GNU Linux-libre 5.10.219-gnu1
[releases.git] / drivers / scsi / scsi_error.c
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 enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *,
64                                                    struct scsi_cmnd *);
65
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  */
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
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
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
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         enum scsi_disposition 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
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  */
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
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 int busy = scsi_host_busy(shost);
245         unsigned long flags;
246
247         spin_lock_irqsave(shost->host_lock, flags);
248         shost->host_failed++;
249         scsi_eh_wakeup(shost, busy);
250         spin_unlock_irqrestore(shost->host_lock, flags);
251 }
252
253 /**
254  * scsi_eh_scmd_add - add scsi cmd to error handling.
255  * @scmd:       scmd to run eh on.
256  */
257 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
258 {
259         struct Scsi_Host *shost = scmd->device->host;
260         unsigned long flags;
261         int ret;
262
263         WARN_ON_ONCE(!shost->ehandler);
264
265         spin_lock_irqsave(shost->host_lock, flags);
266         if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
267                 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
268                 WARN_ON_ONCE(ret);
269         }
270         if (shost->eh_deadline != -1 && !shost->last_reset)
271                 shost->last_reset = jiffies;
272
273         scsi_eh_reset(scmd);
274         list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
275         spin_unlock_irqrestore(shost->host_lock, flags);
276         /*
277          * Ensure that all tasks observe the host state change before the
278          * host_failed change.
279          */
280         call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
281 }
282
283 /**
284  * scsi_times_out - Timeout function for normal scsi commands.
285  * @req:        request that is timing out.
286  *
287  * Notes:
288  *     We do not need to lock this.  There is the potential for a race
289  *     only in that the normal completion handling might run, but if the
290  *     normal completion function determines that the timer has already
291  *     fired, then it mustn't do anything.
292  */
293 enum blk_eh_timer_return scsi_times_out(struct request *req)
294 {
295         struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
296         enum blk_eh_timer_return rtn = BLK_EH_DONE;
297         struct Scsi_Host *host = scmd->device->host;
298
299         trace_scsi_dispatch_cmd_timeout(scmd);
300         scsi_log_completion(scmd, TIMEOUT_ERROR);
301
302         if (host->eh_deadline != -1 && !host->last_reset)
303                 host->last_reset = jiffies;
304
305         if (host->hostt->eh_timed_out)
306                 rtn = host->hostt->eh_timed_out(scmd);
307
308         if (rtn == BLK_EH_DONE) {
309                 /*
310                  * If scsi_done() has already set SCMD_STATE_COMPLETE, do not
311                  * modify *scmd.
312                  */
313                 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
314                         return BLK_EH_DONE;
315                 if (scsi_abort_command(scmd) != SUCCESS) {
316                         set_host_byte(scmd, DID_TIME_OUT);
317                         scsi_eh_scmd_add(scmd);
318                 }
319         }
320
321         return rtn;
322 }
323
324 /**
325  * scsi_block_when_processing_errors - Prevent cmds from being queued.
326  * @sdev:       Device on which we are performing recovery.
327  *
328  * Description:
329  *     We block until the host is out of error recovery, and then check to
330  *     see whether the host or the device is offline.
331  *
332  * Return value:
333  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
334  */
335 int scsi_block_when_processing_errors(struct scsi_device *sdev)
336 {
337         int online;
338
339         wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
340
341         online = scsi_device_online(sdev);
342
343         return online;
344 }
345 EXPORT_SYMBOL(scsi_block_when_processing_errors);
346
347 #ifdef CONFIG_SCSI_LOGGING
348 /**
349  * scsi_eh_prt_fail_stats - Log info on failures.
350  * @shost:      scsi host being recovered.
351  * @work_q:     Queue of scsi cmds to process.
352  */
353 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
354                                           struct list_head *work_q)
355 {
356         struct scsi_cmnd *scmd;
357         struct scsi_device *sdev;
358         int total_failures = 0;
359         int cmd_failed = 0;
360         int cmd_cancel = 0;
361         int devices_failed = 0;
362
363         shost_for_each_device(sdev, shost) {
364                 list_for_each_entry(scmd, work_q, eh_entry) {
365                         if (scmd->device == sdev) {
366                                 ++total_failures;
367                                 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
368                                         ++cmd_cancel;
369                                 else
370                                         ++cmd_failed;
371                         }
372                 }
373
374                 if (cmd_cancel || cmd_failed) {
375                         SCSI_LOG_ERROR_RECOVERY(3,
376                                 shost_printk(KERN_INFO, shost,
377                                             "%s: cmds failed: %d, cancel: %d\n",
378                                             __func__, cmd_failed,
379                                             cmd_cancel));
380                         cmd_cancel = 0;
381                         cmd_failed = 0;
382                         ++devices_failed;
383                 }
384         }
385
386         SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
387                                    "Total of %d commands on %d"
388                                    " devices require eh work\n",
389                                    total_failures, devices_failed));
390 }
391 #endif
392
393  /**
394  * scsi_report_lun_change - Set flag on all *other* devices on the same target
395  *                          to indicate that a UNIT ATTENTION is expected.
