of/platform: Initialise default DMA masks
[linux/fpc-iii.git] / drivers / scsi / scsi_error.c
blob8932ae81a15a7c36bf7e7800cada555d87a6bd26
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/scsi_devinfo.h>
42 #include <scsi/sg.h>
44 #include "scsi_priv.h"
45 #include "scsi_logging.h"
46 #include "scsi_transport_api.h"
48 #include <trace/events/scsi.h>
50 #include <asm/unaligned.h>
52 static void scsi_eh_done(struct scsi_cmnd *scmd);
55 * These should *probably* be handled by the host itself.
56 * Since it is allowed to sleep, it probably should.
58 #define BUS_RESET_SETTLE_TIME (10)
59 #define HOST_RESET_SETTLE_TIME (10)
61 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
62 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
63 struct scsi_cmnd *);
65 void scsi_eh_wakeup(struct Scsi_Host *shost)
67 lockdep_assert_held(shost->host_lock);
69 if (atomic_read(&shost->host_busy) == shost->host_failed) {
70 trace_scsi_eh_wakeup(shost);
71 wake_up_process(shost->ehandler);
72 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
73 "Waking error handler thread\n"));
77 /**
78 * scsi_schedule_eh - schedule EH for SCSI host
79 * @shost: SCSI host to invoke error handling on.
81 * Schedule SCSI EH without scmd.
83 void scsi_schedule_eh(struct Scsi_Host *shost)
85 unsigned long flags;
87 spin_lock_irqsave(shost->host_lock, flags);
89 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
90 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
91 shost->host_eh_scheduled++;
92 scsi_eh_wakeup(shost);
95 spin_unlock_irqrestore(shost->host_lock, flags);
97 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
99 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
101 if (!shost->last_reset || shost->eh_deadline == -1)
102 return 0;
105 * 32bit accesses are guaranteed to be atomic
106 * (on all supported architectures), so instead
107 * of using a spinlock we can as well double check
108 * if eh_deadline has been set to 'off' during the
109 * time_before call.
111 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
112 shost->eh_deadline > -1)
113 return 0;
115 return 1;
119 * scmd_eh_abort_handler - Handle command aborts
120 * @work: command to be aborted.
122 * Note: this function must be called only for a command that has timed out.
123 * Because the block layer marks a request as complete before it calls
124 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
125 * timed out do not have any effect. Hence it is safe to call
126 * scsi_finish_command() from this function.
128 void
129 scmd_eh_abort_handler(struct work_struct *work)
131 struct scsi_cmnd *scmd =
132 container_of(work, struct scsi_cmnd, abort_work.work);
133 struct scsi_device *sdev = scmd->device;
134 int rtn;
136 if (scsi_host_eh_past_deadline(sdev->host)) {
137 SCSI_LOG_ERROR_RECOVERY(3,
138 scmd_printk(KERN_INFO, scmd,
139 "eh timeout, not aborting\n"));
140 } else {
141 SCSI_LOG_ERROR_RECOVERY(3,
142 scmd_printk(KERN_INFO, scmd,
143 "aborting command\n"));
144 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
145 if (rtn == SUCCESS) {
146 set_host_byte(scmd, DID_TIME_OUT);
147 if (scsi_host_eh_past_deadline(sdev->host)) {
148 SCSI_LOG_ERROR_RECOVERY(3,
149 scmd_printk(KERN_INFO, scmd,
150 "eh timeout, not retrying "
151 "aborted command\n"));
152 } else if (!scsi_noretry_cmd(scmd) &&
153 (++scmd->retries <= scmd->allowed)) {
154 SCSI_LOG_ERROR_RECOVERY(3,
155 scmd_printk(KERN_WARNING, scmd,
156 "retry aborted command\n"));
157 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
158 return;
159 } else {
160 SCSI_LOG_ERROR_RECOVERY(3,
161 scmd_printk(KERN_WARNING, scmd,
162 "finish aborted command\n"));
163 scsi_finish_command(scmd);
164 return;
166 } else {
167 SCSI_LOG_ERROR_RECOVERY(3,
168 scmd_printk(KERN_INFO, scmd,
169 "cmd abort %s\n",
170 (rtn == FAST_IO_FAIL) ?
171 "not send" : "failed"));
175 scsi_eh_scmd_add(scmd);
179 * scsi_abort_command - schedule a command abort
180 * @scmd: scmd to abort.
182 * We only need to abort commands after a command timeout
184 static int
185 scsi_abort_command(struct scsi_cmnd *scmd)
187 struct scsi_device *sdev = scmd->device;
188 struct Scsi_Host *shost = sdev->host;
189 unsigned long flags;
191 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
193 * Retry after abort failed, escalate to next level.
195 SCSI_LOG_ERROR_RECOVERY(3,
196 scmd_printk(KERN_INFO, scmd,
197 "previous abort failed\n"));
198 BUG_ON(delayed_work_pending(&scmd->abort_work));
199 return FAILED;
202 spin_lock_irqsave(shost->host_lock, flags);
203 if (shost->eh_deadline != -1 && !shost->last_reset)
204 shost->last_reset = jiffies;
205 spin_unlock_irqrestore(shost->host_lock, flags);
207 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
208 SCSI_LOG_ERROR_RECOVERY(3,
209 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
210 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
211 return SUCCESS;
215 * scsi_eh_reset - call into ->eh_action to reset internal counters
216 * @scmd: scmd to run eh on.
218 * The scsi driver might be carrying internal state about the
219 * devices, so we need to call into the driver to reset the
220 * internal state once the error handler is started.
222 static void scsi_eh_reset(struct scsi_cmnd *scmd)
224 if (!blk_rq_is_passthrough(scmd->request)) {
225 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
226 if (sdrv->eh_reset)
227 sdrv->eh_reset(scmd);
231 static void scsi_eh_inc_host_failed(struct rcu_head *head)
233 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
234 struct Scsi_Host *shost = scmd->device->host;
235 unsigned long flags;
237 spin_lock_irqsave(shost->host_lock, flags);
238 shost->host_failed++;
239 scsi_eh_wakeup(shost);
240 spin_unlock_irqrestore(shost->host_lock, flags);
244 * scsi_eh_scmd_add - add scsi cmd to error handling.
245 * @scmd: scmd to run eh on.
247 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
249 struct Scsi_Host *shost = scmd->device->host;
250 unsigned long flags;
251 int ret;
253 WARN_ON_ONCE(!shost->ehandler);
255 spin_lock_irqsave(shost->host_lock, flags);
256 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
257 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
258 WARN_ON_ONCE(ret);
260 if (shost->eh_deadline != -1 && !shost->last_reset)
261 shost->last_reset = jiffies;
263 scsi_eh_reset(scmd);
264 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
265 spin_unlock_irqrestore(shost->host_lock, flags);
267 * Ensure that all tasks observe the host state change before the
268 * host_failed change.
270 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
274 * scsi_times_out - Timeout function for normal scsi commands.
275 * @req: request that is timing out.
277 * Notes:
278 * We do not need to lock this. There is the potential for a race
279 * only in that the normal completion handling might run, but if the
280 * normal completion function determines that the timer has already
281 * fired, then it mustn't do anything.
283 enum blk_eh_timer_return scsi_times_out(struct request *req)
285 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
286 enum blk_eh_timer_return rtn = BLK_EH_DONE;
287 struct Scsi_Host *host = scmd->device->host;
289 trace_scsi_dispatch_cmd_timeout(scmd);
290 scsi_log_completion(scmd, TIMEOUT_ERROR);
292 if (host->eh_deadline != -1 && !host->last_reset)
293 host->last_reset = jiffies;
295 if (host->hostt->eh_timed_out)
296 rtn = host->hostt->eh_timed_out(scmd);
298 if (rtn == BLK_EH_DONE) {
299 if (scsi_abort_command(scmd) != SUCCESS) {
300 set_host_byte(scmd, DID_TIME_OUT);
301 scsi_eh_scmd_add(scmd);
305 return rtn;
309 * scsi_block_when_processing_errors - Prevent cmds from being queued.
310 * @sdev: Device on which we are performing recovery.
312 * Description:
313 * We block until the host is out of error recovery, and then check to
314 * see whether the host or the device is offline.
316 * Return value:
317 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
319 int scsi_block_when_processing_errors(struct scsi_device *sdev)
321 int online;
323 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
325 online = scsi_device_online(sdev);
327 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
328 "%s: rtn: %d\n", __func__, online));
330 return online;
332 EXPORT_SYMBOL(scsi_block_when_processing_errors);
334 #ifdef CONFIG_SCSI_LOGGING
336 * scsi_eh_prt_fail_stats - Log info on failures.
337 * @shost: scsi host being recovered.
338 * @work_q: Queue of scsi cmds to process.
