Linux 4.16.11
[linux/fpc-iii.git] / drivers / scsi / scsi_error.c
blobca53a5f785ee239cee9b7de7c10edf75b01ff2d8
1 /*
2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
13 * minor cleanups.
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/gfp.h>
20 #include <linux/timer.h>
21 #include <linux/string.h>
22 #include <linux/kernel.h>
23 #include <linux/freezer.h>
24 #include <linux/kthread.h>
25 #include <linux/interrupt.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
28 #include <linux/jiffies.h>
30 #include <scsi/scsi.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_dbg.h>
33 #include <scsi/scsi_device.h>
34 #include <scsi/scsi_driver.h>
35 #include <scsi/scsi_eh.h>
36 #include <scsi/scsi_common.h>
37 #include <scsi/scsi_transport.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_ioctl.h>
40 #include <scsi/scsi_dh.h>
41 #include <scsi/sg.h>
43 #include "scsi_priv.h"
44 #include "scsi_logging.h"
45 #include "scsi_transport_api.h"
47 #include <trace/events/scsi.h>
49 #include <asm/unaligned.h>
51 static void scsi_eh_done(struct scsi_cmnd *scmd);
54 * These should *probably* be handled by the host itself.
55 * Since it is allowed to sleep, it probably should.
57 #define BUS_RESET_SETTLE_TIME (10)
58 #define HOST_RESET_SETTLE_TIME (10)
60 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
61 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
62 struct scsi_cmnd *);
64 void scsi_eh_wakeup(struct Scsi_Host *shost)
66 lockdep_assert_held(shost->host_lock);
68 if (atomic_read(&shost->host_busy) == shost->host_failed) {
69 trace_scsi_eh_wakeup(shost);
70 wake_up_process(shost->ehandler);
71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
72 "Waking error handler thread\n"));
76 /**
77 * scsi_schedule_eh - schedule EH for SCSI host
78 * @shost: SCSI host to invoke error handling on.
80 * Schedule SCSI EH without scmd.
82 void scsi_schedule_eh(struct Scsi_Host *shost)
84 unsigned long flags;
86 spin_lock_irqsave(shost->host_lock, flags);
88 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
89 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
90 shost->host_eh_scheduled++;
91 scsi_eh_wakeup(shost);
94 spin_unlock_irqrestore(shost->host_lock, flags);
96 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
98 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
100 if (!shost->last_reset || shost->eh_deadline == -1)
101 return 0;
104 * 32bit accesses are guaranteed to be atomic
105 * (on all supported architectures), so instead
106 * of using a spinlock we can as well double check
107 * if eh_deadline has been set to 'off' during the
108 * time_before call.
110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
111 shost->eh_deadline > -1)
112 return 0;
114 return 1;
118 * scmd_eh_abort_handler - Handle command aborts
119 * @work: command to be aborted.
121 void
122 scmd_eh_abort_handler(struct work_struct *work)
124 struct scsi_cmnd *scmd =
125 container_of(work, struct scsi_cmnd, abort_work.work);
126 struct scsi_device *sdev = scmd->device;
127 int rtn;
129 if (scsi_host_eh_past_deadline(sdev->host)) {
130 SCSI_LOG_ERROR_RECOVERY(3,
131 scmd_printk(KERN_INFO, scmd,
132 "eh timeout, not aborting\n"));
133 } else {
134 SCSI_LOG_ERROR_RECOVERY(3,
135 scmd_printk(KERN_INFO, scmd,
136 "aborting command\n"));
137 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
138 if (rtn == SUCCESS) {
139 set_host_byte(scmd, DID_TIME_OUT);
140 if (scsi_host_eh_past_deadline(sdev->host)) {
141 SCSI_LOG_ERROR_RECOVERY(3,
142 scmd_printk(KERN_INFO, scmd,
143 "eh timeout, not retrying "
144 "aborted command\n"));
145 } else if (!scsi_noretry_cmd(scmd) &&
146 (++scmd->retries <= scmd->allowed)) {
147 SCSI_LOG_ERROR_RECOVERY(3,
148 scmd_printk(KERN_WARNING, scmd,
149 "retry aborted command\n"));
150 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
151 return;
152 } else {
153 SCSI_LOG_ERROR_RECOVERY(3,
154 scmd_printk(KERN_WARNING, scmd,
155 "finish aborted command\n"));
156 scsi_finish_command(scmd);
157 return;
159 } else {
160 SCSI_LOG_ERROR_RECOVERY(3,
161 scmd_printk(KERN_INFO, scmd,
162 "cmd abort %s\n",
163 (rtn == FAST_IO_FAIL) ?
164 "not send" : "failed"));
168 scsi_eh_scmd_add(scmd);
172 * scsi_abort_command - schedule a command abort
173 * @scmd: scmd to abort.
175 * We only need to abort commands after a command timeout
177 static int
178 scsi_abort_command(struct scsi_cmnd *scmd)
180 struct scsi_device *sdev = scmd->device;
181 struct Scsi_Host *shost = sdev->host;
182 unsigned long flags;
184 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
186 * Retry after abort failed, escalate to next level.
188 SCSI_LOG_ERROR_RECOVERY(3,
189 scmd_printk(KERN_INFO, scmd,
190 "previous abort failed\n"));
191 BUG_ON(delayed_work_pending(&scmd->abort_work));
192 return FAILED;
195 spin_lock_irqsave(shost->host_lock, flags);
196 if (shost->eh_deadline != -1 && !shost->last_reset)
197 shost->last_reset = jiffies;
198 spin_unlock_irqrestore(shost->host_lock, flags);
200 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
201 SCSI_LOG_ERROR_RECOVERY(3,
202 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
203 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
204 return SUCCESS;
208 * scsi_eh_reset - call into ->eh_action to reset internal counters
209 * @scmd: scmd to run eh on.
211 * The scsi driver might be carrying internal state about the
212 * devices, so we need to call into the driver to reset the
213 * internal state once the error handler is started.
215 static void scsi_eh_reset(struct scsi_cmnd *scmd)
217 if (!blk_rq_is_passthrough(scmd->request)) {
218 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
219 if (sdrv->eh_reset)
220 sdrv->eh_reset(scmd);
224 static void scsi_eh_inc_host_failed(struct rcu_head *head)
226 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
227 struct Scsi_Host *shost = scmd->device->host;
228 unsigned long flags;
230 spin_lock_irqsave(shost->host_lock, flags);
231 shost->host_failed++;
232 scsi_eh_wakeup(shost);
233 spin_unlock_irqrestore(shost->host_lock, flags);
237 * scsi_eh_scmd_add - add scsi cmd to error handling.
238 * @scmd: scmd to run eh on.
240 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
242 struct Scsi_Host *shost = scmd->device->host;
243 unsigned long flags;
244 int ret;
246 WARN_ON_ONCE(!shost->ehandler);
248 spin_lock_irqsave(shost->host_lock, flags);
249 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
250 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
251 WARN_ON_ONCE(ret);
253 if (shost->eh_deadline != -1 && !shost->last_reset)
254 shost->last_reset = jiffies;
256 scsi_eh_reset(scmd);
257 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
258 spin_unlock_irqrestore(shost->host_lock, flags);
260 * Ensure that all tasks observe the host state change before the
261 * host_failed change.
263 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
267 * scsi_times_out - Timeout function for normal scsi commands.
268 * @req: request that is timing out.
270 * Notes:
271 * We do not need to lock this. There is the potential for a race
272 * only in that the normal completion handling might run, but if the
273 * normal completion function determines that the timer has already
274 * fired, then it mustn't do anything.
276 enum blk_eh_timer_return scsi_times_out(struct request *req)
278 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
279 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
280 struct Scsi_Host *host = scmd->device->host;
282 trace_scsi_dispatch_cmd_timeout(scmd);
283 scsi_log_completion(scmd, TIMEOUT_ERROR);
285 if (host->eh_deadline != -1 && !host->last_reset)
286 host->last_reset = jiffies;
288 if (host->hostt->eh_timed_out)
289 rtn = host->hostt->eh_timed_out(scmd);
291 if (rtn == BLK_EH_NOT_HANDLED) {
292 if (scsi_abort_command(scmd) != SUCCESS) {
293 set_host_byte(scmd, DID_TIME_OUT);
294 scsi_eh_scmd_add(scmd);
298 return rtn;
302 * scsi_block_when_processing_errors - Prevent cmds from being queued.
303 * @sdev: Device on which we are performing recovery.
305 * Description:
306 * We block until the host is out of error recovery, and then check to
307 * see whether the host or the device is offline.
