1 // SPDX-License-Identifier: GPL-2.0-only
3 * scsi_error.c Copyright (C) 1997 Eric Youngdale
5 * SCSI error/timeout handling
6 * Initial versions: Eric Youngdale. Based upon conversations with
7 * Leonard Zubkoff and David Miller at Linux Expo,
8 * ideas originating from all over the place.
10 * Restructured scsi_unjam_host and associated functions.
11 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
13 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
15 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
18 #include <linux/module.h>
19 #include <linux/sched.h>
20 #include <linux/gfp.h>
21 #include <linux/timer.h>
22 #include <linux/string.h>
23 #include <linux/kernel.h>
24 #include <linux/freezer.h>
25 #include <linux/kthread.h>
26 #include <linux/interrupt.h>
27 #include <linux/blkdev.h>
28 #include <linux/delay.h>
29 #include <linux/jiffies.h>
31 #include <scsi/scsi.h>
32 #include <scsi/scsi_cmnd.h>
33 #include <scsi/scsi_dbg.h>
34 #include <scsi/scsi_device.h>
35 #include <scsi/scsi_driver.h>
36 #include <scsi/scsi_eh.h>
37 #include <scsi/scsi_common.h>
38 #include <scsi/scsi_transport.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_ioctl.h>
41 #include <scsi/scsi_dh.h>
42 #include <scsi/scsi_devinfo.h>
45 #include "scsi_priv.h"
46 #include "scsi_logging.h"
47 #include "scsi_transport_api.h"
49 #include <trace/events/scsi.h>
51 #include <asm/unaligned.h>
53 static void scsi_eh_done(struct scsi_cmnd *scmd);
56 * These should *probably* be handled by the host itself.
57 * Since it is allowed to sleep, it probably should.
59 #define BUS_RESET_SETTLE_TIME (10)
60 #define HOST_RESET_SETTLE_TIME (10)
62 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
63 static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *,
66 void scsi_eh_wakeup(struct Scsi_Host *shost)
68 lockdep_assert_held(shost->host_lock);
70 if (scsi_host_busy(shost) == shost->host_failed) {
71 trace_scsi_eh_wakeup(shost);
72 wake_up_process(shost->ehandler);
73 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
74 "Waking error handler thread\n"));
79 * scsi_schedule_eh - schedule EH for SCSI host
80 * @shost: SCSI host to invoke error handling on.
82 * Schedule SCSI EH without scmd.
84 void scsi_schedule_eh(struct Scsi_Host *shost)
88 spin_lock_irqsave(shost->host_lock, flags);
90 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
91 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
92 shost->host_eh_scheduled++;
93 scsi_eh_wakeup(shost);
96 spin_unlock_irqrestore(shost->host_lock, flags);
98 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
100 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
102 if (!shost->last_reset || shost->eh_deadline == -1)
106 * 32bit accesses are guaranteed to be atomic
107 * (on all supported architectures), so instead
108 * of using a spinlock we can as well double check
109 * if eh_deadline has been set to 'off' during the
112 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
113 shost->eh_deadline > -1)
119 static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd)
121 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
124 return ++cmd->retries <= cmd->allowed;
127 static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd)
129 struct scsi_device *sdev = cmd->device;
130 struct Scsi_Host *host = sdev->host;
132 if (host->hostt->eh_should_retry_cmd)
133 return host->hostt->eh_should_retry_cmd(cmd);
139 * scmd_eh_abort_handler - Handle command aborts
140 * @work: command to be aborted.
142 * Note: this function must be called only for a command that has timed out.
143 * Because the block layer marks a request as complete before it calls
144 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
145 * timed out do not have any effect. Hence it is safe to call
146 * scsi_finish_command() from this function.
149 scmd_eh_abort_handler(struct work_struct *work)
151 struct scsi_cmnd *scmd =
152 container_of(work, struct scsi_cmnd, abort_work.work);
153 struct scsi_device *sdev = scmd->device;
154 enum scsi_disposition rtn;
156 if (scsi_host_eh_past_deadline(sdev->host)) {
157 SCSI_LOG_ERROR_RECOVERY(3,
158 scmd_printk(KERN_INFO, scmd,
159 "eh timeout, not aborting\n"));
161 SCSI_LOG_ERROR_RECOVERY(3,
162 scmd_printk(KERN_INFO, scmd,
163 "aborting command\n"));
164 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
165 if (rtn == SUCCESS) {
166 set_host_byte(scmd, DID_TIME_OUT);
167 if (scsi_host_eh_past_deadline(sdev->host)) {
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_INFO, scmd,
170 "eh timeout, not retrying "
171 "aborted command\n"));
172 } else if (!scsi_noretry_cmd(scmd) &&
173 scsi_cmd_retry_allowed(scmd) &&
174 scsi_eh_should_retry_cmd(scmd)) {
175 SCSI_LOG_ERROR_RECOVERY(3,
176 scmd_printk(KERN_WARNING, scmd,
177 "retry aborted command\n"));
178 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
181 SCSI_LOG_ERROR_RECOVERY(3,
182 scmd_printk(KERN_WARNING, scmd,
183 "finish aborted command\n"));
184 scsi_finish_command(scmd);
188 SCSI_LOG_ERROR_RECOVERY(3,
189 scmd_printk(KERN_INFO, scmd,
191 (rtn == FAST_IO_FAIL) ?
192 "not send" : "failed"));
196 scsi_eh_scmd_add(scmd);
200 * scsi_abort_command - schedule a command abort
201 * @scmd: scmd to abort.
203 * We only need to abort commands after a command timeout
206 scsi_abort_command(struct scsi_cmnd *scmd)
208 struct scsi_device *sdev = scmd->device;
209 struct Scsi_Host *shost = sdev->host;
212 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
214 * Retry after abort failed, escalate to next level.
216 SCSI_LOG_ERROR_RECOVERY(3,
217 scmd_printk(KERN_INFO, scmd,
218 "previous abort failed\n"));
219 BUG_ON(delayed_work_pending(&scmd->abort_work));
223 spin_lock_irqsave(shost->host_lock, flags);
224 if (shost->eh_deadline != -1 && !shost->last_reset)
225 shost->last_reset = jiffies;
226 spin_unlock_irqrestore(shost->host_lock, flags);
228 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
229 SCSI_LOG_ERROR_RECOVERY(3,
230 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
231 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
236 * scsi_eh_reset - call into ->eh_action to reset internal counters
237 * @scmd: scmd to run eh on.
239 * The scsi driver might be carrying internal state about the
240 * devices, so we need to call into the driver to reset the
241 * internal state once the error handler is started.
243 static void scsi_eh_reset(struct scsi_cmnd *scmd)
245 if (!blk_rq_is_passthrough(scmd->request)) {
246 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
248 sdrv->eh_reset(scmd);
252 static void scsi_eh_inc_host_failed(struct rcu_head *head)
254 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
255 struct Scsi_Host *shost = scmd->device->host;
258 spin_lock_irqsave(shost->host_lock, flags);
259 shost->host_failed++;
260 scsi_eh_wakeup(shost);
261 spin_unlock_irqrestore(shost->host_lock, flags);
265 * scsi_eh_scmd_add - add scsi cmd to error handling.
266 * @scmd: scmd to run eh on.
268 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
270 struct Scsi_Host *shost = scmd->device->host;
274 WARN_ON_ONCE(!shost->ehandler);
276 spin_lock_irqsave(shost->host_lock, flags);
277 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
278 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
281 if (shost->eh_deadline != -1 && !shost->last_reset)
282 shost->last_reset = jiffies;
285 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
286 spin_unlock_irqrestore(shost->host_lock, flags);
288 * Ensure that all tasks observe the host state change before the
289 * host_failed change.
291 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
295 * scsi_times_out - Timeout function for normal scsi commands.
296 * @req: request that is timing out.