396  * @sdev:       Device reporting the UNIT ATTENTION
397  */
398 static void scsi_report_lun_change(struct scsi_device *sdev)
399 {
400         sdev->sdev_target->expecting_lun_change = 1;
401 }
402
403 /**
404  * scsi_report_sense - Examine scsi sense information and log messages for
405  *                     certain conditions, also issue uevents for some of them.
406  * @sdev:       Device reporting the sense code
407  * @sshdr:      sshdr to be examined
408  */
409 static void scsi_report_sense(struct scsi_device *sdev,
410                               struct scsi_sense_hdr *sshdr)
411 {
412         enum scsi_device_event evt_type = SDEV_EVT_MAXBITS;     /* i.e. none */
413
414         if (sshdr->sense_key == UNIT_ATTENTION) {
415                 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
416                         evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
417                         sdev_printk(KERN_WARNING, sdev,
418                                     "Inquiry data has changed");
419                 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
420                         evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
421                         scsi_report_lun_change(sdev);
422                         sdev_printk(KERN_WARNING, sdev,
423                                     "Warning! Received an indication that the "
424                                     "LUN assignments on this target have "
425                                     "changed. The Linux SCSI layer does not "
426                                     "automatically remap LUN assignments.\n");
427                 } else if (sshdr->asc == 0x3f)
428                         sdev_printk(KERN_WARNING, sdev,
429                                     "Warning! Received an indication that the "
430                                     "operating parameters on this target have "
431                                     "changed. The Linux SCSI layer does not "
432                                     "automatically adjust these parameters.\n");
433
434                 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
435                         evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
436                         sdev_printk(KERN_WARNING, sdev,
437                                     "Warning! Received an indication that the "
438                                     "LUN reached a thin provisioning soft "
439                                     "threshold.\n");
440                 }
441
442                 if (sshdr->asc == 0x29) {
443                         evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
444                         sdev_printk(KERN_WARNING, sdev,
445                                     "Power-on or device reset occurred\n");
446                 }
447
448                 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
449                         evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
450                         sdev_printk(KERN_WARNING, sdev,
451                                     "Mode parameters changed");
452                 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
453                         evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
454                         sdev_printk(KERN_WARNING, sdev,
455                                     "Asymmetric access state changed");
456                 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
457                         evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
458                         sdev_printk(KERN_WARNING, sdev,
459                                     "Capacity data has changed");
460                 } else if (sshdr->asc == 0x2a)
461                         sdev_printk(KERN_WARNING, sdev,
462                                     "Parameters changed");
463         }
464
465         if (evt_type != SDEV_EVT_MAXBITS) {
466                 set_bit(evt_type, sdev->pending_events);
467                 schedule_work(&sdev->event_work);
468         }
469 }
470
471 /**
472  * scsi_check_sense - Examine scsi cmd sense
473  * @scmd:       Cmd to have sense checked.
474  *
475  * Return value:
476  *      SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
477  *
478  * Notes:
479  *      When a deferred error is detected the current command has
480  *      not been executed and needs retrying.
481  */
482 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
483 {
484         struct scsi_device *sdev = scmd->device;
485         struct scsi_sense_hdr sshdr;
486
487         if (! scsi_command_normalize_sense(scmd, &sshdr))
488                 return FAILED;  /* no valid sense data */
489
490         scsi_report_sense(sdev, &sshdr);
491
492         if (scsi_sense_is_deferred(&sshdr))
493                 return NEEDS_RETRY;
494
495         if (sdev->handler && sdev->handler->check_sense) {
496                 enum scsi_disposition rc;
497
498                 rc = sdev->handler->check_sense(sdev, &sshdr);
499                 if (rc != SCSI_RETURN_NOT_HANDLED)
500                         return rc;
501                 /* handler does not care. Drop down to default handling */
502         }
503
504         if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
505                 /*
506                  * nasty: for mid-layer issued TURs, we need to return the
507                  * actual sense data without any recovery attempt.  For eh
508                  * issued ones, we need to try to recover and interpret
509                  */
510                 return SUCCESS;
511
512         /*
513          * Previous logic looked for FILEMARK, EOM or ILI which are
514          * mainly associated with tapes and returned SUCCESS.