309 * Return value:
310 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
312 int scsi_block_when_processing_errors(struct scsi_device *sdev)
314 int online;
316 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
318 online = scsi_device_online(sdev);
320 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
321 "%s: rtn: %d\n", __func__, online));
323 return online;
325 EXPORT_SYMBOL(scsi_block_when_processing_errors);
327 #ifdef CONFIG_SCSI_LOGGING
329 * scsi_eh_prt_fail_stats - Log info on failures.
330 * @shost: scsi host being recovered.
331 * @work_q: Queue of scsi cmds to process.
333 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
334 struct list_head *work_q)
336 struct scsi_cmnd *scmd;
337 struct scsi_device *sdev;
338 int total_failures = 0;
339 int cmd_failed = 0;
340 int cmd_cancel = 0;
341 int devices_failed = 0;
343 shost_for_each_device(sdev, shost) {
344 list_for_each_entry(scmd, work_q, eh_entry) {
345 if (scmd->device == sdev) {
346 ++total_failures;
347 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
348 ++cmd_cancel;
349 else
350 ++cmd_failed;
354 if (cmd_cancel || cmd_failed) {
355 SCSI_LOG_ERROR_RECOVERY(3,
356 shost_printk(KERN_INFO, shost,
357 "%s: cmds failed: %d, cancel: %d\n",
358 __func__, cmd_failed,
359 cmd_cancel));
360 cmd_cancel = 0;
361 cmd_failed = 0;
362 ++devices_failed;
366 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
367 "Total of %d commands on %d"
368 " devices require eh work\n",
369 total_failures, devices_failed));
371 #endif
374 * scsi_report_lun_change - Set flag on all *other* devices on the same target
375 * to indicate that a UNIT ATTENTION is expected.
376 * @sdev: Device reporting the UNIT ATTENTION
378 static void scsi_report_lun_change(struct scsi_device *sdev)
380 sdev->sdev_target->expecting_lun_change = 1;
384 * scsi_report_sense - Examine scsi sense information and log messages for
385 * certain conditions, also issue uevents for some of them.
386 * @sdev: Device reporting the sense code
387 * @sshdr: sshdr to be examined
389 static void scsi_report_sense(struct scsi_device *sdev,
390 struct scsi_sense_hdr *sshdr)
392 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
394 if (sshdr->sense_key == UNIT_ATTENTION) {
395 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
396 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
397 sdev_printk(KERN_WARNING, sdev,
398 "Inquiry data has changed");
399 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
400 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
401 scsi_report_lun_change(sdev);
402 sdev_printk(KERN_WARNING, sdev,
403 "Warning! Received an indication that the "
404 "LUN assignments on this target have "
405 "changed. The Linux SCSI layer does not "
406 "automatically remap LUN assignments.\n");
407 } else if (sshdr->asc == 0x3f)
408 sdev_printk(KERN_WARNING, sdev,
409 "Warning! Received an indication that the "
410 "operating parameters on this target have "
411 "changed. The Linux SCSI layer does not "
412 "automatically adjust these parameters.\n");
414 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
415 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
416 sdev_printk(KERN_WARNING, sdev,
417 "Warning! Received an indication that the "
418 "LUN reached a thin provisioning soft "
419 "threshold.\n");
422 if (sshdr->asc == 0x29) {
423 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
424 sdev_printk(KERN_WARNING, sdev,
425 "Power-on or device reset occurred\n");
428 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
429 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
430 sdev_printk(KERN_WARNING, sdev,
431 "Mode parameters changed");
432 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
433 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
434 sdev_printk(KERN_WARNING, sdev,
435 "Asymmetric access state changed");
436 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
437 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
438 sdev_printk(KERN_WARNING, sdev,
439 "Capacity data has changed");
440 } else if (sshdr->asc == 0x2a)
441 sdev_printk(KERN_WARNING, sdev,
442 "Parameters changed");
445 if (evt_type != SDEV_EVT_MAXBITS) {
446 set_bit(evt_type, sdev->pending_events);
447 schedule_work(&sdev->event_work);
452 * scsi_check_sense - Examine scsi cmd sense
453 * @scmd: Cmd to have sense checked.
455 * Return value:
456 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
458 * Notes:
459 * When a deferred error is detected the current command has
460 * not been executed and needs retrying.
462 int scsi_check_sense(struct scsi_cmnd *scmd)
464 struct scsi_device *sdev = scmd->device;
465 struct scsi_sense_hdr sshdr;
467 if (! scsi_command_normalize_sense(scmd, &sshdr))
468 return FAILED; /* no valid sense data */
470 scsi_report_sense(sdev, &sshdr);
472 if (scsi_sense_is_deferred(&sshdr))
473 return NEEDS_RETRY;
475 if (sdev->handler && sdev->handler->check_sense) {
476 int rc;
478 rc = sdev->handler->check_sense(sdev, &sshdr);
479 if (rc != SCSI_RETURN_NOT_HANDLED)
480 return rc;
481 /* handler does not care. Drop down to default handling */
484 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
486 * nasty: for mid-layer issued TURs, we need to return the
487 * actual sense data without any recovery attempt. For eh
488 * issued ones, we need to try to recover and interpret
490 return SUCCESS;
493 * Previous logic looked for FILEMARK, EOM or ILI which are
494 * mainly associated with tapes and returned SUCCESS.
496 if (sshdr.response_code == 0x70) {
497 /* fixed format */
498 if (scmd->sense_buffer[2] & 0xe0)
499 return SUCCESS;
500 } else {
502 * descriptor format: look for "stream commands sense data
503 * descriptor" (see SSC-3). Assume single sense data
504 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
506 if ((sshdr.additional_length > 3) &&
507 (scmd->sense_buffer[8] == 0x4) &&
508 (scmd->sense_buffer[11] & 0xe0))
509 return SUCCESS;
512 switch (sshdr.sense_key) {
513 case NO_SENSE:
514 return SUCCESS;
515 case RECOVERED_ERROR:
516 return /* soft_error */ SUCCESS;
518 case ABORTED_COMMAND:
519 if (sshdr.asc == 0x10) /* DIF */
520 return SUCCESS;
522 return NEEDS_RETRY;
523 case NOT_READY:
524 case UNIT_ATTENTION:
526 * if we are expecting a cc/ua because of a bus reset that we
527 * performed, treat this just as a retry. otherwise this is
528 * information that we should pass up to the upper-level driver
529 * so that we can deal with it there.
531 if (scmd->device->expecting_cc_ua) {
533 * Because some device does not queue unit
534 * attentions correctly, we carefully check
535 * additional sense code and qualifier so as
536 * not to squash media change unit attention.
538 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
539 scmd->device->expecting_cc_ua = 0;
540 return NEEDS_RETRY;
544 * we might also expect a cc/ua if another LUN on the target
545 * reported a UA with an ASC/ASCQ of 3F 0E -
546 * REPORTED LUNS DATA HAS CHANGED.
548 if (scmd->device->sdev_target->expecting_lun_change &&
549 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
550 return NEEDS_RETRY;
552 * if the device is in the process of becoming ready, we
553 * should retry.
555 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
556 return NEEDS_RETRY;
558 * if the device is not started, we need to wake
559 * the error handler to start the motor
561 if (scmd->device->allow_restart &&
562 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
563 return FAILED;
565 * Pass the UA upwards for a determination in the completion
566 * functions.
568 return SUCCESS;
570 /* these are not supported */
571 case DATA_PROTECT:
572 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
573 /* Thin provisioning hard threshold reached */
574 set_host_byte(scmd, DID_ALLOC_FAILURE);
575 return SUCCESS;
577 /* FALLTHROUGH */
578 case COPY_ABORTED:
579 case VOLUME_OVERFLOW:
580 case MISCOMPARE:
581 case BLANK_CHECK:
582 set_host_byte(scmd, DID_TARGET_FAILURE);
583 return SUCCESS;
585 case MEDIUM_ERROR:
586 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
587 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
588 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
589 set_host_byte(scmd, DID_MEDIUM_ERROR);
590 return SUCCESS;
592 return NEEDS_RETRY;
594 case HARDWARE_ERROR:
595 if (scmd->device->retry_hwerror)
596 return ADD_TO_MLQUEUE;
597 else
598 set_host_byte(scmd, DID_TARGET_FAILURE);
599 /* FALLTHROUGH */
601 case ILLEGAL_REQUEST:
602 if (sshdr.asc == 0x20 || /* Invalid command operation code */
603 sshdr.asc == 0x21 || /* Logical block address out of range */
604 sshdr.asc == 0x22 || /* Invalid function */
605 sshdr.asc == 0x24 || /* Invalid field in cdb */
606 sshdr.asc == 0x26 || /* Parameter value invalid */
607 sshdr.asc == 0x27) { /* Write protected */
608 set_host_byte(scmd, DID_TARGET_FAILURE);
610 return SUCCESS;
612 default:
613 return SUCCESS;
616 EXPORT_SYMBOL_GPL(scsi_check_sense);
618 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
620 struct scsi_host_template *sht = sdev->host->hostt;
621 struct scsi_device *tmp_sdev;
623 if (!sht->track_queue_depth ||
624 sdev->queue_depth >= sdev->max_queue_depth)
625 return;
627 if (time_before(jiffies,
628 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
629 return;
631 if (time_before(jiffies,
632 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
633 return;
636 * Walk all devices of a target and do
637 * ramp up on them.