299 * We do not need to lock this. There is the potential for a race
300 * only in that the normal completion handling might run, but if the
301 * normal completion function determines that the timer has already
302 * fired, then it mustn't do anything.
304 enum blk_eh_timer_return scsi_times_out(struct request *req)
306 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
307 enum blk_eh_timer_return rtn = BLK_EH_DONE;
308 struct Scsi_Host *host = scmd->device->host;
310 trace_scsi_dispatch_cmd_timeout(scmd);
311 scsi_log_completion(scmd, TIMEOUT_ERROR);
313 if (host->eh_deadline != -1 && !host->last_reset)
314 host->last_reset = jiffies;
316 if (host->hostt->eh_timed_out)
317 rtn = host->hostt->eh_timed_out(scmd);
319 if (rtn == BLK_EH_DONE) {
321 * Set the command to complete first in order to prevent a real
322 * completion from releasing the command while error handling
323 * is using it. If the command was already completed, then the
324 * lower level driver beat the timeout handler, and it is safe
325 * to return without escalating error recovery.
327 * If timeout handling lost the race to a real completion, the
328 * block layer may ignore that due to a fake timeout injection,
329 * so return RESET_TIMER to allow error handling another shot
332 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
333 return BLK_EH_RESET_TIMER;
334 if (scsi_abort_command(scmd) != SUCCESS) {
335 set_host_byte(scmd, DID_TIME_OUT);
336 scsi_eh_scmd_add(scmd);
344 * scsi_block_when_processing_errors - Prevent cmds from being queued.
345 * @sdev: Device on which we are performing recovery.
348 * We block until the host is out of error recovery, and then check to
349 * see whether the host or the device is offline.
352 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
354 int scsi_block_when_processing_errors(struct scsi_device *sdev)
358 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
360 online = scsi_device_online(sdev);
364 EXPORT_SYMBOL(scsi_block_when_processing_errors);
366 #ifdef CONFIG_SCSI_LOGGING
368 * scsi_eh_prt_fail_stats - Log info on failures.
369 * @shost: scsi host being recovered.
370 * @work_q: Queue of scsi cmds to process.
372 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
373 struct list_head *work_q)
375 struct scsi_cmnd *scmd;
376 struct scsi_device *sdev;
377 int total_failures = 0;
380 int devices_failed = 0;
382 shost_for_each_device(sdev, shost) {
383 list_for_each_entry(scmd, work_q, eh_entry) {
384 if (scmd->device == sdev) {
386 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
393 if (cmd_cancel || cmd_failed) {
394 SCSI_LOG_ERROR_RECOVERY(3,
395 shost_printk(KERN_INFO, shost,
396 "%s: cmds failed: %d, cancel: %d\n",
397 __func__, cmd_failed,
405 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
406 "Total of %d commands on %d"
407 " devices require eh work\n",
408 total_failures, devices_failed));
413 * scsi_report_lun_change - Set flag on all *other* devices on the same target
414 * to indicate that a UNIT ATTENTION is expected.
415 * @sdev: Device reporting the UNIT ATTENTION
417 static void scsi_report_lun_change(struct scsi_device *sdev)
419 sdev->sdev_target->expecting_lun_change = 1;
423 * scsi_report_sense - Examine scsi sense information and log messages for
424 * certain conditions, also issue uevents for some of them.
425 * @sdev: Device reporting the sense code
426 * @sshdr: sshdr to be examined
428 static void scsi_report_sense(struct scsi_device *sdev,
429 struct scsi_sense_hdr *sshdr)
431 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
433 if (sshdr->sense_key == UNIT_ATTENTION) {
434 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
435 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
436 sdev_printk(KERN_WARNING, sdev,
437 "Inquiry data has changed");
438 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
439 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
440 scsi_report_lun_change(sdev);
441 sdev_printk(KERN_WARNING, sdev,
442 "Warning! Received an indication that the "
443 "LUN assignments on this target have "
444 "changed. The Linux SCSI layer does not "
445 "automatically remap LUN assignments.\n");
446 } else if (sshdr->asc == 0x3f)
447 sdev_printk(KERN_WARNING, sdev,
448 "Warning! Received an indication that the "
449 "operating parameters on this target have "
450 "changed. The Linux SCSI layer does not "
451 "automatically adjust these parameters.\n");
453 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
454 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
455 sdev_printk(KERN_WARNING, sdev,
456 "Warning! Received an indication that the "
457 "LUN reached a thin provisioning soft "
461 if (sshdr->asc == 0x29) {
462 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
463 sdev_printk(KERN_WARNING, sdev,
464 "Power-on or device reset occurred\n");
467 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
468 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
469 sdev_printk(KERN_WARNING, sdev,
470 "Mode parameters changed");
471 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
472 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
473 sdev_printk(KERN_WARNING, sdev,
474 "Asymmetric access state changed");
475 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
476 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
477 sdev_printk(KERN_WARNING, sdev,
478 "Capacity data has changed");
479 } else if (sshdr->asc == 0x2a)
480 sdev_printk(KERN_WARNING, sdev,
481 "Parameters changed");
484 if (evt_type != SDEV_EVT_MAXBITS) {
485 set_bit(evt_type, sdev->pending_events);
486 schedule_work(&sdev->event_work);
491 * scsi_check_sense - Examine scsi cmd sense
492 * @scmd: Cmd to have sense checked.
495 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
498 * When a deferred error is detected the current command has
499 * not been executed and needs retrying.
501 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
503 struct scsi_device *sdev = scmd->device;
504 struct scsi_sense_hdr sshdr;
506 if (! scsi_command_normalize_sense(scmd, &sshdr))
507 return FAILED; /* no valid sense data */
509 scsi_report_sense(sdev, &sshdr);
511 if (scsi_sense_is_deferred(&sshdr))
514 if (sdev->handler && sdev->handler->check_sense) {
515 enum scsi_disposition rc;
517 rc = sdev->handler->check_sense(sdev, &sshdr);
518 if (rc != SCSI_RETURN_NOT_HANDLED)
520 /* handler does not care. Drop down to default handling */
523 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
525 * nasty: for mid-layer issued TURs, we need to return the
526 * actual sense data without any recovery attempt. For eh
527 * issued ones, we need to try to recover and interpret
532 * Previous logic looked for FILEMARK, EOM or ILI which are
533 * mainly associated with tapes and returned SUCCESS.
535 if (sshdr.response_code == 0x70) {
537 if (scmd->sense_buffer[2] & 0xe0)
541 * descriptor format: look for "stream commands sense data
542 * descriptor" (see SSC-3). Assume single sense data
543 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
545 if ((sshdr.additional_length > 3) &&
546 (scmd->sense_buffer[8] == 0x4) &&
547 (scmd->sense_buffer[11] & 0xe0))
551 switch (sshdr.sense_key) {
554 case RECOVERED_ERROR:
555 return /* soft_error */ SUCCESS;
557 case ABORTED_COMMAND:
558 if (sshdr.asc == 0x10) /* DIF */
561 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
562 return ADD_TO_MLQUEUE;
563 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
564 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
565 return ADD_TO_MLQUEUE;
571 * if we are expecting a cc/ua because of a bus reset that we
572 * performed, treat this just as a retry. otherwise this is
573 * information that we should pass up to the upper-level driver
574 * so that we can deal with it there.
576 if (scmd->device->expecting_cc_ua) {
578 * Because some device does not queue unit
579 * attentions correctly, we carefully check
580 * additional sense code and qualifier so as
581 * not to squash media change unit attention.
583 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
584 scmd->device->expecting_cc_ua = 0;
589 * we might also expect a cc/ua if another LUN on the target
590 * reported a UA with an ASC/ASCQ of 3F 0E -
591 * REPORTED LUNS DATA HAS CHANGED.
