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 static void scsi_eh_done(struct scsi_cmnd *scmd);
54
55 /*
56 * These should *probably* be handled by the host itself.
57 * Since it is allowed to sleep, it probably should.
58 */
59 #define BUS_RESET_SETTLE_TIME (10)
60 #define HOST_RESET_SETTLE_TIME (10)
61
62 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
63 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
64 struct scsi_cmnd *);
65
scsi_eh_wakeup(struct Scsi_Host * shost,unsigned int busy)66 void scsi_eh_wakeup(struct Scsi_Host *shost, unsigned int busy)
67 {
68 lockdep_assert_held(shost->host_lock);
69
70 if (busy == shost->host_failed) {
71 trace_scsi_eh_wakeup(shost);
72 wake_up_process(shost->ehandler);
73 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
74 "Waking error handler thread\n"));
75 }
76 }
77
78 /**
79 * scsi_schedule_eh - schedule EH for SCSI host
80 * @shost: SCSI host to invoke error handling on.
81 *
82 * Schedule SCSI EH without scmd.
83 */
scsi_schedule_eh(struct Scsi_Host * shost)84 void scsi_schedule_eh(struct Scsi_Host *shost)
85 {
86 unsigned long flags;
87
88 spin_lock_irqsave(shost->host_lock, flags);
89
90 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
91 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
92 shost->host_eh_scheduled++;
93 scsi_eh_wakeup(shost, scsi_host_busy(shost));
94 }
95
96 spin_unlock_irqrestore(shost->host_lock, flags);
97 }
98 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
99
scsi_host_eh_past_deadline(struct Scsi_Host * shost)100 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
101 {
102 if (!shost->last_reset || shost->eh_deadline == -1)
103 return 0;
104
105 /*
106 * 32bit accesses are guaranteed to be atomic
107 * (on all supported architectures), so instead
108 * of using a spinlock we can as well double check
109 * if eh_deadline has been set to 'off' during the
110 * time_before call.
111 */
112 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
113 shost->eh_deadline > -1)
114 return 0;
115
116 return 1;
117 }
118
scsi_cmd_retry_allowed(struct scsi_cmnd * cmd)119 static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd)
120 {
121 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
122 return true;
123
124 return ++cmd->retries <= cmd->allowed;
125 }
126
127 /**
128 * scmd_eh_abort_handler - Handle command aborts
129 * @work: command to be aborted.
130 *
131 * Note: this function must be called only for a command that has timed out.
132 * Because the block layer marks a request as complete before it calls
133 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
134 * timed out do not have any effect. Hence it is safe to call
135 * scsi_finish_command() from this function.
136 */
137 void
scmd_eh_abort_handler(struct work_struct * work)138 scmd_eh_abort_handler(struct work_struct *work)
139 {
140 struct scsi_cmnd *scmd =
141 container_of(work, struct scsi_cmnd, abort_work.work);
142 struct scsi_device *sdev = scmd->device;
143 int rtn;
144
145 if (scsi_host_eh_past_deadline(sdev->host)) {
146 SCSI_LOG_ERROR_RECOVERY(3,
147 scmd_printk(KERN_INFO, scmd,
148 "eh timeout, not aborting\n"));
149 } else {
150 SCSI_LOG_ERROR_RECOVERY(3,
151 scmd_printk(KERN_INFO, scmd,
152 "aborting command\n"));
153 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
154 if (rtn == SUCCESS) {
155 set_host_byte(scmd, DID_TIME_OUT);
156 if (scsi_host_eh_past_deadline(sdev->host)) {
157 SCSI_LOG_ERROR_RECOVERY(3,
158 scmd_printk(KERN_INFO, scmd,
159 "eh timeout, not retrying "
160 "aborted command\n"));
161 } else if (!scsi_noretry_cmd(scmd) &&
162 scsi_cmd_retry_allowed(scmd)) {
163 SCSI_LOG_ERROR_RECOVERY(3,
164 scmd_printk(KERN_WARNING, scmd,
165 "retry aborted command\n"));
166 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
167 return;
168 } else {
169 SCSI_LOG_ERROR_RECOVERY(3,
170 scmd_printk(KERN_WARNING, scmd,
171 "finish aborted command\n"));
172 scsi_finish_command(scmd);
173 return;
174 }
175 } else {
176 SCSI_LOG_ERROR_RECOVERY(3,
177 scmd_printk(KERN_INFO, scmd,
178 "cmd abort %s\n",
179 (rtn == FAST_IO_FAIL) ?
180 "not send" : "failed"));
181 }
182 }
183
184 scsi_eh_scmd_add(scmd);
185 }
186
187 /**
188 * scsi_abort_command - schedule a command abort
189 * @scmd: scmd to abort.
190 *
191 * We only need to abort commands after a command timeout
192 */
193 static int
scsi_abort_command(struct scsi_cmnd * scmd)194 scsi_abort_command(struct scsi_cmnd *scmd)
195 {
196 struct scsi_device *sdev = scmd->device;
197 struct Scsi_Host *shost = sdev->host;
198 unsigned long flags;
199
200 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
201 /*
202 * Retry after abort failed, escalate to next level.
203 */
204 SCSI_LOG_ERROR_RECOVERY(3,
205 scmd_printk(KERN_INFO, scmd,
206 "previous abort failed\n"));
207 BUG_ON(delayed_work_pending(&scmd->abort_work));
208 return FAILED;
209 }
210
211 spin_lock_irqsave(shost->host_lock, flags);
212 if (shost->eh_deadline != -1 && !shost->last_reset)
213 shost->last_reset = jiffies;
214 spin_unlock_irqrestore(shost->host_lock, flags);
215
216 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
217 SCSI_LOG_ERROR_RECOVERY(3,
218 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
219 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
220 return SUCCESS;
221 }
222
223 /**
224 * scsi_eh_reset - call into ->eh_action to reset internal counters
225 * @scmd: scmd to run eh on.
226 *
227 * The scsi driver might be carrying internal state about the
228 * devices, so we need to call into the driver to reset the
229 * internal state once the error handler is started.
230 */
scsi_eh_reset(struct scsi_cmnd * scmd)231 static void scsi_eh_reset(struct scsi_cmnd *scmd)
232 {
233 if (!blk_rq_is_passthrough(scmd->request)) {
234 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
235 if (sdrv->eh_reset)
236 sdrv->eh_reset(scmd);
237 }
238 }
239
scsi_eh_inc_host_failed(struct rcu_head * head)240 static void scsi_eh_inc_host_failed(struct rcu_head *head)
241 {
242 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
243 struct Scsi_Host *shost = scmd->device->host;
244 unsigned long flags;
245
246 spin_lock_irqsave(shost->host_lock, flags);
247 shost->host_failed++;
248 scsi_eh_wakeup(shost, scsi_host_busy(shost));
249 spin_unlock_irqrestore(shost->host_lock, flags);
250 }
251
252 /**
253 * scsi_eh_scmd_add - add scsi cmd to error handling.
254 * @scmd: scmd to run eh on.
255 */
scsi_eh_scmd_add(struct scsi_cmnd * scmd)256 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
257 {
258 struct Scsi_Host *shost = scmd->device->host;
259 unsigned long flags;
260 int ret;
261
262 WARN_ON_ONCE(!shost->ehandler);
263
264 spin_lock_irqsave(shost->host_lock, flags);
265 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
266 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
267 WARN_ON_ONCE(ret);
268 }
269 if (shost->eh_deadline != -1 && !shost->last_reset)
270 shost->last_reset = jiffies;
271
272 scsi_eh_reset(scmd);
273 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
274 spin_unlock_irqrestore(shost->host_lock, flags);
275 /*
276 * Ensure that all tasks observe the host state change before the
277 * host_failed change.
278 */
279 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
280 }
281
282 /**
283 * scsi_times_out - Timeout function for normal scsi commands.
284 * @req: request that is timing out.
285 *
286 * Notes:
287 * We do not need to lock this. There is the potential for a race
288 * only in that the normal completion handling might run, but if the
289 * normal completion function determines that the timer has already
290 * fired, then it mustn't do anything.
291 */
scsi_times_out(struct request * req)292 enum blk_eh_timer_return scsi_times_out(struct request *req)
293 {
294 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
295 enum blk_eh_timer_return rtn = BLK_EH_DONE;
296 struct Scsi_Host *host = scmd->device->host;
297
298 trace_scsi_dispatch_cmd_timeout(scmd);
299 scsi_log_completion(scmd, TIMEOUT_ERROR);
300
301 if (host->eh_deadline != -1 && !host->last_reset)
302 host->last_reset = jiffies;
303
304 if (host->hostt->eh_timed_out)
305 rtn = host->hostt->eh_timed_out(scmd);
306
307 if (rtn == BLK_EH_DONE) {
308 /*
309 * If scsi_done() has already set SCMD_STATE_COMPLETE, do not
310 * modify *scmd.
311 */
312 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
313 return BLK_EH_DONE;
314 if (scsi_abort_command(scmd) != SUCCESS) {
315 set_host_byte(scmd, DID_TIME_OUT);
316 scsi_eh_scmd_add(scmd);
317 }
318 }
319
320 return rtn;
321 }
322
323 /**
324 * scsi_block_when_processing_errors - Prevent cmds from being queued.
325 * @sdev: Device on which we are performing recovery.
326 *
327 * Description:
328 * We block until the host is out of error recovery, and then check to
329 * see whether the host or the device is offline.
330 *
331 * Return value:
332 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
333 */
scsi_block_when_processing_errors(struct scsi_device * sdev)334 int scsi_block_when_processing_errors(struct scsi_device *sdev)
335 {
336 int online;
337
338 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
339
340 online = scsi_device_online(sdev);
341
342 return online;
343 }
344 EXPORT_SYMBOL(scsi_block_when_processing_errors);
345
346 #ifdef CONFIG_SCSI_LOGGING
347 /**
348 * scsi_eh_prt_fail_stats - Log info on failures.
349 * @shost: scsi host being recovered.
350 * @work_q: Queue of scsi cmds to process.