515          */
516         if (sshdr.response_code == 0x70) {
517                 /* fixed format */
518                 if (scmd->sense_buffer[2] & 0xe0)
519                         return SUCCESS;
520         } else {
521                 /*
522                  * descriptor format: look for "stream commands sense data
523                  * descriptor" (see SSC-3). Assume single sense data
524                  * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
525                  */
526                 if ((sshdr.additional_length > 3) &&
527                     (scmd->sense_buffer[8] == 0x4) &&
528                     (scmd->sense_buffer[11] & 0xe0))
529                         return SUCCESS;
530         }
531
532         switch (sshdr.sense_key) {
533         case NO_SENSE:
534                 return SUCCESS;
535         case RECOVERED_ERROR:
536                 return /* soft_error */ SUCCESS;
537
538         case ABORTED_COMMAND:
539                 if (sshdr.asc == 0x10) /* DIF */
540                         return SUCCESS;
541
542                 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
543                         return ADD_TO_MLQUEUE;
544                 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
545                     sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
546                         return ADD_TO_MLQUEUE;
547
548                 return NEEDS_RETRY;
549         case NOT_READY:
550         case UNIT_ATTENTION:
551                 /*
552                  * if we are expecting a cc/ua because of a bus reset that we
553                  * performed, treat this just as a retry.  otherwise this is
554                  * information that we should pass up to the upper-level driver
555                  * so that we can deal with it there.
556                  */
557                 if (scmd->device->expecting_cc_ua) {
558                         /*
559                          * Because some device does not queue unit
560                          * attentions correctly, we carefully check
561                          * additional sense code and qualifier so as
562                          * not to squash media change unit attention.
563                          */
564                         if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
565                                 scmd->device->expecting_cc_ua = 0;
566                                 return NEEDS_RETRY;
567                         }
568                 }
569                 /*
570                  * we might also expect a cc/ua if another LUN on the target
571                  * reported a UA with an ASC/ASCQ of 3F 0E -
572                  * REPORTED LUNS DATA HAS CHANGED.
573                  */
574                 if (scmd->device->sdev_target->expecting_lun_change &&
575                     sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
576                         return NEEDS_RETRY;
577                 /*
578                  * if the device is in the process of becoming ready, we
579                  * should retry.
580                  */
581                 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
582                         return NEEDS_RETRY;
583                 /*
584                  * if the device is not started, we need to wake
585                  * the error handler to start the motor
586                  */
587                 if (scmd->device->allow_restart &&
588                     (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
589                         return FAILED;
590                 /*
591                  * Pass the UA upwards for a determination in the completion
592                  * functions.
593                  */
594                 return SUCCESS;
595
596                 /* these are not supported */
597         case DATA_PROTECT:
598                 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
599                         /* Thin provisioning hard threshold reached */
600                         set_host_byte(scmd, DID_ALLOC_FAILURE);
601                         return SUCCESS;
602                 }
603                 fallthrough;
604         case COPY_ABORTED:
605         case VOLUME_OVERFLOW:
606         case MISCOMPARE:
607         case BLANK_CHECK:
608                 set_host_byte(scmd, DID_TARGET_FAILURE);
609                 return SUCCESS;
610
611         case MEDIUM_ERROR:
612                 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
613                     sshdr.asc == 0x13 || /* AMNF DATA FIELD */
614                     sshdr.asc == 0x14) { /* RECORD NOT FOUND */
615                         set_host_byte(scmd, DID_MEDIUM_ERROR);
616                         return SUCCESS;
617                 }
618                 return NEEDS_RETRY;
619
620         case HARDWARE_ERROR:
621                 if (scmd->device->retry_hwerror)
622                         return ADD_TO_MLQUEUE;
623                 else
624                         set_host_byte(scmd, DID_TARGET_FAILURE);
625                 fallthrough;
626
627         case ILLEGAL_REQUEST:
628                 if (sshdr.asc == 0x20 || /* Invalid command operation code */
629                     sshdr.asc == 0x21 || /* Logical block address out of range */
630                     sshdr.asc == 0x22 || /* Invalid function */
631                     sshdr.asc == 0x24 || /* Invalid field in cdb */
632                     sshdr.asc == 0x26 || /* Parameter value invalid */
633                     sshdr.asc == 0x27) { /* Write protected */
634                         set_host_byte(scmd, DID_TARGET_FAILURE);
635                 }
636                 return SUCCESS;
637
638         default:
639                 return SUCCESS;
640         }
641 }
642 EXPORT_SYMBOL_GPL(scsi_check_sense);
643
644 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
645 {
646         struct scsi_host_template *sht = sdev->host->hostt;
647         struct scsi_device *tmp_sdev;
648
649         if (!sht->track_queue_depth ||
650             sdev->queue_depth >= sdev->max_queue_depth)
651                 return;
652
653         if (time_before(jiffies,
654             sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
655                 return;
656
657         if (time_before(jiffies,
658             sdev->last_queue_full_time + sdev->queue_ramp_up_period))
659                 return;
660
661         /*
662          * Walk all devices of a target and do
663          * ramp up on them.