593 if (scmd->device->sdev_target->expecting_lun_change &&
594 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
597 * if the device is in the process of becoming ready, we
600 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
603 * if the device is not started, we need to wake
604 * the error handler to start the motor
606 if (scmd->device->allow_restart &&
607 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
610 * Pass the UA upwards for a determination in the completion
615 /* these are not supported */
617 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
618 /* Thin provisioning hard threshold reached */
619 set_host_byte(scmd, DID_ALLOC_FAILURE);
624 case VOLUME_OVERFLOW:
627 set_host_byte(scmd, DID_TARGET_FAILURE);
631 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
632 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
633 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
634 set_host_byte(scmd, DID_MEDIUM_ERROR);
640 if (scmd->device->retry_hwerror)
641 return ADD_TO_MLQUEUE;
643 set_host_byte(scmd, DID_TARGET_FAILURE);
646 case ILLEGAL_REQUEST:
647 if (sshdr.asc == 0x20 || /* Invalid command operation code */
648 sshdr.asc == 0x21 || /* Logical block address out of range */
649 sshdr.asc == 0x22 || /* Invalid function */
650 sshdr.asc == 0x24 || /* Invalid field in cdb */
651 sshdr.asc == 0x26 || /* Parameter value invalid */
652 sshdr.asc == 0x27) { /* Write protected */
653 set_host_byte(scmd, DID_TARGET_FAILURE);
661 EXPORT_SYMBOL_GPL(scsi_check_sense);
663 static void scsi_handle_queue_ramp_up(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 sdev->queue_depth >= sdev->max_queue_depth)
672 if (time_before(jiffies,
673 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
676 if (time_before(jiffies,
677 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
681 * Walk all devices of a target and do
684 shost_for_each_device(tmp_sdev, sdev->host) {
685 if (tmp_sdev->channel != sdev->channel ||
686 tmp_sdev->id != sdev->id ||
687 tmp_sdev->queue_depth == sdev->max_queue_depth)
690 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
691 sdev->last_queue_ramp_up = jiffies;
695 static void scsi_handle_queue_full(struct scsi_device *sdev)
697 struct scsi_host_template *sht = sdev->host->hostt;
698 struct scsi_device *tmp_sdev;
700 if (!sht->track_queue_depth)
703 shost_for_each_device(tmp_sdev, sdev->host) {
704 if (tmp_sdev->channel != sdev->channel ||
705 tmp_sdev->id != sdev->id)
708 * We do not know the number of commands that were at
709 * the device when we got the queue full so we start
710 * from the highest possible value and work our way down.
712 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
717 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
718 * @scmd: SCSI cmd to examine.
721 * This is *only* called when we are examining the status of commands
722 * queued during error recovery. the main difference here is that we
723 * don't allow for the possibility of retries here, and we are a lot
724 * more restrictive about what we consider acceptable.
726 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
729 * first check the host byte, to see if there is anything in there
730 * that would indicate what we need to do.
732 if (host_byte(scmd->result) == DID_RESET) {
734 * rats. we are already in the error handler, so we now
735 * get to try and figure out what to do next. if the sense
736 * is valid, we have a pretty good idea of what to do.
737 * if not, we mark it as FAILED.
739 return scsi_check_sense(scmd);
741 if (host_byte(scmd->result) != DID_OK)
745 * next, check the message byte.
747 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
751 * now, check the status byte to see if this indicates
754 switch (status_byte(scmd->result)) {
756 scsi_handle_queue_ramp_up(scmd->device);
758 case COMMAND_TERMINATED:
760 case CHECK_CONDITION:
761 return scsi_check_sense(scmd);
763 case INTERMEDIATE_GOOD:
764 case INTERMEDIATE_C_GOOD:
766 * who knows? FIXME(eric)
769 case RESERVATION_CONFLICT:
770 if (scmd->cmnd[0] == TEST_UNIT_READY)
771 /* it is a success, we probed the device and
774 /* otherwise, we failed to send the command */
777 scsi_handle_queue_full(scmd->device);
788 * scsi_eh_done - Completion function for error handling.
789 * @scmd: Cmd that is done.
791 static void scsi_eh_done(struct scsi_cmnd *scmd)
793 struct completion *eh_action;
795 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
796 "%s result: %x\n", __func__, scmd->result));
798 eh_action = scmd->device->host->eh_action;
804 * scsi_try_host_reset - ask host adapter to reset itself
805 * @scmd: SCSI cmd to send host reset.
807 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
810 enum scsi_disposition rtn;
811 struct Scsi_Host *host = scmd->device->host;
812 struct scsi_host_template *hostt = host->hostt;
814 SCSI_LOG_ERROR_RECOVERY(3,
815 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
817 if (!hostt->eh_host_reset_handler)
820 rtn = hostt->eh_host_reset_handler(scmd);
822 if (rtn == SUCCESS) {
823 if (!hostt->skip_settle_delay)
824 ssleep(HOST_RESET_SETTLE_TIME);
825 spin_lock_irqsave(host->host_lock, flags);
826 scsi_report_bus_reset(host, scmd_channel(scmd));
827 spin_unlock_irqrestore(host->host_lock, flags);
834 * scsi_try_bus_reset - ask host to perform a bus reset
835 * @scmd: SCSI cmd to send bus reset.
837 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
840 enum scsi_disposition rtn;
841 struct Scsi_Host *host = scmd->device->host;
842 struct scsi_host_template *hostt = host->hostt;
844 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
845 "%s: Snd Bus RST\n", __func__));
847 if (!hostt->eh_bus_reset_handler)
850 rtn = hostt->eh_bus_reset_handler(scmd);
852 if (rtn == SUCCESS) {
853 if (!hostt->skip_settle_delay)
854 ssleep(BUS_RESET_SETTLE_TIME);
855 spin_lock_irqsave(host->host_lock, flags);
856 scsi_report_bus_reset(host, scmd_channel(scmd));
857 spin_unlock_irqrestore(host->host_lock, flags);
863 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
866 sdev->expecting_cc_ua = 1;
870 * scsi_try_target_reset - Ask host to perform a target reset
871 * @scmd: SCSI cmd used to send a target reset
874 * There is no timeout for this operation. if this operation is
875 * unreliable for a given host, then the host itself needs to put a
876 * timer on it, and set the host back to a consistent state prior to
879 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
882 enum scsi_disposition rtn;
883 struct Scsi_Host *host = scmd->device->host;
884 struct scsi_host_template *hostt = host->hostt;
886 if (!hostt->eh_target_reset_handler)
889 rtn = hostt->eh_target_reset_handler(scmd);
890 if (rtn == SUCCESS) {
891 spin_lock_irqsave(host->host_lock, flags);
892 __starget_for_each_device(scsi_target(scmd->device), NULL,
893 __scsi_report_device_reset);
894 spin_unlock_irqrestore(host->host_lock, flags);
901 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
902 * @scmd: SCSI cmd used to send BDR
905 * There is no timeout for this operation. if this operation is
906 * unreliable for a given host, then the host itself needs to put a
907 * timer on it, and set the host back to a consistent state prior to
910 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
912 enum scsi_disposition rtn;
913 struct scsi_host_template *hostt = scmd->device->host->hostt;
915 if (!hostt->eh_device_reset_handler)
918 rtn = hostt->eh_device_reset_handler(scmd);
920 __scsi_report_device_reset(scmd->device, NULL);
925 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
926 * @hostt: SCSI driver host template
927 * @scmd: SCSI cmd used to send a target reset
930 * SUCCESS, FAILED, or FAST_IO_FAIL
933 * SUCCESS does not necessarily indicate that the command
934 * has been aborted; it only indicates that the LLDDs
935 * has cleared all references to that command.
