Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 *
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.
9 *
10 * Restructured scsi_unjam_host and associated functions.
11 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 *
13 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14 * minor cleanups.
15 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
16 */
17
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>
30
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>
43#include <scsi/sg.h>
44
45#include "scsi_priv.h"
46#include "scsi_logging.h"
47#include "scsi_transport_api.h"
48
49#include <trace/events/scsi.h>
50
51#include <asm/unaligned.h>
52
53/*
54 * These should *probably* be handled by the host itself.
55 * Since it is allowed to sleep, it probably should.
56 */
57#define BUS_RESET_SETTLE_TIME (10)
58#define HOST_RESET_SETTLE_TIME (10)
59
60static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
61static enum scsi_disposition scsi_try_to_abort_cmd(const struct scsi_host_template *,
62 struct scsi_cmnd *);
63
64void scsi_eh_wakeup(struct Scsi_Host *shost)
65{
66 lockdep_assert_held(shost->host_lock);
67
68 if (scsi_host_busy(shost) == 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"));
73 }
74}
75
76/**
77 * scsi_schedule_eh - schedule EH for SCSI host
78 * @shost: SCSI host to invoke error handling on.
79 *
80 * Schedule SCSI EH without scmd.
81 */
82void scsi_schedule_eh(struct Scsi_Host *shost)
83{
84 unsigned long flags;
85
86 spin_lock_irqsave(shost->host_lock, flags);
87
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);
92 }
93
94 spin_unlock_irqrestore(shost->host_lock, flags);
95}
96EXPORT_SYMBOL_GPL(scsi_schedule_eh);
97
98static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
99{
100 if (!shost->last_reset || shost->eh_deadline == -1)
101 return 0;
102
103 /*
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.
109 */
110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
111 shost->eh_deadline > -1)
112 return 0;
113
114 return 1;
115}
116
117static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd)
118{
119 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
120 return true;
121
122 return ++cmd->retries <= cmd->allowed;
123}
124
125static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd)
126{
127 struct scsi_device *sdev = cmd->device;
128 struct Scsi_Host *host = sdev->host;
129
130 if (host->hostt->eh_should_retry_cmd)
131 return host->hostt->eh_should_retry_cmd(cmd);
132
133 return true;
134}
135
136/**
137 * scmd_eh_abort_handler - Handle command aborts
138 * @work: command to be aborted.
139 *
140 * Note: this function must be called only for a command that has timed out.
141 * Because the block layer marks a request as complete before it calls
142 * scsi_timeout(), a .scsi_done() call from the LLD for a command that has
143 * timed out do not have any effect. Hence it is safe to call
144 * scsi_finish_command() from this function.
145 */
146void
147scmd_eh_abort_handler(struct work_struct *work)
148{
149 struct scsi_cmnd *scmd =
150 container_of(work, struct scsi_cmnd, abort_work.work);
151 struct scsi_device *sdev = scmd->device;
152 struct Scsi_Host *shost = sdev->host;
153 enum scsi_disposition rtn;
154 unsigned long flags;
155
156 if (scsi_host_eh_past_deadline(shost)) {
157 SCSI_LOG_ERROR_RECOVERY(3,
158 scmd_printk(KERN_INFO, scmd,
159 "eh timeout, not aborting\n"));
160 goto out;
161 }
162
163 SCSI_LOG_ERROR_RECOVERY(3,
164 scmd_printk(KERN_INFO, scmd,
165 "aborting command\n"));
166 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
167 if (rtn != SUCCESS) {
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_INFO, scmd,
170 "cmd abort %s\n",
171 (rtn == FAST_IO_FAIL) ?
172 "not send" : "failed"));
173 goto out;
174 }
175 set_host_byte(scmd, DID_TIME_OUT);
176 if (scsi_host_eh_past_deadline(shost)) {
177 SCSI_LOG_ERROR_RECOVERY(3,
178 scmd_printk(KERN_INFO, scmd,
179 "eh timeout, not retrying "
180 "aborted command\n"));
181 goto out;
182 }
183
184 spin_lock_irqsave(shost->host_lock, flags);
185 list_del_init(&scmd->eh_entry);
186
187 /*
188 * If the abort succeeds, and there is no further
189 * EH action, clear the ->last_reset time.
190 */
191 if (list_empty(&shost->eh_abort_list) &&
192 list_empty(&shost->eh_cmd_q))
193 if (shost->eh_deadline != -1)
194 shost->last_reset = 0;
195
196 spin_unlock_irqrestore(shost->host_lock, flags);
197
198 if (!scsi_noretry_cmd(scmd) &&
199 scsi_cmd_retry_allowed(scmd) &&
200 scsi_eh_should_retry_cmd(scmd)) {
201 SCSI_LOG_ERROR_RECOVERY(3,
202 scmd_printk(KERN_WARNING, scmd,
203 "retry aborted command\n"));
204 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
205 } else {
206 SCSI_LOG_ERROR_RECOVERY(3,
207 scmd_printk(KERN_WARNING, scmd,
208 "finish aborted command\n"));
209 scsi_finish_command(scmd);
210 }
211 return;
212
213out:
214 spin_lock_irqsave(shost->host_lock, flags);
215 list_del_init(&scmd->eh_entry);
216 spin_unlock_irqrestore(shost->host_lock, flags);
217
218 scsi_eh_scmd_add(scmd);
219}
220
221/**
222 * scsi_abort_command - schedule a command abort
223 * @scmd: scmd to abort.
224 *
225 * We only need to abort commands after a command timeout
226 */
227static int
228scsi_abort_command(struct scsi_cmnd *scmd)
229{
230 struct scsi_device *sdev = scmd->device;
231 struct Scsi_Host *shost = sdev->host;
232 unsigned long flags;
233
234 if (!shost->hostt->eh_abort_handler) {
235 /* No abort handler, fail command directly */
236 return FAILED;
237 }
238
239 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
240 /*
241 * Retry after abort failed, escalate to next level.
242 */
243 SCSI_LOG_ERROR_RECOVERY(3,
244 scmd_printk(KERN_INFO, scmd,
245 "previous abort failed\n"));
246 BUG_ON(delayed_work_pending(&scmd->abort_work));
247 return FAILED;
248 }
249
250 spin_lock_irqsave(shost->host_lock, flags);
251 if (shost->eh_deadline != -1 && !shost->last_reset)
252 shost->last_reset = jiffies;
253 BUG_ON(!list_empty(&scmd->eh_entry));
254 list_add_tail(&scmd->eh_entry, &shost->eh_abort_list);
255 spin_unlock_irqrestore(shost->host_lock, flags);
256
257 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
258 SCSI_LOG_ERROR_RECOVERY(3,
259 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
260 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
261 return SUCCESS;
262}
263
264/**
265 * scsi_eh_reset - call into ->eh_action to reset internal counters
266 * @scmd: scmd to run eh on.
267 *
268 * The scsi driver might be carrying internal state about the
269 * devices, so we need to call into the driver to reset the
270 * internal state once the error handler is started.
271 */
272static void scsi_eh_reset(struct scsi_cmnd *scmd)
273{
274 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
275 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
276 if (sdrv->eh_reset)
277 sdrv->eh_reset(scmd);
278 }
279}
280
281static void scsi_eh_inc_host_failed(struct rcu_head *head)
282{
283 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
284 struct Scsi_Host *shost = scmd->device->host;
285 unsigned long flags;
286
287 spin_lock_irqsave(shost->host_lock, flags);
288 shost->host_failed++;
289 scsi_eh_wakeup(shost);
290 spin_unlock_irqrestore(shost->host_lock, flags);
291}
292
293/**
294 * scsi_eh_scmd_add - add scsi cmd to error handling.
295 * @scmd: scmd to run eh on.
296 */
297void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
298{
299 struct Scsi_Host *shost = scmd->device->host;
300 unsigned long flags;
301 int ret;
302
303 WARN_ON_ONCE(!shost->ehandler);
304
305 spin_lock_irqsave(shost->host_lock, flags);
306 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
307 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
308 WARN_ON_ONCE(ret);
309 }
310 if (shost->eh_deadline != -1 && !shost->last_reset)
311 shost->last_reset = jiffies;
312
313 scsi_eh_reset(scmd);
314 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
315 spin_unlock_irqrestore(shost->host_lock, flags);
316 /*
317 * Ensure that all tasks observe the host state change before the
318 * host_failed change.
319 */
320 call_rcu_hurry(&scmd->rcu, scsi_eh_inc_host_failed);
321}
322
323/**
324 * scsi_timeout - Timeout function for normal scsi commands.
325 * @req: request that is timing out.
326 *
327 * Notes:
328 * We do not need to lock this. There is the potential for a race
329 * only in that the normal completion handling might run, but if the
330 * normal completion function determines that the timer has already
331 * fired, then it mustn't do anything.
332 */
333enum blk_eh_timer_return scsi_timeout(struct request *req)
334{
335 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
336 struct Scsi_Host *host = scmd->device->host;
337
338 trace_scsi_dispatch_cmd_timeout(scmd);
339 scsi_log_completion(scmd, TIMEOUT_ERROR);
340
341 atomic_inc(&scmd->device->iotmo_cnt);
342 if (host->eh_deadline != -1 && !host->last_reset)
343 host->last_reset = jiffies;
344
345 if (host->hostt->eh_timed_out) {
346 switch (host->hostt->eh_timed_out(scmd)) {
347 case SCSI_EH_DONE:
348 return BLK_EH_DONE;
349 case SCSI_EH_RESET_TIMER:
350 return BLK_EH_RESET_TIMER;
351 case SCSI_EH_NOT_HANDLED:
352 break;
353 }
354 }
355
356 /*
357 * If scsi_done() has already set SCMD_STATE_COMPLETE, do not modify
358 * *scmd.
359 */
360 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
361 return BLK_EH_DONE;
362 atomic_inc(&scmd->device->iodone_cnt);
363 if (scsi_abort_command(scmd) != SUCCESS) {
364 set_host_byte(scmd, DID_TIME_OUT);
365 scsi_eh_scmd_add(scmd);
366 }
367
368 return BLK_EH_DONE;
369}
370
371/**
372 * scsi_block_when_processing_errors - Prevent cmds from being queued.
373 * @sdev: Device on which we are performing recovery.