340 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
341 struct list_head *work_q)
343 struct scsi_cmnd *scmd;
344 struct scsi_device *sdev;
345 int total_failures = 0;
346 int cmd_failed = 0;
347 int cmd_cancel = 0;
348 int devices_failed = 0;
350 shost_for_each_device(sdev, shost) {
351 list_for_each_entry(scmd, work_q, eh_entry) {
352 if (scmd->device == sdev) {
353 ++total_failures;
354 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
355 ++cmd_cancel;
356 else
357 ++cmd_failed;
361 if (cmd_cancel || cmd_failed) {
362 SCSI_LOG_ERROR_RECOVERY(3,
363 shost_printk(KERN_INFO, shost,
364 "%s: cmds failed: %d, cancel: %d\n",
365 __func__, cmd_failed,
366 cmd_cancel));
367 cmd_cancel = 0;
368 cmd_failed = 0;
369 ++devices_failed;
373 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
374 "Total of %d commands on %d"
375 " devices require eh work\n",
376 total_failures, devices_failed));
378 #endif
381 * scsi_report_lun_change - Set flag on all *other* devices on the same target
382 * to indicate that a UNIT ATTENTION is expected.
383 * @sdev: Device reporting the UNIT ATTENTION
385 static void scsi_report_lun_change(struct scsi_device *sdev)
387 sdev->sdev_target->expecting_lun_change = 1;
391 * scsi_report_sense - Examine scsi sense information and log messages for
392 * certain conditions, also issue uevents for some of them.
393 * @sdev: Device reporting the sense code
394 * @sshdr: sshdr to be examined
396 static void scsi_report_sense(struct scsi_device *sdev,
397 struct scsi_sense_hdr *sshdr)
399 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
401 if (sshdr->sense_key == UNIT_ATTENTION) {
402 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
403 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
404 sdev_printk(KERN_WARNING, sdev,
405 "Inquiry data has changed");
406 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
407 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
408 scsi_report_lun_change(sdev);
409 sdev_printk(KERN_WARNING, sdev,
410 "Warning! Received an indication that the "
411 "LUN assignments on this target have "
412 "changed. The Linux SCSI layer does not "
413 "automatically remap LUN assignments.\n");
414 } else if (sshdr->asc == 0x3f)
415 sdev_printk(KERN_WARNING, sdev,
416 "Warning! Received an indication that the "
417 "operating parameters on this target have "
418 "changed. The Linux SCSI layer does not "
419 "automatically adjust these parameters.\n");
421 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
422 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
423 sdev_printk(KERN_WARNING, sdev,
424 "Warning! Received an indication that the "
425 "LUN reached a thin provisioning soft "
426 "threshold.\n");
429 if (sshdr->asc == 0x29) {
430 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
431 sdev_printk(KERN_WARNING, sdev,
432 "Power-on or device reset occurred\n");
435 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
436 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
437 sdev_printk(KERN_WARNING, sdev,
438 "Mode parameters changed");
439 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
440 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
441 sdev_printk(KERN_WARNING, sdev,
442 "Asymmetric access state changed");
443 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
444 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
445 sdev_printk(KERN_WARNING, sdev,
446 "Capacity data has changed");
447 } else if (sshdr->asc == 0x2a)
448 sdev_printk(KERN_WARNING, sdev,
449 "Parameters changed");
452 if (evt_type != SDEV_EVT_MAXBITS) {
453 set_bit(evt_type, sdev->pending_events);
454 schedule_work(&sdev->event_work);
459 * scsi_check_sense - Examine scsi cmd sense
460 * @scmd: Cmd to have sense checked.
462 * Return value:
463 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
465 * Notes:
466 * When a deferred error is detected the current command has
467 * not been executed and needs retrying.
469 int scsi_check_sense(struct scsi_cmnd *scmd)
471 struct scsi_device *sdev = scmd->device;
472 struct scsi_sense_hdr sshdr;
474 if (! scsi_command_normalize_sense(scmd, &sshdr))
475 return FAILED; /* no valid sense data */
477 scsi_report_sense(sdev, &sshdr);
479 if (scsi_sense_is_deferred(&sshdr))
480 return NEEDS_RETRY;
482 if (sdev->handler && sdev->handler->check_sense) {
483 int rc;
485 rc = sdev->handler->check_sense(sdev, &sshdr);
486 if (rc != SCSI_RETURN_NOT_HANDLED)
487 return rc;
488 /* handler does not care. Drop down to default handling */
491 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
493 * nasty: for mid-layer issued TURs, we need to return the
494 * actual sense data without any recovery attempt. For eh
495 * issued ones, we need to try to recover and interpret
497 return SUCCESS;
500 * Previous logic looked for FILEMARK, EOM or ILI which are
501 * mainly associated with tapes and returned SUCCESS.
503 if (sshdr.response_code == 0x70) {
504 /* fixed format */
505 if (scmd->sense_buffer[2] & 0xe0)
506 return SUCCESS;
507 } else {
509 * descriptor format: look for "stream commands sense data
510 * descriptor" (see SSC-3). Assume single sense data
511 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
513 if ((sshdr.additional_length > 3) &&
514 (scmd->sense_buffer[8] == 0x4) &&
515 (scmd->sense_buffer[11] & 0xe0))
516 return SUCCESS;
519 switch (sshdr.sense_key) {
520 case NO_SENSE:
521 return SUCCESS;
522 case RECOVERED_ERROR:
523 return /* soft_error */ SUCCESS;
525 case ABORTED_COMMAND:
526 if (sshdr.asc == 0x10) /* DIF */
527 return SUCCESS;
529 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
530 return ADD_TO_MLQUEUE;
531 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
532 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
533 return ADD_TO_MLQUEUE;
535 return NEEDS_RETRY;
536 case NOT_READY:
537 case UNIT_ATTENTION:
539 * if we are expecting a cc/ua because of a bus reset that we
540 * performed, treat this just as a retry. otherwise this is
541 * information that we should pass up to the upper-level driver
542 * so that we can deal with it there.
544 if (scmd->device->expecting_cc_ua) {
546 * Because some device does not queue unit
547 * attentions correctly, we carefully check
548 * additional sense code and qualifier so as
549 * not to squash media change unit attention.
551 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
552 scmd->device->expecting_cc_ua = 0;
553 return NEEDS_RETRY;
557 * we might also expect a cc/ua if another LUN on the target
558 * reported a UA with an ASC/ASCQ of 3F 0E -
559 * REPORTED LUNS DATA HAS CHANGED.
561 if (scmd->device->sdev_target->expecting_lun_change &&
562 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
563 return NEEDS_RETRY;
565 * if the device is in the process of becoming ready, we
566 * should retry.
568 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
569 return NEEDS_RETRY;
571 * if the device is not started, we need to wake
572 * the error handler to start the motor
574 if (scmd->device->allow_restart &&
575 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
576 return FAILED;
578 * Pass the UA upwards for a determination in the completion
579 * functions.
581 return SUCCESS;
583 /* these are not supported */
584 case DATA_PROTECT:
585 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
586 /* Thin provisioning hard threshold reached */
587 set_host_byte(scmd, DID_ALLOC_FAILURE);
588 return SUCCESS;
590 /* FALLTHROUGH */
591 case COPY_ABORTED:
592 case VOLUME_OVERFLOW:
593 case MISCOMPARE:
594 case BLANK_CHECK:
595 set_host_byte(scmd, DID_TARGET_FAILURE);
596 return SUCCESS;
598 case MEDIUM_ERROR:
599 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
600 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
601 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
602 set_host_byte(scmd, DID_MEDIUM_ERROR);
603 return SUCCESS;
605 return NEEDS_RETRY;
607 case HARDWARE_ERROR:
608 if (scmd->device->retry_hwerror)
609 return ADD_TO_MLQUEUE;
610 else
611 set_host_byte(scmd, DID_TARGET_FAILURE);
612 /* FALLTHROUGH */
614 case ILLEGAL_REQUEST:
615 if (sshdr.asc == 0x20 || /* Invalid command operation code */
616 sshdr.asc == 0x21 || /* Logical block address out of range */
617 sshdr.asc == 0x22 || /* Invalid function */
618 sshdr.asc == 0x24 || /* Invalid field in cdb */
619 sshdr.asc == 0x26 || /* Parameter value invalid */
620 sshdr.asc == 0x27) { /* Write protected */
621 set_host_byte(scmd, DID_TARGET_FAILURE);
623 return SUCCESS;
625 default:
626 return SUCCESS;
629 EXPORT_SYMBOL_GPL(scsi_check_sense);
631 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
633 struct scsi_host_template *sht = sdev->host->hostt;
634 struct scsi_device *tmp_sdev;
636 if (!sht->track_queue_depth ||
637 sdev->queue_depth >= sdev->max_queue_depth)
638 return;
640 if (time_before(jiffies,
641 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
642 return;
644 if (time_before(jiffies,
645 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
646 return;
649 * Walk all devices of a target and do
650 * ramp up on them.