936 * LLDDs should return FAILED only if an abort was required
937 * but could not be executed. LLDDs should return FAST_IO_FAIL
938 * if the device is temporarily unavailable (eg due to a
939 * link down on FibreChannel)
941 static enum scsi_disposition
942 scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
944 if (!hostt->eh_abort_handler)
947 return hostt->eh_abort_handler(scmd);
950 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
952 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
953 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
954 if (scsi_try_target_reset(scmd) != SUCCESS)
955 if (scsi_try_bus_reset(scmd) != SUCCESS)
956 scsi_try_host_reset(scmd);
960 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
961 * @scmd: SCSI command structure to hijack
962 * @ses: structure to save restore information
963 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
964 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
965 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
967 * This function is used to save a scsi command information before re-execution
968 * as part of the error recovery process. If @sense_bytes is 0 the command
969 * sent must be one that does not transfer any data. If @sense_bytes != 0
970 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
971 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
973 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
974 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
976 struct scsi_device *sdev = scmd->device;
979 * We need saved copies of a number of fields - this is because
980 * error handling may need to overwrite these with different values
981 * to run different commands, and once error handling is complete,
982 * we will need to restore these values prior to running the actual
985 ses->cmd_len = scmd->cmd_len;
986 ses->cmnd = scmd->cmnd;
987 ses->data_direction = scmd->sc_data_direction;
988 ses->sdb = scmd->sdb;
989 ses->result = scmd->result;
990 ses->resid_len = scmd->req.resid_len;
991 ses->underflow = scmd->underflow;
992 ses->prot_op = scmd->prot_op;
993 ses->eh_eflags = scmd->eh_eflags;
995 scmd->prot_op = SCSI_PROT_NORMAL;
997 scmd->cmnd = ses->eh_cmnd;
998 memset(scmd->cmnd, 0, BLK_MAX_CDB);
999 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1001 scmd->req.resid_len = 0;
1004 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
1006 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
1008 scmd->sdb.table.sgl = &ses->sense_sgl;
1009 scmd->sc_data_direction = DMA_FROM_DEVICE;
1010 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
1011 scmd->cmnd[0] = REQUEST_SENSE;
1012 scmd->cmnd[4] = scmd->sdb.length;
1013 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1015 scmd->sc_data_direction = DMA_NONE;
1017 BUG_ON(cmnd_size > BLK_MAX_CDB);
1018 memcpy(scmd->cmnd, cmnd, cmnd_size);
1019 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1023 scmd->underflow = 0;
1025 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1026 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1027 (sdev->lun << 5 & 0xe0);
1030 * Zero the sense buffer. The scsi spec mandates that any
1031 * untransferred sense data should be interpreted as being zero.
1033 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1035 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1038 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1039 * @scmd: SCSI command structure to restore
1040 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1042 * Undo any damage done by above scsi_eh_prep_cmnd().
1044 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1047 * Restore original data
1049 scmd->cmd_len = ses->cmd_len;
1050 scmd->cmnd = ses->cmnd;
1051 scmd->sc_data_direction = ses->data_direction;
1052 scmd->sdb = ses->sdb;
1053 scmd->result = ses->result;
1054 scmd->req.resid_len = ses->resid_len;
1055 scmd->underflow = ses->underflow;
1056 scmd->prot_op = ses->prot_op;
1057 scmd->eh_eflags = ses->eh_eflags;
1059 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1062 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1063 * @scmd: SCSI command structure to hijack
1064 * @cmnd: CDB to send
1065 * @cmnd_size: size in bytes of @cmnd
1066 * @timeout: timeout for this request
1067 * @sense_bytes: size of sense data to copy or 0
1069 * This function is used to send a scsi command down to a target device
1070 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1073 * SUCCESS or FAILED or NEEDS_RETRY
1075 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1076 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
1078 struct scsi_device *sdev = scmd->device;
1079 struct Scsi_Host *shost = sdev->host;
1080 DECLARE_COMPLETION_ONSTACK(done);
1081 unsigned long timeleft = timeout, delay;
1082 struct scsi_eh_save ses;
1083 const unsigned long stall_for = msecs_to_jiffies(100);
1087 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1088 shost->eh_action = &done;
1090 scsi_log_send(scmd);
1091 scmd->scsi_done = scsi_eh_done;
1094 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1095 * change the SCSI device state after we have examined it and before
1096 * .queuecommand() is called.
1098 mutex_lock(&sdev->state_mutex);
1099 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1100 mutex_unlock(&sdev->state_mutex);
1101 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1102 "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1104 delay = min(timeleft, stall_for);
1106 msleep(jiffies_to_msecs(delay));
1107 mutex_lock(&sdev->state_mutex);
1109 if (sdev->sdev_state != SDEV_BLOCK)
1110 rtn = shost->hostt->queuecommand(shost, scmd);
1113 mutex_unlock(&sdev->state_mutex);
1116 if (timeleft > stall_for) {
1117 scsi_eh_restore_cmnd(scmd, &ses);
1118 timeleft -= stall_for;
1119 msleep(jiffies_to_msecs(stall_for));
1122 /* signal not to enter either branch of the if () below */
1126 timeleft = wait_for_completion_timeout(&done, timeout);
1130 shost->eh_action = NULL;
1132 scsi_log_completion(scmd, rtn);
1134 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1135 "%s timeleft: %ld\n",
1136 __func__, timeleft));
1139 * If there is time left scsi_eh_done got called, and we will examine
1140 * the actual status codes to see whether the command actually did
1141 * complete normally, else if we have a zero return and no time left,
1142 * the command must still be pending, so abort it and return FAILED.
1143 * If we never actually managed to issue the command, because
1144 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1145 * value above (so don't execute either branch of the if)
1148 rtn = scsi_eh_completed_normally(scmd);
1149 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1150 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1157 case ADD_TO_MLQUEUE:
1164 } else if (rtn != FAILED) {
1165 scsi_abort_eh_cmnd(scmd);
1169 scsi_eh_restore_cmnd(scmd, &ses);
1175 * scsi_request_sense - Request sense data from a particular target.
1176 * @scmd: SCSI cmd for request sense.
1179 * Some hosts automatically obtain this information, others require
1180 * that we obtain it on our own. This function will *not* return until
1181 * the command either times out, or it completes.
1183 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
1185 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1188 static enum scsi_disposition
1189 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
1191 if (!blk_rq_is_passthrough(scmd->request)) {
1192 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1193 if (sdrv->eh_action)
1194 rtn = sdrv->eh_action(scmd, rtn);
1200 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1201 * @scmd: Original SCSI cmd that eh has finished.
1202 * @done_q: Queue for processed commands.
1205 * We don't want to use the normal command completion while we are are
1206 * still handling errors - it may cause other commands to be queued,
1207 * and that would disturb what we are doing. Thus we really want to
1208 * keep a list of pending commands for final completion, and once we
1209 * are ready to leave error handling we handle completion for real.
1211 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1213 list_move_tail(&scmd->eh_entry, done_q);
1215 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1218 * scsi_eh_get_sense - Get device sense data.
1219 * @work_q: Queue of commands to process.
1220 * @done_q: Queue of processed commands.
1223 * See if we need to request sense information. if so, then get it
1224 * now, so we have a better idea of what to do.
1227 * This has the unfortunate side effect that if a shost adapter does
1228 * not automatically request sense information, we end up shutting
1229 * it down before we request it.
1231 * All drivers should request sense information internally these days,
1232 * so for now all I have to say is tough noogies if you end up in here.
1234 * XXX: Long term this code should go away, but that needs an audit of
1237 int scsi_eh_get_sense(struct list_head *work_q,
1238 struct list_head *done_q)
1240 struct scsi_cmnd *scmd, *next;
1241 struct Scsi_Host *shost;
1242 enum scsi_disposition rtn;
1245 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1246 * should not get sense.