639 shost_for_each_device(tmp_sdev, sdev->host) {
640 if (tmp_sdev->channel != sdev->channel ||
641 tmp_sdev->id != sdev->id ||
642 tmp_sdev->queue_depth == sdev->max_queue_depth)
643 continue;
645 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
646 sdev->last_queue_ramp_up = jiffies;
650 static void scsi_handle_queue_full(struct scsi_device *sdev)
652 struct scsi_host_template *sht = sdev->host->hostt;
653 struct scsi_device *tmp_sdev;
655 if (!sht->track_queue_depth)
656 return;
658 shost_for_each_device(tmp_sdev, sdev->host) {
659 if (tmp_sdev->channel != sdev->channel ||
660 tmp_sdev->id != sdev->id)
661 continue;
663 * We do not know the number of commands that were at
664 * the device when we got the queue full so we start
665 * from the highest possible value and work our way down.
667 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
672 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
673 * @scmd: SCSI cmd to examine.
675 * Notes:
676 * This is *only* called when we are examining the status of commands
677 * queued during error recovery. the main difference here is that we
678 * don't allow for the possibility of retries here, and we are a lot
679 * more restrictive about what we consider acceptable.
681 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
684 * first check the host byte, to see if there is anything in there
685 * that would indicate what we need to do.
687 if (host_byte(scmd->result) == DID_RESET) {
689 * rats. we are already in the error handler, so we now
690 * get to try and figure out what to do next. if the sense
691 * is valid, we have a pretty good idea of what to do.
692 * if not, we mark it as FAILED.
694 return scsi_check_sense(scmd);
696 if (host_byte(scmd->result) != DID_OK)
697 return FAILED;
700 * next, check the message byte.
702 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
703 return FAILED;
706 * now, check the status byte to see if this indicates
707 * anything special.
709 switch (status_byte(scmd->result)) {
710 case GOOD:
711 scsi_handle_queue_ramp_up(scmd->device);
712 /* FALLTHROUGH */
713 case COMMAND_TERMINATED:
714 return SUCCESS;
715 case CHECK_CONDITION:
716 return scsi_check_sense(scmd);
717 case CONDITION_GOOD:
718 case INTERMEDIATE_GOOD:
719 case INTERMEDIATE_C_GOOD:
721 * who knows? FIXME(eric)
723 return SUCCESS;
724 case RESERVATION_CONFLICT:
725 if (scmd->cmnd[0] == TEST_UNIT_READY)
726 /* it is a success, we probed the device and
727 * found it */
728 return SUCCESS;
729 /* otherwise, we failed to send the command */
730 return FAILED;
731 case QUEUE_FULL:
732 scsi_handle_queue_full(scmd->device);
733 /* fall through */
734 case BUSY:
735 return NEEDS_RETRY;
736 default:
737 return FAILED;
739 return FAILED;
743 * scsi_eh_done - Completion function for error handling.
744 * @scmd: Cmd that is done.
746 static void scsi_eh_done(struct scsi_cmnd *scmd)
748 struct completion *eh_action;
750 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
751 "%s result: %x\n", __func__, scmd->result));
753 eh_action = scmd->device->host->eh_action;
754 if (eh_action)
755 complete(eh_action);
759 * scsi_try_host_reset - ask host adapter to reset itself
760 * @scmd: SCSI cmd to send host reset.
762 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
764 unsigned long flags;
765 int rtn;
766 struct Scsi_Host *host = scmd->device->host;
767 struct scsi_host_template *hostt = host->hostt;
769 SCSI_LOG_ERROR_RECOVERY(3,
770 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
772 if (!hostt->eh_host_reset_handler)
773 return FAILED;
775 rtn = hostt->eh_host_reset_handler(scmd);
777 if (rtn == SUCCESS) {
778 if (!hostt->skip_settle_delay)
779 ssleep(HOST_RESET_SETTLE_TIME);
780 spin_lock_irqsave(host->host_lock, flags);
781 scsi_report_bus_reset(host, scmd_channel(scmd));
782 spin_unlock_irqrestore(host->host_lock, flags);
785 return rtn;
789 * scsi_try_bus_reset - ask host to perform a bus reset
790 * @scmd: SCSI cmd to send bus reset.
792 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
794 unsigned long flags;
795 int rtn;
796 struct Scsi_Host *host = scmd->device->host;
797 struct scsi_host_template *hostt = host->hostt;
799 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
800 "%s: Snd Bus RST\n", __func__));
802 if (!hostt->eh_bus_reset_handler)
803 return FAILED;
805 rtn = hostt->eh_bus_reset_handler(scmd);
807 if (rtn == SUCCESS) {
808 if (!hostt->skip_settle_delay)
809 ssleep(BUS_RESET_SETTLE_TIME);
810 spin_lock_irqsave(host->host_lock, flags);
811 scsi_report_bus_reset(host, scmd_channel(scmd));
812 spin_unlock_irqrestore(host->host_lock, flags);
815 return rtn;
818 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
820 sdev->was_reset = 1;
821 sdev->expecting_cc_ua = 1;
825 * scsi_try_target_reset - Ask host to perform a target reset
826 * @scmd: SCSI cmd used to send a target reset
828 * Notes:
829 * There is no timeout for this operation. if this operation is
830 * unreliable for a given host, then the host itself needs to put a
831 * timer on it, and set the host back to a consistent state prior to
832 * returning.
834 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
836 unsigned long flags;
837 int rtn;
838 struct Scsi_Host *host = scmd->device->host;
839 struct scsi_host_template *hostt = host->hostt;
841 if (!hostt->eh_target_reset_handler)
842 return FAILED;
844 rtn = hostt->eh_target_reset_handler(scmd);
845 if (rtn == SUCCESS) {
846 spin_lock_irqsave(host->host_lock, flags);
847 __starget_for_each_device(scsi_target(scmd->device), NULL,
848 __scsi_report_device_reset);
849 spin_unlock_irqrestore(host->host_lock, flags);
852 return rtn;
856 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
857 * @scmd: SCSI cmd used to send BDR
859 * Notes:
860 * There is no timeout for this operation. if this operation is
861 * unreliable for a given host, then the host itself needs to put a
862 * timer on it, and set the host back to a consistent state prior to
863 * returning.
865 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
867 int rtn;
868 struct scsi_host_template *hostt = scmd->device->host->hostt;
870 if (!hostt->eh_device_reset_handler)
871 return FAILED;
873 rtn = hostt->eh_device_reset_handler(scmd);
874 if (rtn == SUCCESS)
875 __scsi_report_device_reset(scmd->device, NULL);
876 return rtn;
880 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
881 * @hostt: SCSI driver host template
882 * @scmd: SCSI cmd used to send a target reset
884 * Return value:
885 * SUCCESS, FAILED, or FAST_IO_FAIL
887 * Notes:
888 * SUCCESS does not necessarily indicate that the command
889 * has been aborted; it only indicates that the LLDDs
890 * has cleared all references to that command.
891 * LLDDs should return FAILED only if an abort was required
892 * but could not be executed. LLDDs should return FAST_IO_FAIL
893 * if the device is temporarily unavailable (eg due to a
894 * link down on FibreChannel)
896 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
897 struct scsi_cmnd *scmd)
899 if (!hostt->eh_abort_handler)
900 return FAILED;
902 return hostt->eh_abort_handler(scmd);
905 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
907 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
908 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
909 if (scsi_try_target_reset(scmd) != SUCCESS)
910 if (scsi_try_bus_reset(scmd) != SUCCESS)
911 scsi_try_host_reset(scmd);
915 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
916 * @scmd: SCSI command structure to hijack
917 * @ses: structure to save restore information
918 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
919 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
920 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
922 * This function is used to save a scsi command information before re-execution
923 * as part of the error recovery process. If @sense_bytes is 0 the command
924 * sent must be one that does not transfer any data. If @sense_bytes != 0
925 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
926 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
928 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
929 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
931 struct scsi_device *sdev = scmd->device;
934 * We need saved copies of a number of fields - this is because
935 * error handling may need to overwrite these with different values
936 * to run different commands, and once error handling is complete,
937 * we will need to restore these values prior to running the actual
938 * command.