652 shost_for_each_device(tmp_sdev, sdev->host) {
653 if (tmp_sdev->channel != sdev->channel ||
654 tmp_sdev->id != sdev->id ||
655 tmp_sdev->queue_depth == sdev->max_queue_depth)
656 continue;
658 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
659 sdev->last_queue_ramp_up = jiffies;
663 static void scsi_handle_queue_full(struct scsi_device *sdev)
665 struct scsi_host_template *sht = sdev->host->hostt;
666 struct scsi_device *tmp_sdev;
668 if (!sht->track_queue_depth)
669 return;
671 shost_for_each_device(tmp_sdev, sdev->host) {
672 if (tmp_sdev->channel != sdev->channel ||
673 tmp_sdev->id != sdev->id)
674 continue;
676 * We do not know the number of commands that were at
677 * the device when we got the queue full so we start
678 * from the highest possible value and work our way down.
680 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
685 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
686 * @scmd: SCSI cmd to examine.
688 * Notes:
689 * This is *only* called when we are examining the status of commands
690 * queued during error recovery. the main difference here is that we
691 * don't allow for the possibility of retries here, and we are a lot
692 * more restrictive about what we consider acceptable.
694 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
697 * first check the host byte, to see if there is anything in there
698 * that would indicate what we need to do.
700 if (host_byte(scmd->result) == DID_RESET) {
702 * rats. we are already in the error handler, so we now
703 * get to try and figure out what to do next. if the sense
704 * is valid, we have a pretty good idea of what to do.
705 * if not, we mark it as FAILED.
707 return scsi_check_sense(scmd);
709 if (host_byte(scmd->result) != DID_OK)
710 return FAILED;
713 * next, check the message byte.
715 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
716 return FAILED;
719 * now, check the status byte to see if this indicates
720 * anything special.
722 switch (status_byte(scmd->result)) {
723 case GOOD:
724 scsi_handle_queue_ramp_up(scmd->device);
725 /* FALLTHROUGH */
726 case COMMAND_TERMINATED:
727 return SUCCESS;
728 case CHECK_CONDITION:
729 return scsi_check_sense(scmd);
730 case CONDITION_GOOD:
731 case INTERMEDIATE_GOOD:
732 case INTERMEDIATE_C_GOOD:
734 * who knows? FIXME(eric)
736 return SUCCESS;
737 case RESERVATION_CONFLICT:
738 if (scmd->cmnd[0] == TEST_UNIT_READY)
739 /* it is a success, we probed the device and
740 * found it */
741 return SUCCESS;
742 /* otherwise, we failed to send the command */
743 return FAILED;
744 case QUEUE_FULL:
745 scsi_handle_queue_full(scmd->device);
746 /* fall through */
747 case BUSY:
748 return NEEDS_RETRY;
749 default:
750 return FAILED;
752 return FAILED;
756 * scsi_eh_done - Completion function for error handling.
757 * @scmd: Cmd that is done.
759 static void scsi_eh_done(struct scsi_cmnd *scmd)
761 struct completion *eh_action;
763 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
764 "%s result: %x\n", __func__, scmd->result));
766 eh_action = scmd->device->host->eh_action;
767 if (eh_action)
768 complete(eh_action);
772 * scsi_try_host_reset - ask host adapter to reset itself
773 * @scmd: SCSI cmd to send host reset.
775 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
777 unsigned long flags;
778 int rtn;
779 struct Scsi_Host *host = scmd->device->host;
780 struct scsi_host_template *hostt = host->hostt;
782 SCSI_LOG_ERROR_RECOVERY(3,
783 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
785 if (!hostt->eh_host_reset_handler)
786 return FAILED;
788 rtn = hostt->eh_host_reset_handler(scmd);
790 if (rtn == SUCCESS) {
791 if (!hostt->skip_settle_delay)
792 ssleep(HOST_RESET_SETTLE_TIME);
793 spin_lock_irqsave(host->host_lock, flags);
794 scsi_report_bus_reset(host, scmd_channel(scmd));
795 spin_unlock_irqrestore(host->host_lock, flags);
798 return rtn;
802 * scsi_try_bus_reset - ask host to perform a bus reset
803 * @scmd: SCSI cmd to send bus reset.
805 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
807 unsigned long flags;
808 int rtn;
809 struct Scsi_Host *host = scmd->device->host;
810 struct scsi_host_template *hostt = host->hostt;
812 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
813 "%s: Snd Bus RST\n", __func__));
815 if (!hostt->eh_bus_reset_handler)
816 return FAILED;
818 rtn = hostt->eh_bus_reset_handler(scmd);
820 if (rtn == SUCCESS) {
821 if (!hostt->skip_settle_delay)
822 ssleep(BUS_RESET_SETTLE_TIME);
823 spin_lock_irqsave(host->host_lock, flags);
824 scsi_report_bus_reset(host, scmd_channel(scmd));
825 spin_unlock_irqrestore(host->host_lock, flags);
828 return rtn;
831 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
833 sdev->was_reset = 1;
834 sdev->expecting_cc_ua = 1;
838 * scsi_try_target_reset - Ask host to perform a target reset
839 * @scmd: SCSI cmd used to send a target reset
841 * Notes:
842 * There is no timeout for this operation. if this operation is
843 * unreliable for a given host, then the host itself needs to put a
844 * timer on it, and set the host back to a consistent state prior to
845 * returning.
847 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
849 unsigned long flags;
850 int rtn;
851 struct Scsi_Host *host = scmd->device->host;
852 struct scsi_host_template *hostt = host->hostt;
854 if (!hostt->eh_target_reset_handler)
855 return FAILED;
857 rtn = hostt->eh_target_reset_handler(scmd);
858 if (rtn == SUCCESS) {
859 spin_lock_irqsave(host->host_lock, flags);
860 __starget_for_each_device(scsi_target(scmd->device), NULL,
861 __scsi_report_device_reset);
862 spin_unlock_irqrestore(host->host_lock, flags);
865 return rtn;
869 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
870 * @scmd: SCSI cmd used to send BDR
872 * Notes:
873 * There is no timeout for this operation. if this operation is
874 * unreliable for a given host, then the host itself needs to put a
875 * timer on it, and set the host back to a consistent state prior to
876 * returning.
878 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
880 int rtn;
881 struct scsi_host_template *hostt = scmd->device->host->hostt;
883 if (!hostt->eh_device_reset_handler)
884 return FAILED;
886 rtn = hostt->eh_device_reset_handler(scmd);
887 if (rtn == SUCCESS)
888 __scsi_report_device_reset(scmd->device, NULL);
889 return rtn;
893 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
894 * @hostt: SCSI driver host template
895 * @scmd: SCSI cmd used to send a target reset
897 * Return value:
898 * SUCCESS, FAILED, or FAST_IO_FAIL
900 * Notes:
901 * SUCCESS does not necessarily indicate that the command
902 * has been aborted; it only indicates that the LLDDs
903 * has cleared all references to that command.
904 * LLDDs should return FAILED only if an abort was required
905 * but could not be executed. LLDDs should return FAST_IO_FAIL
906 * if the device is temporarily unavailable (eg due to a
907 * link down on FibreChannel)
909 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
910 struct scsi_cmnd *scmd)
912 if (!hostt->eh_abort_handler)
913 return FAILED;
915 return hostt->eh_abort_handler(scmd);
918 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
920 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
921 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
922 if (scsi_try_target_reset(scmd) != SUCCESS)
923 if (scsi_try_bus_reset(scmd) != SUCCESS)
924 scsi_try_host_reset(scmd);
928 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
929 * @scmd: SCSI command structure to hijack
930 * @ses: structure to save restore information
931 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
932 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
933 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
935 * This function is used to save a scsi command information before re-execution
936 * as part of the error recovery process. If @sense_bytes is 0 the command
937 * sent must be one that does not transfer any data. If @sense_bytes != 0
938 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
939 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
941 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
942 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
944 struct scsi_device *sdev = scmd->device;
947 * We need saved copies of a number of fields - this is because
948 * error handling may need to overwrite these with different values
949 * to run different commands, and once error handling is complete,
950 * we will need to restore these values prior to running the actual
951 * command.