351 */
scsi_eh_prt_fail_stats(struct Scsi_Host * shost,struct list_head * work_q)352 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
353 struct list_head *work_q)
354 {
355 struct scsi_cmnd *scmd;
356 struct scsi_device *sdev;
357 int total_failures = 0;
358 int cmd_failed = 0;
359 int cmd_cancel = 0;
360 int devices_failed = 0;
361
362 shost_for_each_device(sdev, shost) {
363 list_for_each_entry(scmd, work_q, eh_entry) {
364 if (scmd->device == sdev) {
365 ++total_failures;
366 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
367 ++cmd_cancel;
368 else
369 ++cmd_failed;
370 }
371 }
372
373 if (cmd_cancel || cmd_failed) {
374 SCSI_LOG_ERROR_RECOVERY(3,
375 shost_printk(KERN_INFO, shost,
376 "%s: cmds failed: %d, cancel: %d\n",
377 __func__, cmd_failed,
378 cmd_cancel));
379 cmd_cancel = 0;
380 cmd_failed = 0;
381 ++devices_failed;
382 }
383 }
384
385 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
386 "Total of %d commands on %d"
387 " devices require eh work\n",
388 total_failures, devices_failed));
389 }
390 #endif
391
392 /**
393 * scsi_report_lun_change - Set flag on all *other* devices on the same target
394 * to indicate that a UNIT ATTENTION is expected.
395 * @sdev: Device reporting the UNIT ATTENTION
396 */
scsi_report_lun_change(struct scsi_device * sdev)397 static void scsi_report_lun_change(struct scsi_device *sdev)
398 {
399 sdev->sdev_target->expecting_lun_change = 1;
400 }
401
402 /**
403 * scsi_report_sense - Examine scsi sense information and log messages for
404 * certain conditions, also issue uevents for some of them.
405 * @sdev: Device reporting the sense code
406 * @sshdr: sshdr to be examined
407 */
scsi_report_sense(struct scsi_device * sdev,struct scsi_sense_hdr * sshdr)408 static void scsi_report_sense(struct scsi_device *sdev,
409 struct scsi_sense_hdr *sshdr)
410 {
411 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
412
413 if (sshdr->sense_key == UNIT_ATTENTION) {
414 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
415 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
416 sdev_printk(KERN_WARNING, sdev,
417 "Inquiry data has changed");
418 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
419 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
420 scsi_report_lun_change(sdev);
421 sdev_printk(KERN_WARNING, sdev,
422 "Warning! Received an indication that the "
423 "LUN assignments on this target have "
424 "changed. The Linux SCSI layer does not "
425 "automatically remap LUN assignments.\n");
426 } else if (sshdr->asc == 0x3f)
427 sdev_printk(KERN_WARNING, sdev,
428 "Warning! Received an indication that the "
429 "operating parameters on this target have "
430 "changed. The Linux SCSI layer does not "
431 "automatically adjust these parameters.\n");
432
433 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
434 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
435 sdev_printk(KERN_WARNING, sdev,
436 "Warning! Received an indication that the "
437 "LUN reached a thin provisioning soft "
438 "threshold.\n");
439 }
440
441 if (sshdr->asc == 0x29) {
442 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
443 sdev_printk(KERN_WARNING, sdev,
444 "Power-on or device reset occurred\n");
445 }
446
447 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
448 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
449 sdev_printk(KERN_WARNING, sdev,
450 "Mode parameters changed");
451 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
452 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
453 sdev_printk(KERN_WARNING, sdev,
454 "Asymmetric access state changed");
455 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
456 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
457 sdev_printk(KERN_WARNING, sdev,
458 "Capacity data has changed");
459 } else if (sshdr->asc == 0x2a)
460 sdev_printk(KERN_WARNING, sdev,
461 "Parameters changed");
462 }
463
464 if (evt_type != SDEV_EVT_MAXBITS) {
465 set_bit(evt_type, sdev->pending_events);
466 schedule_work(&sdev->event_work);
467 }
468 }
469
470 /**
471 * scsi_check_sense - Examine scsi cmd sense
472 * @scmd: Cmd to have sense checked.
473 *
474 * Return value:
475 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
476 *
477 * Notes:
478 * When a deferred error is detected the current command has
479 * not been executed and needs retrying.
480 */
scsi_check_sense(struct scsi_cmnd * scmd)481 int scsi_check_sense(struct scsi_cmnd *scmd)
482 {
483 struct scsi_device *sdev = scmd->device;
484 struct scsi_sense_hdr sshdr;
485
486 if (! scsi_command_normalize_sense(scmd, &sshdr))
487 return FAILED; /* no valid sense data */
488
489 scsi_report_sense(sdev, &sshdr);
490
491 if (scsi_sense_is_deferred(&sshdr))
492 return NEEDS_RETRY;
493
494 if (sdev->handler && sdev->handler->check_sense) {
495 int rc;
496
497 rc = sdev->handler->check_sense(sdev, &sshdr);
498 if (rc != SCSI_RETURN_NOT_HANDLED)
499 return rc;
500 /* handler does not care. Drop down to default handling */
501 }
502
503 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
504 /*
505 * nasty: for mid-layer issued TURs, we need to return the
506 * actual sense data without any recovery attempt. For eh
507 * issued ones, we need to try to recover and interpret
508 */
509 return SUCCESS;
510
511 /*
512 * Previous logic looked for FILEMARK, EOM or ILI which are
513 * mainly associated with tapes and returned SUCCESS.
514 */
515 if (sshdr.response_code == 0x70) {
516 /* fixed format */
517 if (scmd->sense_buffer[2] & 0xe0)
518 return SUCCESS;
519 } else {
520 /*
521 * descriptor format: look for "stream commands sense data
522 * descriptor" (see SSC-3). Assume single sense data
523 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
524 */
525 if ((sshdr.additional_length > 3) &&
526 (scmd->sense_buffer[8] == 0x4) &&
527 (scmd->sense_buffer[11] & 0xe0))
528 return SUCCESS;
529 }
530
531 switch (sshdr.sense_key) {
532 case NO_SENSE:
533 return SUCCESS;
534 case RECOVERED_ERROR:
535 return /* soft_error */ SUCCESS;
536
537 case ABORTED_COMMAND:
538 if (sshdr.asc == 0x10) /* DIF */
539 return SUCCESS;
540
541 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
542 return ADD_TO_MLQUEUE;
543 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
544 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
545 return ADD_TO_MLQUEUE;
546
547 return NEEDS_RETRY;
548 case NOT_READY:
549 case UNIT_ATTENTION:
550 /*
551 * if we are expecting a cc/ua because of a bus reset that we
552 * performed, treat this just as a retry. otherwise this is
553 * information that we should pass up to the upper-level driver
554 * so that we can deal with it there.
555 */
556 if (scmd->device->expecting_cc_ua) {
557 /*
558 * Because some device does not queue unit
559 * attentions correctly, we carefully check
560 * additional sense code and qualifier so as
561 * not to squash media change unit attention.
562 */
563 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
564 scmd->device->expecting_cc_ua = 0;
565 return NEEDS_RETRY;
566 }
567 }
568 /*
569 * we might also expect a cc/ua if another LUN on the target
570 * reported a UA with an ASC/ASCQ of 3F 0E -
571 * REPORTED LUNS DATA HAS CHANGED.
572 */
573 if (scmd->device->sdev_target->expecting_lun_change &&
574 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
575 return NEEDS_RETRY;
576 /*
577 * if the device is in the process of becoming ready, we
578 * should retry.
579 */
580 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
581 return NEEDS_RETRY;
582 /*
583 * if the device is not started, we need to wake
584 * the error handler to start the motor
585 */
586 if (scmd->device->allow_restart &&
587 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
588 return FAILED;
589 /*
590 * Pass the UA upwards for a determination in the completion
591 * functions.
592 */
593 return SUCCESS;
594
595 /* these are not supported */
596 case DATA_PROTECT:
597 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
598 /* Thin provisioning hard threshold reached */
599 set_host_byte(scmd, DID_ALLOC_FAILURE);
600 return SUCCESS;
601 }
602 fallthrough;
603 case COPY_ABORTED:
604 case VOLUME_OVERFLOW:
605 case MISCOMPARE:
606 case BLANK_CHECK:
607 set_host_byte(scmd, DID_TARGET_FAILURE);
608 return SUCCESS;
609
610 case MEDIUM_ERROR:
611 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
612 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
613 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
614 set_host_byte(scmd, DID_MEDIUM_ERROR);
615 return SUCCESS;
616 }
617 return NEEDS_RETRY;
618
619 case HARDWARE_ERROR:
620 if (scmd->device->retry_hwerror)
621 return ADD_TO_MLQUEUE;
622 else
623 set_host_byte(scmd, DID_TARGET_FAILURE);
624 fallthrough;
625
626 case ILLEGAL_REQUEST:
627 if (sshdr.asc == 0x20 || /* Invalid command operation code */
628 sshdr.asc == 0x21 || /* Logical block address out of range */
629 sshdr.asc == 0x22 || /* Invalid function */
630 sshdr.asc == 0x24 || /* Invalid field in cdb */
631 sshdr.asc == 0x26 || /* Parameter value invalid */
632 sshdr.asc == 0x27) { /* Write protected */
633 set_host_byte(scmd, DID_TARGET_FAILURE);
634 }
635 return SUCCESS;
636
637 default:
638 return SUCCESS;
639 }
640 }
641 EXPORT_SYMBOL_GPL(scsi_check_sense);
642
scsi_handle_queue_ramp_up(struct scsi_device * sdev)643 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
644 {
645 struct scsi_host_template *sht = sdev->host->hostt;
646 struct scsi_device *tmp_sdev;
647
648 if (!sht->track_queue_depth ||
649 sdev->queue_depth >= sdev->max_queue_depth)
650 return;
651
652 if (time_before(jiffies,
653 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
654 return;
655
656 if (time_before(jiffies,
657 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
658 return;
659
660 /*
661 * Walk all devices of a target and do
662 * ramp up on them.