1248 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1249 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1250 SCSI_SENSE_VALID(scmd))
1253 shost = scmd->device->host;
1254 if (scsi_host_eh_past_deadline(shost)) {
1255 SCSI_LOG_ERROR_RECOVERY(3,
1256 scmd_printk(KERN_INFO, scmd,
1257 "%s: skip request sense, past eh deadline\n",
1261 if (status_byte(scmd->result) != CHECK_CONDITION)
1263 * don't request sense if there's no check condition
1264 * status because the error we're processing isn't one
1265 * that has a sense code (and some devices get
1266 * confused by sense requests out of the blue)
1270 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1271 "%s: requesting sense\n",
1273 rtn = scsi_request_sense(scmd);
1277 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1278 "sense requested, result %x\n", scmd->result));
1279 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1281 rtn = scsi_decide_disposition(scmd);
1284 * if the result was normal, then just pass it along to the
1289 * We don't want this command reissued, just finished
1290 * with the sense data, so set retries to the max
1291 * allowed to ensure it won't get reissued. If the user
1292 * has requested infinite retries, we also want to
1293 * finish this command, so force completion by setting
1294 * retries and allowed to the same value.
1296 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1297 scmd->retries = scmd->allowed = 1;
1299 scmd->retries = scmd->allowed;
1300 else if (rtn != NEEDS_RETRY)
1303 scsi_eh_finish_cmd(scmd, done_q);
1306 return list_empty(work_q);
1308 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1311 * scsi_eh_tur - Send TUR to device.
1312 * @scmd: &scsi_cmnd to send TUR
1315 * 0 - Device is ready. 1 - Device NOT ready.
1317 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1319 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1321 enum scsi_disposition rtn;
1324 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1325 scmd->device->eh_timeout, 0);
1327 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1328 "%s return: %x\n", __func__, rtn));
1343 * scsi_eh_test_devices - check if devices are responding from error recovery.
1344 * @cmd_list: scsi commands in error recovery.
1345 * @work_q: queue for commands which still need more error recovery
1346 * @done_q: queue for commands which are finished
1347 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1350 * Tests if devices are in a working state. Commands to devices now in
1351 * a working state are sent to the done_q while commands to devices which
1352 * are still failing to respond are returned to the work_q for more
1355 static int scsi_eh_test_devices(struct list_head *cmd_list,
1356 struct list_head *work_q,
1357 struct list_head *done_q, int try_stu)
1359 struct scsi_cmnd *scmd, *next;
1360 struct scsi_device *sdev;
1363 while (!list_empty(cmd_list)) {
1364 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1365 sdev = scmd->device;
1368 if (scsi_host_eh_past_deadline(sdev->host)) {
1369 /* Push items back onto work_q */
1370 list_splice_init(cmd_list, work_q);
1371 SCSI_LOG_ERROR_RECOVERY(3,
1372 sdev_printk(KERN_INFO, sdev,
1373 "%s: skip test device, past eh deadline",
1379 finish_cmds = !scsi_device_online(scmd->device) ||
1380 (try_stu && !scsi_eh_try_stu(scmd) &&
1381 !scsi_eh_tur(scmd)) ||
1384 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1385 if (scmd->device == sdev) {
1388 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1389 scsi_eh_finish_cmd(scmd, done_q);
1391 list_move_tail(&scmd->eh_entry, work_q);
1394 return list_empty(work_q);
1398 * scsi_eh_try_stu - Send START_UNIT to device.
1399 * @scmd: &scsi_cmnd to send START_UNIT
1402 * 0 - Device is ready. 1 - Device NOT ready.
1404 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1406 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1408 if (scmd->device->allow_restart) {
1410 enum scsi_disposition rtn = NEEDS_RETRY;
1412 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1413 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1423 * scsi_eh_stu - send START_UNIT if needed
1424 * @shost: &scsi host being recovered.
1425 * @work_q: &list_head for pending commands.
1426 * @done_q: &list_head for processed commands.
1429 * If commands are failing due to not ready, initializing command required,
1430 * try revalidating the device, which will end up sending a start unit.
1432 static int scsi_eh_stu(struct Scsi_Host *shost,
1433 struct list_head *work_q,
1434 struct list_head *done_q)
1436 struct scsi_cmnd *scmd, *stu_scmd, *next;
1437 struct scsi_device *sdev;
1439 shost_for_each_device(sdev, shost) {
1440 if (scsi_host_eh_past_deadline(shost)) {
1441 SCSI_LOG_ERROR_RECOVERY(3,
1442 sdev_printk(KERN_INFO, sdev,
1443 "%s: skip START_UNIT, past eh deadline\n",
1445 scsi_device_put(sdev);
1449 list_for_each_entry(scmd, work_q, eh_entry)
1450 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1451 scsi_check_sense(scmd) == FAILED ) {
1459 SCSI_LOG_ERROR_RECOVERY(3,
1460 sdev_printk(KERN_INFO, sdev,
1461 "%s: Sending START_UNIT\n",
1464 if (!scsi_eh_try_stu(stu_scmd)) {
1465 if (!scsi_device_online(sdev) ||
1466 !scsi_eh_tur(stu_scmd)) {
1467 list_for_each_entry_safe(scmd, next,
1469 if (scmd->device == sdev &&
1470 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1471 scsi_eh_finish_cmd(scmd, done_q);
1475 SCSI_LOG_ERROR_RECOVERY(3,
1476 sdev_printk(KERN_INFO, sdev,
1477 "%s: START_UNIT failed\n",
1482 return list_empty(work_q);
1487 * scsi_eh_bus_device_reset - send bdr if needed
1488 * @shost: scsi host being recovered.
1489 * @work_q: &list_head for pending commands.
1490 * @done_q: &list_head for processed commands.
1493 * Try a bus device reset. Still, look to see whether we have multiple
1494 * devices that are jammed or not - if we have multiple devices, it
1495 * makes no sense to try bus_device_reset - we really would need to try
1496 * a bus_reset instead.
1498 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1499 struct list_head *work_q,
1500 struct list_head *done_q)
1502 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1503 struct scsi_device *sdev;
1504 enum scsi_disposition rtn;
1506 shost_for_each_device(sdev, shost) {
1507 if (scsi_host_eh_past_deadline(shost)) {
1508 SCSI_LOG_ERROR_RECOVERY(3,
1509 sdev_printk(KERN_INFO, sdev,
1510 "%s: skip BDR, past eh deadline\n",
1512 scsi_device_put(sdev);
1516 list_for_each_entry(scmd, work_q, eh_entry)
1517 if (scmd->device == sdev) {
1525 SCSI_LOG_ERROR_RECOVERY(3,
1526 sdev_printk(KERN_INFO, sdev,
1527 "%s: Sending BDR\n", current->comm));
1528 rtn = scsi_try_bus_device_reset(bdr_scmd);
1529 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1530 if (!scsi_device_online(sdev) ||
1531 rtn == FAST_IO_FAIL ||
1532 !scsi_eh_tur(bdr_scmd)) {
1533 list_for_each_entry_safe(scmd, next,
1535 if (scmd->device == sdev &&
1536 scsi_eh_action(scmd, rtn) != FAILED)
1537 scsi_eh_finish_cmd(scmd,
1542 SCSI_LOG_ERROR_RECOVERY(3,
1543 sdev_printk(KERN_INFO, sdev,
1544 "%s: BDR failed\n", current->comm));
1548 return list_empty(work_q);
1552 * scsi_eh_target_reset - send target reset if needed
1553 * @shost: scsi host being recovered.
1554 * @work_q: &list_head for pending commands.
1555 * @done_q: &list_head for processed commands.
1558 * Try a target reset.