374 *
375 * Description:
376 * We block until the host is out of error recovery, and then check to
377 * see whether the host or the device is offline.
378 *
379 * Return value:
380 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
381 */
382int scsi_block_when_processing_errors(struct scsi_device *sdev)
383{
384 int online;
385
386 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
387
388 online = scsi_device_online(sdev);
389
390 return online;
391}
392EXPORT_SYMBOL(scsi_block_when_processing_errors);
393
394#ifdef CONFIG_SCSI_LOGGING
395/**
396 * scsi_eh_prt_fail_stats - Log info on failures.
397 * @shost: scsi host being recovered.
398 * @work_q: Queue of scsi cmds to process.
399 */
400static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
401 struct list_head *work_q)
402{
403 struct scsi_cmnd *scmd;
404 struct scsi_device *sdev;
405 int total_failures = 0;
406 int cmd_failed = 0;
407 int cmd_cancel = 0;
408 int devices_failed = 0;
409
410 shost_for_each_device(sdev, shost) {
411 list_for_each_entry(scmd, work_q, eh_entry) {
412 if (scmd->device == sdev) {
413 ++total_failures;
414 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
415 ++cmd_cancel;
416 else
417 ++cmd_failed;
418 }
419 }
420
421 if (cmd_cancel || cmd_failed) {
422 SCSI_LOG_ERROR_RECOVERY(3,
423 shost_printk(KERN_INFO, shost,
424 "%s: cmds failed: %d, cancel: %d\n",
425 __func__, cmd_failed,
426 cmd_cancel));
427 cmd_cancel = 0;
428 cmd_failed = 0;
429 ++devices_failed;
430 }
431 }
432
433 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
434 "Total of %d commands on %d"
435 " devices require eh work\n",
436 total_failures, devices_failed));
437}
438#endif
439
440 /**
441 * scsi_report_lun_change - Set flag on all *other* devices on the same target
442 * to indicate that a UNIT ATTENTION is expected.
443 * @sdev: Device reporting the UNIT ATTENTION
444 */
445static void scsi_report_lun_change(struct scsi_device *sdev)
446{
447 sdev->sdev_target->expecting_lun_change = 1;
448}
449
450/**
451 * scsi_report_sense - Examine scsi sense information and log messages for
452 * certain conditions, also issue uevents for some of them.
453 * @sdev: Device reporting the sense code
454 * @sshdr: sshdr to be examined
455 */
456static void scsi_report_sense(struct scsi_device *sdev,
457 struct scsi_sense_hdr *sshdr)
458{
459 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
460
461 if (sshdr->sense_key == UNIT_ATTENTION) {
462 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
463 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
464 sdev_printk(KERN_WARNING, sdev,
465 "Inquiry data has changed");
466 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
467 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
468 scsi_report_lun_change(sdev);
469 sdev_printk(KERN_WARNING, sdev,
470 "LUN assignments on this target have "
471 "changed. The Linux SCSI layer does not "
472 "automatically remap LUN assignments.\n");
473 } else if (sshdr->asc == 0x3f)
474 sdev_printk(KERN_WARNING, sdev,
475 "Operating parameters on this target have "
476 "changed. The Linux SCSI layer does not "
477 "automatically adjust these parameters.\n");
478
479 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
480 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
481 sdev_printk(KERN_WARNING, sdev,
482 "Warning! Received an indication that the "
483 "LUN reached a thin provisioning soft "
484 "threshold.\n");
485 }
486
487 if (sshdr->asc == 0x29) {
488 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
489 /*
490 * Do not print message if it is an expected side-effect
491 * of runtime PM.
492 */
493 if (!sdev->silence_suspend)
494 sdev_printk(KERN_WARNING, sdev,
495 "Power-on or device reset occurred\n");
496 }
497
498 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
499 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
500 sdev_printk(KERN_WARNING, sdev,
501 "Mode parameters changed");
502 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
503 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
504 sdev_printk(KERN_WARNING, sdev,
505 "Asymmetric access state changed");
506 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
507 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
508 sdev_printk(KERN_WARNING, sdev,
509 "Capacity data has changed");
510 } else if (sshdr->asc == 0x2a)
511 sdev_printk(KERN_WARNING, sdev,
512 "Parameters changed");
513 }
514
515 if (evt_type != SDEV_EVT_MAXBITS) {
516 set_bit(evt_type, sdev->pending_events);
517 schedule_work(&sdev->event_work);
518 }
519}
520
521static inline void set_scsi_ml_byte(struct scsi_cmnd *cmd, u8 status)
522{
523 cmd->result = (cmd->result & 0xffff00ff) | (status << 8);
524}
525
526/**
527 * scsi_check_sense - Examine scsi cmd sense
528 * @scmd: Cmd to have sense checked.
529 *
530 * Return value:
531 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
532 *
533 * Notes:
534 * When a deferred error is detected the current command has
535 * not been executed and needs retrying.
536 */
537enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
538{
539 struct request *req = scsi_cmd_to_rq(scmd);
540 struct scsi_device *sdev = scmd->device;
541 struct scsi_sense_hdr sshdr;
542
543 if (! scsi_command_normalize_sense(scmd, &sshdr))
544 return FAILED; /* no valid sense data */
545
546 scsi_report_sense(sdev, &sshdr);
547
548 if (scsi_sense_is_deferred(&sshdr))
549 return NEEDS_RETRY;
550
551 if (sdev->handler && sdev->handler->check_sense) {
552 enum scsi_disposition rc;
553
554 rc = sdev->handler->check_sense(sdev, &sshdr);
555 if (rc != SCSI_RETURN_NOT_HANDLED)
556 return rc;
557 /* handler does not care. Drop down to default handling */
558 }
559
560 if (scmd->cmnd[0] == TEST_UNIT_READY &&
561 scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER)
562 /*
563 * nasty: for mid-layer issued TURs, we need to return the
564 * actual sense data without any recovery attempt. For eh
565 * issued ones, we need to try to recover and interpret
566 */
567 return SUCCESS;
568
569 /*
570 * Previous logic looked for FILEMARK, EOM or ILI which are
571 * mainly associated with tapes and returned SUCCESS.
572 */
573 if (sshdr.response_code == 0x70) {
574 /* fixed format */
575 if (scmd->sense_buffer[2] & 0xe0)
576 return SUCCESS;
577 } else {
578 /*
579 * descriptor format: look for "stream commands sense data
580 * descriptor" (see SSC-3). Assume single sense data
581 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
582 */
583 if ((sshdr.additional_length > 3) &&
584 (scmd->sense_buffer[8] == 0x4) &&
585 (scmd->sense_buffer[11] & 0xe0))
586 return SUCCESS;
587 }
588
589 switch (sshdr.sense_key) {
590 case NO_SENSE:
591 return SUCCESS;
592 case RECOVERED_ERROR:
593 return /* soft_error */ SUCCESS;
594
595 case ABORTED_COMMAND:
596 if (sshdr.asc == 0x10) /* DIF */
597 return SUCCESS;
598
599 /*
600 * Check aborts due to command duration limit policy:
601 * ABORTED COMMAND additional sense code with the
602 * COMMAND TIMEOUT BEFORE PROCESSING or
603 * COMMAND TIMEOUT DURING PROCESSING or
604 * COMMAND TIMEOUT DURING PROCESSING DUE TO ERROR RECOVERY
605 * additional sense code qualifiers.
606 */
607 if (sshdr.asc == 0x2e &&
608 sshdr.ascq >= 0x01 && sshdr.ascq <= 0x03) {
609 set_scsi_ml_byte(scmd, SCSIML_STAT_DL_TIMEOUT);
610 req->cmd_flags |= REQ_FAILFAST_DEV;
611 req->rq_flags |= RQF_QUIET;
612 return SUCCESS;
613 }
614
615 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
616 return ADD_TO_MLQUEUE;
617 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
618 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
619 return ADD_TO_MLQUEUE;
620
621 return NEEDS_RETRY;
622 case NOT_READY:
623 case UNIT_ATTENTION:
624 /*
625 * if we are expecting a cc/ua because of a bus reset that we
626 * performed, treat this just as a retry. otherwise this is
627 * information that we should pass up to the upper-level driver
628 * so that we can deal with it there.
629 */
630 if (scmd->device->expecting_cc_ua) {
631 /*
632 * Because some device does not queue unit
633 * attentions correctly, we carefully check
634 * additional sense code and qualifier so as
635 * not to squash media change unit attention.
636 */
637 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
638 scmd->device->expecting_cc_ua = 0;
639 return NEEDS_RETRY;
640 }
641 }
642 /*
643 * we might also expect a cc/ua if another LUN on the target
644 * reported a UA with an ASC/ASCQ of 3F 0E -
645 * REPORTED LUNS DATA HAS CHANGED.
646 */
647 if (scmd->device->sdev_target->expecting_lun_change &&
648 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
649 return NEEDS_RETRY;
650 /*
651 * if the device is in the process of becoming ready, we
652 * should retry.
653 */
654 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
655 return NEEDS_RETRY;
656 /*
657 * if the device is not started, we need to wake
658 * the error handler to start the motor
659 */
660 if (scmd->device->allow_restart &&
661 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
662 return FAILED;
663 /*
664 * Pass the UA upwards for a determination in the completion
665 * functions.
