642 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1643 scsi_eh_finish_cmd(scmd, done_q);
1645 } else {
1646 SCSI_LOG_ERROR_RECOVERY(3,
1647 shost_printk(KERN_INFO, shost,
1648 "%s: HRST failed\n",
1649 current->comm));
1652 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1656 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1657 * @work_q: &list_head for pending commands.
1658 * @done_q: &list_head for processed commands.
1660 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1661 struct list_head *done_q)
1663 struct scsi_cmnd *scmd, *next;
1664 struct scsi_device *sdev;
1666 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1667 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1668 "not ready after error recovery\n");
1669 sdev = scmd->device;
1671 mutex_lock(&sdev->state_mutex);
1672 scsi_device_set_state(sdev, SDEV_OFFLINE);
1673 mutex_unlock(&sdev->state_mutex);
1675 scsi_eh_finish_cmd(scmd, done_q);
1677 return;
1681 * scsi_noretry_cmd - determine if command should be failed fast
1682 * @scmd: SCSI cmd to examine.
1684 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1686 switch (host_byte(scmd->result)) {
1687 case DID_OK:
1688 break;
1689 case DID_TIME_OUT:
1690 goto check_type;
1691 case DID_BUS_BUSY:
1692 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1693 case DID_PARITY:
1694 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1695 case DID_ERROR:
1696 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1697 status_byte(scmd->result) == RESERVATION_CONFLICT)
1698 return 0;
1699 /* fall through */
1700 case DID_SOFT_ERROR:
1701 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1704 if (status_byte(scmd->result) != CHECK_CONDITION)
1705 return 0;
1707 check_type:
1709 * assume caller has checked sense and determined
1710 * the check condition was retryable.
1712 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1713 blk_rq_is_passthrough(scmd->request))
1714 return 1;
1715 else
1716 return 0;
1720 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1721 * @scmd: SCSI cmd to examine.
1723 * Notes:
1724 * This is *only* called when we are examining the status after sending
1725 * out the actual data command. any commands that are queued for error
1726 * recovery (e.g. test_unit_ready) do *not* come through here.
1728 * When this routine returns failed, it means the error handler thread
1729 * is woken. In cases where the error code indicates an error that
1730 * doesn't require the error handler read (i.e. we don't need to
1731 * abort/reset), this function should return SUCCESS.
1733 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1735 int rtn;
1738 * if the device is offline, then we clearly just pass the result back
1739 * up to the top level.
1741 if (!scsi_device_online(scmd->device)) {
1742 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1743 "%s: device offline - report as SUCCESS\n", __func__));
1744 return SUCCESS;
1748 * first check the host byte, to see if there is anything in there
1749 * that would indicate what we need to do.
1751 switch (host_byte(scmd->result)) {
1752 case DID_PASSTHROUGH:
1754 * no matter what, pass this through to the upper layer.
1755 * nuke this special code so that it looks like we are saying
1756 * did_ok.
1758 scmd->result &= 0xff00ffff;
1759 return SUCCESS;
1760 case DID_OK:
1762 * looks good. drop through, and check the next byte.
1764 break;
1765 case DID_ABORT:
1766 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1767 set_host_byte(scmd, DID_TIME_OUT);
1768 return SUCCESS;
1770 /* FALLTHROUGH */
1771 case DID_NO_CONNECT:
1772 case DID_BAD_TARGET:
1774 * note - this means that we just report the status back
1775 * to the top level driver, not that we actually think
1776 * that it indicates SUCCESS.
1778 return SUCCESS;
1779 case DID_SOFT_ERROR:
1781 * when the low level driver returns did_soft_error,
1782 * it is responsible for keeping an internal retry counter
1783 * in order to avoid endless loops (db)
1785 goto maybe_retry;
1786 case DID_IMM_RETRY:
1787 return NEEDS_RETRY;
1789 case DID_REQUEUE:
1790 return ADD_TO_MLQUEUE;
1791 case DID_TRANSPORT_DISRUPTED:
1793 * LLD/transport was disrupted during processing of the IO.