663 */
664 shost_for_each_device(tmp_sdev, sdev->host) {
665 if (tmp_sdev->channel != sdev->channel ||
666 tmp_sdev->id != sdev->id ||
667 tmp_sdev->queue_depth == sdev->max_queue_depth)
668 continue;
669
670 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
671 sdev->last_queue_ramp_up = jiffies;
672 }
673 }
674
scsi_handle_queue_full(struct scsi_device * sdev)675 static void scsi_handle_queue_full(struct scsi_device *sdev)
676 {
677 struct scsi_host_template *sht = sdev->host->hostt;
678 struct scsi_device *tmp_sdev;
679
680 if (!sht->track_queue_depth)
681 return;
682
683 shost_for_each_device(tmp_sdev, sdev->host) {
684 if (tmp_sdev->channel != sdev->channel ||
685 tmp_sdev->id != sdev->id)
686 continue;
687 /*
688 * We do not know the number of commands that were at
689 * the device when we got the queue full so we start
690 * from the highest possible value and work our way down.
691 */
692 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
693 }
694 }
695
696 /**
697 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
698 * @scmd: SCSI cmd to examine.
699 *
700 * Notes:
701 * This is *only* called when we are examining the status of commands
702 * queued during error recovery. the main difference here is that we
703 * don't allow for the possibility of retries here, and we are a lot
704 * more restrictive about what we consider acceptable.
705 */
scsi_eh_completed_normally(struct scsi_cmnd * scmd)706 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
707 {
708 /*
709 * first check the host byte, to see if there is anything in there
710 * that would indicate what we need to do.
711 */
712 if (host_byte(scmd->result) == DID_RESET) {
713 /*
714 * rats. we are already in the error handler, so we now
715 * get to try and figure out what to do next. if the sense
716 * is valid, we have a pretty good idea of what to do.
717 * if not, we mark it as FAILED.
718 */
719 return scsi_check_sense(scmd);
720 }
721 if (host_byte(scmd->result) != DID_OK)
722 return FAILED;
723
724 /*
725 * next, check the message byte.
726 */
727 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
728 return FAILED;
729
730 /*
731 * now, check the status byte to see if this indicates
732 * anything special.
733 */
734 switch (status_byte(scmd->result)) {
735 case GOOD:
736 scsi_handle_queue_ramp_up(scmd->device);
737 fallthrough;
738 case COMMAND_TERMINATED:
739 return SUCCESS;
740 case CHECK_CONDITION:
741 return scsi_check_sense(scmd);
742 case CONDITION_GOOD:
743 case INTERMEDIATE_GOOD:
744 case INTERMEDIATE_C_GOOD:
745 /*
746 * who knows? FIXME(eric)
747 */
748 return SUCCESS;
749 case RESERVATION_CONFLICT:
750 if (scmd->cmnd[0] == TEST_UNIT_READY)
751 /* it is a success, we probed the device and
752 * found it */
753 return SUCCESS;
754 /* otherwise, we failed to send the command */
755 return FAILED;
756 case QUEUE_FULL:
757 scsi_handle_queue_full(scmd->device);
758 fallthrough;
759 case BUSY:
760 return NEEDS_RETRY;
761 default:
762 return FAILED;
763 }
764 return FAILED;
765 }
766
767 /**
768 * scsi_eh_done - Completion function for error handling.
769 * @scmd: Cmd that is done.
770 */
scsi_eh_done(struct scsi_cmnd * scmd)771 static void scsi_eh_done(struct scsi_cmnd *scmd)
772 {
773 struct completion *eh_action;
774
775 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
776 "%s result: %x\n", __func__, scmd->result));
777
778 eh_action = scmd->device->host->eh_action;
779 if (eh_action)
780 complete(eh_action);
781 }
782
783 /**
784 * scsi_try_host_reset - ask host adapter to reset itself
785 * @scmd: SCSI cmd to send host reset.
786 */
scsi_try_host_reset(struct scsi_cmnd * scmd)787 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
788 {
789 unsigned long flags;
790 int rtn;
791 struct Scsi_Host *host = scmd->device->host;
792 struct scsi_host_template *hostt = host->hostt;
793
794 SCSI_LOG_ERROR_RECOVERY(3,
795 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
796
797 if (!hostt->eh_host_reset_handler)
798 return FAILED;
799
800 rtn = hostt->eh_host_reset_handler(scmd);
801
802 if (rtn == SUCCESS) {
803 if (!hostt->skip_settle_delay)
804 ssleep(HOST_RESET_SETTLE_TIME);
805 spin_lock_irqsave(host->host_lock, flags);
806 scsi_report_bus_reset(host, scmd_channel(scmd));
807 spin_unlock_irqrestore(host->host_lock, flags);
808 }
809
810 return rtn;
811 }
812
813 /**
814 * scsi_try_bus_reset - ask host to perform a bus reset
815 * @scmd: SCSI cmd to send bus reset.
816 */
scsi_try_bus_reset(struct scsi_cmnd * scmd)817 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
818 {
819 unsigned long flags;
820 int rtn;
821 struct Scsi_Host *host = scmd->device->host;
822 struct scsi_host_template *hostt = host->hostt;
823
824 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
825 "%s: Snd Bus RST\n", __func__));
826
827 if (!hostt->eh_bus_reset_handler)
828 return FAILED;
829
830 rtn = hostt->eh_bus_reset_handler(scmd);
831
832 if (rtn == SUCCESS) {
833 if (!hostt->skip_settle_delay)
834 ssleep(BUS_RESET_SETTLE_TIME);
835 spin_lock_irqsave(host->host_lock, flags);
836 scsi_report_bus_reset(host, scmd_channel(scmd));
837 spin_unlock_irqrestore(host->host_lock, flags);
838 }
839
840 return rtn;
841 }
842
__scsi_report_device_reset(struct scsi_device * sdev,void * data)843 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
844 {
845 sdev->was_reset = 1;
846 sdev->expecting_cc_ua = 1;
847 }
848
849 /**
850 * scsi_try_target_reset - Ask host to perform a target reset
851 * @scmd: SCSI cmd used to send a target reset
852 *
853 * Notes:
854 * There is no timeout for this operation. if this operation is
855 * unreliable for a given host, then the host itself needs to put a
856 * timer on it, and set the host back to a consistent state prior to
857 * returning.
858 */
scsi_try_target_reset(struct scsi_cmnd * scmd)859 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
860 {
861 unsigned long flags;
862 int rtn;
863 struct Scsi_Host *host = scmd->device->host;
864 struct scsi_host_template *hostt = host->hostt;
865
866 if (!hostt->eh_target_reset_handler)
867 return FAILED;
868
869 rtn = hostt->eh_target_reset_handler(scmd);
870 if (rtn == SUCCESS) {
871 spin_lock_irqsave(host->host_lock, flags);
872 __starget_for_each_device(scsi_target(scmd->device), NULL,
873 __scsi_report_device_reset);
874 spin_unlock_irqrestore(host->host_lock, flags);
875 }
876
877 return rtn;
878 }
879
880 /**
881 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
882 * @scmd: SCSI cmd used to send BDR
883 *
884 * Notes:
885 * There is no timeout for this operation. if this operation is
886 * unreliable for a given host, then the host itself needs to put a
887 * timer on it, and set the host back to a consistent state prior to
888 * returning.
889 */
scsi_try_bus_device_reset(struct scsi_cmnd * scmd)890 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
891 {
892 int rtn;
893 struct scsi_host_template *hostt = scmd->device->host->hostt;
894
895 if (!hostt->eh_device_reset_handler)
896 return FAILED;
897
898 rtn = hostt->eh_device_reset_handler(scmd);
899 if (rtn == SUCCESS)
900 __scsi_report_device_reset(scmd->device, NULL);
901 return rtn;
902 }
903
904 /**
905 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
906 * @hostt: SCSI driver host template
907 * @scmd: SCSI cmd used to send a target reset
908 *
909 * Return value:
910 * SUCCESS, FAILED, or FAST_IO_FAIL
911 *
912 * Notes:
913 * SUCCESS does not necessarily indicate that the command
914 * has been aborted; it only indicates that the LLDDs
915 * has cleared all references to that command.
916 * LLDDs should return FAILED only if an abort was required
917 * but could not be executed. LLDDs should return FAST_IO_FAIL
918 * if the device is temporarily unavailable (eg due to a
919 * link down on FibreChannel)
920 */
scsi_try_to_abort_cmd(struct scsi_host_template * hostt,struct scsi_cmnd * scmd)921 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
922 struct scsi_cmnd *scmd)
923 {
924 if (!hostt->eh_abort_handler)
925 return FAILED;
926
927 return hostt->eh_abort_handler(scmd);
928 }
929
scsi_abort_eh_cmnd(struct scsi_cmnd * scmd)930 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
931 {
932 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
933 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
934 if (scsi_try_target_reset(scmd) != SUCCESS)
935 if (scsi_try_bus_reset(scmd) != SUCCESS)
936 scsi_try_host_reset(scmd);
937 }
938
939 /**
940 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
941 * @scmd: SCSI command structure to hijack
942 * @ses: structure to save restore information
943 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
944 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
945 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
946 *
947 * This function is used to save a scsi command information before re-execution
948 * as part of the error recovery process. If @sense_bytes is 0 the command
949 * sent must be one that does not transfer any data. If @sense_bytes != 0
950 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
951 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
952 */
scsi_eh_prep_cmnd(struct scsi_cmnd * scmd,struct scsi_eh_save * ses,unsigned char * cmnd,int cmnd_size,unsigned sense_bytes)953 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
954 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
955 {
956 struct scsi_device *sdev = scmd->device;
957
958 /*
959 * We need saved copies of a number of fields - this is because
960 * error handling may need to overwrite these with different values
961 * to run different commands, and once error handling is complete,
962 * we will need to restore these values prior to running the actual
963 * command.