940 ses->cmd_len = scmd->cmd_len;
941 ses->cmnd = scmd->cmnd;
942 ses->data_direction = scmd->sc_data_direction;
943 ses->sdb = scmd->sdb;
944 ses->next_rq = scmd->request->next_rq;
945 ses->result = scmd->result;
946 ses->underflow = scmd->underflow;
947 ses->prot_op = scmd->prot_op;
948 ses->eh_eflags = scmd->eh_eflags;
950 scmd->prot_op = SCSI_PROT_NORMAL;
951 scmd->eh_eflags = 0;
952 scmd->cmnd = ses->eh_cmnd;
953 memset(scmd->cmnd, 0, BLK_MAX_CDB);
954 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
955 scmd->request->next_rq = NULL;
956 scmd->result = 0;
958 if (sense_bytes) {
959 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
960 sense_bytes);
961 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
962 scmd->sdb.length);
963 scmd->sdb.table.sgl = &ses->sense_sgl;
964 scmd->sc_data_direction = DMA_FROM_DEVICE;
965 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
966 scmd->cmnd[0] = REQUEST_SENSE;
967 scmd->cmnd[4] = scmd->sdb.length;
968 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
969 } else {
970 scmd->sc_data_direction = DMA_NONE;
971 if (cmnd) {
972 BUG_ON(cmnd_size > BLK_MAX_CDB);
973 memcpy(scmd->cmnd, cmnd, cmnd_size);
974 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
978 scmd->underflow = 0;
980 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
981 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
982 (sdev->lun << 5 & 0xe0);
985 * Zero the sense buffer. The scsi spec mandates that any
986 * untransferred sense data should be interpreted as being zero.
988 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
990 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
993 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
994 * @scmd: SCSI command structure to restore
995 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
997 * Undo any damage done by above scsi_eh_prep_cmnd().
999 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1002 * Restore original data
1004 scmd->cmd_len = ses->cmd_len;
1005 scmd->cmnd = ses->cmnd;
1006 scmd->sc_data_direction = ses->data_direction;
1007 scmd->sdb = ses->sdb;
1008 scmd->request->next_rq = ses->next_rq;
1009 scmd->result = ses->result;
1010 scmd->underflow = ses->underflow;
1011 scmd->prot_op = ses->prot_op;
1012 scmd->eh_eflags = ses->eh_eflags;
1014 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1017 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1018 * @scmd: SCSI command structure to hijack
1019 * @cmnd: CDB to send
1020 * @cmnd_size: size in bytes of @cmnd
1021 * @timeout: timeout for this request
1022 * @sense_bytes: size of sense data to copy or 0
1024 * This function is used to send a scsi command down to a target device
1025 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1027 * Return value:
1028 * SUCCESS or FAILED or NEEDS_RETRY
1030 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1031 int cmnd_size, int timeout, unsigned sense_bytes)
1033 struct scsi_device *sdev = scmd->device;
1034 struct Scsi_Host *shost = sdev->host;
1035 DECLARE_COMPLETION_ONSTACK(done);
1036 unsigned long timeleft = timeout;
1037 struct scsi_eh_save ses;
1038 const unsigned long stall_for = msecs_to_jiffies(100);
1039 int rtn;
1041 retry:
1042 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1043 shost->eh_action = &done;
1045 scsi_log_send(scmd);
1046 scmd->scsi_done = scsi_eh_done;
1047 rtn = shost->hostt->queuecommand(shost, scmd);
1048 if (rtn) {
1049 if (timeleft > stall_for) {
1050 scsi_eh_restore_cmnd(scmd, &ses);
1051 timeleft -= stall_for;
1052 msleep(jiffies_to_msecs(stall_for));
1053 goto retry;
1055 /* signal not to enter either branch of the if () below */
1056 timeleft = 0;
1057 rtn = FAILED;
1058 } else {
1059 timeleft = wait_for_completion_timeout(&done, timeout);
1060 rtn = SUCCESS;
1063 shost->eh_action = NULL;
1065 scsi_log_completion(scmd, rtn);
1067 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1068 "%s timeleft: %ld\n",
1069 __func__, timeleft));
1072 * If there is time left scsi_eh_done got called, and we will examine
1073 * the actual status codes to see whether the command actually did
1074 * complete normally, else if we have a zero return and no time left,
1075 * the command must still be pending, so abort it and return FAILED.
1076 * If we never actually managed to issue the command, because
1077 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1078 * value above (so don't execute either branch of the if)
1080 if (timeleft) {
1081 rtn = scsi_eh_completed_normally(scmd);
1082 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1083 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1085 switch (rtn) {
1086 case SUCCESS:
1087 case NEEDS_RETRY:
1088 case FAILED:
1089 break;
1090 case ADD_TO_MLQUEUE:
1091 rtn = NEEDS_RETRY;
1092 break;
1093 default:
1094 rtn = FAILED;
1095 break;
1097 } else if (rtn != FAILED) {
1098 scsi_abort_eh_cmnd(scmd);
1099 rtn = FAILED;
1102 scsi_eh_restore_cmnd(scmd, &ses);
1104 return rtn;
1108 * scsi_request_sense - Request sense data from a particular target.
1109 * @scmd: SCSI cmd for request sense.
1111 * Notes:
1112 * Some hosts automatically obtain this information, others require
1113 * that we obtain it on our own. This function will *not* return until
1114 * the command either times out, or it completes.
1116 static int scsi_request_sense(struct scsi_cmnd *scmd)
1118 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1121 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1123 if (!blk_rq_is_passthrough(scmd->request)) {
1124 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1125 if (sdrv->eh_action)
1126 rtn = sdrv->eh_action(scmd, rtn);
1128 return rtn;
1132 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1133 * @scmd: Original SCSI cmd that eh has finished.
1134 * @done_q: Queue for processed commands.
1136 * Notes:
1137 * We don't want to use the normal command completion while we are are
1138 * still handling errors - it may cause other commands to be queued,
1139 * and that would disturb what we are doing. Thus we really want to
1140 * keep a list of pending commands for final completion, and once we
1141 * are ready to leave error handling we handle completion for real.
1143 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1145 list_move_tail(&scmd->eh_entry, done_q);
1147 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1150 * scsi_eh_get_sense - Get device sense data.
1151 * @work_q: Queue of commands to process.
1152 * @done_q: Queue of processed commands.
1154 * Description:
1155 * See if we need to request sense information. if so, then get it
1156 * now, so we have a better idea of what to do.
1158 * Notes:
1159 * This has the unfortunate side effect that if a shost adapter does
1160 * not automatically request sense information, we end up shutting
1161 * it down before we request it.
1163 * All drivers should request sense information internally these days,
1164 * so for now all I have to say is tough noogies if you end up in here.
1166 * XXX: Long term this code should go away, but that needs an audit of
1167 * all LLDDs first.
1169 int scsi_eh_get_sense(struct list_head *work_q,
1170 struct list_head *done_q)
1172 struct scsi_cmnd *scmd, *next;
1173 struct Scsi_Host *shost;
1174 int rtn;
1177 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1178 * should not get sense.
1180 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1181 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1182 SCSI_SENSE_VALID(scmd))
1183 continue;
1185 shost = scmd->device->host;
1186 if (scsi_host_eh_past_deadline(shost)) {
1187 SCSI_LOG_ERROR_RECOVERY(3,
1188 scmd_printk(KERN_INFO, scmd,
1189 "%s: skip request sense, past eh deadline\n",
1190 current->comm));
1191 break;
1193 if (status_byte(scmd->result) != CHECK_CONDITION)
1195 * don't request sense if there's no check condition
1196 * status because the error we're processing isn't one
1197 * that has a sense code (and some devices get
1198 * confused by sense requests out of the blue)
1200 continue;
1202 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1203 "%s: requesting sense\n",
1204 current->comm));
1205 rtn = scsi_request_sense(scmd);
1206 if (rtn != SUCCESS)
1207 continue;
1209 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1210 "sense requested, result %x\n", scmd->result));
1211 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1213 rtn = scsi_decide_disposition(scmd);
1216 * if the result was normal, then just pass it along to the
1217 * upper level.
1219 if (rtn == SUCCESS)
1220 /* we don't want this command reissued, just
1221 * finished with the sense data, so set
1222 * retries to the max allowed to ensure it
1223 * won't get reissued */
1224 scmd->retries = scmd->allowed;
1225 else if (rtn != NEEDS_RETRY)
1226 continue;
1228 scsi_eh_finish_cmd(scmd, done_q);
1231 return list_empty(work_q);
1233 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1236 * scsi_eh_tur - Send TUR to device.
1237 * @scmd: &scsi_cmnd to send TUR
1239 * Return value:
1240 * 0 - Device is ready. 1 - Device NOT ready.