666 */
667 return SUCCESS;
668
669 /* these are not supported */
670 case DATA_PROTECT:
671 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
672 /* Thin provisioning hard threshold reached */
673 set_scsi_ml_byte(scmd, SCSIML_STAT_NOSPC);
674 return SUCCESS;
675 }
676 fallthrough;
677 case COPY_ABORTED:
678 case VOLUME_OVERFLOW:
679 case MISCOMPARE:
680 case BLANK_CHECK:
681 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
682 return SUCCESS;
683
684 case MEDIUM_ERROR:
685 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
686 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
687 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
688 set_scsi_ml_byte(scmd, SCSIML_STAT_MED_ERROR);
689 return SUCCESS;
690 }
691 return NEEDS_RETRY;
692
693 case HARDWARE_ERROR:
694 if (scmd->device->retry_hwerror)
695 return ADD_TO_MLQUEUE;
696 else
697 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
698 fallthrough;
699
700 case ILLEGAL_REQUEST:
701 if (sshdr.asc == 0x20 || /* Invalid command operation code */
702 sshdr.asc == 0x21 || /* Logical block address out of range */
703 sshdr.asc == 0x22 || /* Invalid function */
704 sshdr.asc == 0x24 || /* Invalid field in cdb */
705 sshdr.asc == 0x26 || /* Parameter value invalid */
706 sshdr.asc == 0x27) { /* Write protected */
707 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
708 }
709 return SUCCESS;
710
711 case COMPLETED:
712 if (sshdr.asc == 0x55 && sshdr.ascq == 0x0a) {
713 set_scsi_ml_byte(scmd, SCSIML_STAT_DL_TIMEOUT);
714 req->cmd_flags |= REQ_FAILFAST_DEV;
715 req->rq_flags |= RQF_QUIET;
716 }
717 return SUCCESS;
718
719 default:
720 return SUCCESS;
721 }
722}
723EXPORT_SYMBOL_GPL(scsi_check_sense);
724
725static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
726{
727 const struct scsi_host_template *sht = sdev->host->hostt;
728 struct scsi_device *tmp_sdev;
729
730 if (!sht->track_queue_depth ||
731 sdev->queue_depth >= sdev->max_queue_depth)
732 return;
733
734 if (time_before(jiffies,
735 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
736 return;
737
738 if (time_before(jiffies,
739 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
740 return;
741
742 /*
743 * Walk all devices of a target and do
744 * ramp up on them.
745 */
746 shost_for_each_device(tmp_sdev, sdev->host) {
747 if (tmp_sdev->channel != sdev->channel ||
748 tmp_sdev->id != sdev->id ||
749 tmp_sdev->queue_depth == sdev->max_queue_depth)
750 continue;
751
752 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
753 sdev->last_queue_ramp_up = jiffies;
754 }
755}
756
757static void scsi_handle_queue_full(struct scsi_device *sdev)
758{
759 const struct scsi_host_template *sht = sdev->host->hostt;
760 struct scsi_device *tmp_sdev;
761
762 if (!sht->track_queue_depth)
763 return;
764
765 shost_for_each_device(tmp_sdev, sdev->host) {
766 if (tmp_sdev->channel != sdev->channel ||
767 tmp_sdev->id != sdev->id)
768 continue;
769 /*
770 * We do not know the number of commands that were at
771 * the device when we got the queue full so we start
772 * from the highest possible value and work our way down.
773 */
774 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
775 }
776}
777
778/**
779 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
780 * @scmd: SCSI cmd to examine.
781 *
782 * Notes:
783 * This is *only* called when we are examining the status of commands
784 * queued during error recovery. the main difference here is that we
785 * don't allow for the possibility of retries here, and we are a lot
786 * more restrictive about what we consider acceptable.
787 */
788static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
789{
790 /*
791 * first check the host byte, to see if there is anything in there
792 * that would indicate what we need to do.
793 */
794 if (host_byte(scmd->result) == DID_RESET) {
795 /*
796 * rats. we are already in the error handler, so we now
797 * get to try and figure out what to do next. if the sense
798 * is valid, we have a pretty good idea of what to do.
799 * if not, we mark it as FAILED.
800 */
801 return scsi_check_sense(scmd);
802 }
803 if (host_byte(scmd->result) != DID_OK)
804 return FAILED;
805
806 /*
807 * now, check the status byte to see if this indicates
808 * anything special.
809 */
810 switch (get_status_byte(scmd)) {
811 case SAM_STAT_GOOD:
812 scsi_handle_queue_ramp_up(scmd->device);
813 if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd))
814 /*
815 * If we have sense data, call scsi_check_sense() in
816 * order to set the correct SCSI ML byte (if any).
817 * No point in checking the return value, since the
818 * command has already completed successfully.
819 */
820 scsi_check_sense(scmd);
821 fallthrough;
822 case SAM_STAT_COMMAND_TERMINATED:
823 return SUCCESS;
824 case SAM_STAT_CHECK_CONDITION:
825 return scsi_check_sense(scmd);
826 case SAM_STAT_CONDITION_MET:
827 case SAM_STAT_INTERMEDIATE:
828 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
829 /*
830 * who knows? FIXME(eric)
831 */
832 return SUCCESS;
833 case SAM_STAT_RESERVATION_CONFLICT:
834 if (scmd->cmnd[0] == TEST_UNIT_READY)
835 /* it is a success, we probed the device and
836 * found it */
837 return SUCCESS;
838 /* otherwise, we failed to send the command */
839 return FAILED;
840 case SAM_STAT_TASK_SET_FULL:
841 scsi_handle_queue_full(scmd->device);
842 fallthrough;
843 case SAM_STAT_BUSY:
844 return NEEDS_RETRY;
845 default:
846 return FAILED;
847 }
848 return FAILED;
849}
850
851/**
852 * scsi_eh_done - Completion function for error handling.
853 * @scmd: Cmd that is done.
854 */
855void scsi_eh_done(struct scsi_cmnd *scmd)
856{
857 struct completion *eh_action;
858
859 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
860 "%s result: %x\n", __func__, scmd->result));
861
862 eh_action = scmd->device->host->eh_action;
863 if (eh_action)
864 complete(eh_action);
865}
866
867/**
868 * scsi_try_host_reset - ask host adapter to reset itself
869 * @scmd: SCSI cmd to send host reset.
870 */
871static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
872{
873 unsigned long flags;
874 enum scsi_disposition rtn;
875 struct Scsi_Host *host = scmd->device->host;
876 const struct scsi_host_template *hostt = host->hostt;
877
878 SCSI_LOG_ERROR_RECOVERY(3,
879 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
880
881 if (!hostt->eh_host_reset_handler)
882 return FAILED;
883
884 rtn = hostt->eh_host_reset_handler(scmd);
885
886 if (rtn == SUCCESS) {
887 if (!hostt->skip_settle_delay)
888 ssleep(HOST_RESET_SETTLE_TIME);
889 spin_lock_irqsave(host->host_lock, flags);
890 scsi_report_bus_reset(host, scmd_channel(scmd));
891 spin_unlock_irqrestore(host->host_lock, flags);
892 }
893
894 return rtn;
895}
896
897/**
898 * scsi_try_bus_reset - ask host to perform a bus reset
899 * @scmd: SCSI cmd to send bus reset.
900 */
901static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
902{
903 unsigned long flags;
904 enum scsi_disposition rtn;
905 struct Scsi_Host *host = scmd->device->host;
906 const struct scsi_host_template *hostt = host->hostt;
907
908 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
909 "%s: Snd Bus RST\n", __func__));
910
911 if (!hostt->eh_bus_reset_handler)
912 return FAILED;
913
914 rtn = hostt->eh_bus_reset_handler(scmd);
915
916 if (rtn == SUCCESS) {
917 if (!hostt->skip_settle_delay)
918 ssleep(BUS_RESET_SETTLE_TIME);
919 spin_lock_irqsave(host->host_lock, flags);
920 scsi_report_bus_reset(host, scmd_channel(scmd));
921 spin_unlock_irqrestore(host->host_lock, flags);
922 }
923
924 return rtn;
925}
926
927static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
928{
929 sdev->was_reset = 1;
930 sdev->expecting_cc_ua = 1;
931}
932
933/**
934 * scsi_try_target_reset - Ask host to perform a target reset
935 * @scmd: SCSI cmd used to send a target reset
936 *
937 * Notes:
938 * There is no timeout for this operation. if this operation is
939 * unreliable for a given host, then the host itself needs to put a
940 * timer on it, and set the host back to a consistent state prior to
941 * returning.
942 */
943static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
944{
945 unsigned long flags;
946 enum scsi_disposition rtn;
947 struct Scsi_Host *host = scmd->device->host;
948 const struct scsi_host_template *hostt = host->hostt;
949
950 if (!hostt->eh_target_reset_handler)
951 return FAILED;
952
953 rtn = hostt->eh_target_reset_handler(scmd);
954 if (rtn == SUCCESS) {
955 spin_lock_irqsave(host->host_lock, flags);
956 __starget_for_each_device(scsi_target(scmd->device), NULL,
957 __scsi_report_device_reset);
958 spin_unlock_irqrestore(host->host_lock, flags);
959 }
960
961 return rtn;
962}
963
964/**
965 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
966 * @scmd: SCSI cmd used to send BDR
967 *
968 * Notes:
969 * There is no timeout for this operation. if this operation is
970 * unreliable for a given host, then the host itself needs to put a
971 * timer on it, and set the host back to a consistent state prior to
972 * returning.
973 */
974static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
975{
976 enum scsi_disposition rtn;
977 const struct scsi_host_template *hostt = scmd->device->host->hostt;
978
979 if (!hostt->eh_device_reset_handler)
980 return FAILED;
981
982 rtn = hostt->eh_device_reset_handler(scmd);
983 if (rtn == SUCCESS)
984 __scsi_report_device_reset(scmd->device, NULL);
985 return rtn;
986}
987
988/**
989 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
990 * @hostt: SCSI driver host template
991 * @scmd: SCSI cmd used to send a target reset
992 *
993 * Return value:
994 * SUCCESS, FAILED, or FAST_IO_FAIL
995 *
996 * Notes:
997 * SUCCESS does not necessarily indicate that the command
998 * has been aborted; it only indicates that the LLDDs
999 * has cleared all references to that command.
1000 * LLDDs should return FAILED only if an abort was required
1001 * but could not be executed. LLDDs should return FAST_IO_FAIL
1002 * if the device is temporarily unavailable (eg due to a
1003 * link down on FibreChannel)
1004 */
1005static enum scsi_disposition
1006scsi_try_to_abort_cmd(const struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
1007{
1008 if (!hostt->eh_abort_handler)
1009 return FAILED;
1010
1011 return hostt->eh_abort_handler(scmd);
1012}
1013
1014static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
1015{
1016 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
1017 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
1018 if (scsi_try_target_reset(scmd) != SUCCESS)
1019 if (scsi_try_bus_reset(scmd) != SUCCESS)
1020 scsi_try_host_reset(scmd);
1021}
1022
1023/**
1024 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
1025 * @scmd: SCSI command structure to hijack
1026 * @ses: structure to save restore information
1027 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
1028 * @cmnd_size: size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE)
1029 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
1030 *
1031 * This function is used to save a scsi command information before re-execution
1032 * as part of the error recovery process. If @sense_bytes is 0 the command
1033 * sent must be one that does not transfer any data. If @sense_bytes != 0
1034 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
1035 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
1036 */
1037void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
1038 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
1039{
1040 struct scsi_device *sdev = scmd->device;
1041
1042 /*
1043 * We need saved copies of a number of fields - this is because
1044 * error handling may need to overwrite these with different values
1045 * to run different commands, and once error handling is complete,
1046 * we will need to restore these values prior to running the actual
1047 * command.