1794 * The transport class is now blocked/blocking,
1795 * and the transport will decide what to do with the IO
1796 * based on its timers and recovery capablilities if
1797 * there are enough retries.
1799 goto maybe_retry;
1800 case DID_TRANSPORT_FAILFAST:
1802 * The transport decided to failfast the IO (most likely
1803 * the fast io fail tmo fired), so send IO directly upwards.
1805 return SUCCESS;
1806 case DID_ERROR:
1807 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1808 status_byte(scmd->result) == RESERVATION_CONFLICT)
1810 * execute reservation conflict processing code
1811 * lower down
1813 break;
1814 /* fallthrough */
1815 case DID_BUS_BUSY:
1816 case DID_PARITY:
1817 goto maybe_retry;
1818 case DID_TIME_OUT:
1820 * when we scan the bus, we get timeout messages for
1821 * these commands if there is no device available.
1822 * other hosts report did_no_connect for the same thing.
1824 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1825 scmd->cmnd[0] == INQUIRY)) {
1826 return SUCCESS;
1827 } else {
1828 return FAILED;
1830 case DID_RESET:
1831 return SUCCESS;
1832 default:
1833 return FAILED;
1837 * next, check the message byte.
1839 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1840 return FAILED;
1843 * check the status byte to see if this indicates anything special.
1845 switch (status_byte(scmd->result)) {
1846 case QUEUE_FULL:
1847 scsi_handle_queue_full(scmd->device);
1849 * the case of trying to send too many commands to a
1850 * tagged queueing device.
1852 /* FALLTHROUGH */
1853 case BUSY:
1855 * device can't talk to us at the moment. Should only
1856 * occur (SAM-3) when the task queue is empty, so will cause
1857 * the empty queue handling to trigger a stall in the
1858 * device.
1860 return ADD_TO_MLQUEUE;
1861 case GOOD:
1862 if (scmd->cmnd[0] == REPORT_LUNS)
1863 scmd->device->sdev_target->expecting_lun_change = 0;
1864 scsi_handle_queue_ramp_up(scmd->device);
1865 /* FALLTHROUGH */
1866 case COMMAND_TERMINATED:
1867 return SUCCESS;
1868 case TASK_ABORTED:
1869 goto maybe_retry;
1870 case CHECK_CONDITION:
1871 rtn = scsi_check_sense(scmd);
1872 if (rtn == NEEDS_RETRY)
1873 goto maybe_retry;
1874 /* if rtn == FAILED, we have no sense information;
1875 * returning FAILED will wake the error handler thread
1876 * to collect the sense and redo the decide
1877 * disposition */
1878 return rtn;
1879 case CONDITION_GOOD:
1880 case INTERMEDIATE_GOOD:
1881 case INTERMEDIATE_C_GOOD:
1882 case ACA_ACTIVE:
1884 * who knows? FIXME(eric)
1886 return SUCCESS;
1888 case RESERVATION_CONFLICT:
1889 sdev_printk(KERN_INFO, scmd->device,
1890 "reservation conflict\n");
1891 set_host_byte(scmd, DID_NEXUS_FAILURE);
1892 return SUCCESS; /* causes immediate i/o error */
1893 default:
1894 return FAILED;
1896 return FAILED;
1898 maybe_retry:
1900 /* we requeue for retry because the error was retryable, and
1901 * the request was not marked fast fail. Note that above,
1902 * even if the request is marked fast fail, we still requeue
1903 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1904 if ((++scmd->retries) <= scmd->allowed
1905 && !scsi_noretry_cmd(scmd)) {
1906 return NEEDS_RETRY;
1907 } else {
1909 * no more retries - report this one back to upper level.