953 ses->cmd_len = scmd->cmd_len;
954 ses->cmnd = scmd->cmnd;
955 ses->data_direction = scmd->sc_data_direction;
956 ses->sdb = scmd->sdb;
957 ses->next_rq = scmd->request->next_rq;
958 ses->result = scmd->result;
959 ses->underflow = scmd->underflow;
960 ses->prot_op = scmd->prot_op;
961 ses->eh_eflags = scmd->eh_eflags;
963 scmd->prot_op = SCSI_PROT_NORMAL;
964 scmd->eh_eflags = 0;
965 scmd->cmnd = ses->eh_cmnd;
966 memset(scmd->cmnd, 0, BLK_MAX_CDB);
967 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
968 scmd->request->next_rq = NULL;
969 scmd->result = 0;
971 if (sense_bytes) {
972 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
973 sense_bytes);
974 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
975 scmd->sdb.length);
976 scmd->sdb.table.sgl = &ses->sense_sgl;
977 scmd->sc_data_direction = DMA_FROM_DEVICE;
978 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
979 scmd->cmnd[0] = REQUEST_SENSE;
980 scmd->cmnd[4] = scmd->sdb.length;
981 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
982 } else {
983 scmd->sc_data_direction = DMA_NONE;
984 if (cmnd) {
985 BUG_ON(cmnd_size > BLK_MAX_CDB);
986 memcpy(scmd->cmnd, cmnd, cmnd_size);
987 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
991 scmd->underflow = 0;
993 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
994 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
995 (sdev->lun << 5 & 0xe0);
998 * Zero the sense buffer. The scsi spec mandates that any
999 * untransferred sense data should be interpreted as being zero.
1001 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1003 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1006 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1007 * @scmd: SCSI command structure to restore
1008 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1010 * Undo any damage done by above scsi_eh_prep_cmnd().
1012 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1015 * Restore original data
1017 scmd->cmd_len = ses->cmd_len;
1018 scmd->cmnd = ses->cmnd;
1019 scmd->sc_data_direction = ses->data_direction;
1020 scmd->sdb = ses->sdb;
1021 scmd->request->next_rq = ses->next_rq;
1022 scmd->result = ses->result;
1023 scmd->underflow = ses->underflow;
1024 scmd->prot_op = ses->prot_op;
1025 scmd->eh_eflags = ses->eh_eflags;
1027 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1030 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1031 * @scmd: SCSI command structure to hijack
1032 * @cmnd: CDB to send
1033 * @cmnd_size: size in bytes of @cmnd
1034 * @timeout: timeout for this request
1035 * @sense_bytes: size of sense data to copy or 0
1037 * This function is used to send a scsi command down to a target device
1038 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1040 * Return value:
1041 * SUCCESS or FAILED or NEEDS_RETRY
1043 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1044 int cmnd_size, int timeout, unsigned sense_bytes)
1046 struct scsi_device *sdev = scmd->device;
1047 struct Scsi_Host *shost = sdev->host;
1048 DECLARE_COMPLETION_ONSTACK(done);
1049 unsigned long timeleft = timeout;
1050 struct scsi_eh_save ses;
1051 const unsigned long stall_for = msecs_to_jiffies(100);
1052 int rtn;
1054 retry:
1055 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1056 shost->eh_action = &done;
1058 scsi_log_send(scmd);
1059 scmd->scsi_done = scsi_eh_done;
1060 rtn = shost->hostt->queuecommand(shost, scmd);
1061 if (rtn) {
1062 if (timeleft > stall_for) {
1063 scsi_eh_restore_cmnd(scmd, &ses);
1064 timeleft -= stall_for;
1065 msleep(jiffies_to_msecs(stall_for));
1066 goto retry;
1068 /* signal not to enter either branch of the if () below */
1069 timeleft = 0;
1070 rtn = FAILED;
1071 } else {
1072 timeleft = wait_for_completion_timeout(&done, timeout);
1073 rtn = SUCCESS;
1076 shost->eh_action = NULL;
1078 scsi_log_completion(scmd, rtn);
1080 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1081 "%s timeleft: %ld\n",
1082 __func__, timeleft));
1085 * If there is time left scsi_eh_done got called, and we will examine
1086 * the actual status codes to see whether the command actually did
1087 * complete normally, else if we have a zero return and no time left,
1088 * the command must still be pending, so abort it and return FAILED.
1089 * If we never actually managed to issue the command, because
1090 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1091 * value above (so don't execute either branch of the if)
1093 if (timeleft) {
1094 rtn = scsi_eh_completed_normally(scmd);
1095 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1096 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1098 switch (rtn) {
1099 case SUCCESS:
1100 case NEEDS_RETRY:
1101 case FAILED:
1102 break;
1103 case ADD_TO_MLQUEUE:
1104 rtn = NEEDS_RETRY;
1105 break;
1106 default:
1107 rtn = FAILED;
1108 break;
1110 } else if (rtn != FAILED) {
1111 scsi_abort_eh_cmnd(scmd);
1112 rtn = FAILED;
1115 scsi_eh_restore_cmnd(scmd, &ses);
1117 return rtn;
1121 * scsi_request_sense - Request sense data from a particular target.
1122 * @scmd: SCSI cmd for request sense.
1124 * Notes:
1125 * Some hosts automatically obtain this information, others require
1126 * that we obtain it on our own. This function will *not* return until
1127 * the command either times out, or it completes.
1129 static int scsi_request_sense(struct scsi_cmnd *scmd)
1131 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1134 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1136 if (!blk_rq_is_passthrough(scmd->request)) {
1137 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1138 if (sdrv->eh_action)
1139 rtn = sdrv->eh_action(scmd, rtn);
1141 return rtn;
1145 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1146 * @scmd: Original SCSI cmd that eh has finished.
1147 * @done_q: Queue for processed commands.
1149 * Notes:
1150 * We don't want to use the normal command completion while we are are
1151 * still handling errors - it may cause other commands to be queued,
1152 * and that would disturb what we are doing. Thus we really want to
1153 * keep a list of pending commands for final completion, and once we
1154 * are ready to leave error handling we handle completion for real.
1156 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1158 list_move_tail(&scmd->eh_entry, done_q);
1160 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1163 * scsi_eh_get_sense - Get device sense data.
1164 * @work_q: Queue of commands to process.
1165 * @done_q: Queue of processed commands.
1167 * Description:
1168 * See if we need to request sense information. if so, then get it
1169 * now, so we have a better idea of what to do.
1171 * Notes:
1172 * This has the unfortunate side effect that if a shost adapter does
1173 * not automatically request sense information, we end up shutting
1174 * it down before we request it.
1176 * All drivers should request sense information internally these days,
1177 * so for now all I have to say is tough noogies if you end up in here.
1179 * XXX: Long term this code should go away, but that needs an audit of
1180 * all LLDDs first.
1182 int scsi_eh_get_sense(struct list_head *work_q,
1183 struct list_head *done_q)
1185 struct scsi_cmnd *scmd, *next;
1186 struct Scsi_Host *shost;
1187 int rtn;
1190 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1191 * should not get sense.
1193 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1194 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1195 SCSI_SENSE_VALID(scmd))
1196 continue;
1198 shost = scmd->device->host;
1199 if (scsi_host_eh_past_deadline(shost)) {
1200 SCSI_LOG_ERROR_RECOVERY(3,
1201 scmd_printk(KERN_INFO, scmd,
1202 "%s: skip request sense, past eh deadline\n",
1203 current->comm));
1204 break;
1206 if (status_byte(scmd->result) != CHECK_CONDITION)
1208 * don't request sense if there's no check condition
1209 * status because the error we're processing isn't one
1210 * that has a sense code (and some devices get
1211 * confused by sense requests out of the blue)
1213 continue;
1215 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1216 "%s: requesting sense\n",
1217 current->comm));
1218 rtn = scsi_request_sense(scmd);
1219 if (rtn != SUCCESS)
1220 continue;
1222 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1223 "sense requested, result %x\n", scmd->result));
1224 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1226 rtn = scsi_decide_disposition(scmd);
1229 * if the result was normal, then just pass it along to the
1230 * upper level.
1232 if (rtn == SUCCESS)
1233 /* we don't want this command reissued, just
1234 * finished with the sense data, so set
1235 * retries to the max allowed to ensure it
1236 * won't get reissued */
1237 scmd->retries = scmd->allowed;
1238 else if (rtn != NEEDS_RETRY)
1239 continue;
1241 scsi_eh_finish_cmd(scmd, done_q);
1244 return list_empty(work_q);
1246 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1249 * scsi_eh_tur - Send TUR to device.
1250 * @scmd: &scsi_cmnd to send TUR
1252 * Return value:
1253 * 0 - Device is ready. 1 - Device NOT ready.
1255 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1257 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1258 int retry_cnt = 1, rtn;
1260 retry_tur:
1261 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1262 scmd->device->eh_timeout, 0);
1264 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1265 "%s return: %x\n", __func__, rtn));
1267 switch (rtn) {
1268 case NEEDS_RETRY:
1269 if (retry_cnt--)
1270 goto retry_tur;
1271 /*FALLTHRU*/
1272 case SUCCESS:
1273 return 0;
1274 default:
1275 return 1;
1280 * scsi_eh_test_devices - check if devices are responding from error recovery.