664          */
665         shost_for_each_device(tmp_sdev, sdev->host) {
666                 if (tmp_sdev->channel != sdev->channel ||
667                     tmp_sdev->id != sdev->id ||
668                     tmp_sdev->queue_depth == sdev->max_queue_depth)
669                         continue;
670
671                 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
672                 sdev->last_queue_ramp_up = jiffies;
673         }
674 }
675
676 static void scsi_handle_queue_full(struct scsi_device *sdev)
677 {
678         struct scsi_host_template *sht = sdev->host->hostt;
679         struct scsi_device *tmp_sdev;
680
681         if (!sht->track_queue_depth)
682                 return;
683
684         shost_for_each_device(tmp_sdev, sdev->host) {
685                 if (tmp_sdev->channel != sdev->channel ||
686                     tmp_sdev->id != sdev->id)
687                         continue;
688                 /*
689                  * We do not know the number of commands that were at
690                  * the device when we got the queue full so we start
691                  * from the highest possible value and work our way down.
692                  */
693                 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
694         }
695 }
696
697 /**
698  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
699  * @scmd:       SCSI cmd to examine.
700  *
701  * Notes:
702  *    This is *only* called when we are examining the status of commands
703  *    queued during error recovery.  the main difference here is that we
704  *    don't allow for the possibility of retries here, and we are a lot
705  *    more restrictive about what we consider acceptable.
706  */
707 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
708 {
709         /*
710          * first check the host byte, to see if there is anything in there
711          * that would indicate what we need to do.
712          */
713         if (host_byte(scmd->result) == DID_RESET) {
714                 /*
715                  * rats.  we are already in the error handler, so we now
716                  * get to try and figure out what to do next.  if the sense
717                  * is valid, we have a pretty good idea of what to do.
718                  * if not, we mark it as FAILED.
719                  */
720                 return scsi_check_sense(scmd);
721         }
722         if (host_byte(scmd->result) != DID_OK)
723                 return FAILED;
724
725         /*
726          * next, check the message byte.
727          */
728         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
729                 return FAILED;
730
731         /*
732          * now, check the status byte to see if this indicates
733          * anything special.
734          */
735         switch (status_byte(scmd->result)) {
736         case GOOD:
737                 scsi_handle_queue_ramp_up(scmd->device);
738                 fallthrough;
739         case COMMAND_TERMINATED:
740                 return SUCCESS;
741         case CHECK_CONDITION:
742                 return scsi_check_sense(scmd);
743         case CONDITION_GOOD:
744         case INTERMEDIATE_GOOD:
745         case INTERMEDIATE_C_GOOD:
746                 /*
747                  * who knows?  FIXME(eric)
748                  */
749                 return SUCCESS;
750         case RESERVATION_CONFLICT:
751                 if (scmd->cmnd[0] == TEST_UNIT_READY)
752                         /* it is a success, we probed the device and
753                          * found it */
754                         return SUCCESS;
755                 /* otherwise, we failed to send the command */
756                 return FAILED;
757         case QUEUE_FULL:
758                 scsi_handle_queue_full(scmd->device);
759                 fallthrough;
760         case BUSY:
761                 return NEEDS_RETRY;
762         default:
763                 return FAILED;
764         }
765         return FAILED;
766 }
767
768 /**
769  * scsi_eh_done - Completion function for error handling.
770  * @scmd:       Cmd that is done.
771  */
772 static void scsi_eh_done(struct scsi_cmnd *scmd)
773 {
774         struct completion *eh_action;
775
776         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
777                         "%s result: %x\n", __func__, scmd->result));
778
779         eh_action = scmd->device->host->eh_action;
780         if (eh_action)
781                 complete(eh_action);
782 }
783
784 /**
785  * scsi_try_host_reset - ask host adapter to reset itself
786  * @scmd:       SCSI cmd to send host reset.
787  */
788 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
789 {
790         unsigned long flags;
791         enum scsi_disposition rtn;
792         struct Scsi_Host *host = scmd->device->host;
793         struct scsi_host_template *hostt = host->hostt;
794
795         SCSI_LOG_ERROR_RECOVERY(3,
796                 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
797
798         if (!hostt->eh_host_reset_handler)
799                 return FAILED;
800
801         rtn = hostt->eh_host_reset_handler(scmd);
802
803         if (rtn == SUCCESS) {
804                 if (!hostt->skip_settle_delay)
805                         ssleep(HOST_RESET_SETTLE_TIME);
806                 spin_lock_irqsave(host->host_lock, flags);
807                 scsi_report_bus_reset(host, scmd_channel(scmd));
808                 spin_unlock_irqrestore(host->host_lock, flags);
809         }
810
811         return rtn;
812 }
813
814 /**
815  * scsi_try_bus_reset - ask host to perform a bus reset
816  * @scmd:       SCSI cmd to send bus reset.