1911 return SUCCESS;
1915 static void eh_lock_door_done(struct request *req, blk_status_t status)
1917 __blk_put_request(req->q, req);
1921 * scsi_eh_lock_door - Prevent medium removal for the specified device
1922 * @sdev: SCSI device to prevent medium removal
1924 * Locking:
1925 * We must be called from process context.
1927 * Notes:
1928 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1929 * head of the devices request queue, and continue.
1931 static void scsi_eh_lock_door(struct scsi_device *sdev)
1933 struct request *req;
1934 struct scsi_request *rq;
1937 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a
1938 * request becomes available
1940 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, GFP_KERNEL);
1941 if (IS_ERR(req))
1942 return;
1943 rq = scsi_req(req);
1945 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1946 rq->cmd[1] = 0;
1947 rq->cmd[2] = 0;
1948 rq->cmd[3] = 0;
1949 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
1950 rq->cmd[5] = 0;
1951 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
1953 req->rq_flags |= RQF_QUIET;
1954 req->timeout = 10 * HZ;
1955 rq->retries = 5;
1957 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1961 * scsi_restart_operations - restart io operations to the specified host.
1962 * @shost: Host we are restarting.
1964 * Notes:
1965 * When we entered the error handler, we blocked all further i/o to
1966 * this device. we need to 'reverse' this process.
1968 static void scsi_restart_operations(struct Scsi_Host *shost)
1970 struct scsi_device *sdev;
1971 unsigned long flags;
1974 * If the door was locked, we need to insert a door lock request
1975 * onto the head of the SCSI request queue for the device. There
1976 * is no point trying to lock the door of an off-line device.
1978 shost_for_each_device(sdev, shost) {
1979 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
1980 scsi_eh_lock_door(sdev);
1981 sdev->was_reset = 0;
1986 * next free up anything directly waiting upon the host. this
1987 * will be requests for character device operations, and also for
1988 * ioctls to queued block devices.
1990 SCSI_LOG_ERROR_RECOVERY(3,
1991 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
1993 spin_lock_irqsave(shost->host_lock, flags);
1994 if (scsi_host_set_state(shost, SHOST_RUNNING))
1995 if (scsi_host_set_state(shost, SHOST_CANCEL))
1996 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1997 spin_unlock_irqrestore(shost->host_lock, flags);
1999 wake_up(&shost->host_wait);
2002 * finally we need to re-initiate requests that may be pending. we will
2003 * have had everything blocked while error handling is taking place, and
2004 * now that error recovery is done, we will need to ensure that these
2005 * requests are started.
2007 scsi_run_host_queues(shost);
2010 * if eh is active and host_eh_scheduled is pending we need to re-run
2011 * recovery. we do this check after scsi_run_host_queues() to allow
2012 * everything pent up since the last eh run a chance to make forward
2013 * progress before we sync again. Either we'll immediately re-run
2014 * recovery or scsi_device_unbusy() will wake us again when these
2015 * pending commands complete.
2017 spin_lock_irqsave(shost->host_lock, flags);
2018 if (shost->host_eh_scheduled)
2019 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2020 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2021 spin_unlock_irqrestore(shost->host_lock, flags);
2025 * scsi_eh_ready_devs - check device ready state and recover if not.
2026 * @shost: host to be recovered.
2027 * @work_q: &list_head for pending commands.
2028 * @done_q: &list_head for processed commands.
2030 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2031 struct list_head *work_q,
2032 struct list_head *done_q)
2034 if (!scsi_eh_stu(shost, work_q, done_q))
2035 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2036 if (!scsi_eh_target_reset(shost, work_q, done_q))
2037 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2038 if (!scsi_eh_host_reset(shost, work_q, done_q))
2039 scsi_eh_offline_sdevs(work_q,
2040 done_q);
2042 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2045 * scsi_eh_flush_done_q - finish processed commands or retry them.
2046 * @done_q: list_head of processed commands.