964 */
965 ses->cmd_len = scmd->cmd_len;
966 ses->cmnd = scmd->cmnd;
967 ses->data_direction = scmd->sc_data_direction;
968 ses->sdb = scmd->sdb;
969 ses->result = scmd->result;
970 ses->resid_len = scmd->req.resid_len;
971 ses->underflow = scmd->underflow;
972 ses->prot_op = scmd->prot_op;
973 ses->eh_eflags = scmd->eh_eflags;
974
975 scmd->prot_op = SCSI_PROT_NORMAL;
976 scmd->eh_eflags = 0;
977 scmd->cmnd = ses->eh_cmnd;
978 memset(scmd->cmnd, 0, BLK_MAX_CDB);
979 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
980 scmd->result = 0;
981 scmd->req.resid_len = 0;
982
983 if (sense_bytes) {
984 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
985 sense_bytes);
986 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
987 scmd->sdb.length);
988 scmd->sdb.table.sgl = &ses->sense_sgl;
989 scmd->sc_data_direction = DMA_FROM_DEVICE;
990 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
991 scmd->cmnd[0] = REQUEST_SENSE;
992 scmd->cmnd[4] = scmd->sdb.length;
993 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
994 } else {
995 scmd->sc_data_direction = DMA_NONE;
996 if (cmnd) {
997 BUG_ON(cmnd_size > BLK_MAX_CDB);
998 memcpy(scmd->cmnd, cmnd, cmnd_size);
999 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1000 }
1001 }
1002
1003 scmd->underflow = 0;
1004
1005 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1006 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1007 (sdev->lun << 5 & 0xe0);
1008
1009 /*
1010 * Zero the sense buffer. The scsi spec mandates that any
1011 * untransferred sense data should be interpreted as being zero.
1012 */
1013 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1014 }
1015 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1016
1017 /**
1018 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1019 * @scmd: SCSI command structure to restore
1020 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1021 *
1022 * Undo any damage done by above scsi_eh_prep_cmnd().
1023 */
scsi_eh_restore_cmnd(struct scsi_cmnd * scmd,struct scsi_eh_save * ses)1024 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1025 {
1026 /*
1027 * Restore original data
1028 */
1029 scmd->cmd_len = ses->cmd_len;
1030 scmd->cmnd = ses->cmnd;
1031 scmd->sc_data_direction = ses->data_direction;
1032 scmd->sdb = ses->sdb;
1033 scmd->result = ses->result;
1034 scmd->req.resid_len = ses->resid_len;
1035 scmd->underflow = ses->underflow;
1036 scmd->prot_op = ses->prot_op;
1037 scmd->eh_eflags = ses->eh_eflags;
1038 }
1039 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1040
1041 /**
1042 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1043 * @scmd: SCSI command structure to hijack
1044 * @cmnd: CDB to send
1045 * @cmnd_size: size in bytes of @cmnd
1046 * @timeout: timeout for this request
1047 * @sense_bytes: size of sense data to copy or 0
1048 *
1049 * This function is used to send a scsi command down to a target device
1050 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1051 *
1052 * Return value:
1053 * SUCCESS or FAILED or NEEDS_RETRY
1054 */
scsi_send_eh_cmnd(struct scsi_cmnd * scmd,unsigned char * cmnd,int cmnd_size,int timeout,unsigned sense_bytes)1055 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1056 int cmnd_size, int timeout, unsigned sense_bytes)
1057 {
1058 struct scsi_device *sdev = scmd->device;
1059 struct Scsi_Host *shost = sdev->host;
1060 DECLARE_COMPLETION_ONSTACK(done);
1061 unsigned long timeleft = timeout, delay;
1062 struct scsi_eh_save ses;
1063 const unsigned long stall_for = msecs_to_jiffies(100);
1064 int rtn;
1065
1066 retry:
1067 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1068 shost->eh_action = &done;
1069
1070 scsi_log_send(scmd);
1071 scmd->scsi_done = scsi_eh_done;
1072
1073 /*
1074 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1075 * change the SCSI device state after we have examined it and before
1076 * .queuecommand() is called.
1077 */
1078 mutex_lock(&sdev->state_mutex);
1079 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1080 mutex_unlock(&sdev->state_mutex);
1081 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1082 "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1083 SDEV_BLOCK));
1084 delay = min(timeleft, stall_for);
1085 timeleft -= delay;
1086 msleep(jiffies_to_msecs(delay));
1087 mutex_lock(&sdev->state_mutex);
1088 }
1089 if (sdev->sdev_state != SDEV_BLOCK)
1090 rtn = shost->hostt->queuecommand(shost, scmd);
1091 else
1092 rtn = SCSI_MLQUEUE_DEVICE_BUSY;
1093 mutex_unlock(&sdev->state_mutex);
1094
1095 if (rtn) {
1096 if (timeleft > stall_for) {
1097 scsi_eh_restore_cmnd(scmd, &ses);
1098 timeleft -= stall_for;
1099 msleep(jiffies_to_msecs(stall_for));
1100 goto retry;
1101 }
1102 /* signal not to enter either branch of the if () below */
1103 timeleft = 0;
1104 rtn = FAILED;
1105 } else {
1106 timeleft = wait_for_completion_timeout(&done, timeout);
1107 rtn = SUCCESS;
1108 }
1109
1110 shost->eh_action = NULL;
1111
1112 scsi_log_completion(scmd, rtn);
1113
1114 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1115 "%s timeleft: %ld\n",
1116 __func__, timeleft));
1117
1118 /*
1119 * If there is time left scsi_eh_done got called, and we will examine
1120 * the actual status codes to see whether the command actually did
1121 * complete normally, else if we have a zero return and no time left,
1122 * the command must still be pending, so abort it and return FAILED.
1123 * If we never actually managed to issue the command, because
1124 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1125 * value above (so don't execute either branch of the if)
1126 */
1127 if (timeleft) {
1128 rtn = scsi_eh_completed_normally(scmd);
1129 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1130 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1131
1132 switch (rtn) {
1133 case SUCCESS:
1134 case NEEDS_RETRY:
1135 case FAILED:
1136 break;
1137 case ADD_TO_MLQUEUE:
1138 rtn = NEEDS_RETRY;
1139 break;
1140 default:
1141 rtn = FAILED;
1142 break;
1143 }
1144 } else if (rtn != FAILED) {
1145 scsi_abort_eh_cmnd(scmd);
1146 rtn = FAILED;
1147 }
1148
1149 scsi_eh_restore_cmnd(scmd, &ses);
1150
1151 return rtn;
1152 }
1153
1154 /**
1155 * scsi_request_sense - Request sense data from a particular target.
1156 * @scmd: SCSI cmd for request sense.
1157 *
1158 * Notes:
1159 * Some hosts automatically obtain this information, others require
1160 * that we obtain it on our own. This function will *not* return until
1161 * the command either times out, or it completes.
1162 */
scsi_request_sense(struct scsi_cmnd * scmd)1163 static int scsi_request_sense(struct scsi_cmnd *scmd)
1164 {
1165 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1166 }
1167
scsi_eh_action(struct scsi_cmnd * scmd,int rtn)1168 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1169 {
1170 if (!blk_rq_is_passthrough(scmd->request)) {
1171 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1172 if (sdrv->eh_action)
1173 rtn = sdrv->eh_action(scmd, rtn);
1174 }
1175 return rtn;
1176 }
1177
1178 /**
1179 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1180 * @scmd: Original SCSI cmd that eh has finished.
1181 * @done_q: Queue for processed commands.
1182 *
1183 * Notes:
1184 * We don't want to use the normal command completion while we are are
1185 * still handling errors - it may cause other commands to be queued,
1186 * and that would disturb what we are doing. Thus we really want to
1187 * keep a list of pending commands for final completion, and once we
1188 * are ready to leave error handling we handle completion for real.
1189 */
scsi_eh_finish_cmd(struct scsi_cmnd * scmd,struct list_head * done_q)1190 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1191 {
1192 list_move_tail(&scmd->eh_entry, done_q);
1193 }
1194 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1195
1196 /**
1197 * scsi_eh_get_sense - Get device sense data.
1198 * @work_q: Queue of commands to process.
1199 * @done_q: Queue of processed commands.
1200 *
1201 * Description:
1202 * See if we need to request sense information. if so, then get it
1203 * now, so we have a better idea of what to do.
1204 *
1205 * Notes:
1206 * This has the unfortunate side effect that if a shost adapter does
1207 * not automatically request sense information, we end up shutting
1208 * it down before we request it.
1209 *
1210 * All drivers should request sense information internally these days,
1211 * so for now all I have to say is tough noogies if you end up in here.
1212 *
1213 * XXX: Long term this code should go away, but that needs an audit of
1214 * all LLDDs first.
1215 */
scsi_eh_get_sense(struct list_head * work_q,struct list_head * done_q)1216 int scsi_eh_get_sense(struct list_head *work_q,
1217 struct list_head *done_q)
1218 {
1219 struct scsi_cmnd *scmd, *next;
1220 struct Scsi_Host *shost;
1221 int rtn;
1222
1223 /*
1224 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1225 * should not get sense.
1226 */
1227 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1228 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1229 SCSI_SENSE_VALID(scmd))
1230 continue;
1231
1232 shost = scmd->device->host;
1233 if (scsi_host_eh_past_deadline(shost)) {
1234 SCSI_LOG_ERROR_RECOVERY(3,
1235 scmd_printk(KERN_INFO, scmd,
1236 "%s: skip request sense, past eh deadline\n",
1237 current->comm));
1238 break;
1239 }
1240 if (status_byte(scmd->result) != CHECK_CONDITION)
1241 /*
1242 * don't request sense if there's no check condition
1243 * status because the error we're processing isn't one
1244 * that has a sense code (and some devices get
1245 * confused by sense requests out of the blue)
1246 */
1247 continue;
1248
1249 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1250 "%s: requesting sense\n",
1251 current->comm));
1252 rtn = scsi_request_sense(scmd);
1253 if (rtn != SUCCESS)
1254 continue;
1255
1256 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1257 "sense requested, result %x\n", scmd->result));
1258 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1259
1260 rtn = scsi_decide_disposition(scmd);
1261
1262 /*
1263 * if the result was normal, then just pass it along to the
1264 * upper level.
1265 */
1266 if (rtn == SUCCESS)
1267 /*
1268 * We don't want this command reissued, just finished
1269 * with the sense data, so set retries to the max
1270 * allowed to ensure it won't get reissued. If the user
1271 * has requested infinite retries, we also want to
1272 * finish this command, so force completion by setting
1273 * retries and allowed to the same value.