1242 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1244 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1245 int retry_cnt = 1, rtn;
1247 retry_tur:
1248 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1249 scmd->device->eh_timeout, 0);
1251 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1252 "%s return: %x\n", __func__, rtn));
1254 switch (rtn) {
1255 case NEEDS_RETRY:
1256 if (retry_cnt--)
1257 goto retry_tur;
1258 /*FALLTHRU*/
1259 case SUCCESS:
1260 return 0;
1261 default:
1262 return 1;
1267 * scsi_eh_test_devices - check if devices are responding from error recovery.
1268 * @cmd_list: scsi commands in error recovery.
1269 * @work_q: queue for commands which still need more error recovery
1270 * @done_q: queue for commands which are finished
1271 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1273 * Decription:
1274 * Tests if devices are in a working state. Commands to devices now in
1275 * a working state are sent to the done_q while commands to devices which
1276 * are still failing to respond are returned to the work_q for more
1277 * processing.
1279 static int scsi_eh_test_devices(struct list_head *cmd_list,
1280 struct list_head *work_q,
1281 struct list_head *done_q, int try_stu)
1283 struct scsi_cmnd *scmd, *next;
1284 struct scsi_device *sdev;
1285 int finish_cmds;
1287 while (!list_empty(cmd_list)) {
1288 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1289 sdev = scmd->device;
1291 if (!try_stu) {
1292 if (scsi_host_eh_past_deadline(sdev->host)) {
1293 /* Push items back onto work_q */
1294 list_splice_init(cmd_list, work_q);
1295 SCSI_LOG_ERROR_RECOVERY(3,
1296 sdev_printk(KERN_INFO, sdev,
1297 "%s: skip test device, past eh deadline",
1298 current->comm));
1299 break;
1303 finish_cmds = !scsi_device_online(scmd->device) ||
1304 (try_stu && !scsi_eh_try_stu(scmd) &&
1305 !scsi_eh_tur(scmd)) ||
1306 !scsi_eh_tur(scmd);
1308 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1309 if (scmd->device == sdev) {
1310 if (finish_cmds &&
1311 (try_stu ||
1312 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1313 scsi_eh_finish_cmd(scmd, done_q);
1314 else
1315 list_move_tail(&scmd->eh_entry, work_q);
1318 return list_empty(work_q);
1322 * scsi_eh_try_stu - Send START_UNIT to device.
1323 * @scmd: &scsi_cmnd to send START_UNIT
1325 * Return value:
1326 * 0 - Device is ready. 1 - Device NOT ready.
1328 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1330 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1332 if (scmd->device->allow_restart) {
1333 int i, rtn = NEEDS_RETRY;
1335 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1336 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1338 if (rtn == SUCCESS)
1339 return 0;
1342 return 1;
1346 * scsi_eh_stu - send START_UNIT if needed
1347 * @shost: &scsi host being recovered.
1348 * @work_q: &list_head for pending commands.
1349 * @done_q: &list_head for processed commands.
1351 * Notes:
1352 * If commands are failing due to not ready, initializing command required,
1353 * try revalidating the device, which will end up sending a start unit.
1355 static int scsi_eh_stu(struct Scsi_Host *shost,
1356 struct list_head *work_q,
1357 struct list_head *done_q)
1359 struct scsi_cmnd *scmd, *stu_scmd, *next;
1360 struct scsi_device *sdev;
1362 shost_for_each_device(sdev, shost) {
1363 if (scsi_host_eh_past_deadline(shost)) {
1364 SCSI_LOG_ERROR_RECOVERY(3,
1365 sdev_printk(KERN_INFO, sdev,
1366 "%s: skip START_UNIT, past eh deadline\n",
1367 current->comm));
1368 break;
1370 stu_scmd = NULL;
1371 list_for_each_entry(scmd, work_q, eh_entry)
1372 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1373 scsi_check_sense(scmd) == FAILED ) {
1374 stu_scmd = scmd;
1375 break;
1378 if (!stu_scmd)
1379 continue;
1381 SCSI_LOG_ERROR_RECOVERY(3,
1382 sdev_printk(KERN_INFO, sdev,
1383 "%s: Sending START_UNIT\n",
1384 current->comm));
1386 if (!scsi_eh_try_stu(stu_scmd)) {
1387 if (!scsi_device_online(sdev) ||
1388 !scsi_eh_tur(stu_scmd)) {
1389 list_for_each_entry_safe(scmd, next,
1390 work_q, eh_entry) {
1391 if (scmd->device == sdev &&
1392 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1393 scsi_eh_finish_cmd(scmd, done_q);
1396 } else {
1397 SCSI_LOG_ERROR_RECOVERY(3,
1398 sdev_printk(KERN_INFO, sdev,
1399 "%s: START_UNIT failed\n",
1400 current->comm));
1404 return list_empty(work_q);
1409 * scsi_eh_bus_device_reset - send bdr if needed
1410 * @shost: scsi host being recovered.
1411 * @work_q: &list_head for pending commands.
1412 * @done_q: &list_head for processed commands.
1414 * Notes:
1415 * Try a bus device reset. Still, look to see whether we have multiple
1416 * devices that are jammed or not - if we have multiple devices, it
1417 * makes no sense to try bus_device_reset - we really would need to try
1418 * a bus_reset instead.
1420 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1421 struct list_head *work_q,
1422 struct list_head *done_q)
1424 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1425 struct scsi_device *sdev;
1426 int rtn;
1428 shost_for_each_device(sdev, shost) {
1429 if (scsi_host_eh_past_deadline(shost)) {
1430 SCSI_LOG_ERROR_RECOVERY(3,
1431 sdev_printk(KERN_INFO, sdev,
1432 "%s: skip BDR, past eh deadline\n",
1433 current->comm));
1434 break;
1436 bdr_scmd = NULL;
1437 list_for_each_entry(scmd, work_q, eh_entry)
1438 if (scmd->device == sdev) {
1439 bdr_scmd = scmd;
1440 break;
1443 if (!bdr_scmd)
1444 continue;
1446 SCSI_LOG_ERROR_RECOVERY(3,
1447 sdev_printk(KERN_INFO, sdev,
1448 "%s: Sending BDR\n", current->comm));
1449 rtn = scsi_try_bus_device_reset(bdr_scmd);
1450 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1451 if (!scsi_device_online(sdev) ||
1452 rtn == FAST_IO_FAIL ||
1453 !scsi_eh_tur(bdr_scmd)) {
1454 list_for_each_entry_safe(scmd, next,
1455 work_q, eh_entry) {
1456 if (scmd->device == sdev &&
1457 scsi_eh_action(scmd, rtn) != FAILED)
1458 scsi_eh_finish_cmd(scmd,
1459 done_q);
1462 } else {
1463 SCSI_LOG_ERROR_RECOVERY(3,
1464 sdev_printk(KERN_INFO, sdev,
1465 "%s: BDR failed\n", current->comm));
1469 return list_empty(work_q);
1473 * scsi_eh_target_reset - send target reset if needed
1474 * @shost: scsi host being recovered.
1475 * @work_q: &list_head for pending commands.
1476 * @done_q: &list_head for processed commands.
1478 * Notes:
1479 * Try a target reset.
1481 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1482 struct list_head *work_q,
1483 struct list_head *done_q)
1485 LIST_HEAD(tmp_list);
1486 LIST_HEAD(check_list);
1488 list_splice_init(work_q, &tmp_list);
1490 while (!list_empty(&tmp_list)) {
1491 struct scsi_cmnd *next, *scmd;
1492 int rtn;
1493 unsigned int id;
1495 if (scsi_host_eh_past_deadline(shost)) {
1496 /* push back on work queue for further processing */
1497 list_splice_init(&check_list, work_q);
1498 list_splice_init(&tmp_list, work_q);
1499 SCSI_LOG_ERROR_RECOVERY(3,
1500 shost_printk(KERN_INFO, shost,
1501 "%s: Skip target reset, past eh deadline\n",
1502 current->comm));
1503 return list_empty(work_q);
1506 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1507 id = scmd_id(scmd);
1509 SCSI_LOG_ERROR_RECOVERY(3,
1510 shost_printk(KERN_INFO, shost,
1511 "%s: Sending target reset to target %d\n",
1512 current->comm, id));
1513 rtn = scsi_try_target_reset(scmd);
1514 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1515 SCSI_LOG_ERROR_RECOVERY(3,
1516 shost_printk(KERN_INFO, shost,
1517 "%s: Target reset failed"
1518 " target: %d\n",
1519 current->comm, id));
1520 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1521 if (scmd_id(scmd) != id)
1522 continue;
1524 if (rtn == SUCCESS)
1525 list_move_tail(&scmd->eh_entry, &check_list);
1526 else if (rtn == FAST_IO_FAIL)
1527 scsi_eh_finish_cmd(scmd, done_q);
1528 else
1529 /* push back on work queue for further processing */
1530 list_move(&scmd->eh_entry, work_q);
1534 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1538 * scsi_eh_bus_reset - send a bus reset
1539 * @shost: &scsi host being recovered.