1048 */
1049 ses->cmd_len = scmd->cmd_len;
1050 ses->data_direction = scmd->sc_data_direction;
1051 ses->sdb = scmd->sdb;
1052 ses->result = scmd->result;
1053 ses->resid_len = scmd->resid_len;
1054 ses->underflow = scmd->underflow;
1055 ses->prot_op = scmd->prot_op;
1056 ses->eh_eflags = scmd->eh_eflags;
1057
1058 scmd->prot_op = SCSI_PROT_NORMAL;
1059 scmd->eh_eflags = 0;
1060 memcpy(ses->cmnd, scmd->cmnd, sizeof(ses->cmnd));
1061 memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
1062 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1063 scmd->result = 0;
1064 scmd->resid_len = 0;
1065
1066 if (sense_bytes) {
1067 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
1068 sense_bytes);
1069 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
1070 scmd->sdb.length);
1071 scmd->sdb.table.sgl = &ses->sense_sgl;
1072 scmd->sc_data_direction = DMA_FROM_DEVICE;
1073 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
1074 scmd->cmnd[0] = REQUEST_SENSE;
1075 scmd->cmnd[4] = scmd->sdb.length;
1076 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1077 } else {
1078 scmd->sc_data_direction = DMA_NONE;
1079 if (cmnd) {
1080 BUG_ON(cmnd_size > sizeof(scmd->cmnd));
1081 memcpy(scmd->cmnd, cmnd, cmnd_size);
1082 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1083 }
1084 }
1085
1086 scmd->underflow = 0;
1087
1088 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1089 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1090 (sdev->lun << 5 & 0xe0);
1091
1092 /*
1093 * Zero the sense buffer. The scsi spec mandates that any
1094 * untransferred sense data should be interpreted as being zero.
1095 */
1096 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1097}
1098EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1099
1100/**
1101 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1102 * @scmd: SCSI command structure to restore
1103 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1104 *
1105 * Undo any damage done by above scsi_eh_prep_cmnd().
1106 */
1107void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1108{
1109 /*
1110 * Restore original data
1111 */
1112 scmd->cmd_len = ses->cmd_len;
1113 memcpy(scmd->cmnd, ses->cmnd, sizeof(ses->cmnd));
1114 scmd->sc_data_direction = ses->data_direction;
1115 scmd->sdb = ses->sdb;
1116 scmd->result = ses->result;
1117 scmd->resid_len = ses->resid_len;
1118 scmd->underflow = ses->underflow;
1119 scmd->prot_op = ses->prot_op;
1120 scmd->eh_eflags = ses->eh_eflags;
1121}
1122EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1123
1124/**
1125 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1126 * @scmd: SCSI command structure to hijack
1127 * @cmnd: CDB to send
1128 * @cmnd_size: size in bytes of @cmnd
1129 * @timeout: timeout for this request
1130 * @sense_bytes: size of sense data to copy or 0
1131 *
1132 * This function is used to send a scsi command down to a target device
1133 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1134 *
1135 * Return value:
1136 * SUCCESS or FAILED or NEEDS_RETRY
1137 */
1138static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1139 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
1140{
1141 struct scsi_device *sdev = scmd->device;
1142 struct Scsi_Host *shost = sdev->host;
1143 DECLARE_COMPLETION_ONSTACK(done);
1144 unsigned long timeleft = timeout, delay;
1145 struct scsi_eh_save ses;
1146 const unsigned long stall_for = msecs_to_jiffies(100);
1147 int rtn;
1148
1149retry:
1150 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1151 shost->eh_action = &done;
1152
1153 scsi_log_send(scmd);
1154 scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER;
1155 scmd->flags |= SCMD_LAST;
1156
1157 /*
1158 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1159 * change the SCSI device state after we have examined it and before
1160 * .queuecommand() is called.
1161 */
1162 mutex_lock(&sdev->state_mutex);
1163 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1164 mutex_unlock(&sdev->state_mutex);
1165 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1166 "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1167 SDEV_BLOCK));
1168 delay = min(timeleft, stall_for);
1169 timeleft -= delay;
1170 msleep(jiffies_to_msecs(delay));
1171 mutex_lock(&sdev->state_mutex);
1172 }
1173 if (sdev->sdev_state != SDEV_BLOCK)
1174 rtn = shost->hostt->queuecommand(shost, scmd);
1175 else
1176 rtn = FAILED;
1177 mutex_unlock(&sdev->state_mutex);
1178
1179 if (rtn) {
1180 if (timeleft > stall_for) {
1181 scsi_eh_restore_cmnd(scmd, &ses);
1182
1183 timeleft -= stall_for;
1184 msleep(jiffies_to_msecs(stall_for));
1185 goto retry;
1186 }
1187 /* signal not to enter either branch of the if () below */
1188 timeleft = 0;
1189 rtn = FAILED;
1190 } else {
1191 timeleft = wait_for_completion_timeout(&done, timeout);
1192 rtn = SUCCESS;
1193 }
1194
1195 shost->eh_action = NULL;
1196
1197 scsi_log_completion(scmd, rtn);
1198
1199 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1200 "%s timeleft: %ld\n",
1201 __func__, timeleft));
1202
1203 /*
1204 * If there is time left scsi_eh_done got called, and we will examine
1205 * the actual status codes to see whether the command actually did
1206 * complete normally, else if we have a zero return and no time left,
1207 * the command must still be pending, so abort it and return FAILED.
1208 * If we never actually managed to issue the command, because
1209 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1210 * value above (so don't execute either branch of the if)
1211 */
1212 if (timeleft) {
1213 rtn = scsi_eh_completed_normally(scmd);
1214 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1215 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1216
1217 switch (rtn) {
1218 case SUCCESS:
1219 case NEEDS_RETRY:
1220 case FAILED:
1221 break;
1222 case ADD_TO_MLQUEUE:
1223 rtn = NEEDS_RETRY;
1224 break;
1225 default:
1226 rtn = FAILED;
1227 break;
1228 }
1229 } else if (rtn != FAILED) {
1230 scsi_abort_eh_cmnd(scmd);
1231 rtn = FAILED;
1232 }
1233
1234 scsi_eh_restore_cmnd(scmd, &ses);
1235
1236 return rtn;
1237}
1238
1239/**
1240 * scsi_request_sense - Request sense data from a particular target.
1241 * @scmd: SCSI cmd for request sense.
1242 *
1243 * Notes:
1244 * Some hosts automatically obtain this information, others require
1245 * that we obtain it on our own. This function will *not* return until
1246 * the command either times out, or it completes.
1247 */
1248static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
1249{
1250 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1251}
1252
1253static enum scsi_disposition
1254scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
1255{
1256 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
1257 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1258 if (sdrv->eh_action)
1259 rtn = sdrv->eh_action(scmd, rtn);
1260 }
1261 return rtn;
1262}
1263
1264/**
1265 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1266 * @scmd: Original SCSI cmd that eh has finished.
1267 * @done_q: Queue for processed commands.
1268 *
1269 * Notes:
1270 * We don't want to use the normal command completion while we are are
1271 * still handling errors - it may cause other commands to be queued,
1272 * and that would disturb what we are doing. Thus we really want to
1273 * keep a list of pending commands for final completion, and once we
1274 * are ready to leave error handling we handle completion for real.
1275 */
1276void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1277{
1278 list_move_tail(&scmd->eh_entry, done_q);
1279}
1280EXPORT_SYMBOL(scsi_eh_finish_cmd);
1281
1282/**
1283 * scsi_eh_get_sense - Get device sense data.
1284 * @work_q: Queue of commands to process.
1285 * @done_q: Queue of processed commands.
1286 *
1287 * Description:
1288 * See if we need to request sense information. if so, then get it
1289 * now, so we have a better idea of what to do.
1290 *
1291 * Notes:
1292 * This has the unfortunate side effect that if a shost adapter does
1293 * not automatically request sense information, we end up shutting
1294 * it down before we request it.
1295 *
1296 * All drivers should request sense information internally these days,
1297 * so for now all I have to say is tough noogies if you end up in here.
1298 *
1299 * XXX: Long term this code should go away, but that needs an audit of
1300 * all LLDDs first.
1301 */
1302int scsi_eh_get_sense(struct list_head *work_q,
1303 struct list_head *done_q)
1304{
1305 struct scsi_cmnd *scmd, *next;
1306 struct Scsi_Host *shost;
1307 enum scsi_disposition rtn;
1308
1309 /*
1310 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1311 * should not get sense.
1312 */
1313 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1314 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1315 SCSI_SENSE_VALID(scmd))
1316 continue;
1317
1318 shost = scmd->device->host;
1319 if (scsi_host_eh_past_deadline(shost)) {
1320 SCSI_LOG_ERROR_RECOVERY(3,
1321 scmd_printk(KERN_INFO, scmd,
1322 "%s: skip request sense, past eh deadline\n",
1323 current->comm));
1324 break;
1325 }
1326 if (!scsi_status_is_check_condition(scmd->result))
1327 /*
1328 * don't request sense if there's no check condition
1329 * status because the error we're processing isn't one
1330 * that has a sense code (and some devices get
1331 * confused by sense requests out of the blue)
1332 */
1333 continue;
1334
1335 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1336 "%s: requesting sense\n",
1337 current->comm));
1338 rtn = scsi_request_sense(scmd);
1339 if (rtn != SUCCESS)
1340 continue;
1341
1342 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1343 "sense requested, result %x\n", scmd->result));
1344 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1345
1346 rtn = scsi_decide_disposition(scmd);
1347
1348 /*
1349 * if the result was normal, then just pass it along to the
1350 * upper level.