2048 void scsi_eh_flush_done_q(struct list_head *done_q)
2050 struct scsi_cmnd *scmd, *next;
2052 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2053 list_del_init(&scmd->eh_entry);
2054 if (scsi_device_online(scmd->device) &&
2055 !scsi_noretry_cmd(scmd) &&
2056 (++scmd->retries <= scmd->allowed)) {
2057 SCSI_LOG_ERROR_RECOVERY(3,
2058 scmd_printk(KERN_INFO, scmd,
2059 "%s: flush retry cmd\n",
2060 current->comm));
2061 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2062 } else {
2064 * If just we got sense for the device (called
2065 * scsi_eh_get_sense), scmd->result is already
2066 * set, do not set DRIVER_TIMEOUT.
2068 if (!scmd->result)
2069 scmd->result |= (DRIVER_TIMEOUT << 24);
2070 SCSI_LOG_ERROR_RECOVERY(3,
2071 scmd_printk(KERN_INFO, scmd,
2072 "%s: flush finish cmd\n",
2073 current->comm));
2074 scsi_finish_command(scmd);
2078 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2081 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2082 * @shost: Host to unjam.
2084 * Notes:
2085 * When we come in here, we *know* that all commands on the bus have
2086 * either completed, failed or timed out. we also know that no further
2087 * commands are being sent to the host, so things are relatively quiet
2088 * and we have freedom to fiddle with things as we wish.
2090 * This is only the *default* implementation. it is possible for
2091 * individual drivers to supply their own version of this function, and
2092 * if the maintainer wishes to do this, it is strongly suggested that
2093 * this function be taken as a template and modified. this function
2094 * was designed to correctly handle problems for about 95% of the
2095 * different cases out there, and it should always provide at least a
2096 * reasonable amount of error recovery.
2098 * Any command marked 'failed' or 'timeout' must eventually have
2099 * scsi_finish_cmd() called for it. we do all of the retry stuff
2100 * here, so when we restart the host after we return it should have an
2101 * empty queue.
2103 static void scsi_unjam_host(struct Scsi_Host *shost)
2105 unsigned long flags;
2106 LIST_HEAD(eh_work_q);
2107 LIST_HEAD(eh_done_q);
2109 spin_lock_irqsave(shost->host_lock, flags);
2110 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2111 spin_unlock_irqrestore(shost->host_lock, flags);
2113 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2115 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2116 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2118 spin_lock_irqsave(shost->host_lock, flags);
2119 if (shost->eh_deadline != -1)
2120 shost->last_reset = 0;
2121 spin_unlock_irqrestore(shost->host_lock, flags);
2122 scsi_eh_flush_done_q(&eh_done_q);
2126 * scsi_error_handler - SCSI error handler thread
2127 * @data: Host for which we are running.
2129 * Notes:
2130 * This is the main error handling loop. This is run as a kernel thread
2131 * for every SCSI host and handles all error handling activity.
2133 int scsi_error_handler(void *data)
2135 struct Scsi_Host *shost = data;
2138 * We use TASK_INTERRUPTIBLE so that the thread is not
2139 * counted against the load average as a running process.
2140 * We never actually get interrupted because kthread_run
2141 * disables signal delivery for the created thread.
2143 while (true) {
2145 * The sequence in kthread_stop() sets the stop flag first
2146 * then wakes the process. To avoid missed wakeups, the task
2147 * should always be in a non running state before the stop
2148 * flag is checked
2150 set_current_state(TASK_INTERRUPTIBLE);
2151 if (kthread_should_stop())
2152 break;
2154 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2155 shost->host_failed != atomic_read(&shost->host_busy)) {
2156 SCSI_LOG_ERROR_RECOVERY(1,
2157 shost_printk(KERN_INFO, shost,
2158 "scsi_eh_%d: sleeping\n",
2159 shost->host_no));
2160 schedule();
2161 continue;
2164 __set_current_state(TASK_RUNNING);
2165 SCSI_LOG_ERROR_RECOVERY(1,
2166 shost_printk(KERN_INFO, shost,
2167 "scsi_eh_%d: waking up %d/%d/%d\n",
2168 shost->host_no, shost->host_eh_scheduled,
2169 shost->host_failed,
2170 atomic_read(&shost->host_busy)));
2173 * We have a host that is failing for some reason. Figure out
2174 * what we need to do to get it up and online again (if we can).