1281 * @cmd_list: scsi commands in error recovery.
1282 * @work_q: queue for commands which still need more error recovery
1283 * @done_q: queue for commands which are finished
1284 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1286 * Decription:
1287 * Tests if devices are in a working state. Commands to devices now in
1288 * a working state are sent to the done_q while commands to devices which
1289 * are still failing to respond are returned to the work_q for more
1290 * processing.
1292 static int scsi_eh_test_devices(struct list_head *cmd_list,
1293 struct list_head *work_q,
1294 struct list_head *done_q, int try_stu)
1296 struct scsi_cmnd *scmd, *next;
1297 struct scsi_device *sdev;
1298 int finish_cmds;
1300 while (!list_empty(cmd_list)) {
1301 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1302 sdev = scmd->device;
1304 if (!try_stu) {
1305 if (scsi_host_eh_past_deadline(sdev->host)) {
1306 /* Push items back onto work_q */
1307 list_splice_init(cmd_list, work_q);
1308 SCSI_LOG_ERROR_RECOVERY(3,
1309 sdev_printk(KERN_INFO, sdev,
1310 "%s: skip test device, past eh deadline",
1311 current->comm));
1312 break;
1316 finish_cmds = !scsi_device_online(scmd->device) ||
1317 (try_stu && !scsi_eh_try_stu(scmd) &&
1318 !scsi_eh_tur(scmd)) ||
1319 !scsi_eh_tur(scmd);
1321 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1322 if (scmd->device == sdev) {
1323 if (finish_cmds &&
1324 (try_stu ||
1325 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1326 scsi_eh_finish_cmd(scmd, done_q);
1327 else
1328 list_move_tail(&scmd->eh_entry, work_q);
1331 return list_empty(work_q);
1335 * scsi_eh_try_stu - Send START_UNIT to device.
1336 * @scmd: &scsi_cmnd to send START_UNIT
1338 * Return value:
1339 * 0 - Device is ready. 1 - Device NOT ready.
1341 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1343 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1345 if (scmd->device->allow_restart) {
1346 int i, rtn = NEEDS_RETRY;
1348 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1349 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1351 if (rtn == SUCCESS)
1352 return 0;
1355 return 1;
1359 * scsi_eh_stu - send START_UNIT if needed
1360 * @shost: &scsi host being recovered.
1361 * @work_q: &list_head for pending commands.
1362 * @done_q: &list_head for processed commands.
1364 * Notes:
1365 * If commands are failing due to not ready, initializing command required,
1366 * try revalidating the device, which will end up sending a start unit.
1368 static int scsi_eh_stu(struct Scsi_Host *shost,
1369 struct list_head *work_q,
1370 struct list_head *done_q)
1372 struct scsi_cmnd *scmd, *stu_scmd, *next;
1373 struct scsi_device *sdev;
1375 shost_for_each_device(sdev, shost) {
1376 if (scsi_host_eh_past_deadline(shost)) {
1377 SCSI_LOG_ERROR_RECOVERY(3,
1378 sdev_printk(KERN_INFO, sdev,
1379 "%s: skip START_UNIT, past eh deadline\n",
1380 current->comm));
1381 break;
1383 stu_scmd = NULL;
1384 list_for_each_entry(scmd, work_q, eh_entry)
1385 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1386 scsi_check_sense(scmd) == FAILED ) {
1387 stu_scmd = scmd;
1388 break;
1391 if (!stu_scmd)
1392 continue;
1394 SCSI_LOG_ERROR_RECOVERY(3,
1395 sdev_printk(KERN_INFO, sdev,
1396 "%s: Sending START_UNIT\n",
1397 current->comm));
1399 if (!scsi_eh_try_stu(stu_scmd)) {
1400 if (!scsi_device_online(sdev) ||
1401 !scsi_eh_tur(stu_scmd)) {
1402 list_for_each_entry_safe(scmd, next,
1403 work_q, eh_entry) {
1404 if (scmd->device == sdev &&
1405 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1406 scsi_eh_finish_cmd(scmd, done_q);
1409 } else {
1410 SCSI_LOG_ERROR_RECOVERY(3,
1411 sdev_printk(KERN_INFO, sdev,
1412 "%s: START_UNIT failed\n",
1413 current->comm));
1417 return list_empty(work_q);
1422 * scsi_eh_bus_device_reset - send bdr if needed
1423 * @shost: scsi host being recovered.
1424 * @work_q: &list_head for pending commands.
1425 * @done_q: &list_head for processed commands.
1427 * Notes:
1428 * Try a bus device reset. Still, look to see whether we have multiple
1429 * devices that are jammed or not - if we have multiple devices, it
1430 * makes no sense to try bus_device_reset - we really would need to try
1431 * a bus_reset instead.
1433 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1434 struct list_head *work_q,
1435 struct list_head *done_q)
1437 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1438 struct scsi_device *sdev;
1439 int rtn;
1441 shost_for_each_device(sdev, shost) {
1442 if (scsi_host_eh_past_deadline(shost)) {
1443 SCSI_LOG_ERROR_RECOVERY(3,
1444 sdev_printk(KERN_INFO, sdev,
1445 "%s: skip BDR, past eh deadline\n",
1446 current->comm));
1447 break;
1449 bdr_scmd = NULL;
1450 list_for_each_entry(scmd, work_q, eh_entry)
1451 if (scmd->device == sdev) {
1452 bdr_scmd = scmd;
1453 break;
1456 if (!bdr_scmd)
1457 continue;
1459 SCSI_LOG_ERROR_RECOVERY(3,
1460 sdev_printk(KERN_INFO, sdev,
1461 "%s: Sending BDR\n", current->comm));
1462 rtn = scsi_try_bus_device_reset(bdr_scmd);
1463 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1464 if (!scsi_device_online(sdev) ||
1465 rtn == FAST_IO_FAIL ||
1466 !scsi_eh_tur(bdr_scmd)) {
1467 list_for_each_entry_safe(scmd, next,
1468 work_q, eh_entry) {
1469 if (scmd->device == sdev &&
1470 scsi_eh_action(scmd, rtn) != FAILED)
1471 scsi_eh_finish_cmd(scmd,
1472 done_q);
1475 } else {
1476 SCSI_LOG_ERROR_RECOVERY(3,
1477 sdev_printk(KERN_INFO, sdev,
1478 "%s: BDR failed\n", current->comm));
1482 return list_empty(work_q);
1486 * scsi_eh_target_reset - send target reset if needed
1487 * @shost: scsi host being recovered.
1488 * @work_q: &list_head for pending commands.
1489 * @done_q: &list_head for processed commands.
1491 * Notes:
1492 * Try a target reset.
1494 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1495 struct list_head *work_q,
1496 struct list_head *done_q)
1498 LIST_HEAD(tmp_list);
1499 LIST_HEAD(check_list);
1501 list_splice_init(work_q, &tmp_list);
1503 while (!list_empty(&tmp_list)) {
1504 struct scsi_cmnd *next, *scmd;
1505 int rtn;
1506 unsigned int id;
1508 if (scsi_host_eh_past_deadline(shost)) {
1509 /* push back on work queue for further processing */
1510 list_splice_init(&check_list, work_q);
1511 list_splice_init(&tmp_list, work_q);
1512 SCSI_LOG_ERROR_RECOVERY(3,
1513 shost_printk(KERN_INFO, shost,
1514 "%s: Skip target reset, past eh deadline\n",
1515 current->comm));
1516 return list_empty(work_q);
1519 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1520 id = scmd_id(scmd);
1522 SCSI_LOG_ERROR_RECOVERY(3,
1523 shost_printk(KERN_INFO, shost,
1524 "%s: Sending target reset to target %d\n",
1525 current->comm, id));
1526 rtn = scsi_try_target_reset(scmd);
1527 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1528 SCSI_LOG_ERROR_RECOVERY(3,
1529 shost_printk(KERN_INFO, shost,
1530 "%s: Target reset failed"
1531 " target: %d\n",
1532 current->comm, id));
1533 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1534 if (scmd_id(scmd) != id)
1535 continue;
1537 if (rtn == SUCCESS)
1538 list_move_tail(&scmd->eh_entry, &check_list);
1539 else if (rtn == FAST_IO_FAIL)
1540 scsi_eh_finish_cmd(scmd, done_q);
1541 else
1542 /* push back on work queue for further processing */
1543 list_move(&scmd->eh_entry, work_q);
1547 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1551 * scsi_eh_bus_reset - send a bus reset
1552 * @shost: &scsi host being recovered.