1351 */
1352 if (rtn == SUCCESS)
1353 /*
1354 * We don't want this command reissued, just finished
1355 * with the sense data, so set retries to the max
1356 * allowed to ensure it won't get reissued. If the user
1357 * has requested infinite retries, we also want to
1358 * finish this command, so force completion by setting
1359 * retries and allowed to the same value.
1360 */
1361 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1362 scmd->retries = scmd->allowed = 1;
1363 else
1364 scmd->retries = scmd->allowed;
1365 else if (rtn != NEEDS_RETRY)
1366 continue;
1367
1368 scsi_eh_finish_cmd(scmd, done_q);
1369 }
1370
1371 return list_empty(work_q);
1372}
1373EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1374
1375/**
1376 * scsi_eh_tur - Send TUR to device.
1377 * @scmd: &scsi_cmnd to send TUR
1378 *
1379 * Return value:
1380 * 0 - Device is ready. 1 - Device NOT ready.
1381 */
1382static int scsi_eh_tur(struct scsi_cmnd *scmd)
1383{
1384 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1385 int retry_cnt = 1;
1386 enum scsi_disposition rtn;
1387
1388retry_tur:
1389 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1390 scmd->device->eh_timeout, 0);
1391
1392 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1393 "%s return: %x\n", __func__, rtn));
1394
1395 switch (rtn) {
1396 case NEEDS_RETRY:
1397 if (retry_cnt--)
1398 goto retry_tur;
1399 fallthrough;
1400 case SUCCESS:
1401 return 0;
1402 default:
1403 return 1;
1404 }
1405}
1406
1407/**
1408 * scsi_eh_test_devices - check if devices are responding from error recovery.
1409 * @cmd_list: scsi commands in error recovery.
1410 * @work_q: queue for commands which still need more error recovery
1411 * @done_q: queue for commands which are finished
1412 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1413 *
1414 * Decription:
1415 * Tests if devices are in a working state. Commands to devices now in
1416 * a working state are sent to the done_q while commands to devices which
1417 * are still failing to respond are returned to the work_q for more
1418 * processing.
1419 **/
1420static int scsi_eh_test_devices(struct list_head *cmd_list,
1421 struct list_head *work_q,
1422 struct list_head *done_q, int try_stu)
1423{
1424 struct scsi_cmnd *scmd, *next;
1425 struct scsi_device *sdev;
1426 int finish_cmds;
1427
1428 while (!list_empty(cmd_list)) {
1429 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1430 sdev = scmd->device;
1431
1432 if (!try_stu) {
1433 if (scsi_host_eh_past_deadline(sdev->host)) {
1434 /* Push items back onto work_q */
1435 list_splice_init(cmd_list, work_q);
1436 SCSI_LOG_ERROR_RECOVERY(3,
1437 sdev_printk(KERN_INFO, sdev,
1438 "%s: skip test device, past eh deadline",
1439 current->comm));
1440 break;
1441 }
1442 }
1443
1444 finish_cmds = !scsi_device_online(scmd->device) ||
1445 (try_stu && !scsi_eh_try_stu(scmd) &&
1446 !scsi_eh_tur(scmd)) ||
1447 !scsi_eh_tur(scmd);
1448
1449 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1450 if (scmd->device == sdev) {
1451 if (finish_cmds &&
1452 (try_stu ||
1453 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1454 scsi_eh_finish_cmd(scmd, done_q);
1455 else
1456 list_move_tail(&scmd->eh_entry, work_q);
1457 }
1458 }
1459 return list_empty(work_q);
1460}
1461
1462/**
1463 * scsi_eh_try_stu - Send START_UNIT to device.
1464 * @scmd: &scsi_cmnd to send START_UNIT
1465 *
1466 * Return value:
1467 * 0 - Device is ready. 1 - Device NOT ready.
1468 */
1469static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1470{
1471 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1472
1473 if (scmd->device->allow_restart) {
1474 int i;
1475 enum scsi_disposition rtn = NEEDS_RETRY;
1476
1477 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1478 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
1479 scmd->device->eh_timeout, 0);
1480
1481 if (rtn == SUCCESS)
1482 return 0;
1483 }
1484
1485 return 1;
1486}
1487
1488 /**
1489 * scsi_eh_stu - send START_UNIT if needed
1490 * @shost: &scsi host being recovered.
1491 * @work_q: &list_head for pending commands.
1492 * @done_q: &list_head for processed commands.
1493 *
1494 * Notes:
1495 * If commands are failing due to not ready, initializing command required,
1496 * try revalidating the device, which will end up sending a start unit.
1497 */
1498static int scsi_eh_stu(struct Scsi_Host *shost,
1499 struct list_head *work_q,
1500 struct list_head *done_q)
1501{
1502 struct scsi_cmnd *scmd, *stu_scmd, *next;
1503 struct scsi_device *sdev;
1504
1505 shost_for_each_device(sdev, shost) {
1506 if (scsi_host_eh_past_deadline(shost)) {
1507 SCSI_LOG_ERROR_RECOVERY(3,
1508 sdev_printk(KERN_INFO, sdev,
1509 "%s: skip START_UNIT, past eh deadline\n",
1510 current->comm));
1511 scsi_device_put(sdev);
1512 break;
1513 }
1514 stu_scmd = NULL;
1515 list_for_each_entry(scmd, work_q, eh_entry)
1516 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1517 scsi_check_sense(scmd) == FAILED ) {
1518 stu_scmd = scmd;
1519 break;
1520 }
1521
1522 if (!stu_scmd)
1523 continue;
1524
1525 SCSI_LOG_ERROR_RECOVERY(3,
1526 sdev_printk(KERN_INFO, sdev,
1527 "%s: Sending START_UNIT\n",
1528 current->comm));
1529
1530 if (!scsi_eh_try_stu(stu_scmd)) {
1531 if (!scsi_device_online(sdev) ||
1532 !scsi_eh_tur(stu_scmd)) {
1533 list_for_each_entry_safe(scmd, next,
1534 work_q, eh_entry) {
1535 if (scmd->device == sdev &&
1536 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1537 scsi_eh_finish_cmd(scmd, done_q);
1538 }
1539 }
1540 } else {
1541 SCSI_LOG_ERROR_RECOVERY(3,
1542 sdev_printk(KERN_INFO, sdev,
1543 "%s: START_UNIT failed\n",
1544 current->comm));
1545 }
1546 }
1547
1548 return list_empty(work_q);
1549}
1550
1551
1552/**
1553 * scsi_eh_bus_device_reset - send bdr if needed
1554 * @shost: scsi host being recovered.
1555 * @work_q: &list_head for pending commands.
1556 * @done_q: &list_head for processed commands.
1557 *
1558 * Notes:
1559 * Try a bus device reset. Still, look to see whether we have multiple
1560 * devices that are jammed or not - if we have multiple devices, it
1561 * makes no sense to try bus_device_reset - we really would need to try
1562 * a bus_reset instead.
1563 */
1564static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1565 struct list_head *work_q,
1566 struct list_head *done_q)
1567{
1568 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1569 struct scsi_device *sdev;
1570 enum scsi_disposition rtn;
1571
1572 shost_for_each_device(sdev, shost) {
1573 if (scsi_host_eh_past_deadline(shost)) {
1574 SCSI_LOG_ERROR_RECOVERY(3,
1575 sdev_printk(KERN_INFO, sdev,
1576 "%s: skip BDR, past eh deadline\n",
1577 current->comm));
1578 scsi_device_put(sdev);
1579 break;
1580 }
1581 bdr_scmd = NULL;
1582 list_for_each_entry(scmd, work_q, eh_entry)
1583 if (scmd->device == sdev) {
1584 bdr_scmd = scmd;
1585 break;
1586 }
1587
1588 if (!bdr_scmd)
1589 continue;
1590
1591 SCSI_LOG_ERROR_RECOVERY(3,
1592 sdev_printk(KERN_INFO, sdev,
1593 "%s: Sending BDR\n", current->comm));
1594 rtn = scsi_try_bus_device_reset(bdr_scmd);
1595 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1596 if (!scsi_device_online(sdev) ||
1597 rtn == FAST_IO_FAIL ||
1598 !scsi_eh_tur(bdr_scmd)) {
1599 list_for_each_entry_safe(scmd, next,
1600 work_q, eh_entry) {
1601 if (scmd->device == sdev &&
1602 scsi_eh_action(scmd, rtn) != FAILED)
1603 scsi_eh_finish_cmd(scmd,
1604 done_q);
1605 }
1606 }
1607 } else {
1608 SCSI_LOG_ERROR_RECOVERY(3,
1609 sdev_printk(KERN_INFO, sdev,
1610 "%s: BDR failed\n", current->comm));
1611 }
1612 }
1613
1614 return list_empty(work_q);
1615}
1616
1617/**
1618 * scsi_eh_target_reset - send target reset if needed
1619 * @shost: scsi host being recovered.
1620 * @work_q: &list_head for pending commands.
1621 * @done_q: &list_head for processed commands.
1622 *
1623 * Notes:
1624 * Try a target reset.
1625 */
1626static int scsi_eh_target_reset(struct Scsi_Host *shost,
1627 struct list_head *work_q,
1628 struct list_head *done_q)
1629{
1630 LIST_HEAD(tmp_list);
1631 LIST_HEAD(check_list);
1632
1633 list_splice_init(work_q, &tmp_list);
1634
1635 while (!list_empty(&tmp_list)) {
1636 struct scsi_cmnd *next, *scmd;
1637 enum scsi_disposition rtn;
1638 unsigned int id;
1639
1640 if (scsi_host_eh_past_deadline(shost)) {
1641 /* push back on work queue for further processing */
1642 list_splice_init(&check_list, work_q);
1643 list_splice_init(&tmp_list, work_q);
1644 SCSI_LOG_ERROR_RECOVERY(3,
1645 shost_printk(KERN_INFO, shost,
1646 "%s: Skip target reset, past eh deadline\n",
1647 current->comm));
1648 return list_empty(work_q);
1649 }
1650
1651 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1652 id = scmd_id(scmd);
1653
1654 SCSI_LOG_ERROR_RECOVERY(3,
1655 shost_printk(KERN_INFO, shost,
1656 "%s: Sending target reset to target %d\n",
1657 current->comm, id));
1658 rtn = scsi_try_target_reset(scmd);
1659 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1660 SCSI_LOG_ERROR_RECOVERY(3,
1661 shost_printk(KERN_INFO, shost,
1662 "%s: Target reset failed"
1663 " target: %d\n",
1664 current->comm, id));
1665 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1666 if (scmd_id(scmd) != id)
1667 continue;
1668
1669 if (rtn == SUCCESS)
1670 list_move_tail(&scmd->eh_entry, &check_list);
1671 else if (rtn == FAST_IO_FAIL)
1672 scsi_eh_finish_cmd(scmd, done_q);
1673 else
1674 /* push back on work queue for further processing */
1675 list_move(&scmd->eh_entry, work_q);
1676 }
1677 }
1678
1679 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1680}
1681
1682/**
1683 * scsi_eh_bus_reset - send a bus reset
1684 * @shost: &scsi host being recovered.