2175 * If we fail, we end up taking the thing offline.
2177 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2178 SCSI_LOG_ERROR_RECOVERY(1,
2179 shost_printk(KERN_ERR, shost,
2180 "scsi_eh_%d: unable to autoresume\n",
2181 shost->host_no));
2182 continue;
2185 if (shost->transportt->eh_strategy_handler)
2186 shost->transportt->eh_strategy_handler(shost);
2187 else
2188 scsi_unjam_host(shost);
2190 /* All scmds have been handled */
2191 shost->host_failed = 0;
2194 * Note - if the above fails completely, the action is to take
2195 * individual devices offline and flush the queue of any
2196 * outstanding requests that may have been pending. When we
2197 * restart, we restart any I/O to any other devices on the bus
2198 * which are still online.
2200 scsi_restart_operations(shost);
2201 if (!shost->eh_noresume)
2202 scsi_autopm_put_host(shost);
2204 __set_current_state(TASK_RUNNING);
2206 SCSI_LOG_ERROR_RECOVERY(1,
2207 shost_printk(KERN_INFO, shost,
2208 "Error handler scsi_eh_%d exiting\n",
2209 shost->host_no));
2210 shost->ehandler = NULL;
2211 return 0;
2215 * Function: scsi_report_bus_reset()
2217 * Purpose: Utility function used by low-level drivers to report that
2218 * they have observed a bus reset on the bus being handled.
2220 * Arguments: shost - Host in question
2221 * channel - channel on which reset was observed.
2223 * Returns: Nothing
2225 * Lock status: Host lock must be held.
2227 * Notes: This only needs to be called if the reset is one which
2228 * originates from an unknown location. Resets originated
2229 * by the mid-level itself don't need to call this, but there
2230 * should be no harm.
2232 * The main purpose of this is to make sure that a CHECK_CONDITION
2233 * is properly treated.
2235 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2237 struct scsi_device *sdev;
2239 __shost_for_each_device(sdev, shost) {
2240 if (channel == sdev_channel(sdev))
2241 __scsi_report_device_reset(sdev, NULL);
2244 EXPORT_SYMBOL(scsi_report_bus_reset);
2247 * Function: scsi_report_device_reset()
2249 * Purpose: Utility function used by low-level drivers to report that
2250 * they have observed a device reset on the device being handled.
2252 * Arguments: shost - Host in question
2253 * channel - channel on which reset was observed
2254 * target - target on which reset was observed
2256 * Returns: Nothing
2258 * Lock status: Host lock must be held
2260 * Notes: This only needs to be called if the reset is one which
2261 * originates from an unknown location. Resets originated
2262 * by the mid-level itself don't need to call this, but there
2263 * should be no harm.
2265 * The main purpose of this is to make sure that a CHECK_CONDITION
2266 * is properly treated.