1553 * @work_q: &list_head for pending commands.
1554 * @done_q: &list_head for processed commands.
1556 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1557 struct list_head *work_q,
1558 struct list_head *done_q)
1560 struct scsi_cmnd *scmd, *chan_scmd, *next;
1561 LIST_HEAD(check_list);
1562 unsigned int channel;
1563 int rtn;
1566 * we really want to loop over the various channels, and do this on
1567 * a channel by channel basis. we should also check to see if any
1568 * of the failed commands are on soft_reset devices, and if so, skip
1569 * the reset.
1572 for (channel = 0; channel <= shost->max_channel; channel++) {
1573 if (scsi_host_eh_past_deadline(shost)) {
1574 list_splice_init(&check_list, work_q);
1575 SCSI_LOG_ERROR_RECOVERY(3,
1576 shost_printk(KERN_INFO, shost,
1577 "%s: skip BRST, past eh deadline\n",
1578 current->comm));
1579 return list_empty(work_q);
1582 chan_scmd = NULL;
1583 list_for_each_entry(scmd, work_q, eh_entry) {
1584 if (channel == scmd_channel(scmd)) {
1585 chan_scmd = scmd;
1586 break;
1588 * FIXME add back in some support for
1589 * soft_reset devices.
1594 if (!chan_scmd)
1595 continue;
1596 SCSI_LOG_ERROR_RECOVERY(3,
1597 shost_printk(KERN_INFO, shost,
1598 "%s: Sending BRST chan: %d\n",
1599 current->comm, channel));
1600 rtn = scsi_try_bus_reset(chan_scmd);
1601 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1602 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1603 if (channel == scmd_channel(scmd)) {
1604 if (rtn == FAST_IO_FAIL)
1605 scsi_eh_finish_cmd(scmd,
1606 done_q);
1607 else
1608 list_move_tail(&scmd->eh_entry,
1609 &check_list);
1612 } else {
1613 SCSI_LOG_ERROR_RECOVERY(3,
1614 shost_printk(KERN_INFO, shost,
1615 "%s: BRST failed chan: %d\n",
1616 current->comm, channel));
1619 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1623 * scsi_eh_host_reset - send a host reset
1624 * @shost: host to be reset.
1625 * @work_q: &list_head for pending commands.
1626 * @done_q: &list_head for processed commands.
1628 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1629 struct list_head *work_q,
1630 struct list_head *done_q)
1632 struct scsi_cmnd *scmd, *next;
1633 LIST_HEAD(check_list);
1634 int rtn;
1636 if (!list_empty(work_q)) {
1637 scmd = list_entry(work_q->next,
1638 struct scsi_cmnd, eh_entry);
1640 SCSI_LOG_ERROR_RECOVERY(3,
1641 shost_printk(KERN_INFO, shost,
1642 "%s: Sending HRST\n",
1643 current->comm));
1645 rtn = scsi_try_host_reset(scmd);
1646 if (rtn == SUCCESS) {
1647 list_splice_init(work_q, &check_list);
1648 } else if (rtn == FAST_IO_FAIL) {
1649 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1650 scsi_eh_finish_cmd(scmd, done_q);
1652 } else {
1653 SCSI_LOG_ERROR_RECOVERY(3,
1654 shost_printk(KERN_INFO, shost,
1655 "%s: HRST failed\n",
1656 current->comm));
1659 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1663 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1664 * @work_q: &list_head for pending commands.
1665 * @done_q: &list_head for processed commands.
1667 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1668 struct list_head *done_q)
1670 struct scsi_cmnd *scmd, *next;
1671 struct scsi_device *sdev;
1673 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1674 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1675 "not ready after error recovery\n");
1676 sdev = scmd->device;
1678 mutex_lock(&sdev->state_mutex);
1679 scsi_device_set_state(sdev, SDEV_OFFLINE);
1680 mutex_unlock(&sdev->state_mutex);
1682 scsi_eh_finish_cmd(scmd, done_q);
1684 return;
1688 * scsi_noretry_cmd - determine if command should be failed fast
1689 * @scmd: SCSI cmd to examine.
1691 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1693 switch (host_byte(scmd->result)) {
1694 case DID_OK:
1695 break;
1696 case DID_TIME_OUT:
1697 goto check_type;
1698 case DID_BUS_BUSY:
1699 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1700 case DID_PARITY:
1701 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1702 case DID_ERROR:
1703 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1704 status_byte(scmd->result) == RESERVATION_CONFLICT)
1705 return 0;
1706 /* fall through */
1707 case DID_SOFT_ERROR:
1708 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1711 if (status_byte(scmd->result) != CHECK_CONDITION)
1712 return 0;
1714 check_type:
1716 * assume caller has checked sense and determined
1717 * the check condition was retryable.
1719 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1720 blk_rq_is_passthrough(scmd->request))
1721 return 1;
1722 else
1723 return 0;
1727 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1728 * @scmd: SCSI cmd to examine.
1730 * Notes:
1731 * This is *only* called when we are examining the status after sending
1732 * out the actual data command. any commands that are queued for error
1733 * recovery (e.g. test_unit_ready) do *not* come through here.
1735 * When this routine returns failed, it means the error handler thread
1736 * is woken. In cases where the error code indicates an error that
1737 * doesn't require the error handler read (i.e. we don't need to
1738 * abort/reset), this function should return SUCCESS.
1740 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1742 int rtn;
1745 * if the device is offline, then we clearly just pass the result back
1746 * up to the top level.
1748 if (!scsi_device_online(scmd->device)) {
1749 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1750 "%s: device offline - report as SUCCESS\n", __func__));
1751 return SUCCESS;
1755 * first check the host byte, to see if there is anything in there
1756 * that would indicate what we need to do.
1758 switch (host_byte(scmd->result)) {
1759 case DID_PASSTHROUGH:
1761 * no matter what, pass this through to the upper layer.
1762 * nuke this special code so that it looks like we are saying
1763 * did_ok.
1765 scmd->result &= 0xff00ffff;
1766 return SUCCESS;
1767 case DID_OK:
1769 * looks good. drop through, and check the next byte.
1771 break;
1772 case DID_ABORT:
1773 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1774 set_host_byte(scmd, DID_TIME_OUT);
1775 return SUCCESS;
1777 /* FALLTHROUGH */
1778 case DID_NO_CONNECT:
1779 case DID_BAD_TARGET:
1781 * note - this means that we just report the status back
1782 * to the top level driver, not that we actually think
1783 * that it indicates SUCCESS.
1785 return SUCCESS;
1786 case DID_SOFT_ERROR:
1788 * when the low level driver returns did_soft_error,
1789 * it is responsible for keeping an internal retry counter
1790 * in order to avoid endless loops (db)
1792 goto maybe_retry;
1793 case DID_IMM_RETRY:
1794 return NEEDS_RETRY;
1796 case DID_REQUEUE:
1797 return ADD_TO_MLQUEUE;
1798 case DID_TRANSPORT_DISRUPTED:
1800 * LLD/transport was disrupted during processing of the IO.
1801 * The transport class is now blocked/blocking,
1802 * and the transport will decide what to do with the IO
1803 * based on its timers and recovery capablilities if
1804 * there are enough retries.
1806 goto maybe_retry;
1807 case DID_TRANSPORT_FAILFAST:
1809 * The transport decided to failfast the IO (most likely
1810 * the fast io fail tmo fired), so send IO directly upwards.
1812 return SUCCESS;
1813 case DID_ERROR:
1814 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1815 status_byte(scmd->result) == RESERVATION_CONFLICT)
1817 * execute reservation conflict processing code
1818 * lower down
1820 break;
1821 /* fallthrough */
1822 case DID_BUS_BUSY:
1823 case DID_PARITY:
1824 goto maybe_retry;
1825 case DID_TIME_OUT:
1827 * when we scan the bus, we get timeout messages for
1828 * these commands if there is no device available.
1829 * other hosts report did_no_connect for the same thing.
1831 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1832 scmd->cmnd[0] == INQUIRY)) {
1833 return SUCCESS;
1834 } else {
1835 return FAILED;
1837 case DID_RESET:
1838 return SUCCESS;
1839 default:
1840 return FAILED;
1844 * next, check the message byte.
1846 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1847 return FAILED;
1850 * check the status byte to see if this indicates anything special.
1852 switch (status_byte(scmd->result)) {
1853 case QUEUE_FULL:
1854 scsi_handle_queue_full(scmd->device);
1856 * the case of trying to send too many commands to a
1857 * tagged queueing device.
1859 /* FALLTHROUGH */
1860 case BUSY:
1862 * device can't talk to us at the moment. Should only
1863 * occur (SAM-3) when the task queue is empty, so will cause
1864 * the empty queue handling to trigger a stall in the
1865 * device.