1540 * @work_q: &list_head for pending commands.
1541 * @done_q: &list_head for processed commands.
1543 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1544 struct list_head *work_q,
1545 struct list_head *done_q)
1547 struct scsi_cmnd *scmd, *chan_scmd, *next;
1548 LIST_HEAD(check_list);
1549 unsigned int channel;
1550 int rtn;
1553 * we really want to loop over the various channels, and do this on
1554 * a channel by channel basis. we should also check to see if any
1555 * of the failed commands are on soft_reset devices, and if so, skip
1556 * the reset.
1559 for (channel = 0; channel <= shost->max_channel; channel++) {
1560 if (scsi_host_eh_past_deadline(shost)) {
1561 list_splice_init(&check_list, work_q);
1562 SCSI_LOG_ERROR_RECOVERY(3,
1563 shost_printk(KERN_INFO, shost,
1564 "%s: skip BRST, past eh deadline\n",
1565 current->comm));
1566 return list_empty(work_q);
1569 chan_scmd = NULL;
1570 list_for_each_entry(scmd, work_q, eh_entry) {
1571 if (channel == scmd_channel(scmd)) {
1572 chan_scmd = scmd;
1573 break;
1575 * FIXME add back in some support for
1576 * soft_reset devices.
1581 if (!chan_scmd)
1582 continue;
1583 SCSI_LOG_ERROR_RECOVERY(3,
1584 shost_printk(KERN_INFO, shost,
1585 "%s: Sending BRST chan: %d\n",
1586 current->comm, channel));
1587 rtn = scsi_try_bus_reset(chan_scmd);
1588 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1589 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1590 if (channel == scmd_channel(scmd)) {
1591 if (rtn == FAST_IO_FAIL)
1592 scsi_eh_finish_cmd(scmd,
1593 done_q);
1594 else
1595 list_move_tail(&scmd->eh_entry,
1596 &check_list);
1599 } else {
1600 SCSI_LOG_ERROR_RECOVERY(3,
1601 shost_printk(KERN_INFO, shost,
1602 "%s: BRST failed chan: %d\n",
1603 current->comm, channel));
1606 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1610 * scsi_eh_host_reset - send a host reset
1611 * @shost: host to be reset.
1612 * @work_q: &list_head for pending commands.
1613 * @done_q: &list_head for processed commands.
1615 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1616 struct list_head *work_q,
1617 struct list_head *done_q)
1619 struct scsi_cmnd *scmd, *next;
1620 LIST_HEAD(check_list);
1621 int rtn;
1623 if (!list_empty(work_q)) {
1624 scmd = list_entry(work_q->next,
1625 struct scsi_cmnd, eh_entry);
1627 SCSI_LOG_ERROR_RECOVERY(3,
1628 shost_printk(KERN_INFO, shost,
1629 "%s: Sending HRST\n",
1630 current->comm));
1632 rtn = scsi_try_host_reset(scmd);
1633 if (rtn == SUCCESS) {
1634 list_splice_init(work_q, &check_list);
1635 } else if (rtn == FAST_IO_FAIL) {
1636 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1637 scsi_eh_finish_cmd(scmd, done_q);
1639 } else {
1640 SCSI_LOG_ERROR_RECOVERY(3,
1641 shost_printk(KERN_INFO, shost,
1642 "%s: HRST failed\n",
1643 current->comm));
1646 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1650 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1651 * @work_q: &list_head for pending commands.
1652 * @done_q: &list_head for processed commands.
1654 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1655 struct list_head *done_q)
1657 struct scsi_cmnd *scmd, *next;
1658 struct scsi_device *sdev;
1660 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1661 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1662 "not ready after error recovery\n");
1663 sdev = scmd->device;
1665 mutex_lock(&sdev->state_mutex);
1666 scsi_device_set_state(sdev, SDEV_OFFLINE);
1667 mutex_unlock(&sdev->state_mutex);
1669 scsi_eh_finish_cmd(scmd, done_q);
1671 return;
1675 * scsi_noretry_cmd - determine if command should be failed fast
1676 * @scmd: SCSI cmd to examine.
1678 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1680 switch (host_byte(scmd->result)) {
1681 case DID_OK:
1682 break;
1683 case DID_TIME_OUT:
1684 goto check_type;
1685 case DID_BUS_BUSY:
1686 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1687 case DID_PARITY:
1688 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1689 case DID_ERROR:
1690 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1691 status_byte(scmd->result) == RESERVATION_CONFLICT)
1692 return 0;
1693 /* fall through */
1694 case DID_SOFT_ERROR:
1695 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1698 if (status_byte(scmd->result) != CHECK_CONDITION)
1699 return 0;
1701 check_type:
1703 * assume caller has checked sense and determined
1704 * the check condition was retryable.
1706 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1707 blk_rq_is_passthrough(scmd->request))
1708 return 1;
1709 else
1710 return 0;
1714 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1715 * @scmd: SCSI cmd to examine.
1717 * Notes:
1718 * This is *only* called when we are examining the status after sending
1719 * out the actual data command. any commands that are queued for error
1720 * recovery (e.g. test_unit_ready) do *not* come through here.
1722 * When this routine returns failed, it means the error handler thread
1723 * is woken. In cases where the error code indicates an error that
1724 * doesn't require the error handler read (i.e. we don't need to
1725 * abort/reset), this function should return SUCCESS.
1727 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1729 int rtn;
1732 * if the device is offline, then we clearly just pass the result back
1733 * up to the top level.
1735 if (!scsi_device_online(scmd->device)) {
1736 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1737 "%s: device offline - report as SUCCESS\n", __func__));
1738 return SUCCESS;
1742 * first check the host byte, to see if there is anything in there
1743 * that would indicate what we need to do.
1745 switch (host_byte(scmd->result)) {
1746 case DID_PASSTHROUGH:
1748 * no matter what, pass this through to the upper layer.
1749 * nuke this special code so that it looks like we are saying
1750 * did_ok.
1752 scmd->result &= 0xff00ffff;
1753 return SUCCESS;
1754 case DID_OK:
1756 * looks good. drop through, and check the next byte.
1758 break;
1759 case DID_ABORT:
1760 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1761 set_host_byte(scmd, DID_TIME_OUT);
1762 return SUCCESS;
1764 /* FALLTHROUGH */
1765 case DID_NO_CONNECT:
1766 case DID_BAD_TARGET:
1768 * note - this means that we just report the status back
1769 * to the top level driver, not that we actually think
1770 * that it indicates SUCCESS.
1772 return SUCCESS;
1773 case DID_SOFT_ERROR:
1775 * when the low level driver returns did_soft_error,
1776 * it is responsible for keeping an internal retry counter
1777 * in order to avoid endless loops (db)
1779 goto maybe_retry;
1780 case DID_IMM_RETRY:
1781 return NEEDS_RETRY;
1783 case DID_REQUEUE:
1784 return ADD_TO_MLQUEUE;
1785 case DID_TRANSPORT_DISRUPTED:
1787 * LLD/transport was disrupted during processing of the IO.
1788 * The transport class is now blocked/blocking,
1789 * and the transport will decide what to do with the IO
1790 * based on its timers and recovery capablilities if
1791 * there are enough retries.
1793 goto maybe_retry;
1794 case DID_TRANSPORT_FAILFAST:
1796 * The transport decided to failfast the IO (most likely
1797 * the fast io fail tmo fired), so send IO directly upwards.
1799 return SUCCESS;
1800 case DID_ERROR:
1801 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1802 status_byte(scmd->result) == RESERVATION_CONFLICT)
1804 * execute reservation conflict processing code
1805 * lower down
1807 break;
1808 /* fallthrough */
1809 case DID_BUS_BUSY:
1810 case DID_PARITY:
1811 goto maybe_retry;
1812 case DID_TIME_OUT:
1814 * when we scan the bus, we get timeout messages for
1815 * these commands if there is no device available.
1816 * other hosts report did_no_connect for the same thing.
1818 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1819 scmd->cmnd[0] == INQUIRY)) {
1820 return SUCCESS;
1821 } else {
1822 return FAILED;
1824 case DID_RESET:
1825 return SUCCESS;
1826 default:
1827 return FAILED;
1831 * next, check the message byte.
1833 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1834 return FAILED;
1837 * check the status byte to see if this indicates anything special.
1839 switch (status_byte(scmd->result)) {
1840 case QUEUE_FULL:
1841 scsi_handle_queue_full(scmd->device);
1843 * the case of trying to send too many commands to a
1844 * tagged queueing device.
1846 /* FALLTHROUGH */
1847 case BUSY:
1849 * device can't talk to us at the moment. Should only
1850 * occur (SAM-3) when the task queue is empty, so will cause
1851 * the empty queue handling to trigger a stall in the
1852 * device.