817  */
818 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
819 {
820         unsigned long flags;
821         enum scsi_disposition rtn;
822         struct Scsi_Host *host = scmd->device->host;
823         struct scsi_host_template *hostt = host->hostt;
824
825         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
826                 "%s: Snd Bus RST\n", __func__));
827
828         if (!hostt->eh_bus_reset_handler)
829                 return FAILED;
830
831         rtn = hostt->eh_bus_reset_handler(scmd);
832
833         if (rtn == SUCCESS) {
834                 if (!hostt->skip_settle_delay)
835                         ssleep(BUS_RESET_SETTLE_TIME);
836                 spin_lock_irqsave(host->host_lock, flags);
837                 scsi_report_bus_reset(host, scmd_channel(scmd));
838                 spin_unlock_irqrestore(host->host_lock, flags);
839         }
840
841         return rtn;
842 }
843
844 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
845 {
846         sdev->was_reset = 1;
847         sdev->expecting_cc_ua = 1;
848 }
849
850 /**
851  * scsi_try_target_reset - Ask host to perform a target reset
852  * @scmd:       SCSI cmd used to send a target reset
853  *
854  * Notes:
855  *    There is no timeout for this operation.  if this operation is
856  *    unreliable for a given host, then the host itself needs to put a
857  *    timer on it, and set the host back to a consistent state prior to
858  *    returning.
859  */
860 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
861 {
862         unsigned long flags;
863         enum scsi_disposition rtn;
864         struct Scsi_Host *host = scmd->device->host;
865         struct scsi_host_template *hostt = host->hostt;
866
867         if (!hostt->eh_target_reset_handler)
868                 return FAILED;
869
870         rtn = hostt->eh_target_reset_handler(scmd);
871         if (rtn == SUCCESS) {
872                 spin_lock_irqsave(host->host_lock, flags);
873                 __starget_for_each_device(scsi_target(scmd->device), NULL,
874                                           __scsi_report_device_reset);
875                 spin_unlock_irqrestore(host->host_lock, flags);
876         }
877
878         return rtn;
879 }
880
881 /**
882  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
883  * @scmd:       SCSI cmd used to send BDR
884  *
885  * Notes:
886  *    There is no timeout for this operation.  if this operation is
887  *    unreliable for a given host, then the host itself needs to put a
888  *    timer on it, and set the host back to a consistent state prior to
889  *    returning.
890  */
891 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
892 {
893         enum scsi_disposition rtn;
894         struct scsi_host_template *hostt = scmd->device->host->hostt;
895
896         if (!hostt->eh_device_reset_handler)
897                 return FAILED;
898
899         rtn = hostt->eh_device_reset_handler(scmd);
900         if (rtn == SUCCESS)
901                 __scsi_report_device_reset(scmd->device, NULL);
902         return rtn;
903 }
904
905 /**
906  * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
907  * @hostt:      SCSI driver host template
908  * @scmd:       SCSI cmd used to send a target reset
909  *
910  * Return value:
911  *      SUCCESS, FAILED, or FAST_IO_FAIL
912  *
913  * Notes:
914  *    SUCCESS does not necessarily indicate that the command
915  *    has been aborted; it only indicates that the LLDDs
916  *    has cleared all references to that command.
917  *    LLDDs should return FAILED only if an abort was required
918  *    but could not be executed. LLDDs should return FAST_IO_FAIL
919  *    if the device is temporarily unavailable (eg due to a
920  *    link down on FibreChannel)
921  */
922 static enum scsi_disposition
923 scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
924 {
925         if (!hostt->eh_abort_handler)
926                 return FAILED;
927
928         return hostt->eh_abort_handler(scmd);
929 }
930
931 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
932 {
933         if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
934                 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
935                         if (scsi_try_target_reset(scmd) != SUCCESS)
936                                 if (scsi_try_bus_reset(scmd) != SUCCESS)
937                                         scsi_try_host_reset(scmd);
938 }
939
940 /**
941  * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recovery
942  * @scmd:       SCSI command structure to hijack
943  * @ses:        structure to save restore information
944  * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
945  * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
946  * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
947  *
948  * This function is used to save a scsi command information before re-execution
949  * as part of the error recovery process.  If @sense_bytes is 0 the command
950  * sent must be one that does not transfer any data.  If @sense_bytes != 0
951  * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
952  * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
953  */
954 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
955                         unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
956 {
957         struct scsi_device *sdev = scmd->device;
958
959         /*
960          * We need saved copies of a number of fields - this is because
961          * error handling may need to overwrite these with different values
962          * to run different commands, and once error handling is complete,
963          * we will need to restore these values prior to running the actual
964          * command.