1867 return ADD_TO_MLQUEUE;
1868 case GOOD:
1869 if (scmd->cmnd[0] == REPORT_LUNS)
1870 scmd->device->sdev_target->expecting_lun_change = 0;
1871 scsi_handle_queue_ramp_up(scmd->device);
1872 /* FALLTHROUGH */
1873 case COMMAND_TERMINATED:
1874 return SUCCESS;
1875 case TASK_ABORTED:
1876 goto maybe_retry;
1877 case CHECK_CONDITION:
1878 rtn = scsi_check_sense(scmd);
1879 if (rtn == NEEDS_RETRY)
1880 goto maybe_retry;
1881 /* if rtn == FAILED, we have no sense information;
1882 * returning FAILED will wake the error handler thread
1883 * to collect the sense and redo the decide
1884 * disposition */
1885 return rtn;
1886 case CONDITION_GOOD:
1887 case INTERMEDIATE_GOOD:
1888 case INTERMEDIATE_C_GOOD:
1889 case ACA_ACTIVE:
1891 * who knows? FIXME(eric)
1893 return SUCCESS;
1895 case RESERVATION_CONFLICT:
1896 sdev_printk(KERN_INFO, scmd->device,
1897 "reservation conflict\n");
1898 set_host_byte(scmd, DID_NEXUS_FAILURE);
1899 return SUCCESS; /* causes immediate i/o error */
1900 default:
1901 return FAILED;
1903 return FAILED;
1905 maybe_retry:
1907 /* we requeue for retry because the error was retryable, and
1908 * the request was not marked fast fail. Note that above,
1909 * even if the request is marked fast fail, we still requeue
1910 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1911 if ((++scmd->retries) <= scmd->allowed
1912 && !scsi_noretry_cmd(scmd)) {
1913 return NEEDS_RETRY;
1914 } else {
1916 * no more retries - report this one back to upper level.
1918 return SUCCESS;
1922 static void eh_lock_door_done(struct request *req, blk_status_t status)
1924 __blk_put_request(req->q, req);
1928 * scsi_eh_lock_door - Prevent medium removal for the specified device
1929 * @sdev: SCSI device to prevent medium removal
1931 * Locking:
1932 * We must be called from process context.
1934 * Notes:
1935 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1936 * head of the devices request queue, and continue.
1938 static void scsi_eh_lock_door(struct scsi_device *sdev)
1940 struct request *req;
1941 struct scsi_request *rq;
1943 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, 0);
1944 if (IS_ERR(req))
1945 return;
1946 rq = scsi_req(req);
1948 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1949 rq->cmd[1] = 0;
1950 rq->cmd[2] = 0;
1951 rq->cmd[3] = 0;
1952 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
1953 rq->cmd[5] = 0;
1954 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
1956 req->rq_flags |= RQF_QUIET;
1957 req->timeout = 10 * HZ;
1958 rq->retries = 5;
1960 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1964 * scsi_restart_operations - restart io operations to the specified host.
1965 * @shost: Host we are restarting.
1967 * Notes:
1968 * When we entered the error handler, we blocked all further i/o to
1969 * this device. we need to 'reverse' this process.
1971 static void scsi_restart_operations(struct Scsi_Host *shost)
1973 struct scsi_device *sdev;
1974 unsigned long flags;
1977 * If the door was locked, we need to insert a door lock request
1978 * onto the head of the SCSI request queue for the device. There
1979 * is no point trying to lock the door of an off-line device.
1981 shost_for_each_device(sdev, shost) {
1982 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
1983 scsi_eh_lock_door(sdev);
1984 sdev->was_reset = 0;
1989 * next free up anything directly waiting upon the host. this
1990 * will be requests for character device operations, and also for
1991 * ioctls to queued block devices.
1993 SCSI_LOG_ERROR_RECOVERY(3,
1994 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
1996 spin_lock_irqsave(shost->host_lock, flags);
1997 if (scsi_host_set_state(shost, SHOST_RUNNING))
1998 if (scsi_host_set_state(shost, SHOST_CANCEL))
1999 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2000 spin_unlock_irqrestore(shost->host_lock, flags);
2002 wake_up(&shost->host_wait);
2005 * finally we need to re-initiate requests that may be pending. we will
2006 * have had everything blocked while error handling is taking place, and
2007 * now that error recovery is done, we will need to ensure that these
2008 * requests are started.
2010 scsi_run_host_queues(shost);
2013 * if eh is active and host_eh_scheduled is pending we need to re-run
2014 * recovery. we do this check after scsi_run_host_queues() to allow
2015 * everything pent up since the last eh run a chance to make forward
2016 * progress before we sync again. Either we'll immediately re-run
2017 * recovery or scsi_device_unbusy() will wake us again when these
2018 * pending commands complete.
2020 spin_lock_irqsave(shost->host_lock, flags);
2021 if (shost->host_eh_scheduled)
2022 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2023 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2024 spin_unlock_irqrestore(shost->host_lock, flags);
2028 * scsi_eh_ready_devs - check device ready state and recover if not.
2029 * @shost: host to be recovered.
2030 * @work_q: &list_head for pending commands.
2031 * @done_q: &list_head for processed commands.
2033 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2034 struct list_head *work_q,
2035 struct list_head *done_q)
2037 if (!scsi_eh_stu(shost, work_q, done_q))
2038 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2039 if (!scsi_eh_target_reset(shost, work_q, done_q))
2040 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2041 if (!scsi_eh_host_reset(shost, work_q, done_q))
2042 scsi_eh_offline_sdevs(work_q,
2043 done_q);
2045 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2048 * scsi_eh_flush_done_q - finish processed commands or retry them.
2049 * @done_q: list_head of processed commands.
2051 void scsi_eh_flush_done_q(struct list_head *done_q)
2053 struct scsi_cmnd *scmd, *next;
2055 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2056 list_del_init(&scmd->eh_entry);
2057 if (scsi_device_online(scmd->device) &&
2058 !scsi_noretry_cmd(scmd) &&
2059 (++scmd->retries <= scmd->allowed)) {
2060 SCSI_LOG_ERROR_RECOVERY(3,
2061 scmd_printk(KERN_INFO, scmd,
2062 "%s: flush retry cmd\n",
2063 current->comm));
2064 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2065 } else {
2067 * If just we got sense for the device (called
2068 * scsi_eh_get_sense), scmd->result is already
2069 * set, do not set DRIVER_TIMEOUT.
2071 if (!scmd->result)
2072 scmd->result |= (DRIVER_TIMEOUT << 24);
2073 SCSI_LOG_ERROR_RECOVERY(3,
2074 scmd_printk(KERN_INFO, scmd,
2075 "%s: flush finish cmd\n",
2076 current->comm));
2077 scsi_finish_command(scmd);
2081 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2084 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2085 * @shost: Host to unjam.
2087 * Notes:
2088 * When we come in here, we *know* that all commands on the bus have
2089 * either completed, failed or timed out. we also know that no further
2090 * commands are being sent to the host, so things are relatively quiet
2091 * and we have freedom to fiddle with things as we wish.
2093 * This is only the *default* implementation. it is possible for
2094 * individual drivers to supply their own version of this function, and
2095 * if the maintainer wishes to do this, it is strongly suggested that
2096 * this function be taken as a template and modified. this function
2097 * was designed to correctly handle problems for about 95% of the
2098 * different cases out there, and it should always provide at least a
2099 * reasonable amount of error recovery.
2101 * Any command marked 'failed' or 'timeout' must eventually have
2102 * scsi_finish_cmd() called for it. we do all of the retry stuff
2103 * here, so when we restart the host after we return it should have an
2104 * empty queue.
2106 static void scsi_unjam_host(struct Scsi_Host *shost)
2108 unsigned long flags;
2109 LIST_HEAD(eh_work_q);
2110 LIST_HEAD(eh_done_q);
2112 spin_lock_irqsave(shost->host_lock, flags);
2113 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2114 spin_unlock_irqrestore(shost->host_lock, flags);
2116 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2118 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2119 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2121 spin_lock_irqsave(shost->host_lock, flags);
2122 if (shost->eh_deadline != -1)
2123 shost->last_reset = 0;
2124 spin_unlock_irqrestore(shost->host_lock, flags);
2125 scsi_eh_flush_done_q(&eh_done_q);
2129 * scsi_error_handler - SCSI error handler thread
2130 * @data: Host for which we are running.
2132 * Notes:
2133 * This is the main error handling loop. This is run as a kernel thread
2134 * for every SCSI host and handles all error handling activity.
2136 int scsi_error_handler(void *data)
2138 struct Scsi_Host *shost = data;
2141 * We use TASK_INTERRUPTIBLE so that the thread is not
2142 * counted against the load average as a running process.