1854 return ADD_TO_MLQUEUE;
1855 case GOOD:
1856 if (scmd->cmnd[0] == REPORT_LUNS)
1857 scmd->device->sdev_target->expecting_lun_change = 0;
1858 scsi_handle_queue_ramp_up(scmd->device);
1859 /* FALLTHROUGH */
1860 case COMMAND_TERMINATED:
1861 return SUCCESS;
1862 case TASK_ABORTED:
1863 goto maybe_retry;
1864 case CHECK_CONDITION:
1865 rtn = scsi_check_sense(scmd);
1866 if (rtn == NEEDS_RETRY)
1867 goto maybe_retry;
1868 /* if rtn == FAILED, we have no sense information;
1869 * returning FAILED will wake the error handler thread
1870 * to collect the sense and redo the decide
1871 * disposition */
1872 return rtn;
1873 case CONDITION_GOOD:
1874 case INTERMEDIATE_GOOD:
1875 case INTERMEDIATE_C_GOOD:
1876 case ACA_ACTIVE:
1878 * who knows? FIXME(eric)
1880 return SUCCESS;
1882 case RESERVATION_CONFLICT:
1883 sdev_printk(KERN_INFO, scmd->device,
1884 "reservation conflict\n");
1885 set_host_byte(scmd, DID_NEXUS_FAILURE);
1886 return SUCCESS; /* causes immediate i/o error */
1887 default:
1888 return FAILED;
1890 return FAILED;
1892 maybe_retry:
1894 /* we requeue for retry because the error was retryable, and
1895 * the request was not marked fast fail. Note that above,
1896 * even if the request is marked fast fail, we still requeue
1897 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1898 if ((++scmd->retries) <= scmd->allowed
1899 && !scsi_noretry_cmd(scmd)) {
1900 return NEEDS_RETRY;
1901 } else {
1903 * no more retries - report this one back to upper level.
1905 return SUCCESS;
1909 static void eh_lock_door_done(struct request *req, blk_status_t status)
1911 __blk_put_request(req->q, req);
1915 * scsi_eh_lock_door - Prevent medium removal for the specified device
1916 * @sdev: SCSI device to prevent medium removal
1918 * Locking:
1919 * We must be called from process context.
1921 * Notes:
1922 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1923 * head of the devices request queue, and continue.
1925 static void scsi_eh_lock_door(struct scsi_device *sdev)
1927 struct request *req;
1928 struct scsi_request *rq;
1931 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a
1932 * request becomes available
1934 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, GFP_KERNEL);
1935 if (IS_ERR(req))
1936 return;
1937 rq = scsi_req(req);
1939 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1940 rq->cmd[1] = 0;
1941 rq->cmd[2] = 0;
1942 rq->cmd[3] = 0;
1943 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
1944 rq->cmd[5] = 0;
1945 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
1947 req->rq_flags |= RQF_QUIET;
1948 req->timeout = 10 * HZ;
1949 rq->retries = 5;
1951 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1955 * scsi_restart_operations - restart io operations to the specified host.
1956 * @shost: Host we are restarting.
1958 * Notes:
1959 * When we entered the error handler, we blocked all further i/o to
1960 * this device. we need to 'reverse' this process.
1962 static void scsi_restart_operations(struct Scsi_Host *shost)
1964 struct scsi_device *sdev;
1965 unsigned long flags;
1968 * If the door was locked, we need to insert a door lock request
1969 * onto the head of the SCSI request queue for the device. There
1970 * is no point trying to lock the door of an off-line device.
1972 shost_for_each_device(sdev, shost) {
1973 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
1974 scsi_eh_lock_door(sdev);
1975 sdev->was_reset = 0;
1980 * next free up anything directly waiting upon the host. this
1981 * will be requests for character device operations, and also for
1982 * ioctls to queued block devices.
1984 SCSI_LOG_ERROR_RECOVERY(3,
1985 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
1987 spin_lock_irqsave(shost->host_lock, flags);
1988 if (scsi_host_set_state(shost, SHOST_RUNNING))
1989 if (scsi_host_set_state(shost, SHOST_CANCEL))
1990 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1991 spin_unlock_irqrestore(shost->host_lock, flags);
1993 wake_up(&shost->host_wait);
1996 * finally we need to re-initiate requests that may be pending. we will
1997 * have had everything blocked while error handling is taking place, and
1998 * now that error recovery is done, we will need to ensure that these
1999 * requests are started.
2001 scsi_run_host_queues(shost);
2004 * if eh is active and host_eh_scheduled is pending we need to re-run
2005 * recovery. we do this check after scsi_run_host_queues() to allow
2006 * everything pent up since the last eh run a chance to make forward
2007 * progress before we sync again. Either we'll immediately re-run
2008 * recovery or scsi_device_unbusy() will wake us again when these
2009 * pending commands complete.
2011 spin_lock_irqsave(shost->host_lock, flags);
2012 if (shost->host_eh_scheduled)
2013 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2014 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2015 spin_unlock_irqrestore(shost->host_lock, flags);
2019 * scsi_eh_ready_devs - check device ready state and recover if not.
2020 * @shost: host to be recovered.
2021 * @work_q: &list_head for pending commands.
2022 * @done_q: &list_head for processed commands.
2024 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2025 struct list_head *work_q,
2026 struct list_head *done_q)
2028 if (!scsi_eh_stu(shost, work_q, done_q))
2029 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2030 if (!scsi_eh_target_reset(shost, work_q, done_q))
2031 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2032 if (!scsi_eh_host_reset(shost, work_q, done_q))
2033 scsi_eh_offline_sdevs(work_q,
2034 done_q);
2036 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2039 * scsi_eh_flush_done_q - finish processed commands or retry them.
2040 * @done_q: list_head of processed commands.
2042 void scsi_eh_flush_done_q(struct list_head *done_q)
2044 struct scsi_cmnd *scmd, *next;
2046 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2047 list_del_init(&scmd->eh_entry);
2048 if (scsi_device_online(scmd->device) &&
2049 !scsi_noretry_cmd(scmd) &&
2050 (++scmd->retries <= scmd->allowed)) {
2051 SCSI_LOG_ERROR_RECOVERY(3,
2052 scmd_printk(KERN_INFO, scmd,
2053 "%s: flush retry cmd\n",
2054 current->comm));
2055 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2056 } else {
2058 * If just we got sense for the device (called
2059 * scsi_eh_get_sense), scmd->result is already
2060 * set, do not set DRIVER_TIMEOUT.
2062 if (!scmd->result)
2063 scmd->result |= (DRIVER_TIMEOUT << 24);
2064 SCSI_LOG_ERROR_RECOVERY(3,
2065 scmd_printk(KERN_INFO, scmd,
2066 "%s: flush finish cmd\n",
2067 current->comm));
2068 scsi_finish_command(scmd);
2072 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2075 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2076 * @shost: Host to unjam.
2078 * Notes:
2079 * When we come in here, we *know* that all commands on the bus have
2080 * either completed, failed or timed out. we also know that no further
2081 * commands are being sent to the host, so things are relatively quiet
2082 * and we have freedom to fiddle with things as we wish.
2084 * This is only the *default* implementation. it is possible for
2085 * individual drivers to supply their own version of this function, and
2086 * if the maintainer wishes to do this, it is strongly suggested that
2087 * this function be taken as a template and modified. this function
2088 * was designed to correctly handle problems for about 95% of the
2089 * different cases out there, and it should always provide at least a
2090 * reasonable amount of error recovery.
2092 * Any command marked 'failed' or 'timeout' must eventually have
2093 * scsi_finish_cmd() called for it. we do all of the retry stuff
2094 * here, so when we restart the host after we return it should have an
2095 * empty queue.
2097 static void scsi_unjam_host(struct Scsi_Host *shost)
2099 unsigned long flags;
2100 LIST_HEAD(eh_work_q);
2101 LIST_HEAD(eh_done_q);
2103 spin_lock_irqsave(shost->host_lock, flags);
2104 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2105 spin_unlock_irqrestore(shost->host_lock, flags);
2107 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2109 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2110 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2112 spin_lock_irqsave(shost->host_lock, flags);
2113 if (shost->eh_deadline != -1)
2114 shost->last_reset = 0;
2115 spin_unlock_irqrestore(shost->host_lock, flags);
2116 scsi_eh_flush_done_q(&eh_done_q);
2120 * scsi_error_handler - SCSI error handler thread
2121 * @data: Host for which we are running.
2123 * Notes:
2124 * This is the main error handling loop. This is run as a kernel thread
2125 * for every SCSI host and handles all error handling activity.
2127 int scsi_error_handler(void *data)
2129 struct Scsi_Host *shost = data;
2132 * We use TASK_INTERRUPTIBLE so that the thread is not
2133 * counted against the load average as a running process.