1685 * @work_q: &list_head for pending commands.
1686 * @done_q: &list_head for processed commands.
1687 */
1688static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1689 struct list_head *work_q,
1690 struct list_head *done_q)
1691{
1692 struct scsi_cmnd *scmd, *chan_scmd, *next;
1693 LIST_HEAD(check_list);
1694 unsigned int channel;
1695 enum scsi_disposition rtn;
1696
1697 /*
1698 * we really want to loop over the various channels, and do this on
1699 * a channel by channel basis. we should also check to see if any
1700 * of the failed commands are on soft_reset devices, and if so, skip
1701 * the reset.
1702 */
1703
1704 for (channel = 0; channel <= shost->max_channel; channel++) {
1705 if (scsi_host_eh_past_deadline(shost)) {
1706 list_splice_init(&check_list, work_q);
1707 SCSI_LOG_ERROR_RECOVERY(3,
1708 shost_printk(KERN_INFO, shost,
1709 "%s: skip BRST, past eh deadline\n",
1710 current->comm));
1711 return list_empty(work_q);
1712 }
1713
1714 chan_scmd = NULL;
1715 list_for_each_entry(scmd, work_q, eh_entry) {
1716 if (channel == scmd_channel(scmd)) {
1717 chan_scmd = scmd;
1718 break;
1719 /*
1720 * FIXME add back in some support for
1721 * soft_reset devices.
1722 */
1723 }
1724 }
1725
1726 if (!chan_scmd)
1727 continue;
1728 SCSI_LOG_ERROR_RECOVERY(3,
1729 shost_printk(KERN_INFO, shost,
1730 "%s: Sending BRST chan: %d\n",
1731 current->comm, channel));
1732 rtn = scsi_try_bus_reset(chan_scmd);
1733 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1734 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1735 if (channel == scmd_channel(scmd)) {
1736 if (rtn == FAST_IO_FAIL)
1737 scsi_eh_finish_cmd(scmd,
1738 done_q);
1739 else
1740 list_move_tail(&scmd->eh_entry,
1741 &check_list);
1742 }
1743 }
1744 } else {
1745 SCSI_LOG_ERROR_RECOVERY(3,
1746 shost_printk(KERN_INFO, shost,
1747 "%s: BRST failed chan: %d\n",
1748 current->comm, channel));
1749 }
1750 }
1751 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1752}
1753
1754/**
1755 * scsi_eh_host_reset - send a host reset
1756 * @shost: host to be reset.
1757 * @work_q: &list_head for pending commands.
1758 * @done_q: &list_head for processed commands.
1759 */
1760static int scsi_eh_host_reset(struct Scsi_Host *shost,
1761 struct list_head *work_q,
1762 struct list_head *done_q)
1763{
1764 struct scsi_cmnd *scmd, *next;
1765 LIST_HEAD(check_list);
1766 enum scsi_disposition rtn;
1767
1768 if (!list_empty(work_q)) {
1769 scmd = list_entry(work_q->next,
1770 struct scsi_cmnd, eh_entry);
1771
1772 SCSI_LOG_ERROR_RECOVERY(3,
1773 shost_printk(KERN_INFO, shost,
1774 "%s: Sending HRST\n",
1775 current->comm));
1776
1777 rtn = scsi_try_host_reset(scmd);
1778 if (rtn == SUCCESS) {
1779 list_splice_init(work_q, &check_list);
1780 } else if (rtn == FAST_IO_FAIL) {
1781 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1782 scsi_eh_finish_cmd(scmd, done_q);
1783 }
1784 } else {
1785 SCSI_LOG_ERROR_RECOVERY(3,
1786 shost_printk(KERN_INFO, shost,
1787 "%s: HRST failed\n",
1788 current->comm));
1789 }
1790 }
1791 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1792}
1793
1794/**
1795 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1796 * @work_q: &list_head for pending commands.
1797 * @done_q: &list_head for processed commands.
1798 */
1799static void scsi_eh_offline_sdevs(struct list_head *work_q,
1800 struct list_head *done_q)
1801{
1802 struct scsi_cmnd *scmd, *next;
1803 struct scsi_device *sdev;
1804
1805 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1806 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1807 "not ready after error recovery\n");
1808 sdev = scmd->device;
1809
1810 mutex_lock(&sdev->state_mutex);
1811 scsi_device_set_state(sdev, SDEV_OFFLINE);
1812 mutex_unlock(&sdev->state_mutex);
1813
1814 scsi_eh_finish_cmd(scmd, done_q);
1815 }
1816 return;
1817}
1818
1819/**
1820 * scsi_noretry_cmd - determine if command should be failed fast
1821 * @scmd: SCSI cmd to examine.
1822 */
1823bool scsi_noretry_cmd(struct scsi_cmnd *scmd)
1824{
1825 struct request *req = scsi_cmd_to_rq(scmd);
1826
1827 switch (host_byte(scmd->result)) {
1828 case DID_OK:
1829 break;
1830 case DID_TIME_OUT:
1831 goto check_type;
1832 case DID_BUS_BUSY:
1833 return !!(req->cmd_flags & REQ_FAILFAST_TRANSPORT);
1834 case DID_PARITY:
1835 return !!(req->cmd_flags & REQ_FAILFAST_DEV);
1836 case DID_ERROR:
1837 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1838 return false;
1839 fallthrough;
1840 case DID_SOFT_ERROR:
1841 return !!(req->cmd_flags & REQ_FAILFAST_DRIVER);
1842 }
1843
1844 /* Never retry commands aborted due to a duration limit timeout */
1845 if (scsi_ml_byte(scmd->result) == SCSIML_STAT_DL_TIMEOUT)
1846 return true;
1847
1848 if (!scsi_status_is_check_condition(scmd->result))
1849 return false;
1850
1851check_type:
1852 /*
1853 * assume caller has checked sense and determined
1854 * the check condition was retryable.
1855 */
1856 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req))
1857 return true;
1858
1859 return false;
1860}
1861
1862/**
1863 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1864 * @scmd: SCSI cmd to examine.
1865 *
1866 * Notes:
1867 * This is *only* called when we are examining the status after sending
1868 * out the actual data command. any commands that are queued for error
1869 * recovery (e.g. test_unit_ready) do *not* come through here.
1870 *
1871 * When this routine returns failed, it means the error handler thread
1872 * is woken. In cases where the error code indicates an error that
1873 * doesn't require the error handler read (i.e. we don't need to
1874 * abort/reset), this function should return SUCCESS.
1875 */
1876enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
1877{
1878 enum scsi_disposition rtn;
1879
1880 /*
1881 * if the device is offline, then we clearly just pass the result back
1882 * up to the top level.
1883 */
1884 if (!scsi_device_online(scmd->device)) {
1885 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1886 "%s: device offline - report as SUCCESS\n", __func__));
1887 return SUCCESS;
1888 }
1889
1890 /*
1891 * first check the host byte, to see if there is anything in there
1892 * that would indicate what we need to do.
1893 */
1894 switch (host_byte(scmd->result)) {
1895 case DID_PASSTHROUGH:
1896 /*
1897 * no matter what, pass this through to the upper layer.
1898 * nuke this special code so that it looks like we are saying
1899 * did_ok.
1900 */
1901 scmd->result &= 0xff00ffff;
1902 return SUCCESS;
1903 case DID_OK:
1904 /*
1905 * looks good. drop through, and check the next byte.
1906 */
1907 break;
1908 case DID_ABORT:
1909 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1910 set_host_byte(scmd, DID_TIME_OUT);
1911 return SUCCESS;
1912 }
1913 fallthrough;
1914 case DID_NO_CONNECT:
1915 case DID_BAD_TARGET:
1916 /*
1917 * note - this means that we just report the status back
1918 * to the top level driver, not that we actually think
1919 * that it indicates SUCCESS.
1920 */
1921 return SUCCESS;
1922 case DID_SOFT_ERROR:
1923 /*
1924 * when the low level driver returns did_soft_error,
1925 * it is responsible for keeping an internal retry counter
1926 * in order to avoid endless loops (db)
1927 */
1928 goto maybe_retry;
1929 case DID_IMM_RETRY:
1930 return NEEDS_RETRY;
1931
1932 case DID_REQUEUE:
1933 return ADD_TO_MLQUEUE;
1934 case DID_TRANSPORT_DISRUPTED:
1935 /*
1936 * LLD/transport was disrupted during processing of the IO.
1937 * The transport class is now blocked/blocking,
1938 * and the transport will decide what to do with the IO
1939 * based on its timers and recovery capablilities if
1940 * there are enough retries.
1941 */
1942 goto maybe_retry;
1943 case DID_TRANSPORT_FAILFAST:
1944 /*
1945 * The transport decided to failfast the IO (most likely
1946 * the fast io fail tmo fired), so send IO directly upwards.
1947 */
1948 return SUCCESS;
1949 case DID_TRANSPORT_MARGINAL:
1950 /*
1951 * caller has decided not to do retries on
1952 * abort success, so send IO directly upwards
1953 */
1954 return SUCCESS;
1955 case DID_ERROR:
1956 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1957 /*
1958 * execute reservation conflict processing code
1959 * lower down
1960 */
1961 break;
1962 fallthrough;
1963 case DID_BUS_BUSY:
1964 case DID_PARITY:
1965 goto maybe_retry;
1966 case DID_TIME_OUT:
1967 /*
1968 * when we scan the bus, we get timeout messages for
1969 * these commands if there is no device available.
1970 * other hosts report did_no_connect for the same thing.