965          */
966         ses->cmd_len = scmd->cmd_len;
967         ses->cmnd = scmd->cmnd;
968         ses->data_direction = scmd->sc_data_direction;
969         ses->sdb = scmd->sdb;
970         ses->result = scmd->result;
971         ses->resid_len = scmd->req.resid_len;
972         ses->underflow = scmd->underflow;
973         ses->prot_op = scmd->prot_op;
974         ses->eh_eflags = scmd->eh_eflags;
975
976         scmd->prot_op = SCSI_PROT_NORMAL;
977         scmd->eh_eflags = 0;
978         scmd->cmnd = ses->eh_cmnd;
979         memset(scmd->cmnd, 0, BLK_MAX_CDB);
980         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
981         scmd->result = 0;
982         scmd->req.resid_len = 0;
983
984         if (sense_bytes) {
985                 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
986                                          sense_bytes);
987                 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
988                             scmd->sdb.length);
989                 scmd->sdb.table.sgl = &ses->sense_sgl;
990                 scmd->sc_data_direction = DMA_FROM_DEVICE;
991                 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
992                 scmd->cmnd[0] = REQUEST_SENSE;
993                 scmd->cmnd[4] = scmd->sdb.length;
994                 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
995         } else {
996                 scmd->sc_data_direction = DMA_NONE;
997                 if (cmnd) {
998                         BUG_ON(cmnd_size > BLK_MAX_CDB);
999                         memcpy(scmd->cmnd, cmnd, cmnd_size);
1000                         scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1001                 }
1002         }
1003
1004         scmd->underflow = 0;
1005
1006         if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1007                 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1008                         (sdev->lun << 5 & 0xe0);
1009
1010         /*
1011          * Zero the sense buffer.  The scsi spec mandates that any
1012          * untransferred sense data should be interpreted as being zero.
1013          */
1014         memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1015 }
1016 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1017
1018 /**
1019  * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recovery
1020  * @scmd:       SCSI command structure to restore
1021  * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
1022  *
1023  * Undo any damage done by above scsi_eh_prep_cmnd().
1024  */
1025 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1026 {
1027         /*
1028          * Restore original data
1029          */
1030         scmd->cmd_len = ses->cmd_len;
1031         scmd->cmnd = ses->cmnd;
1032         scmd->sc_data_direction = ses->data_direction;
1033         scmd->sdb = ses->sdb;
1034         scmd->result = ses->result;
1035         scmd->req.resid_len = ses->resid_len;
1036         scmd->underflow = ses->underflow;
1037         scmd->prot_op = ses->prot_op;
1038         scmd->eh_eflags = ses->eh_eflags;
1039 }
1040 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1041
1042 /**
1043  * scsi_send_eh_cmnd  - submit a scsi command as part of error recovery
1044  * @scmd:       SCSI command structure to hijack
1045  * @cmnd:       CDB to send
1046  * @cmnd_size:  size in bytes of @cmnd
1047  * @timeout:    timeout for this request
1048  * @sense_bytes: size of sense data to copy or 0
1049  *
1050  * This function is used to send a scsi command down to a target device
1051  * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1052  *
1053  * Return value:
1054  *    SUCCESS or FAILED or NEEDS_RETRY
1055  */
1056 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1057         unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
1058 {
1059         struct scsi_device *sdev = scmd->device;
1060         struct Scsi_Host *shost = sdev->host;
1061         DECLARE_COMPLETION_ONSTACK(done);
1062         unsigned long timeleft = timeout, delay;
1063         struct scsi_eh_save ses;
1064         const unsigned long stall_for = msecs_to_jiffies(100);
1065         int rtn;
1066
1067 retry:
1068         scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1069         shost->eh_action = &done;
1070
1071         scsi_log_send(scmd);
1072         scmd->scsi_done = scsi_eh_done;
1073         scmd->flags |= SCMD_LAST;
1074
1075         /*
1076          * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1077          * change the SCSI device state after we have examined it and before
1078          * .queuecommand() is called.
1079          */
1080         mutex_lock(&sdev->state_mutex);
1081         while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1082                 mutex_unlock(&sdev->state_mutex);
1083                 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1084                         "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1085                         SDEV_BLOCK));
1086                 delay = min(timeleft, stall_for);
1087                 timeleft -= delay;
1088                 msleep(jiffies_to_msecs(delay));
1089                 mutex_lock(&sdev->state_mutex);
1090         }
1091         if (sdev->sdev_state != SDEV_BLOCK)
1092                 rtn = shost->hostt->queuecommand(shost, scmd);
1093         else
1094                 rtn = SCSI_MLQUEUE_DEVICE_BUSY;
1095         mutex_unlock(&sdev->state_mutex);
1096
1097         if (rtn) {
1098                 if (timeleft > stall_for) {
1099                         scsi_eh_restore_cmnd(scmd, &ses);
1100                         timeleft -= stall_for;
1101                         msleep(jiffies_to_msecs(stall_for));
1102                         goto retry;
1103                 }
1104                 /* signal not to enter either branch of the if () below */
1105                 timeleft = 0;
1106                 rtn = FAILED;
1107         } else {
1108                 timeleft = wait_for_completion_timeout(&done, timeout);
1109                 rtn = SUCCESS;
1110         }
1111
1112         shost->eh_action = NULL;
1113
1114         scsi_log_completion(scmd, rtn);
1115
1116         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1117                         "%s timeleft: %ld\n",
1118                         __func__, timeleft));
1119
1120         /*
1121          * If there is time left scsi_eh_done got called, and we will examine
1122          * the actual status codes to see whether the command actually did
1123          * complete normally, else if we have a zero return and no time left,
1124          * the command must still be pending, so abort it and return FAILED.