2268 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2270 struct scsi_device *sdev;
2272 __shost_for_each_device(sdev, shost) {
2273 if (channel == sdev_channel(sdev) &&
2274 target == sdev_id(sdev))
2275 __scsi_report_device_reset(sdev, NULL);
2278 EXPORT_SYMBOL(scsi_report_device_reset);
2280 static void
2281 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2286 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2287 * @dev: scsi_device to operate on
2288 * @arg: reset type (see sg.h)
2291 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2293 struct scsi_cmnd *scmd;
2294 struct Scsi_Host *shost = dev->host;
2295 struct request *rq;
2296 unsigned long flags;
2297 int error = 0, rtn, val;
2299 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2300 return -EACCES;
2302 error = get_user(val, arg);
2303 if (error)
2304 return error;
2306 if (scsi_autopm_get_host(shost) < 0)
2307 return -EIO;
2309 error = -EIO;
2310 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2311 shost->hostt->cmd_size, GFP_KERNEL);
2312 if (!rq)
2313 goto out_put_autopm_host;
2314 blk_rq_init(NULL, rq);
2316 scmd = (struct scsi_cmnd *)(rq + 1);
2317 scsi_init_command(dev, scmd);
2318 scmd->request = rq;
2319 scmd->cmnd = scsi_req(rq)->cmd;
2321 scmd->scsi_done = scsi_reset_provider_done_command;
2322 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2324 scmd->cmd_len = 0;
2326 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2328 spin_lock_irqsave(shost->host_lock, flags);
2329 shost->tmf_in_progress = 1;
2330 spin_unlock_irqrestore(shost->host_lock, flags);
2332 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2333 case SG_SCSI_RESET_NOTHING:
2334 rtn = SUCCESS;
2335 break;
2336 case SG_SCSI_RESET_DEVICE:
2337 rtn = scsi_try_bus_device_reset(scmd);
2338 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2339 break;
2340 /* FALLTHROUGH */
2341 case SG_SCSI_RESET_TARGET:
2342 rtn = scsi_try_target_reset(scmd);
2343 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2344 break;
2345 /* FALLTHROUGH */
2346 case SG_SCSI_RESET_BUS:
2347 rtn = scsi_try_bus_reset(scmd);
2348 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2349 break;
2350 /* FALLTHROUGH */
2351 case SG_SCSI_RESET_HOST:
2352 rtn = scsi_try_host_reset(scmd);
2353 if (rtn == SUCCESS)
2354 break;
2355 /* FALLTHROUGH */
2356 default:
2357 rtn = FAILED;
2358 break;
2361 error = (rtn == SUCCESS) ? 0 : -EIO;
2363 spin_lock_irqsave(shost->host_lock, flags);
2364 shost->tmf_in_progress = 0;
2365 spin_unlock_irqrestore(shost->host_lock, flags);
2368 * be sure to wake up anyone who was sleeping or had their queue
2369 * suspended while we performed the TMF.
2371 SCSI_LOG_ERROR_RECOVERY(3,
2372 shost_printk(KERN_INFO, shost,
2373 "waking up host to restart after TMF\n"));
2375 wake_up(&shost->host_wait);
2376 scsi_run_host_queues(shost);
2378 scsi_put_command(scmd);
2379 kfree(rq);
2381 out_put_autopm_host:
2382 scsi_autopm_put_host(shost);
2383 return error;
2385 EXPORT_SYMBOL(scsi_ioctl_reset);
2387 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2388 struct scsi_sense_hdr *sshdr)
2390 return scsi_normalize_sense(cmd->sense_buffer,
2391 SCSI_SENSE_BUFFERSIZE, sshdr);
2393 EXPORT_SYMBOL(scsi_command_normalize_sense);
2396 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2397 * @sense_buffer: byte array of sense data
2398 * @sb_len: number of valid bytes in sense_buffer
2399 * @info_out: pointer to 64 integer where 8 or 4 byte information
2400 * field will be placed if found.
2402 * Return value:
2403 * true if information field found, false if not found.
2405 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2406 u64 *info_out)
2408 const u8 * ucp;
2410 if (sb_len < 7)
2411 return false;
2412 switch (sense_buffer[0] & 0x7f) {
2413 case 0x70:
2414 case 0x71:
2415 if (sense_buffer[0] & 0x80) {
2416 *info_out = get_unaligned_be32(&sense_buffer[3]);
2417 return true;
2419 return false;
2420 case 0x72:
2421 case 0x73:
2422 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2423 0 /* info desc */);
2424 if (ucp && (0xa == ucp[1])) {
2425 *info_out = get_unaligned_be64(&ucp[4]);
2426 return true;
2428 return false;
2429 default:
2430 return false;
2433 EXPORT_SYMBOL(scsi_get_sense_info_fld);