1274 */
1275 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1276 scmd->retries = scmd->allowed = 1;
1277 else
1278 scmd->retries = scmd->allowed;
1279 else if (rtn != NEEDS_RETRY)
1280 continue;
1281
1282 scsi_eh_finish_cmd(scmd, done_q);
1283 }
1284
1285 return list_empty(work_q);
1286 }
1287 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1288
1289 /**
1290 * scsi_eh_tur - Send TUR to device.
1291 * @scmd: &scsi_cmnd to send TUR
1292 *
1293 * Return value:
1294 * 0 - Device is ready. 1 - Device NOT ready.
1295 */
scsi_eh_tur(struct scsi_cmnd * scmd)1296 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1297 {
1298 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1299 int retry_cnt = 1, rtn;
1300
1301 retry_tur:
1302 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1303 scmd->device->eh_timeout, 0);
1304
1305 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1306 "%s return: %x\n", __func__, rtn));
1307
1308 switch (rtn) {
1309 case NEEDS_RETRY:
1310 if (retry_cnt--)
1311 goto retry_tur;
1312 fallthrough;
1313 case SUCCESS:
1314 return 0;
1315 default:
1316 return 1;
1317 }
1318 }
1319
1320 /**
1321 * scsi_eh_test_devices - check if devices are responding from error recovery.
1322 * @cmd_list: scsi commands in error recovery.
1323 * @work_q: queue for commands which still need more error recovery
1324 * @done_q: queue for commands which are finished
1325 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1326 *
1327 * Decription:
1328 * Tests if devices are in a working state. Commands to devices now in
1329 * a working state are sent to the done_q while commands to devices which
1330 * are still failing to respond are returned to the work_q for more
1331 * processing.
1332 **/
scsi_eh_test_devices(struct list_head * cmd_list,struct list_head * work_q,struct list_head * done_q,int try_stu)1333 static int scsi_eh_test_devices(struct list_head *cmd_list,
1334 struct list_head *work_q,
1335 struct list_head *done_q, int try_stu)
1336 {
1337 struct scsi_cmnd *scmd, *next;
1338 struct scsi_device *sdev;
1339 int finish_cmds;
1340
1341 while (!list_empty(cmd_list)) {
1342 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1343 sdev = scmd->device;
1344
1345 if (!try_stu) {
1346 if (scsi_host_eh_past_deadline(sdev->host)) {
1347 /* Push items back onto work_q */
1348 list_splice_init(cmd_list, work_q);
1349 SCSI_LOG_ERROR_RECOVERY(3,
1350 sdev_printk(KERN_INFO, sdev,
1351 "%s: skip test device, past eh deadline",
1352 current->comm));
1353 break;
1354 }
1355 }
1356
1357 finish_cmds = !scsi_device_online(scmd->device) ||
1358 (try_stu && !scsi_eh_try_stu(scmd) &&
1359 !scsi_eh_tur(scmd)) ||
1360 !scsi_eh_tur(scmd);
1361
1362 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1363 if (scmd->device == sdev) {
1364 if (finish_cmds &&
1365 (try_stu ||
1366 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1367 scsi_eh_finish_cmd(scmd, done_q);
1368 else
1369 list_move_tail(&scmd->eh_entry, work_q);
1370 }
1371 }
1372 return list_empty(work_q);
1373 }
1374
1375 /**
1376 * scsi_eh_try_stu - Send START_UNIT to device.
1377 * @scmd: &scsi_cmnd to send START_UNIT
1378 *
1379 * Return value:
1380 * 0 - Device is ready. 1 - Device NOT ready.
1381 */
scsi_eh_try_stu(struct scsi_cmnd * scmd)1382 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1383 {
1384 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1385
1386 if (scmd->device->allow_restart) {
1387 int i, rtn = NEEDS_RETRY;
1388
1389 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1390 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1391
1392 if (rtn == SUCCESS)
1393 return 0;
1394 }
1395
1396 return 1;
1397 }
1398
1399 /**
1400 * scsi_eh_stu - send START_UNIT if needed
1401 * @shost: &scsi host being recovered.
1402 * @work_q: &list_head for pending commands.
1403 * @done_q: &list_head for processed commands.
1404 *
1405 * Notes:
1406 * If commands are failing due to not ready, initializing command required,
1407 * try revalidating the device, which will end up sending a start unit.
1408 */
scsi_eh_stu(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1409 static int scsi_eh_stu(struct Scsi_Host *shost,
1410 struct list_head *work_q,
1411 struct list_head *done_q)
1412 {
1413 struct scsi_cmnd *scmd, *stu_scmd, *next;
1414 struct scsi_device *sdev;
1415
1416 shost_for_each_device(sdev, shost) {
1417 if (scsi_host_eh_past_deadline(shost)) {
1418 SCSI_LOG_ERROR_RECOVERY(3,
1419 sdev_printk(KERN_INFO, sdev,
1420 "%s: skip START_UNIT, past eh deadline\n",
1421 current->comm));
1422 scsi_device_put(sdev);
1423 break;
1424 }
1425 stu_scmd = NULL;
1426 list_for_each_entry(scmd, work_q, eh_entry)
1427 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1428 scsi_check_sense(scmd) == FAILED ) {
1429 stu_scmd = scmd;
1430 break;
1431 }
1432
1433 if (!stu_scmd)
1434 continue;
1435
1436 SCSI_LOG_ERROR_RECOVERY(3,
1437 sdev_printk(KERN_INFO, sdev,
1438 "%s: Sending START_UNIT\n",
1439 current->comm));
1440
1441 if (!scsi_eh_try_stu(stu_scmd)) {
1442 if (!scsi_device_online(sdev) ||
1443 !scsi_eh_tur(stu_scmd)) {
1444 list_for_each_entry_safe(scmd, next,
1445 work_q, eh_entry) {
1446 if (scmd->device == sdev &&
1447 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1448 scsi_eh_finish_cmd(scmd, done_q);
1449 }
1450 }
1451 } else {
1452 SCSI_LOG_ERROR_RECOVERY(3,
1453 sdev_printk(KERN_INFO, sdev,
1454 "%s: START_UNIT failed\n",
1455 current->comm));
1456 }
1457 }
1458
1459 return list_empty(work_q);
1460 }
1461
1462
1463 /**
1464 * scsi_eh_bus_device_reset - send bdr if needed
1465 * @shost: scsi host being recovered.
1466 * @work_q: &list_head for pending commands.
1467 * @done_q: &list_head for processed commands.
1468 *
1469 * Notes:
1470 * Try a bus device reset. Still, look to see whether we have multiple
1471 * devices that are jammed or not - if we have multiple devices, it
1472 * makes no sense to try bus_device_reset - we really would need to try
1473 * a bus_reset instead.
1474 */
scsi_eh_bus_device_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1475 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1476 struct list_head *work_q,
1477 struct list_head *done_q)
1478 {
1479 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1480 struct scsi_device *sdev;
1481 int rtn;
1482
1483 shost_for_each_device(sdev, shost) {
1484 if (scsi_host_eh_past_deadline(shost)) {
1485 SCSI_LOG_ERROR_RECOVERY(3,
1486 sdev_printk(KERN_INFO, sdev,
1487 "%s: skip BDR, past eh deadline\n",
1488 current->comm));
1489 scsi_device_put(sdev);
1490 break;
1491 }
1492 bdr_scmd = NULL;
1493 list_for_each_entry(scmd, work_q, eh_entry)
1494 if (scmd->device == sdev) {
1495 bdr_scmd = scmd;
1496 break;
1497 }
1498
1499 if (!bdr_scmd)
1500 continue;
1501
1502 SCSI_LOG_ERROR_RECOVERY(3,
1503 sdev_printk(KERN_INFO, sdev,
1504 "%s: Sending BDR\n", current->comm));
1505 rtn = scsi_try_bus_device_reset(bdr_scmd);
1506 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1507 if (!scsi_device_online(sdev) ||
1508 rtn == FAST_IO_FAIL ||
1509 !scsi_eh_tur(bdr_scmd)) {
1510 list_for_each_entry_safe(scmd, next,
1511 work_q, eh_entry) {
1512 if (scmd->device == sdev &&
1513 scsi_eh_action(scmd, rtn) != FAILED)
1514 scsi_eh_finish_cmd(scmd,
1515 done_q);
1516 }
1517 }
1518 } else {
1519 SCSI_LOG_ERROR_RECOVERY(3,
1520 sdev_printk(KERN_INFO, sdev,
1521 "%s: BDR failed\n", current->comm));
1522 }
1523 }
1524
1525 return list_empty(work_q);
1526 }
1527
1528 /**
1529 * scsi_eh_target_reset - send target reset if needed
1530 * @shost: scsi host being recovered.
1531 * @work_q: &list_head for pending commands.
1532 * @done_q: &list_head for processed commands.
1533 *
1534 * Notes:
1535 * Try a target reset.
1536 */
scsi_eh_target_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1537 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1538 struct list_head *work_q,
1539 struct list_head *done_q)
1540 {
1541 LIST_HEAD(tmp_list);
1542 LIST_HEAD(check_list);
1543
1544 list_splice_init(work_q, &tmp_list);
1545
1546 while (!list_empty(&tmp_list)) {
1547 struct scsi_cmnd *next, *scmd;
1548 int rtn;
1549 unsigned int id;
1550
1551 if (scsi_host_eh_past_deadline(shost)) {
1552 /* push back on work queue for further processing */
1553 list_splice_init(&check_list, work_q);
1554 list_splice_init(&tmp_list, work_q);
1555 SCSI_LOG_ERROR_RECOVERY(3,
1556 shost_printk(KERN_INFO, shost,
1557 "%s: Skip target reset, past eh deadline\n",
1558 current->comm));
1559 return list_empty(work_q);
1560 }
1561
1562 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1563 id = scmd_id(scmd);
1564
1565 SCSI_LOG_ERROR_RECOVERY(3,
1566 shost_printk(KERN_INFO, shost,
1567 "%s: Sending target reset to target %d\n",
1568 current->comm, id));
1569 rtn = scsi_try_target_reset(scmd);
1570 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1571 SCSI_LOG_ERROR_RECOVERY(3,
1572 shost_printk(KERN_INFO, shost,
1573 "%s: Target reset failed"
1574 " target: %d\n",
1575 current->comm, id));
1576 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1577 if (scmd_id(scmd) != id)
1578 continue;
1579
1580 if (rtn == SUCCESS)
1581 list_move_tail(&scmd->eh_entry, &check_list);
1582 else if (rtn == FAST_IO_FAIL)
1583 scsi_eh_finish_cmd(scmd, done_q);
1584 else
1585 /* push back on work queue for further processing */
1586 list_move(&scmd->eh_entry, work_q);
1587 }
1588 }
1589
1590 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1591 }
1592
1593 /**
1594 * scsi_eh_bus_reset - send a bus reset
1595 * @shost: &scsi host being recovered.