1125          * If we never actually managed to issue the command, because
1126          * ->queuecommand() kept returning non zero, use the rtn = FAILED
1127          * value above (so don't execute either branch of the if)
1128          */
1129         if (timeleft) {
1130                 rtn = scsi_eh_completed_normally(scmd);
1131                 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1132                         "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1133
1134                 switch (rtn) {
1135                 case SUCCESS:
1136                 case NEEDS_RETRY:
1137                 case FAILED:
1138                         break;
1139                 case ADD_TO_MLQUEUE:
1140                         rtn = NEEDS_RETRY;
1141                         break;
1142                 default:
1143                         rtn = FAILED;
1144                         break;
1145                 }
1146         } else if (rtn != FAILED) {
1147                 scsi_abort_eh_cmnd(scmd);
1148                 rtn = FAILED;
1149         }
1150
1151         scsi_eh_restore_cmnd(scmd, &ses);
1152
1153         return rtn;
1154 }
1155
1156 /**
1157  * scsi_request_sense - Request sense data from a particular target.
1158  * @scmd:       SCSI cmd for request sense.
1159  *
1160  * Notes:
1161  *    Some hosts automatically obtain this information, others require
1162  *    that we obtain it on our own. This function will *not* return until
1163  *    the command either times out, or it completes.
1164  */
1165 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
1166 {
1167         return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1168 }
1169
1170 static enum scsi_disposition
1171 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
1172 {
1173         if (!blk_rq_is_passthrough(scmd->request)) {
1174                 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1175                 if (sdrv->eh_action)
1176                         rtn = sdrv->eh_action(scmd, rtn);
1177         }
1178         return rtn;
1179 }
1180
1181 /**
1182  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1183  * @scmd:       Original SCSI cmd that eh has finished.
1184  * @done_q:     Queue for processed commands.
1185  *
1186  * Notes:
1187  *    We don't want to use the normal command completion while we are are
1188  *    still handling errors - it may cause other commands to be queued,
1189  *    and that would disturb what we are doing.  Thus we really want to
1190  *    keep a list of pending commands for final completion, and once we
1191  *    are ready to leave error handling we handle completion for real.
1192  */
1193 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1194 {
1195         list_move_tail(&scmd->eh_entry, done_q);
1196 }
1197 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1198
1199 /**
1200  * scsi_eh_get_sense - Get device sense data.
1201  * @work_q:     Queue of commands to process.
1202  * @done_q:     Queue of processed commands.
1203  *
1204  * Description:
1205  *    See if we need to request sense information.  if so, then get it
1206  *    now, so we have a better idea of what to do.
1207  *
1208  * Notes:
1209  *    This has the unfortunate side effect that if a shost adapter does
1210  *    not automatically request sense information, we end up shutting
1211  *    it down before we request it.
1212  *
1213  *    All drivers should request sense information internally these days,
1214  *    so for now all I have to say is tough noogies if you end up in here.
1215  *
1216  *    XXX: Long term this code should go away, but that needs an audit of
1217  *         all LLDDs first.
1218  */
1219 int scsi_eh_get_sense(struct list_head *work_q,
1220                       struct list_head *done_q)
1221 {
1222         struct scsi_cmnd *scmd, *next;
1223         struct Scsi_Host *shost;
1224         enum scsi_disposition rtn;
1225
1226         /*
1227          * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1228          * should not get sense.
1229          */
1230         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1231                 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1232                     SCSI_SENSE_VALID(scmd))
1233                         continue;
1234
1235                 shost = scmd->device->host;
1236                 if (scsi_host_eh_past_deadline(shost)) {
1237                         SCSI_LOG_ERROR_RECOVERY(3,
1238                                 scmd_printk(KERN_INFO, scmd,
1239                                             "%s: skip request sense, past eh deadline\n",
1240                                              current->comm));
1241                         break;
1242                 }
1243                 if (status_byte(scmd->result) != CHECK_CONDITION)
1244                         /*
1245                          * don't request sense if there's no check condition
1246                          * status because the error we're processing isn't one
1247                          * that has a sense code (and some devices get
1248                          * confused by sense requests out of the blue)
1249                          */
1250                         continue;
1251
1252                 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1253                                                   "%s: requesting sense\n",
1254                                                   current->comm));
1255                 rtn = scsi_request_sense(scmd);
1256                 if (rtn != SUCCESS)
1257                         continue;
1258
1259                 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1260                         "sense requested, result %x\n", scmd->result));
1261                 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1262
1263                 rtn = scsi_decide_disposition(scmd);
1264
1265                 /*
1266                  * if the result was normal, then just pass it along to the
1267                  * upper level.