1560 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1561 struct list_head *work_q,
1562 struct list_head *done_q)
1564 LIST_HEAD(tmp_list);
1565 LIST_HEAD(check_list);
1567 list_splice_init(work_q, &tmp_list);
1569 while (!list_empty(&tmp_list)) {
1570 struct scsi_cmnd *next, *scmd;
1571 enum scsi_disposition rtn;
1574 if (scsi_host_eh_past_deadline(shost)) {
1575 /* push back on work queue for further processing */
1576 list_splice_init(&check_list, work_q);
1577 list_splice_init(&tmp_list, work_q);
1578 SCSI_LOG_ERROR_RECOVERY(3,
1579 shost_printk(KERN_INFO, shost,
1580 "%s: Skip target reset, past eh deadline\n",
1582 return list_empty(work_q);
1585 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1588 SCSI_LOG_ERROR_RECOVERY(3,
1589 shost_printk(KERN_INFO, shost,
1590 "%s: Sending target reset to target %d\n",
1591 current->comm, id));
1592 rtn = scsi_try_target_reset(scmd);
1593 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1594 SCSI_LOG_ERROR_RECOVERY(3,
1595 shost_printk(KERN_INFO, shost,
1596 "%s: Target reset failed"
1598 current->comm, id));
1599 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1600 if (scmd_id(scmd) != id)
1604 list_move_tail(&scmd->eh_entry, &check_list);
1605 else if (rtn == FAST_IO_FAIL)
1606 scsi_eh_finish_cmd(scmd, done_q);
1608 /* push back on work queue for further processing */
1609 list_move(&scmd->eh_entry, work_q);
1613 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1617 * scsi_eh_bus_reset - send a bus reset
1618 * @shost: &scsi host being recovered.
1619 * @work_q: &list_head for pending commands.
1620 * @done_q: &list_head for processed commands.
1622 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1623 struct list_head *work_q,
1624 struct list_head *done_q)
1626 struct scsi_cmnd *scmd, *chan_scmd, *next;
1627 LIST_HEAD(check_list);
1628 unsigned int channel;
1629 enum scsi_disposition rtn;
1632 * we really want to loop over the various channels, and do this on
1633 * a channel by channel basis. we should also check to see if any
1634 * of the failed commands are on soft_reset devices, and if so, skip
1638 for (channel = 0; channel <= shost->max_channel; channel++) {
1639 if (scsi_host_eh_past_deadline(shost)) {
1640 list_splice_init(&check_list, work_q);
1641 SCSI_LOG_ERROR_RECOVERY(3,
1642 shost_printk(KERN_INFO, shost,
1643 "%s: skip BRST, past eh deadline\n",
1645 return list_empty(work_q);
1649 list_for_each_entry(scmd, work_q, eh_entry) {
1650 if (channel == scmd_channel(scmd)) {
1654 * FIXME add back in some support for
1655 * soft_reset devices.
1662 SCSI_LOG_ERROR_RECOVERY(3,
1663 shost_printk(KERN_INFO, shost,
1664 "%s: Sending BRST chan: %d\n",
1665 current->comm, channel));
1666 rtn = scsi_try_bus_reset(chan_scmd);
1667 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1668 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1669 if (channel == scmd_channel(scmd)) {
1670 if (rtn == FAST_IO_FAIL)
1671 scsi_eh_finish_cmd(scmd,
1674 list_move_tail(&scmd->eh_entry,
1679 SCSI_LOG_ERROR_RECOVERY(3,
1680 shost_printk(KERN_INFO, shost,
1681 "%s: BRST failed chan: %d\n",
1682 current->comm, channel));
1685 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1689 * scsi_eh_host_reset - send a host reset
1690 * @shost: host to be reset.
1691 * @work_q: &list_head for pending commands.
1692 * @done_q: &list_head for processed commands.
1694 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1695 struct list_head *work_q,
1696 struct list_head *done_q)
1698 struct scsi_cmnd *scmd, *next;
1699 LIST_HEAD(check_list);
1700 enum scsi_disposition rtn;
1702 if (!list_empty(work_q)) {
1703 scmd = list_entry(work_q->next,
1704 struct scsi_cmnd, eh_entry);
1706 SCSI_LOG_ERROR_RECOVERY(3,
1707 shost_printk(KERN_INFO, shost,
1708 "%s: Sending HRST\n",
1711 rtn = scsi_try_host_reset(scmd);
1712 if (rtn == SUCCESS) {
1713 list_splice_init(work_q, &check_list);
1714 } else if (rtn == FAST_IO_FAIL) {
1715 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1716 scsi_eh_finish_cmd(scmd, done_q);
1719 SCSI_LOG_ERROR_RECOVERY(3,
1720 shost_printk(KERN_INFO, shost,
1721 "%s: HRST failed\n",
1725 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1729 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1730 * @work_q: &list_head for pending commands.
1731 * @done_q: &list_head for processed commands.
1733 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1734 struct list_head *done_q)
1736 struct scsi_cmnd *scmd, *next;
1737 struct scsi_device *sdev;
1739 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1740 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1741 "not ready after error recovery\n");
1742 sdev = scmd->device;
1744 mutex_lock(&sdev->state_mutex);
1745 scsi_device_set_state(sdev, SDEV_OFFLINE);
1746 mutex_unlock(&sdev->state_mutex);
1748 scsi_eh_finish_cmd(scmd, done_q);
1754 * scsi_noretry_cmd - determine if command should be failed fast
1755 * @scmd: SCSI cmd to examine.
1757 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1759 switch (host_byte(scmd->result)) {
1765 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1767 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1769 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1770 status_byte(scmd->result) == RESERVATION_CONFLICT)
1773 case DID_SOFT_ERROR:
1774 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1777 if (status_byte(scmd->result) != CHECK_CONDITION)
1782 * assume caller has checked sense and determined
1783 * the check condition was retryable.
1785 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1786 blk_rq_is_passthrough(scmd->request))
1793 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1794 * @scmd: SCSI cmd to examine.
1797 * This is *only* called when we are examining the status after sending
1798 * out the actual data command. any commands that are queued for error
1799 * recovery (e.g. test_unit_ready) do *not* come through here.
1801 * When this routine returns failed, it means the error handler thread
1802 * is woken. In cases where the error code indicates an error that
1803 * doesn't require the error handler read (i.e. we don't need to
1804 * abort/reset), this function should return SUCCESS.
1806 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
1808 enum scsi_disposition rtn;
1811 * if the device is offline, then we clearly just pass the result back
1812 * up to the top level.
1814 if (!scsi_device_online(scmd->device)) {
1815 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1816 "%s: device offline - report as SUCCESS\n", __func__));
1821 * first check the host byte, to see if there is anything in there
1822 * that would indicate what we need to do.
1824 switch (host_byte(scmd->result)) {
1825 case DID_PASSTHROUGH:
1827 * no matter what, pass this through to the upper layer.
1828 * nuke this special code so that it looks like we are saying
1831 scmd->result &= 0xff00ffff;
1835 * looks good. drop through, and check the next byte.
1839 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1840 set_host_byte(scmd, DID_TIME_OUT);
1844 case DID_NO_CONNECT:
1845 case DID_BAD_TARGET:
1847 * note - this means that we just report the status back
1848 * to the top level driver, not that we actually think
1849 * that it indicates SUCCESS.
1852 case DID_SOFT_ERROR:
1854 * when the low level driver returns did_soft_error,
1855 * it is responsible for keeping an internal retry counter
1856 * in order to avoid endless loops (db)
1863 return ADD_TO_MLQUEUE;
1864 case DID_TRANSPORT_DISRUPTED:
1866 * LLD/transport was disrupted during processing of the IO.
1867 * The transport class is now blocked/blocking,
1868 * and the transport will decide what to do with the IO
1869 * based on its timers and recovery capablilities if
1870 * there are enough retries.
1873 case DID_TRANSPORT_FAILFAST:
1875 * The transport decided to failfast the IO (most likely
1876 * the fast io fail tmo fired), so send IO directly upwards.
1879 case DID_TRANSPORT_MARGINAL:
1881 * caller has decided not to do retries on
1882 * abort success, so send IO directly upwards
1886 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1887 status_byte(scmd->result) == RESERVATION_CONFLICT)
1889 * execute reservation conflict processing code
1899 * when we scan the bus, we get timeout messages for
1900 * these commands if there is no device available.