1596 * @work_q: &list_head for pending commands.
1597 * @done_q: &list_head for processed commands.
1598 */
scsi_eh_bus_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1599 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1600 struct list_head *work_q,
1601 struct list_head *done_q)
1602 {
1603 struct scsi_cmnd *scmd, *chan_scmd, *next;
1604 LIST_HEAD(check_list);
1605 unsigned int channel;
1606 int rtn;
1607
1608 /*
1609 * we really want to loop over the various channels, and do this on
1610 * a channel by channel basis. we should also check to see if any
1611 * of the failed commands are on soft_reset devices, and if so, skip
1612 * the reset.
1613 */
1614
1615 for (channel = 0; channel <= shost->max_channel; channel++) {
1616 if (scsi_host_eh_past_deadline(shost)) {
1617 list_splice_init(&check_list, work_q);
1618 SCSI_LOG_ERROR_RECOVERY(3,
1619 shost_printk(KERN_INFO, shost,
1620 "%s: skip BRST, past eh deadline\n",
1621 current->comm));
1622 return list_empty(work_q);
1623 }
1624
1625 chan_scmd = NULL;
1626 list_for_each_entry(scmd, work_q, eh_entry) {
1627 if (channel == scmd_channel(scmd)) {
1628 chan_scmd = scmd;
1629 break;
1630 /*
1631 * FIXME add back in some support for
1632 * soft_reset devices.
1633 */
1634 }
1635 }
1636
1637 if (!chan_scmd)
1638 continue;
1639 SCSI_LOG_ERROR_RECOVERY(3,
1640 shost_printk(KERN_INFO, shost,
1641 "%s: Sending BRST chan: %d\n",
1642 current->comm, channel));
1643 rtn = scsi_try_bus_reset(chan_scmd);
1644 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1645 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1646 if (channel == scmd_channel(scmd)) {
1647 if (rtn == FAST_IO_FAIL)
1648 scsi_eh_finish_cmd(scmd,
1649 done_q);
1650 else
1651 list_move_tail(&scmd->eh_entry,
1652 &check_list);
1653 }
1654 }
1655 } else {
1656 SCSI_LOG_ERROR_RECOVERY(3,
1657 shost_printk(KERN_INFO, shost,
1658 "%s: BRST failed chan: %d\n",
1659 current->comm, channel));
1660 }
1661 }
1662 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1663 }
1664
1665 /**
1666 * scsi_eh_host_reset - send a host reset
1667 * @shost: host to be reset.
1668 * @work_q: &list_head for pending commands.
1669 * @done_q: &list_head for processed commands.
1670 */
scsi_eh_host_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1671 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1672 struct list_head *work_q,
1673 struct list_head *done_q)
1674 {
1675 struct scsi_cmnd *scmd, *next;
1676 LIST_HEAD(check_list);
1677 int rtn;
1678
1679 if (!list_empty(work_q)) {
1680 scmd = list_entry(work_q->next,
1681 struct scsi_cmnd, eh_entry);
1682
1683 SCSI_LOG_ERROR_RECOVERY(3,
1684 shost_printk(KERN_INFO, shost,
1685 "%s: Sending HRST\n",
1686 current->comm));
1687
1688 rtn = scsi_try_host_reset(scmd);
1689 if (rtn == SUCCESS) {
1690 list_splice_init(work_q, &check_list);
1691 } else if (rtn == FAST_IO_FAIL) {
1692 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1693 scsi_eh_finish_cmd(scmd, done_q);
1694 }
1695 } else {
1696 SCSI_LOG_ERROR_RECOVERY(3,
1697 shost_printk(KERN_INFO, shost,
1698 "%s: HRST failed\n",
1699 current->comm));
1700 }
1701 }
1702 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1703 }
1704
1705 /**
1706 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1707 * @work_q: &list_head for pending commands.
1708 * @done_q: &list_head for processed commands.
1709 */
scsi_eh_offline_sdevs(struct list_head * work_q,struct list_head * done_q)1710 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1711 struct list_head *done_q)
1712 {
1713 struct scsi_cmnd *scmd, *next;
1714 struct scsi_device *sdev;
1715
1716 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1717 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1718 "not ready after error recovery\n");
1719 sdev = scmd->device;
1720
1721 mutex_lock(&sdev->state_mutex);
1722 scsi_device_set_state(sdev, SDEV_OFFLINE);
1723 mutex_unlock(&sdev->state_mutex);
1724
1725 scsi_eh_finish_cmd(scmd, done_q);
1726 }
1727 return;
1728 }
1729
1730 /**
1731 * scsi_noretry_cmd - determine if command should be failed fast
1732 * @scmd: SCSI cmd to examine.
1733 */
scsi_noretry_cmd(struct scsi_cmnd * scmd)1734 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1735 {
1736 switch (host_byte(scmd->result)) {
1737 case DID_OK:
1738 break;
1739 case DID_TIME_OUT:
1740 goto check_type;
1741 case DID_BUS_BUSY:
1742 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1743 case DID_PARITY:
1744 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1745 case DID_ERROR:
1746 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1747 status_byte(scmd->result) == RESERVATION_CONFLICT)
1748 return 0;
1749 fallthrough;
1750 case DID_SOFT_ERROR:
1751 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1752 }
1753
1754 if (status_byte(scmd->result) != CHECK_CONDITION)
1755 return 0;
1756
1757 check_type:
1758 /*
1759 * assume caller has checked sense and determined
1760 * the check condition was retryable.
1761 */
1762 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1763 blk_rq_is_passthrough(scmd->request))
1764 return 1;
1765
1766 return 0;
1767 }
1768
1769 /**
1770 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1771 * @scmd: SCSI cmd to examine.
1772 *
1773 * Notes:
1774 * This is *only* called when we are examining the status after sending
1775 * out the actual data command. any commands that are queued for error
1776 * recovery (e.g. test_unit_ready) do *not* come through here.
1777 *
1778 * When this routine returns failed, it means the error handler thread
1779 * is woken. In cases where the error code indicates an error that
1780 * doesn't require the error handler read (i.e. we don't need to
1781 * abort/reset), this function should return SUCCESS.
1782 */
scsi_decide_disposition(struct scsi_cmnd * scmd)1783 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1784 {
1785 int rtn;
1786
1787 /*
1788 * if the device is offline, then we clearly just pass the result back
1789 * up to the top level.
1790 */
1791 if (!scsi_device_online(scmd->device)) {
1792 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1793 "%s: device offline - report as SUCCESS\n", __func__));
1794 return SUCCESS;
1795 }
1796
1797 /*
1798 * first check the host byte, to see if there is anything in there
1799 * that would indicate what we need to do.
1800 */
1801 switch (host_byte(scmd->result)) {
1802 case DID_PASSTHROUGH:
1803 /*
1804 * no matter what, pass this through to the upper layer.
1805 * nuke this special code so that it looks like we are saying
1806 * did_ok.
1807 */
1808 scmd->result &= 0xff00ffff;
1809 return SUCCESS;
1810 case DID_OK:
1811 /*
1812 * looks good. drop through, and check the next byte.
1813 */
1814 break;
1815 case DID_ABORT:
1816 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1817 set_host_byte(scmd, DID_TIME_OUT);
1818 return SUCCESS;
1819 }
1820 fallthrough;
1821 case DID_NO_CONNECT:
1822 case DID_BAD_TARGET:
1823 /*
1824 * note - this means that we just report the status back
1825 * to the top level driver, not that we actually think
1826 * that it indicates SUCCESS.
1827 */
1828 return SUCCESS;
1829 case DID_SOFT_ERROR:
1830 /*
1831 * when the low level driver returns did_soft_error,
1832 * it is responsible for keeping an internal retry counter
1833 * in order to avoid endless loops (db)
1834 */
1835 goto maybe_retry;
1836 case DID_IMM_RETRY:
1837 return NEEDS_RETRY;
1838
1839 case DID_REQUEUE:
1840 return ADD_TO_MLQUEUE;
1841 case DID_TRANSPORT_DISRUPTED:
1842 /*
1843 * LLD/transport was disrupted during processing of the IO.
1844 * The transport class is now blocked/blocking,
1845 * and the transport will decide what to do with the IO
1846 * based on its timers and recovery capablilities if
1847 * there are enough retries.
1848 */
1849 goto maybe_retry;
1850 case DID_TRANSPORT_FAILFAST:
1851 /*
1852 * The transport decided to failfast the IO (most likely
1853 * the fast io fail tmo fired), so send IO directly upwards.
1854 */
1855 return SUCCESS;
1856 case DID_ERROR:
1857 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1858 status_byte(scmd->result) == RESERVATION_CONFLICT)
1859 /*
1860 * execute reservation conflict processing code
1861 * lower down
1862 */
1863 break;
1864 fallthrough;
1865 case DID_BUS_BUSY:
1866 case DID_PARITY:
1867 goto maybe_retry;
1868 case DID_TIME_OUT:
1869 /*
1870 * when we scan the bus, we get timeout messages for
1871 * these commands if there is no device available.
1872 * other hosts report did_no_connect for the same thing.
1873 */
1874 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1875 scmd->cmnd[0] == INQUIRY)) {
1876 return SUCCESS;
1877 } else {
1878 return FAILED;
1879 }
1880 case DID_RESET:
1881 return SUCCESS;
1882 default:
1883 return FAILED;
1884 }
1885
1886 /*
1887 * next, check the message byte.