1971 */
1972 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1973 scmd->cmnd[0] == INQUIRY)) {
1974 return SUCCESS;
1975 } else {
1976 return FAILED;
1977 }
1978 case DID_RESET:
1979 return SUCCESS;
1980 default:
1981 return FAILED;
1982 }
1983
1984 /*
1985 * check the status byte to see if this indicates anything special.
1986 */
1987 switch (get_status_byte(scmd)) {
1988 case SAM_STAT_TASK_SET_FULL:
1989 scsi_handle_queue_full(scmd->device);
1990 /*
1991 * the case of trying to send too many commands to a
1992 * tagged queueing device.
1993 */
1994 fallthrough;
1995 case SAM_STAT_BUSY:
1996 /*
1997 * device can't talk to us at the moment. Should only
1998 * occur (SAM-3) when the task queue is empty, so will cause
1999 * the empty queue handling to trigger a stall in the
2000 * device.
2001 */
2002 return ADD_TO_MLQUEUE;
2003 case SAM_STAT_GOOD:
2004 if (scmd->cmnd[0] == REPORT_LUNS)
2005 scmd->device->sdev_target->expecting_lun_change = 0;
2006 scsi_handle_queue_ramp_up(scmd->device);
2007 if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd))
2008 /*
2009 * If we have sense data, call scsi_check_sense() in
2010 * order to set the correct SCSI ML byte (if any).
2011 * No point in checking the return value, since the
2012 * command has already completed successfully.
2013 */
2014 scsi_check_sense(scmd);
2015 fallthrough;
2016 case SAM_STAT_COMMAND_TERMINATED:
2017 return SUCCESS;
2018 case SAM_STAT_TASK_ABORTED:
2019 goto maybe_retry;
2020 case SAM_STAT_CHECK_CONDITION:
2021 rtn = scsi_check_sense(scmd);
2022 if (rtn == NEEDS_RETRY)
2023 goto maybe_retry;
2024 /* if rtn == FAILED, we have no sense information;
2025 * returning FAILED will wake the error handler thread
2026 * to collect the sense and redo the decide
2027 * disposition */
2028 return rtn;
2029 case SAM_STAT_CONDITION_MET:
2030 case SAM_STAT_INTERMEDIATE:
2031 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
2032 case SAM_STAT_ACA_ACTIVE:
2033 /*
2034 * who knows? FIXME(eric)
2035 */
2036 return SUCCESS;
2037
2038 case SAM_STAT_RESERVATION_CONFLICT:
2039 sdev_printk(KERN_INFO, scmd->device,
2040 "reservation conflict\n");
2041 set_scsi_ml_byte(scmd, SCSIML_STAT_RESV_CONFLICT);
2042 return SUCCESS; /* causes immediate i/o error */
2043 }
2044 return FAILED;
2045
2046maybe_retry:
2047
2048 /* we requeue for retry because the error was retryable, and
2049 * the request was not marked fast fail. Note that above,
2050 * even if the request is marked fast fail, we still requeue
2051 * for queue congestion conditions (QUEUE_FULL or BUSY) */
2052 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
2053 return NEEDS_RETRY;
2054 } else {
2055 /*
2056 * no more retries - report this one back to upper level.
2057 */
2058 return SUCCESS;
2059 }
2060}
2061
2062static enum rq_end_io_ret eh_lock_door_done(struct request *req,
2063 blk_status_t status)
2064{
2065 blk_mq_free_request(req);
2066 return RQ_END_IO_NONE;
2067}
2068
2069/**
2070 * scsi_eh_lock_door - Prevent medium removal for the specified device
2071 * @sdev: SCSI device to prevent medium removal
2072 *
2073 * Locking:
2074 * We must be called from process context.
2075 *
2076 * Notes:
2077 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
2078 * head of the devices request queue, and continue.
2079 */
2080static void scsi_eh_lock_door(struct scsi_device *sdev)
2081{
2082 struct scsi_cmnd *scmd;
2083 struct request *req;
2084
2085 req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, 0);
2086 if (IS_ERR(req))
2087 return;
2088 scmd = blk_mq_rq_to_pdu(req);
2089
2090 scmd->cmnd[0] = ALLOW_MEDIUM_REMOVAL;
2091 scmd->cmnd[1] = 0;
2092 scmd->cmnd[2] = 0;
2093 scmd->cmnd[3] = 0;
2094 scmd->cmnd[4] = SCSI_REMOVAL_PREVENT;
2095 scmd->cmnd[5] = 0;
2096 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
2097 scmd->allowed = 5;
2098
2099 req->rq_flags |= RQF_QUIET;
2100 req->timeout = 10 * HZ;
2101 req->end_io = eh_lock_door_done;
2102
2103 blk_execute_rq_nowait(req, true);
2104}
2105
2106/**
2107 * scsi_restart_operations - restart io operations to the specified host.
2108 * @shost: Host we are restarting.
2109 *
2110 * Notes:
2111 * When we entered the error handler, we blocked all further i/o to
2112 * this device. we need to 'reverse' this process.
2113 */
2114static void scsi_restart_operations(struct Scsi_Host *shost)
2115{
2116 struct scsi_device *sdev;
2117 unsigned long flags;
2118
2119 /*
2120 * If the door was locked, we need to insert a door lock request
2121 * onto the head of the SCSI request queue for the device. There
2122 * is no point trying to lock the door of an off-line device.
2123 */
2124 shost_for_each_device(sdev, shost) {
2125 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2126 scsi_eh_lock_door(sdev);
2127 sdev->was_reset = 0;
2128 }
2129 }
2130
2131 /*
2132 * next free up anything directly waiting upon the host. this
2133 * will be requests for character device operations, and also for
2134 * ioctls to queued block devices.
2135 */
2136 SCSI_LOG_ERROR_RECOVERY(3,
2137 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2138
2139 spin_lock_irqsave(shost->host_lock, flags);
2140 if (scsi_host_set_state(shost, SHOST_RUNNING))
2141 if (scsi_host_set_state(shost, SHOST_CANCEL))
2142 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2143 spin_unlock_irqrestore(shost->host_lock, flags);
2144
2145 wake_up(&shost->host_wait);
2146
2147 /*
2148 * finally we need to re-initiate requests that may be pending. we will
2149 * have had everything blocked while error handling is taking place, and
2150 * now that error recovery is done, we will need to ensure that these
2151 * requests are started.
2152 */
2153 scsi_run_host_queues(shost);
2154
2155 /*
2156 * if eh is active and host_eh_scheduled is pending we need to re-run
2157 * recovery. we do this check after scsi_run_host_queues() to allow
2158 * everything pent up since the last eh run a chance to make forward
2159 * progress before we sync again. Either we'll immediately re-run
2160 * recovery or scsi_device_unbusy() will wake us again when these
2161 * pending commands complete.
2162 */
2163 spin_lock_irqsave(shost->host_lock, flags);
2164 if (shost->host_eh_scheduled)
2165 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2166 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2167 spin_unlock_irqrestore(shost->host_lock, flags);
2168}
2169
2170/**
2171 * scsi_eh_ready_devs - check device ready state and recover if not.
2172 * @shost: host to be recovered.
2173 * @work_q: &list_head for pending commands.
2174 * @done_q: &list_head for processed commands.
2175 */
2176void scsi_eh_ready_devs(struct Scsi_Host *shost,
2177 struct list_head *work_q,
2178 struct list_head *done_q)
2179{
2180 if (!scsi_eh_stu(shost, work_q, done_q))
2181 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2182 if (!scsi_eh_target_reset(shost, work_q, done_q))
2183 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2184 if (!scsi_eh_host_reset(shost, work_q, done_q))
2185 scsi_eh_offline_sdevs(work_q,
2186 done_q);
2187}
2188EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2189
2190/**
2191 * scsi_eh_flush_done_q - finish processed commands or retry them.
2192 * @done_q: list_head of processed commands.
2193 */
2194void scsi_eh_flush_done_q(struct list_head *done_q)
2195{
2196 struct scsi_cmnd *scmd, *next;
2197
2198 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2199 list_del_init(&scmd->eh_entry);
2200 if (scsi_device_online(scmd->device) &&
2201 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) &&
2202 scsi_eh_should_retry_cmd(scmd)) {
2203 SCSI_LOG_ERROR_RECOVERY(3,
2204 scmd_printk(KERN_INFO, scmd,
2205 "%s: flush retry cmd\n",
2206 current->comm));
2207 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2208 } else {
2209 /*
2210 * If just we got sense for the device (called
2211 * scsi_eh_get_sense), scmd->result is already
2212 * set, do not set DID_TIME_OUT.
2213 */
2214 if (!scmd->result &&
2215 !(scmd->flags & SCMD_FORCE_EH_SUCCESS))
2216 scmd->result |= (DID_TIME_OUT << 16);
2217 SCSI_LOG_ERROR_RECOVERY(3,
2218 scmd_printk(KERN_INFO, scmd,
2219 "%s: flush finish cmd\n",
2220 current->comm));
2221 scsi_finish_command(scmd);
2222 }
2223 }
2224}
2225EXPORT_SYMBOL(scsi_eh_flush_done_q);
2226
2227/**
2228 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2229 * @shost: Host to unjam.
2230 *
2231 * Notes:
2232 * When we come in here, we *know* that all commands on the bus have
2233 * either completed, failed or timed out. we also know that no further
2234 * commands are being sent to the host, so things are relatively quiet
2235 * and we have freedom to fiddle with things as we wish.
2236 *
2237 * This is only the *default* implementation. it is possible for
2238 * individual drivers to supply their own version of this function, and
2239 * if the maintainer wishes to do this, it is strongly suggested that
2240 * this function be taken as a template and modified. this function
2241 * was designed to correctly handle problems for about 95% of the
2242 * different cases out there, and it should always provide at least a
2243 * reasonable amount of error recovery.
2244 *
2245 * Any command marked 'failed' or 'timeout' must eventually have
2246 * scsi_finish_cmd() called for it. we do all of the retry stuff
2247 * here, so when we restart the host after we return it should have an
2248 * empty queue.