1901 * other hosts report did_no_connect for the same thing.
1903 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1904 scmd->cmnd[0] == INQUIRY)) {
1916 * next, check the message byte.
1918 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1922 * check the status byte to see if this indicates anything special.
1924 switch (status_byte(scmd->result)) {
1926 scsi_handle_queue_full(scmd->device);
1928 * the case of trying to send too many commands to a
1929 * tagged queueing device.
1934 * device can't talk to us at the moment. Should only
1935 * occur (SAM-3) when the task queue is empty, so will cause
1936 * the empty queue handling to trigger a stall in the
1939 return ADD_TO_MLQUEUE;
1941 if (scmd->cmnd[0] == REPORT_LUNS)
1942 scmd->device->sdev_target->expecting_lun_change = 0;
1943 scsi_handle_queue_ramp_up(scmd->device);
1945 case COMMAND_TERMINATED:
1949 case CHECK_CONDITION:
1950 rtn = scsi_check_sense(scmd);
1951 if (rtn == NEEDS_RETRY)
1953 /* if rtn == FAILED, we have no sense information;
1954 * returning FAILED will wake the error handler thread
1955 * to collect the sense and redo the decide
1958 case CONDITION_GOOD:
1959 case INTERMEDIATE_GOOD:
1960 case INTERMEDIATE_C_GOOD:
1963 * who knows? FIXME(eric)
1967 case RESERVATION_CONFLICT:
1968 sdev_printk(KERN_INFO, scmd->device,
1969 "reservation conflict\n");
1970 set_host_byte(scmd, DID_NEXUS_FAILURE);
1971 return SUCCESS; /* causes immediate i/o error */
1979 /* we requeue for retry because the error was retryable, and
1980 * the request was not marked fast fail. Note that above,
1981 * even if the request is marked fast fail, we still requeue
1982 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1983 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
1987 * no more retries - report this one back to upper level.
1993 static void eh_lock_door_done(struct request *req, blk_status_t status)
1995 blk_put_request(req);
1999 * scsi_eh_lock_door - Prevent medium removal for the specified device
2000 * @sdev: SCSI device to prevent medium removal
2003 * We must be called from process context.
2006 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
2007 * head of the devices request queue, and continue.
2009 static void scsi_eh_lock_door(struct scsi_device *sdev)
2011 struct request *req;
2012 struct scsi_request *rq;
2014 req = blk_get_request(sdev->request_queue, REQ_OP_DRV_IN, 0);
2019 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
2023 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
2025 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
2027 req->rq_flags |= RQF_QUIET;
2028 req->timeout = 10 * HZ;
2031 blk_execute_rq_nowait(NULL, req, 1, eh_lock_door_done);
2035 * scsi_restart_operations - restart io operations to the specified host.
2036 * @shost: Host we are restarting.
2039 * When we entered the error handler, we blocked all further i/o to
2040 * this device. we need to 'reverse' this process.
2042 static void scsi_restart_operations(struct Scsi_Host *shost)
2044 struct scsi_device *sdev;
2045 unsigned long flags;
2048 * If the door was locked, we need to insert a door lock request
2049 * onto the head of the SCSI request queue for the device. There
2050 * is no point trying to lock the door of an off-line device.
2052 shost_for_each_device(sdev, shost) {
2053 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2054 scsi_eh_lock_door(sdev);
2055 sdev->was_reset = 0;
2060 * next free up anything directly waiting upon the host. this
2061 * will be requests for character device operations, and also for
2062 * ioctls to queued block devices.
2064 SCSI_LOG_ERROR_RECOVERY(3,
2065 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2067 spin_lock_irqsave(shost->host_lock, flags);
2068 if (scsi_host_set_state(shost, SHOST_RUNNING))
2069 if (scsi_host_set_state(shost, SHOST_CANCEL))
2070 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2071 spin_unlock_irqrestore(shost->host_lock, flags);
2073 wake_up(&shost->host_wait);
2076 * finally we need to re-initiate requests that may be pending. we will
2077 * have had everything blocked while error handling is taking place, and
2078 * now that error recovery is done, we will need to ensure that these
2079 * requests are started.
2081 scsi_run_host_queues(shost);
2084 * if eh is active and host_eh_scheduled is pending we need to re-run
2085 * recovery. we do this check after scsi_run_host_queues() to allow
2086 * everything pent up since the last eh run a chance to make forward
2087 * progress before we sync again. Either we'll immediately re-run
2088 * recovery or scsi_device_unbusy() will wake us again when these
2089 * pending commands complete.
2091 spin_lock_irqsave(shost->host_lock, flags);
2092 if (shost->host_eh_scheduled)
2093 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2094 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2095 spin_unlock_irqrestore(shost->host_lock, flags);
2099 * scsi_eh_ready_devs - check device ready state and recover if not.
2100 * @shost: host to be recovered.
2101 * @work_q: &list_head for pending commands.
2102 * @done_q: &list_head for processed commands.
2104 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2105 struct list_head *work_q,
2106 struct list_head *done_q)
2108 if (!scsi_eh_stu(shost, work_q, done_q))
2109 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2110 if (!scsi_eh_target_reset(shost, work_q, done_q))
2111 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2112 if (!scsi_eh_host_reset(shost, work_q, done_q))
2113 scsi_eh_offline_sdevs(work_q,
2116 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2119 * scsi_eh_flush_done_q - finish processed commands or retry them.
2120 * @done_q: list_head of processed commands.
2122 void scsi_eh_flush_done_q(struct list_head *done_q)
2124 struct scsi_cmnd *scmd, *next;
2126 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2127 list_del_init(&scmd->eh_entry);
2128 if (scsi_device_online(scmd->device) &&
2129 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) &&
2130 scsi_eh_should_retry_cmd(scmd)) {
2131 SCSI_LOG_ERROR_RECOVERY(3,
2132 scmd_printk(KERN_INFO, scmd,
2133 "%s: flush retry cmd\n",
2135 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2138 * If just we got sense for the device (called
2139 * scsi_eh_get_sense), scmd->result is already
2140 * set, do not set DRIVER_TIMEOUT.
2143 scmd->result |= (DRIVER_TIMEOUT << 24);
2144 SCSI_LOG_ERROR_RECOVERY(3,
2145 scmd_printk(KERN_INFO, scmd,
2146 "%s: flush finish cmd\n",
2148 scsi_finish_command(scmd);
2152 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2155 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2156 * @shost: Host to unjam.
2159 * When we come in here, we *know* that all commands on the bus have
2160 * either completed, failed or timed out. we also know that no further
2161 * commands are being sent to the host, so things are relatively quiet
2162 * and we have freedom to fiddle with things as we wish.
2164 * This is only the *default* implementation. it is possible for
2165 * individual drivers to supply their own version of this function, and
2166 * if the maintainer wishes to do this, it is strongly suggested that
2167 * this function be taken as a template and modified. this function
2168 * was designed to correctly handle problems for about 95% of the
2169 * different cases out there, and it should always provide at least a
2170 * reasonable amount of error recovery.
2172 * Any command marked 'failed' or 'timeout' must eventually have
2173 * scsi_finish_cmd() called for it. we do all of the retry stuff
2174 * here, so when we restart the host after we return it should have an
2177 static void scsi_unjam_host(struct Scsi_Host *shost)
2179 unsigned long flags;
2180 LIST_HEAD(eh_work_q);
2181 LIST_HEAD(eh_done_q);
2183 spin_lock_irqsave(shost->host_lock, flags);
2184 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2185 spin_unlock_irqrestore(shost->host_lock, flags);
2187 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2189 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2190 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2192 spin_lock_irqsave(shost->host_lock, flags);
2193 if (shost->eh_deadline != -1)
2194 shost->last_reset = 0;
2195 spin_unlock_irqrestore(shost->host_lock, flags);
2196 scsi_eh_flush_done_q(&eh_done_q);
2200 * scsi_error_handler - SCSI error handler thread
2201 * @data: Host for which we are running.