1888 */
1889 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1890 return FAILED;
1891
1892 /*
1893 * check the status byte to see if this indicates anything special.
1894 */
1895 switch (status_byte(scmd->result)) {
1896 case QUEUE_FULL:
1897 scsi_handle_queue_full(scmd->device);
1898 /*
1899 * the case of trying to send too many commands to a
1900 * tagged queueing device.
1901 */
1902 fallthrough;
1903 case BUSY:
1904 /*
1905 * device can't talk to us at the moment. Should only
1906 * occur (SAM-3) when the task queue is empty, so will cause
1907 * the empty queue handling to trigger a stall in the
1908 * device.
1909 */
1910 return ADD_TO_MLQUEUE;
1911 case GOOD:
1912 if (scmd->cmnd[0] == REPORT_LUNS)
1913 scmd->device->sdev_target->expecting_lun_change = 0;
1914 scsi_handle_queue_ramp_up(scmd->device);
1915 fallthrough;
1916 case COMMAND_TERMINATED:
1917 return SUCCESS;
1918 case TASK_ABORTED:
1919 goto maybe_retry;
1920 case CHECK_CONDITION:
1921 rtn = scsi_check_sense(scmd);
1922 if (rtn == NEEDS_RETRY)
1923 goto maybe_retry;
1924 /* if rtn == FAILED, we have no sense information;
1925 * returning FAILED will wake the error handler thread
1926 * to collect the sense and redo the decide
1927 * disposition */
1928 return rtn;
1929 case CONDITION_GOOD:
1930 case INTERMEDIATE_GOOD:
1931 case INTERMEDIATE_C_GOOD:
1932 case ACA_ACTIVE:
1933 /*
1934 * who knows? FIXME(eric)
1935 */
1936 return SUCCESS;
1937
1938 case RESERVATION_CONFLICT:
1939 sdev_printk(KERN_INFO, scmd->device,
1940 "reservation conflict\n");
1941 set_host_byte(scmd, DID_NEXUS_FAILURE);
1942 return SUCCESS; /* causes immediate i/o error */
1943 default:
1944 return FAILED;
1945 }
1946 return FAILED;
1947
1948 maybe_retry:
1949
1950 /* we requeue for retry because the error was retryable, and
1951 * the request was not marked fast fail. Note that above,
1952 * even if the request is marked fast fail, we still requeue
1953 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1954 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
1955 return NEEDS_RETRY;
1956 } else {
1957 /*
1958 * no more retries - report this one back to upper level.
1959 */
1960 return SUCCESS;
1961 }
1962 }
1963
eh_lock_door_done(struct request * req,blk_status_t status)1964 static void eh_lock_door_done(struct request *req, blk_status_t status)
1965 {
1966 blk_put_request(req);
1967 }
1968
1969 /**
1970 * scsi_eh_lock_door - Prevent medium removal for the specified device
1971 * @sdev: SCSI device to prevent medium removal
1972 *
1973 * Locking:
1974 * We must be called from process context.
1975 *
1976 * Notes:
1977 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1978 * head of the devices request queue, and continue.
1979 */
scsi_eh_lock_door(struct scsi_device * sdev)1980 static void scsi_eh_lock_door(struct scsi_device *sdev)
1981 {
1982 struct request *req;
1983 struct scsi_request *rq;
1984
1985 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, 0);
1986 if (IS_ERR(req))
1987 return;
1988 rq = scsi_req(req);
1989
1990 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1991 rq->cmd[1] = 0;
1992 rq->cmd[2] = 0;
1993 rq->cmd[3] = 0;
1994 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
1995 rq->cmd[5] = 0;
1996 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
1997
1998 req->rq_flags |= RQF_QUIET;
1999 req->timeout = 10 * HZ;
2000 rq->retries = 5;
2001
2002 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
2003 }
2004
2005 /**
2006 * scsi_restart_operations - restart io operations to the specified host.
2007 * @shost: Host we are restarting.
2008 *
2009 * Notes:
2010 * When we entered the error handler, we blocked all further i/o to
2011 * this device. we need to 'reverse' this process.
2012 */
scsi_restart_operations(struct Scsi_Host * shost)2013 static void scsi_restart_operations(struct Scsi_Host *shost)
2014 {
2015 struct scsi_device *sdev;
2016 unsigned long flags;
2017
2018 /*
2019 * If the door was locked, we need to insert a door lock request
2020 * onto the head of the SCSI request queue for the device. There
2021 * is no point trying to lock the door of an off-line device.
2022 */
2023 shost_for_each_device(sdev, shost) {
2024 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2025 scsi_eh_lock_door(sdev);
2026 sdev->was_reset = 0;
2027 }
2028 }
2029
2030 /*
2031 * next free up anything directly waiting upon the host. this
2032 * will be requests for character device operations, and also for
2033 * ioctls to queued block devices.
2034 */
2035 SCSI_LOG_ERROR_RECOVERY(3,
2036 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2037
2038 spin_lock_irqsave(shost->host_lock, flags);
2039 if (scsi_host_set_state(shost, SHOST_RUNNING))
2040 if (scsi_host_set_state(shost, SHOST_CANCEL))
2041 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2042 spin_unlock_irqrestore(shost->host_lock, flags);
2043
2044 wake_up(&shost->host_wait);
2045
2046 /*
2047 * finally we need to re-initiate requests that may be pending. we will
2048 * have had everything blocked while error handling is taking place, and
2049 * now that error recovery is done, we will need to ensure that these
2050 * requests are started.
2051 */
2052 scsi_run_host_queues(shost);
2053
2054 /*
2055 * if eh is active and host_eh_scheduled is pending we need to re-run
2056 * recovery. we do this check after scsi_run_host_queues() to allow
2057 * everything pent up since the last eh run a chance to make forward
2058 * progress before we sync again. Either we'll immediately re-run
2059 * recovery or scsi_device_unbusy() will wake us again when these
2060 * pending commands complete.
2061 */
2062 spin_lock_irqsave(shost->host_lock, flags);
2063 if (shost->host_eh_scheduled)
2064 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2065 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2066 spin_unlock_irqrestore(shost->host_lock, flags);
2067 }
2068
2069 /**
2070 * scsi_eh_ready_devs - check device ready state and recover if not.
2071 * @shost: host to be recovered.
2072 * @work_q: &list_head for pending commands.
2073 * @done_q: &list_head for processed commands.
2074 */
scsi_eh_ready_devs(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)2075 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2076 struct list_head *work_q,
2077 struct list_head *done_q)
2078 {
2079 if (!scsi_eh_stu(shost, work_q, done_q))
2080 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2081 if (!scsi_eh_target_reset(shost, work_q, done_q))
2082 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2083 if (!scsi_eh_host_reset(shost, work_q, done_q))
2084 scsi_eh_offline_sdevs(work_q,
2085 done_q);
2086 }
2087 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2088
2089 /**
2090 * scsi_eh_flush_done_q - finish processed commands or retry them.
2091 * @done_q: list_head of processed commands.
2092 */
scsi_eh_flush_done_q(struct list_head * done_q)2093 void scsi_eh_flush_done_q(struct list_head *done_q)
2094 {
2095 struct scsi_cmnd *scmd, *next;
2096
2097 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2098 list_del_init(&scmd->eh_entry);
2099 if (scsi_device_online(scmd->device) &&
2100 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd)) {
2101 SCSI_LOG_ERROR_RECOVERY(3,
2102 scmd_printk(KERN_INFO, scmd,
2103 "%s: flush retry cmd\n",
2104 current->comm));
2105 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2106 } else {
2107 /*
2108 * If just we got sense for the device (called
2109 * scsi_eh_get_sense), scmd->result is already
2110 * set, do not set DRIVER_TIMEOUT.
2111 */
2112 if (!scmd->result)
2113 scmd->result |= (DRIVER_TIMEOUT << 24);
2114 SCSI_LOG_ERROR_RECOVERY(3,
2115 scmd_printk(KERN_INFO, scmd,
2116 "%s: flush finish cmd\n",
2117 current->comm));
2118 scsi_finish_command(scmd);
2119 }
2120 }
2121 }
2122 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2123
2124 /**
2125 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2126 * @shost: Host to unjam.
2127 *
2128 * Notes:
2129 * When we come in here, we *know* that all commands on the bus have
2130 * either completed, failed or timed out. we also know that no further
2131 * commands are being sent to the host, so things are relatively quiet
2132 * and we have freedom to fiddle with things as we wish.
2133 *
2134 * This is only the *default* implementation. it is possible for
2135 * individual drivers to supply their own version of this function, and
2136 * if the maintainer wishes to do this, it is strongly suggested that
2137 * this function be taken as a template and modified. this function
2138 * was designed to correctly handle problems for about 95% of the
2139 * different cases out there, and it should always provide at least a
2140 * reasonable amount of error recovery.
2141 *
2142 * Any command marked 'failed' or 'timeout' must eventually have
2143 * scsi_finish_cmd() called for it. we do all of the retry stuff
2144 * here, so when we restart the host after we return it should have an
2145 * empty queue.
2146 */
scsi_unjam_host(struct Scsi_Host * shost)2147 static void scsi_unjam_host(struct Scsi_Host *shost)
2148 {
2149 unsigned long flags;
2150 LIST_HEAD(eh_work_q);
2151 LIST_HEAD(eh_done_q);
2152
2153 spin_lock_irqsave(shost->host_lock, flags);
2154 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2155 spin_unlock_irqrestore(shost->host_lock, flags);
2156
2157 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2158
2159 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2160 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2161
2162 spin_lock_irqsave(shost->host_lock, flags);
2163 if (shost->eh_deadline != -1)
2164 shost->last_reset = 0;
2165 spin_unlock_irqrestore(shost->host_lock, flags);
2166 scsi_eh_flush_done_q(&eh_done_q);
2167 }
2168
2169 /**
2170 * scsi_error_handler - SCSI error handler thread
2171 * @data: Host for which we are running.
2172 *
2173 * Notes:
2174 * This is the main error handling loop. This is run as a kernel thread
2175 * for every SCSI host and handles all error handling activity.