2249 */
2250static void scsi_unjam_host(struct Scsi_Host *shost)
2251{
2252 unsigned long flags;
2253 LIST_HEAD(eh_work_q);
2254 LIST_HEAD(eh_done_q);
2255
2256 spin_lock_irqsave(shost->host_lock, flags);
2257 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2258 spin_unlock_irqrestore(shost->host_lock, flags);
2259
2260 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2261
2262 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2263 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2264
2265 spin_lock_irqsave(shost->host_lock, flags);
2266 if (shost->eh_deadline != -1)
2267 shost->last_reset = 0;
2268 spin_unlock_irqrestore(shost->host_lock, flags);
2269 scsi_eh_flush_done_q(&eh_done_q);
2270}
2271
2272/**
2273 * scsi_error_handler - SCSI error handler thread
2274 * @data: Host for which we are running.
2275 *
2276 * Notes:
2277 * This is the main error handling loop. This is run as a kernel thread
2278 * for every SCSI host and handles all error handling activity.
2279 */
2280int scsi_error_handler(void *data)
2281{
2282 struct Scsi_Host *shost = data;
2283
2284 /*
2285 * We use TASK_INTERRUPTIBLE so that the thread is not
2286 * counted against the load average as a running process.
2287 * We never actually get interrupted because kthread_run
2288 * disables signal delivery for the created thread.
2289 */
2290 while (true) {
2291 /*
2292 * The sequence in kthread_stop() sets the stop flag first
2293 * then wakes the process. To avoid missed wakeups, the task
2294 * should always be in a non running state before the stop
2295 * flag is checked
2296 */
2297 set_current_state(TASK_INTERRUPTIBLE);
2298 if (kthread_should_stop())
2299 break;
2300
2301 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2302 shost->host_failed != scsi_host_busy(shost)) {
2303 SCSI_LOG_ERROR_RECOVERY(1,
2304 shost_printk(KERN_INFO, shost,
2305 "scsi_eh_%d: sleeping\n",
2306 shost->host_no));
2307 schedule();
2308 continue;
2309 }
2310
2311 __set_current_state(TASK_RUNNING);
2312 SCSI_LOG_ERROR_RECOVERY(1,
2313 shost_printk(KERN_INFO, shost,
2314 "scsi_eh_%d: waking up %d/%d/%d\n",
2315 shost->host_no, shost->host_eh_scheduled,
2316 shost->host_failed,
2317 scsi_host_busy(shost)));
2318
2319 /*
2320 * We have a host that is failing for some reason. Figure out
2321 * what we need to do to get it up and online again (if we can).
2322 * If we fail, we end up taking the thing offline.
2323 */
2324 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2325 SCSI_LOG_ERROR_RECOVERY(1,
2326 shost_printk(KERN_ERR, shost,
2327 "scsi_eh_%d: unable to autoresume\n",
2328 shost->host_no));
2329 continue;
2330 }
2331
2332 if (shost->transportt->eh_strategy_handler)
2333 shost->transportt->eh_strategy_handler(shost);
2334 else
2335 scsi_unjam_host(shost);
2336
2337 /* All scmds have been handled */
2338 shost->host_failed = 0;
2339
2340 /*
2341 * Note - if the above fails completely, the action is to take
2342 * individual devices offline and flush the queue of any
2343 * outstanding requests that may have been pending. When we
2344 * restart, we restart any I/O to any other devices on the bus
2345 * which are still online.
2346 */
2347 scsi_restart_operations(shost);
2348 if (!shost->eh_noresume)
2349 scsi_autopm_put_host(shost);
2350 }
2351 __set_current_state(TASK_RUNNING);
2352
2353 SCSI_LOG_ERROR_RECOVERY(1,
2354 shost_printk(KERN_INFO, shost,
2355 "Error handler scsi_eh_%d exiting\n",
2356 shost->host_no));
2357 shost->ehandler = NULL;
2358 return 0;
2359}
2360
2361/*
2362 * Function: scsi_report_bus_reset()
2363 *
2364 * Purpose: Utility function used by low-level drivers to report that
2365 * they have observed a bus reset on the bus being handled.
2366 *
2367 * Arguments: shost - Host in question
2368 * channel - channel on which reset was observed.
2369 *
2370 * Returns: Nothing
2371 *
2372 * Lock status: Host lock must be held.
2373 *
2374 * Notes: This only needs to be called if the reset is one which
2375 * originates from an unknown location. Resets originated
2376 * by the mid-level itself don't need to call this, but there
2377 * should be no harm.
2378 *
2379 * The main purpose of this is to make sure that a CHECK_CONDITION
2380 * is properly treated.
2381 */
2382void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2383{
2384 struct scsi_device *sdev;
2385
2386 __shost_for_each_device(sdev, shost) {
2387 if (channel == sdev_channel(sdev))
2388 __scsi_report_device_reset(sdev, NULL);
2389 }
2390}
2391EXPORT_SYMBOL(scsi_report_bus_reset);
2392
2393/*
2394 * Function: scsi_report_device_reset()
2395 *
2396 * Purpose: Utility function used by low-level drivers to report that
2397 * they have observed a device reset on the device being handled.
2398 *
2399 * Arguments: shost - Host in question
2400 * channel - channel on which reset was observed
2401 * target - target on which reset was observed
2402 *
2403 * Returns: Nothing
2404 *
2405 * Lock status: Host lock must be held
2406 *
2407 * Notes: This only needs to be called if the reset is one which
2408 * originates from an unknown location. Resets originated
2409 * by the mid-level itself don't need to call this, but there
2410 * should be no harm.
2411 *
2412 * The main purpose of this is to make sure that a CHECK_CONDITION
2413 * is properly treated.
2414 */
2415void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2416{
2417 struct scsi_device *sdev;
2418
2419 __shost_for_each_device(sdev, shost) {
2420 if (channel == sdev_channel(sdev) &&
2421 target == sdev_id(sdev))
2422 __scsi_report_device_reset(sdev, NULL);
2423 }
2424}
2425EXPORT_SYMBOL(scsi_report_device_reset);
2426
2427/**
2428 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2429 * @dev: scsi_device to operate on
2430 * @arg: reset type (see sg.h)
2431 */
2432int
2433scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2434{
2435 struct scsi_cmnd *scmd;
2436 struct Scsi_Host *shost = dev->host;
2437 struct request *rq;
2438 unsigned long flags;
2439 int error = 0, val;
2440 enum scsi_disposition rtn;
2441
2442 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2443 return -EACCES;
2444
2445 error = get_user(val, arg);
2446 if (error)
2447 return error;
2448
2449 if (scsi_autopm_get_host(shost) < 0)
2450 return -EIO;
2451
2452 error = -EIO;
2453 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2454 shost->hostt->cmd_size, GFP_KERNEL);
2455 if (!rq)
2456 goto out_put_autopm_host;
2457 blk_rq_init(NULL, rq);
2458
2459 scmd = (struct scsi_cmnd *)(rq + 1);
2460 scsi_init_command(dev, scmd);
2461
2462 scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL;
2463 scmd->flags |= SCMD_LAST;
2464 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2465
2466 scmd->cmd_len = 0;
2467
2468 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2469
2470 spin_lock_irqsave(shost->host_lock, flags);
2471 shost->tmf_in_progress = 1;
2472 spin_unlock_irqrestore(shost->host_lock, flags);
2473
2474 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2475 case SG_SCSI_RESET_NOTHING:
2476 rtn = SUCCESS;
2477 break;
2478 case SG_SCSI_RESET_DEVICE:
2479 rtn = scsi_try_bus_device_reset(scmd);
2480 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2481 break;
2482 fallthrough;
2483 case SG_SCSI_RESET_TARGET:
2484 rtn = scsi_try_target_reset(scmd);
2485 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2486 break;
2487 fallthrough;
2488 case SG_SCSI_RESET_BUS:
2489 rtn = scsi_try_bus_reset(scmd);
2490 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2491 break;
2492 fallthrough;
2493 case SG_SCSI_RESET_HOST:
2494 rtn = scsi_try_host_reset(scmd);
2495 if (rtn == SUCCESS)
2496 break;
2497 fallthrough;
2498 default:
2499 rtn = FAILED;
2500 break;
2501 }
2502
2503 error = (rtn == SUCCESS) ? 0 : -EIO;
2504
2505 spin_lock_irqsave(shost->host_lock, flags);
2506 shost->tmf_in_progress = 0;
2507 spin_unlock_irqrestore(shost->host_lock, flags);
2508
2509 /*
2510 * be sure to wake up anyone who was sleeping or had their queue
2511 * suspended while we performed the TMF.
2512 */
2513 SCSI_LOG_ERROR_RECOVERY(3,
2514 shost_printk(KERN_INFO, shost,
2515 "waking up host to restart after TMF\n"));
2516
2517 wake_up(&shost->host_wait);
2518 scsi_run_host_queues(shost);
2519
2520 kfree(rq);
2521
2522out_put_autopm_host:
2523 scsi_autopm_put_host(shost);
2524 return error;
2525}
2526
2527bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2528 struct scsi_sense_hdr *sshdr)
2529{
2530 return scsi_normalize_sense(cmd->sense_buffer,
2531 SCSI_SENSE_BUFFERSIZE, sshdr);
2532}
2533EXPORT_SYMBOL(scsi_command_normalize_sense);
2534
2535/**
2536 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2537 * @sense_buffer: byte array of sense data
2538 * @sb_len: number of valid bytes in sense_buffer
2539 * @info_out: pointer to 64 integer where 8 or 4 byte information
2540 * field will be placed if found.
2541 *
2542 * Return value:
2543 * true if information field found, false if not found.
2544 */
2545bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2546 u64 *info_out)
2547{
2548 const u8 * ucp;
2549
2550 if (sb_len < 7)
2551 return false;
2552 switch (sense_buffer[0] & 0x7f) {
2553 case 0x70:
2554 case 0x71:
2555 if (sense_buffer[0] & 0x80) {
2556 *info_out = get_unaligned_be32(&sense_buffer[3]);
2557 return true;
2558 }
2559 return false;
2560 case 0x72:
2561 case 0x73:
2562 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2563 0 /* info desc */);
2564 if (ucp && (0xa == ucp[1])) {
2565 *info_out = get_unaligned_be64(&ucp[4]);
2566 return true;
2567 }
2568 return false;
2569 default:
2570 return false;
2571 }
2572}
2573EXPORT_SYMBOL(scsi_get_sense_info_fld);