2134 * We never actually get interrupted because kthread_run
2135 * disables signal delivery for the created thread.
2137 while (true) {
2139 * The sequence in kthread_stop() sets the stop flag first
2140 * then wakes the process. To avoid missed wakeups, the task
2141 * should always be in a non running state before the stop
2142 * flag is checked
2144 set_current_state(TASK_INTERRUPTIBLE);
2145 if (kthread_should_stop())
2146 break;
2148 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2149 shost->host_failed != atomic_read(&shost->host_busy)) {
2150 SCSI_LOG_ERROR_RECOVERY(1,
2151 shost_printk(KERN_INFO, shost,
2152 "scsi_eh_%d: sleeping\n",
2153 shost->host_no));
2154 schedule();
2155 continue;
2158 __set_current_state(TASK_RUNNING);
2159 SCSI_LOG_ERROR_RECOVERY(1,
2160 shost_printk(KERN_INFO, shost,
2161 "scsi_eh_%d: waking up %d/%d/%d\n",
2162 shost->host_no, shost->host_eh_scheduled,
2163 shost->host_failed,
2164 atomic_read(&shost->host_busy)));
2167 * We have a host that is failing for some reason. Figure out
2168 * what we need to do to get it up and online again (if we can).
2169 * If we fail, we end up taking the thing offline.
2171 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2172 SCSI_LOG_ERROR_RECOVERY(1,
2173 shost_printk(KERN_ERR, shost,
2174 "scsi_eh_%d: unable to autoresume\n",
2175 shost->host_no));
2176 continue;
2179 if (shost->transportt->eh_strategy_handler)
2180 shost->transportt->eh_strategy_handler(shost);
2181 else
2182 scsi_unjam_host(shost);
2184 /* All scmds have been handled */
2185 shost->host_failed = 0;
2188 * Note - if the above fails completely, the action is to take
2189 * individual devices offline and flush the queue of any
2190 * outstanding requests that may have been pending. When we
2191 * restart, we restart any I/O to any other devices on the bus
2192 * which are still online.
2194 scsi_restart_operations(shost);
2195 if (!shost->eh_noresume)
2196 scsi_autopm_put_host(shost);
2198 __set_current_state(TASK_RUNNING);
2200 SCSI_LOG_ERROR_RECOVERY(1,
2201 shost_printk(KERN_INFO, shost,
2202 "Error handler scsi_eh_%d exiting\n",
2203 shost->host_no));
2204 shost->ehandler = NULL;
2205 return 0;
2209 * Function: scsi_report_bus_reset()
2211 * Purpose: Utility function used by low-level drivers to report that
2212 * they have observed a bus reset on the bus being handled.
2214 * Arguments: shost - Host in question
2215 * channel - channel on which reset was observed.
2217 * Returns: Nothing
2219 * Lock status: Host lock must be held.
2221 * Notes: This only needs to be called if the reset is one which
2222 * originates from an unknown location. Resets originated
2223 * by the mid-level itself don't need to call this, but there
2224 * should be no harm.
2226 * The main purpose of this is to make sure that a CHECK_CONDITION
2227 * is properly treated.
2229 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2231 struct scsi_device *sdev;
2233 __shost_for_each_device(sdev, shost) {
2234 if (channel == sdev_channel(sdev))
2235 __scsi_report_device_reset(sdev, NULL);
2238 EXPORT_SYMBOL(scsi_report_bus_reset);
2241 * Function: scsi_report_device_reset()
2243 * Purpose: Utility function used by low-level drivers to report that
2244 * they have observed a device reset on the device being handled.
2246 * Arguments: shost - Host in question
2247 * channel - channel on which reset was observed
2248 * target - target on which reset was observed
2250 * Returns: Nothing
2252 * Lock status: Host lock must be held
2254 * Notes: This only needs to be called if the reset is one which
2255 * originates from an unknown location. Resets originated
2256 * by the mid-level itself don't need to call this, but there
2257 * should be no harm.
2259 * The main purpose of this is to make sure that a CHECK_CONDITION
2260 * is properly treated.
2262 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2264 struct scsi_device *sdev;
2266 __shost_for_each_device(sdev, shost) {
2267 if (channel == sdev_channel(sdev) &&
2268 target == sdev_id(sdev))
2269 __scsi_report_device_reset(sdev, NULL);
2272 EXPORT_SYMBOL(scsi_report_device_reset);
2274 static void
2275 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2280 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2281 * @dev: scsi_device to operate on
2282 * @arg: reset type (see sg.h)
2285 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2287 struct scsi_cmnd *scmd;
2288 struct Scsi_Host *shost = dev->host;
2289 struct request *rq;
2290 unsigned long flags;
2291 int error = 0, rtn, val;
2293 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2294 return -EACCES;
2296 error = get_user(val, arg);
2297 if (error)
2298 return error;
2300 if (scsi_autopm_get_host(shost) < 0)
2301 return -EIO;
2303 error = -EIO;
2304 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2305 shost->hostt->cmd_size, GFP_KERNEL);
2306 if (!rq)
2307 goto out_put_autopm_host;
2308 blk_rq_init(NULL, rq);
2310 scmd = (struct scsi_cmnd *)(rq + 1);
2311 scsi_init_command(dev, scmd);
2312 scmd->request = rq;
2313 scmd->cmnd = scsi_req(rq)->cmd;
2315 scmd->scsi_done = scsi_reset_provider_done_command;
2316 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2318 scmd->cmd_len = 0;
2320 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2322 spin_lock_irqsave(shost->host_lock, flags);
2323 shost->tmf_in_progress = 1;
2324 spin_unlock_irqrestore(shost->host_lock, flags);
2326 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2327 case SG_SCSI_RESET_NOTHING:
2328 rtn = SUCCESS;
2329 break;
2330 case SG_SCSI_RESET_DEVICE:
2331 rtn = scsi_try_bus_device_reset(scmd);
2332 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2333 break;
2334 /* FALLTHROUGH */
2335 case SG_SCSI_RESET_TARGET:
2336 rtn = scsi_try_target_reset(scmd);
2337 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2338 break;
2339 /* FALLTHROUGH */
2340 case SG_SCSI_RESET_BUS:
2341 rtn = scsi_try_bus_reset(scmd);
2342 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2343 break;
2344 /* FALLTHROUGH */
2345 case SG_SCSI_RESET_HOST:
2346 rtn = scsi_try_host_reset(scmd);
2347 if (rtn == SUCCESS)
2348 break;
2349 /* FALLTHROUGH */
2350 default:
2351 rtn = FAILED;
2352 break;
2355 error = (rtn == SUCCESS) ? 0 : -EIO;
2357 spin_lock_irqsave(shost->host_lock, flags);
2358 shost->tmf_in_progress = 0;
2359 spin_unlock_irqrestore(shost->host_lock, flags);
2362 * be sure to wake up anyone who was sleeping or had their queue
2363 * suspended while we performed the TMF.
2365 SCSI_LOG_ERROR_RECOVERY(3,
2366 shost_printk(KERN_INFO, shost,
2367 "waking up host to restart after TMF\n"));
2369 wake_up(&shost->host_wait);
2370 scsi_run_host_queues(shost);
2372 scsi_put_command(scmd);
2373 kfree(rq);
2375 out_put_autopm_host:
2376 scsi_autopm_put_host(shost);
2377 return error;
2379 EXPORT_SYMBOL(scsi_ioctl_reset);
2381 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2382 struct scsi_sense_hdr *sshdr)
2384 return scsi_normalize_sense(cmd->sense_buffer,
2385 SCSI_SENSE_BUFFERSIZE, sshdr);
2387 EXPORT_SYMBOL(scsi_command_normalize_sense);
2390 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2391 * @sense_buffer: byte array of sense data
2392 * @sb_len: number of valid bytes in sense_buffer
2393 * @info_out: pointer to 64 integer where 8 or 4 byte information
2394 * field will be placed if found.
2396 * Return value:
2397 * true if information field found, false if not found.
2399 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2400 u64 *info_out)
2402 const u8 * ucp;
2404 if (sb_len < 7)
2405 return false;
2406 switch (sense_buffer[0] & 0x7f) {
2407 case 0x70:
2408 case 0x71:
2409 if (sense_buffer[0] & 0x80) {
2410 *info_out = get_unaligned_be32(&sense_buffer[3]);
2411 return true;
2413 return false;
2414 case 0x72:
2415 case 0x73:
2416 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2417 0 /* info desc */);
2418 if (ucp && (0xa == ucp[1])) {
2419 *info_out = get_unaligned_be64(&ucp[4]);
2420 return true;
2422 return false;
2423 default:
2424 return false;
2427 EXPORT_SYMBOL(scsi_get_sense_info_fld);