2176 */
scsi_error_handler(void * data)2177 int scsi_error_handler(void *data)
2178 {
2179 struct Scsi_Host *shost = data;
2180
2181 /*
2182 * We use TASK_INTERRUPTIBLE so that the thread is not
2183 * counted against the load average as a running process.
2184 * We never actually get interrupted because kthread_run
2185 * disables signal delivery for the created thread.
2186 */
2187 while (true) {
2188 /*
2189 * The sequence in kthread_stop() sets the stop flag first
2190 * then wakes the process. To avoid missed wakeups, the task
2191 * should always be in a non running state before the stop
2192 * flag is checked
2193 */
2194 set_current_state(TASK_INTERRUPTIBLE);
2195 if (kthread_should_stop())
2196 break;
2197
2198 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2199 shost->host_failed != scsi_host_busy(shost)) {
2200 SCSI_LOG_ERROR_RECOVERY(1,
2201 shost_printk(KERN_INFO, shost,
2202 "scsi_eh_%d: sleeping\n",
2203 shost->host_no));
2204 schedule();
2205 continue;
2206 }
2207
2208 __set_current_state(TASK_RUNNING);
2209 SCSI_LOG_ERROR_RECOVERY(1,
2210 shost_printk(KERN_INFO, shost,
2211 "scsi_eh_%d: waking up %d/%d/%d\n",
2212 shost->host_no, shost->host_eh_scheduled,
2213 shost->host_failed,
2214 scsi_host_busy(shost)));
2215
2216 /*
2217 * We have a host that is failing for some reason. Figure out
2218 * what we need to do to get it up and online again (if we can).
2219 * If we fail, we end up taking the thing offline.
2220 */
2221 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2222 SCSI_LOG_ERROR_RECOVERY(1,
2223 shost_printk(KERN_ERR, shost,
2224 "scsi_eh_%d: unable to autoresume\n",
2225 shost->host_no));
2226 continue;
2227 }
2228
2229 if (shost->transportt->eh_strategy_handler)
2230 shost->transportt->eh_strategy_handler(shost);
2231 else
2232 scsi_unjam_host(shost);
2233
2234 /* All scmds have been handled */
2235 shost->host_failed = 0;
2236
2237 /*
2238 * Note - if the above fails completely, the action is to take
2239 * individual devices offline and flush the queue of any
2240 * outstanding requests that may have been pending. When we
2241 * restart, we restart any I/O to any other devices on the bus
2242 * which are still online.
2243 */
2244 scsi_restart_operations(shost);
2245 if (!shost->eh_noresume)
2246 scsi_autopm_put_host(shost);
2247 }
2248 __set_current_state(TASK_RUNNING);
2249
2250 SCSI_LOG_ERROR_RECOVERY(1,
2251 shost_printk(KERN_INFO, shost,
2252 "Error handler scsi_eh_%d exiting\n",
2253 shost->host_no));
2254 shost->ehandler = NULL;
2255 return 0;
2256 }
2257
2258 /*
2259 * Function: scsi_report_bus_reset()
2260 *
2261 * Purpose: Utility function used by low-level drivers to report that
2262 * they have observed a bus reset on the bus being handled.
2263 *
2264 * Arguments: shost - Host in question
2265 * channel - channel on which reset was observed.
2266 *
2267 * Returns: Nothing
2268 *
2269 * Lock status: Host lock must be held.
2270 *
2271 * Notes: This only needs to be called if the reset is one which
2272 * originates from an unknown location. Resets originated
2273 * by the mid-level itself don't need to call this, but there
2274 * should be no harm.
2275 *
2276 * The main purpose of this is to make sure that a CHECK_CONDITION
2277 * is properly treated.
2278 */
scsi_report_bus_reset(struct Scsi_Host * shost,int channel)2279 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2280 {
2281 struct scsi_device *sdev;
2282
2283 __shost_for_each_device(sdev, shost) {
2284 if (channel == sdev_channel(sdev))
2285 __scsi_report_device_reset(sdev, NULL);
2286 }
2287 }
2288 EXPORT_SYMBOL(scsi_report_bus_reset);
2289
2290 /*
2291 * Function: scsi_report_device_reset()
2292 *
2293 * Purpose: Utility function used by low-level drivers to report that
2294 * they have observed a device reset on the device being handled.
2295 *
2296 * Arguments: shost - Host in question
2297 * channel - channel on which reset was observed
2298 * target - target on which reset was observed
2299 *
2300 * Returns: Nothing
2301 *
2302 * Lock status: Host lock must be held
2303 *
2304 * Notes: This only needs to be called if the reset is one which
2305 * originates from an unknown location. Resets originated
2306 * by the mid-level itself don't need to call this, but there
2307 * should be no harm.
2308 *
2309 * The main purpose of this is to make sure that a CHECK_CONDITION
2310 * is properly treated.
2311 */
scsi_report_device_reset(struct Scsi_Host * shost,int channel,int target)2312 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2313 {
2314 struct scsi_device *sdev;
2315
2316 __shost_for_each_device(sdev, shost) {
2317 if (channel == sdev_channel(sdev) &&
2318 target == sdev_id(sdev))
2319 __scsi_report_device_reset(sdev, NULL);
2320 }
2321 }
2322 EXPORT_SYMBOL(scsi_report_device_reset);
2323
2324 static void
scsi_reset_provider_done_command(struct scsi_cmnd * scmd)2325 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2326 {
2327 }
2328
2329 /**
2330 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2331 * @dev: scsi_device to operate on
2332 * @arg: reset type (see sg.h)
2333 */
2334 int
scsi_ioctl_reset(struct scsi_device * dev,int __user * arg)2335 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2336 {
2337 struct scsi_cmnd *scmd;
2338 struct Scsi_Host *shost = dev->host;
2339 struct request *rq;
2340 unsigned long flags;
2341 int error = 0, rtn, val;
2342
2343 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2344 return -EACCES;
2345
2346 error = get_user(val, arg);
2347 if (error)
2348 return error;
2349
2350 if (scsi_autopm_get_host(shost) < 0)
2351 return -EIO;
2352
2353 error = -EIO;
2354 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2355 shost->hostt->cmd_size, GFP_KERNEL);
2356 if (!rq)
2357 goto out_put_autopm_host;
2358 blk_rq_init(NULL, rq);
2359
2360 scmd = (struct scsi_cmnd *)(rq + 1);
2361 scsi_init_command(dev, scmd);
2362 scmd->request = rq;
2363 scmd->cmnd = scsi_req(rq)->cmd;
2364
2365 scmd->scsi_done = scsi_reset_provider_done_command;
2366 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2367
2368 scmd->cmd_len = 0;
2369
2370 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2371
2372 spin_lock_irqsave(shost->host_lock, flags);
2373 shost->tmf_in_progress = 1;
2374 spin_unlock_irqrestore(shost->host_lock, flags);
2375
2376 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2377 case SG_SCSI_RESET_NOTHING:
2378 rtn = SUCCESS;
2379 break;
2380 case SG_SCSI_RESET_DEVICE:
2381 rtn = scsi_try_bus_device_reset(scmd);
2382 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2383 break;
2384 fallthrough;
2385 case SG_SCSI_RESET_TARGET:
2386 rtn = scsi_try_target_reset(scmd);
2387 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2388 break;
2389 fallthrough;
2390 case SG_SCSI_RESET_BUS:
2391 rtn = scsi_try_bus_reset(scmd);
2392 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2393 break;
2394 fallthrough;
2395 case SG_SCSI_RESET_HOST:
2396 rtn = scsi_try_host_reset(scmd);
2397 if (rtn == SUCCESS)
2398 break;
2399 fallthrough;
2400 default:
2401 rtn = FAILED;
2402 break;
2403 }
2404
2405 error = (rtn == SUCCESS) ? 0 : -EIO;
2406
2407 spin_lock_irqsave(shost->host_lock, flags);
2408 shost->tmf_in_progress = 0;
2409 spin_unlock_irqrestore(shost->host_lock, flags);
2410
2411 /*
2412 * be sure to wake up anyone who was sleeping or had their queue
2413 * suspended while we performed the TMF.
2414 */
2415 SCSI_LOG_ERROR_RECOVERY(3,
2416 shost_printk(KERN_INFO, shost,
2417 "waking up host to restart after TMF\n"));
2418
2419 wake_up(&shost->host_wait);
2420 scsi_run_host_queues(shost);
2421
2422 kfree(rq);
2423
2424 out_put_autopm_host:
2425 scsi_autopm_put_host(shost);
2426 return error;
2427 }
2428
scsi_command_normalize_sense(const struct scsi_cmnd * cmd,struct scsi_sense_hdr * sshdr)2429 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2430 struct scsi_sense_hdr *sshdr)
2431 {
2432 return scsi_normalize_sense(cmd->sense_buffer,
2433 SCSI_SENSE_BUFFERSIZE, sshdr);
2434 }
2435 EXPORT_SYMBOL(scsi_command_normalize_sense);
2436
2437 /**
2438 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2439 * @sense_buffer: byte array of sense data
2440 * @sb_len: number of valid bytes in sense_buffer
2441 * @info_out: pointer to 64 integer where 8 or 4 byte information
2442 * field will be placed if found.
2443 *
2444 * Return value:
2445 * true if information field found, false if not found.
2446 */
scsi_get_sense_info_fld(const u8 * sense_buffer,int sb_len,u64 * info_out)2447 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2448 u64 *info_out)
2449 {
2450 const u8 * ucp;
2451
2452 if (sb_len < 7)
2453 return false;
2454 switch (sense_buffer[0] & 0x7f) {
2455 case 0x70:
2456 case 0x71:
2457 if (sense_buffer[0] & 0x80) {
2458 *info_out = get_unaligned_be32(&sense_buffer[3]);
2459 return true;
2460 }
2461 return false;
2462 case 0x72:
2463 case 0x73:
2464 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2465 0 /* info desc */);
2466 if (ucp && (0xa == ucp[1])) {
2467 *info_out = get_unaligned_be64(&ucp[4]);
2468 return true;
2469 }
2470 return false;
2471 default:
2472 return false;
2473 }
2474 }
2475 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2476