2143 * We never actually get interrupted because kthread_run
2144 * disables signal delivery for the created thread.
2146 while (true) {
2148 * The sequence in kthread_stop() sets the stop flag first
2149 * then wakes the process. To avoid missed wakeups, the task
2150 * should always be in a non running state before the stop
2151 * flag is checked
2153 set_current_state(TASK_INTERRUPTIBLE);
2154 if (kthread_should_stop())
2155 break;
2157 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2158 shost->host_failed != atomic_read(&shost->host_busy)) {
2159 SCSI_LOG_ERROR_RECOVERY(1,
2160 shost_printk(KERN_INFO, shost,
2161 "scsi_eh_%d: sleeping\n",
2162 shost->host_no));
2163 schedule();
2164 continue;
2167 __set_current_state(TASK_RUNNING);
2168 SCSI_LOG_ERROR_RECOVERY(1,
2169 shost_printk(KERN_INFO, shost,
2170 "scsi_eh_%d: waking up %d/%d/%d\n",
2171 shost->host_no, shost->host_eh_scheduled,
2172 shost->host_failed,
2173 atomic_read(&shost->host_busy)));
2176 * We have a host that is failing for some reason. Figure out
2177 * what we need to do to get it up and online again (if we can).
2178 * If we fail, we end up taking the thing offline.
2180 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2181 SCSI_LOG_ERROR_RECOVERY(1,
2182 shost_printk(KERN_ERR, shost,
2183 "scsi_eh_%d: unable to autoresume\n",
2184 shost->host_no));
2185 continue;
2188 if (shost->transportt->eh_strategy_handler)
2189 shost->transportt->eh_strategy_handler(shost);
2190 else
2191 scsi_unjam_host(shost);
2193 /* All scmds have been handled */
2194 shost->host_failed = 0;
2197 * Note - if the above fails completely, the action is to take
2198 * individual devices offline and flush the queue of any
2199 * outstanding requests that may have been pending. When we
2200 * restart, we restart any I/O to any other devices on the bus
2201 * which are still online.
2203 scsi_restart_operations(shost);
2204 if (!shost->eh_noresume)
2205 scsi_autopm_put_host(shost);
2207 __set_current_state(TASK_RUNNING);
2209 SCSI_LOG_ERROR_RECOVERY(1,
2210 shost_printk(KERN_INFO, shost,
2211 "Error handler scsi_eh_%d exiting\n",
2212 shost->host_no));
2213 shost->ehandler = NULL;
2214 return 0;
2218 * Function: scsi_report_bus_reset()
2220 * Purpose: Utility function used by low-level drivers to report that
2221 * they have observed a bus reset on the bus being handled.
2223 * Arguments: shost - Host in question
2224 * channel - channel on which reset was observed.
2226 * Returns: Nothing
2228 * Lock status: Host lock must be held.
2230 * Notes: This only needs to be called if the reset is one which
2231 * originates from an unknown location. Resets originated
2232 * by the mid-level itself don't need to call this, but there
2233 * should be no harm.
2235 * The main purpose of this is to make sure that a CHECK_CONDITION
2236 * is properly treated.
2238 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2240 struct scsi_device *sdev;
2242 __shost_for_each_device(sdev, shost) {
2243 if (channel == sdev_channel(sdev))
2244 __scsi_report_device_reset(sdev, NULL);
2247 EXPORT_SYMBOL(scsi_report_bus_reset);
2250 * Function: scsi_report_device_reset()
2252 * Purpose: Utility function used by low-level drivers to report that
2253 * they have observed a device reset on the device being handled.
2255 * Arguments: shost - Host in question
2256 * channel - channel on which reset was observed
2257 * target - target on which reset was observed
2259 * Returns: Nothing
2261 * Lock status: Host lock must be held
2263 * Notes: This only needs to be called if the reset is one which
2264 * originates from an unknown location. Resets originated
2265 * by the mid-level itself don't need to call this, but there
2266 * should be no harm.
2268 * The main purpose of this is to make sure that a CHECK_CONDITION
2269 * is properly treated.
2271 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2273 struct scsi_device *sdev;
2275 __shost_for_each_device(sdev, shost) {
2276 if (channel == sdev_channel(sdev) &&
2277 target == sdev_id(sdev))
2278 __scsi_report_device_reset(sdev, NULL);
2281 EXPORT_SYMBOL(scsi_report_device_reset);
2283 static void
2284 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2289 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2290 * @dev: scsi_device to operate on
2291 * @arg: reset type (see sg.h)
2294 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2296 struct scsi_cmnd *scmd;
2297 struct Scsi_Host *shost = dev->host;
2298 struct request *rq;
2299 unsigned long flags;
2300 int error = 0, rtn, val;
2302 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2303 return -EACCES;
2305 error = get_user(val, arg);
2306 if (error)
2307 return error;
2309 if (scsi_autopm_get_host(shost) < 0)
2310 return -EIO;
2312 error = -EIO;
2313 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2314 shost->hostt->cmd_size, GFP_KERNEL);
2315 if (!rq)
2316 goto out_put_autopm_host;
2317 blk_rq_init(NULL, rq);
2319 scmd = (struct scsi_cmnd *)(rq + 1);
2320 scsi_init_command(dev, scmd);
2321 scmd->request = rq;
2322 scmd->cmnd = scsi_req(rq)->cmd;
2324 scmd->scsi_done = scsi_reset_provider_done_command;
2325 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2327 scmd->cmd_len = 0;
2329 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2331 spin_lock_irqsave(shost->host_lock, flags);
2332 shost->tmf_in_progress = 1;
2333 spin_unlock_irqrestore(shost->host_lock, flags);
2335 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2336 case SG_SCSI_RESET_NOTHING:
2337 rtn = SUCCESS;
2338 break;
2339 case SG_SCSI_RESET_DEVICE:
2340 rtn = scsi_try_bus_device_reset(scmd);
2341 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2342 break;
2343 /* FALLTHROUGH */
2344 case SG_SCSI_RESET_TARGET:
2345 rtn = scsi_try_target_reset(scmd);
2346 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2347 break;
2348 /* FALLTHROUGH */
2349 case SG_SCSI_RESET_BUS:
2350 rtn = scsi_try_bus_reset(scmd);
2351 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2352 break;
2353 /* FALLTHROUGH */
2354 case SG_SCSI_RESET_HOST:
2355 rtn = scsi_try_host_reset(scmd);
2356 if (rtn == SUCCESS)
2357 break;
2358 /* FALLTHROUGH */
2359 default:
2360 rtn = FAILED;
2361 break;
2364 error = (rtn == SUCCESS) ? 0 : -EIO;
2366 spin_lock_irqsave(shost->host_lock, flags);
2367 shost->tmf_in_progress = 0;
2368 spin_unlock_irqrestore(shost->host_lock, flags);
2371 * be sure to wake up anyone who was sleeping or had their queue
2372 * suspended while we performed the TMF.
2374 SCSI_LOG_ERROR_RECOVERY(3,
2375 shost_printk(KERN_INFO, shost,
2376 "waking up host to restart after TMF\n"));
2378 wake_up(&shost->host_wait);
2379 scsi_run_host_queues(shost);
2381 scsi_put_command(scmd);
2382 kfree(rq);
2384 out_put_autopm_host:
2385 scsi_autopm_put_host(shost);
2386 return error;
2388 EXPORT_SYMBOL(scsi_ioctl_reset);
2390 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2391 struct scsi_sense_hdr *sshdr)
2393 return scsi_normalize_sense(cmd->sense_buffer,
2394 SCSI_SENSE_BUFFERSIZE, sshdr);
2396 EXPORT_SYMBOL(scsi_command_normalize_sense);
2399 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2400 * @sense_buffer: byte array of sense data
2401 * @sb_len: number of valid bytes in sense_buffer
2402 * @info_out: pointer to 64 integer where 8 or 4 byte information
2403 * field will be placed if found.
2405 * Return value:
2406 * true if information field found, false if not found.
2408 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2409 u64 *info_out)
2411 const u8 * ucp;
2413 if (sb_len < 7)
2414 return false;
2415 switch (sense_buffer[0] & 0x7f) {
2416 case 0x70:
2417 case 0x71:
2418 if (sense_buffer[0] & 0x80) {
2419 *info_out = get_unaligned_be32(&sense_buffer[3]);
2420 return true;
2422 return false;
2423 case 0x72:
2424 case 0x73:
2425 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2426 0 /* info desc */);
2427 if (ucp && (0xa == ucp[1])) {
2428 *info_out = get_unaligned_be64(&ucp[4]);
2429 return true;
2431 return false;
2432 default:
2433 return false;
2436 EXPORT_SYMBOL(scsi_get_sense_info_fld);