2204 * This is the main error handling loop. This is run as a kernel thread
2205 * for every SCSI host and handles all error handling activity.
2207 int scsi_error_handler(void *data)
2209 struct Scsi_Host *shost = data;
2212 * We use TASK_INTERRUPTIBLE so that the thread is not
2213 * counted against the load average as a running process.
2214 * We never actually get interrupted because kthread_run
2215 * disables signal delivery for the created thread.
2219 * The sequence in kthread_stop() sets the stop flag first
2220 * then wakes the process. To avoid missed wakeups, the task
2221 * should always be in a non running state before the stop
2224 set_current_state(TASK_INTERRUPTIBLE);
2225 if (kthread_should_stop())
2228 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2229 shost->host_failed != scsi_host_busy(shost)) {
2230 SCSI_LOG_ERROR_RECOVERY(1,
2231 shost_printk(KERN_INFO, shost,
2232 "scsi_eh_%d: sleeping\n",
2238 __set_current_state(TASK_RUNNING);
2239 SCSI_LOG_ERROR_RECOVERY(1,
2240 shost_printk(KERN_INFO, shost,
2241 "scsi_eh_%d: waking up %d/%d/%d\n",
2242 shost->host_no, shost->host_eh_scheduled,
2244 scsi_host_busy(shost)));
2247 * We have a host that is failing for some reason. Figure out
2248 * what we need to do to get it up and online again (if we can).
2249 * If we fail, we end up taking the thing offline.
2251 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2252 SCSI_LOG_ERROR_RECOVERY(1,
2253 shost_printk(KERN_ERR, shost,
2254 "scsi_eh_%d: unable to autoresume\n",
2259 if (shost->transportt->eh_strategy_handler)
2260 shost->transportt->eh_strategy_handler(shost);
2262 scsi_unjam_host(shost);
2264 /* All scmds have been handled */
2265 shost->host_failed = 0;
2268 * Note - if the above fails completely, the action is to take
2269 * individual devices offline and flush the queue of any
2270 * outstanding requests that may have been pending. When we
2271 * restart, we restart any I/O to any other devices on the bus
2272 * which are still online.
2274 scsi_restart_operations(shost);
2275 if (!shost->eh_noresume)
2276 scsi_autopm_put_host(shost);
2278 __set_current_state(TASK_RUNNING);
2280 SCSI_LOG_ERROR_RECOVERY(1,
2281 shost_printk(KERN_INFO, shost,
2282 "Error handler scsi_eh_%d exiting\n",
2284 shost->ehandler = NULL;
2289 * Function: scsi_report_bus_reset()
2291 * Purpose: Utility function used by low-level drivers to report that
2292 * they have observed a bus reset on the bus being handled.
2294 * Arguments: shost - Host in question
2295 * channel - channel on which reset was observed.
2299 * Lock status: Host lock must be held.
2301 * Notes: This only needs to be called if the reset is one which
2302 * originates from an unknown location. Resets originated
2303 * by the mid-level itself don't need to call this, but there
2304 * should be no harm.
2306 * The main purpose of this is to make sure that a CHECK_CONDITION
2307 * is properly treated.
2309 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2311 struct scsi_device *sdev;
2313 __shost_for_each_device(sdev, shost) {
2314 if (channel == sdev_channel(sdev))
2315 __scsi_report_device_reset(sdev, NULL);
2318 EXPORT_SYMBOL(scsi_report_bus_reset);
2321 * Function: scsi_report_device_reset()
2323 * Purpose: Utility function used by low-level drivers to report that
2324 * they have observed a device reset on the device being handled.
2326 * Arguments: shost - Host in question
2327 * channel - channel on which reset was observed
2328 * target - target on which reset was observed
2332 * Lock status: Host lock must be held
2334 * Notes: This only needs to be called if the reset is one which
2335 * originates from an unknown location. Resets originated
2336 * by the mid-level itself don't need to call this, but there
2337 * should be no harm.
2339 * The main purpose of this is to make sure that a CHECK_CONDITION
2340 * is properly treated.
2342 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2344 struct scsi_device *sdev;
2346 __shost_for_each_device(sdev, shost) {
2347 if (channel == sdev_channel(sdev) &&
2348 target == sdev_id(sdev))
2349 __scsi_report_device_reset(sdev, NULL);
2352 EXPORT_SYMBOL(scsi_report_device_reset);
2355 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2360 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2361 * @dev: scsi_device to operate on
2362 * @arg: reset type (see sg.h)
2365 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2367 struct scsi_cmnd *scmd;
2368 struct Scsi_Host *shost = dev->host;
2370 unsigned long flags;
2372 enum scsi_disposition rtn;
2374 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2377 error = get_user(val, arg);
2381 if (scsi_autopm_get_host(shost) < 0)
2385 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2386 shost->hostt->cmd_size, GFP_KERNEL);
2388 goto out_put_autopm_host;
2389 blk_rq_init(NULL, rq);
2391 scmd = (struct scsi_cmnd *)(rq + 1);
2392 scsi_init_command(dev, scmd);
2394 scmd->cmnd = scsi_req(rq)->cmd;
2396 scmd->scsi_done = scsi_reset_provider_done_command;
2397 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2401 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2403 spin_lock_irqsave(shost->host_lock, flags);
2404 shost->tmf_in_progress = 1;
2405 spin_unlock_irqrestore(shost->host_lock, flags);
2407 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2408 case SG_SCSI_RESET_NOTHING:
2411 case SG_SCSI_RESET_DEVICE:
2412 rtn = scsi_try_bus_device_reset(scmd);
2413 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2416 case SG_SCSI_RESET_TARGET:
2417 rtn = scsi_try_target_reset(scmd);
2418 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2421 case SG_SCSI_RESET_BUS:
2422 rtn = scsi_try_bus_reset(scmd);
2423 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2426 case SG_SCSI_RESET_HOST:
2427 rtn = scsi_try_host_reset(scmd);
2436 error = (rtn == SUCCESS) ? 0 : -EIO;
2438 spin_lock_irqsave(shost->host_lock, flags);
2439 shost->tmf_in_progress = 0;
2440 spin_unlock_irqrestore(shost->host_lock, flags);
2443 * be sure to wake up anyone who was sleeping or had their queue
2444 * suspended while we performed the TMF.
2446 SCSI_LOG_ERROR_RECOVERY(3,
2447 shost_printk(KERN_INFO, shost,
2448 "waking up host to restart after TMF\n"));
2450 wake_up(&shost->host_wait);
2451 scsi_run_host_queues(shost);
2455 out_put_autopm_host:
2456 scsi_autopm_put_host(shost);
2460 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2461 struct scsi_sense_hdr *sshdr)
2463 return scsi_normalize_sense(cmd->sense_buffer,
2464 SCSI_SENSE_BUFFERSIZE, sshdr);
2466 EXPORT_SYMBOL(scsi_command_normalize_sense);
2469 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2470 * @sense_buffer: byte array of sense data
2471 * @sb_len: number of valid bytes in sense_buffer
2472 * @info_out: pointer to 64 integer where 8 or 4 byte information
2473 * field will be placed if found.
2476 * true if information field found, false if not found.
2478 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2485 switch (sense_buffer[0] & 0x7f) {
2488 if (sense_buffer[0] & 0x80) {
2489 *info_out = get_unaligned_be32(&sense_buffer[3]);
2495 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2497 if (ucp && (0xa == ucp[1])) {
2498 *info_out = get_unaligned_be64(&ucp[4]);
2506 EXPORT_SYMBOL(scsi_get_sense_info_fld);