1268                  */
1269                 if (rtn == SUCCESS)
1270                         /*
1271                          * We don't want this command reissued, just finished
1272                          * with the sense data, so set retries to the max
1273                          * allowed to ensure it won't get reissued. If the user
1274                          * has requested infinite retries, we also want to
1275                          * finish this command, so force completion by setting
1276                          * retries and allowed to the same value.
1277                          */
1278                         if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1279                                 scmd->retries = scmd->allowed = 1;
1280                         else
1281                                 scmd->retries = scmd->allowed;
1282                 else if (rtn != NEEDS_RETRY)
1283                         continue;
1284
1285                 scsi_eh_finish_cmd(scmd, done_q);
1286         }
1287
1288         return list_empty(work_q);
1289 }
1290 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1291
1292 /**
1293  * scsi_eh_tur - Send TUR to device.
1294  * @scmd:       &scsi_cmnd to send TUR
1295  *
1296  * Return value:
1297  *    0 - Device is ready. 1 - Device NOT ready.
1298  */
1299 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1300 {
1301         static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1302         int retry_cnt = 1;
1303         enum scsi_disposition rtn;
1304
1305 retry_tur:
1306         rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1307                                 scmd->device->eh_timeout, 0);
1308
1309         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1310                 "%s return: %x\n", __func__, rtn));
1311
1312         switch (rtn) {
1313         case NEEDS_RETRY:
1314                 if (retry_cnt--)
1315                         goto retry_tur;
1316                 fallthrough;
1317         case SUCCESS:
1318                 return 0;
1319         default:
1320                 return 1;
1321         }
1322 }
1323
1324 /**
1325  * scsi_eh_test_devices - check if devices are responding from error recovery.
1326  * @cmd_list:   scsi commands in error recovery.
1327  * @work_q:     queue for commands which still need more error recovery
1328  * @done_q:     queue for commands which are finished
1329  * @try_stu:    boolean on if a STU command should be tried in addition to TUR.
1330  *
1331  * Decription:
1332  *    Tests if devices are in a working state.  Commands to devices now in
1333  *    a working state are sent to the done_q while commands to devices which
1334  *    are still failing to respond are returned to the work_q for more
1335  *    processing.
1336  **/
1337 static int scsi_eh_test_devices(struct list_head *cmd_list,
1338                                 struct list_head *work_q,
1339                                 struct list_head *done_q, int try_stu)
1340 {
1341         struct scsi_cmnd *scmd, *next;
1342         struct scsi_device *sdev;
1343         int finish_cmds;
1344
1345         while (!list_empty(cmd_list)) {
1346                 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1347                 sdev = scmd->device;
1348
1349                 if (!try_stu) {
1350                         if (scsi_host_eh_past_deadline(sdev->host)) {
1351                                 /* Push items back onto work_q */
1352                                 list_splice_init(cmd_list, work_q);
1353                                 SCSI_LOG_ERROR_RECOVERY(3,
1354                                         sdev_printk(KERN_INFO, sdev,
1355                                                     "%s: skip test device, past eh deadline",
1356                                                     current->comm));
1357                                 break;
1358                         }
1359                 }
1360
1361                 finish_cmds = !scsi_device_online(scmd->device) ||
1362                         (try_stu && !scsi_eh_try_stu(scmd) &&
1363                          !scsi_eh_tur(scmd)) ||
1364                         !scsi_eh_tur(scmd);
1365
1366                 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1367                         if (scmd->device == sdev) {
1368                                 if (finish_cmds &&
1369                                     (try_stu ||
1370                                      scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1371                                         scsi_eh_finish_cmd(scmd, done_q);
1372                                 else
1373                                         list_move_tail(&scmd->eh_entry, work_q);
1374                         }
1375         }
1376         return list_empty(work_q);
1377 }
1378
1379 /**
1380  * scsi_eh_try_stu - Send START_UNIT to device.
1381  * @scmd:       &scsi_cmnd to send START_UNIT
1382  *
1383  * Return value:
1384  *    0 - Device is ready. 1 - Device NOT ready.
1385  */
1386 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1387 {
1388         static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1389
1390         if (scmd->device->allow_restart) {
1391                 int i;
1392                 enum scsi_disposition rtn = NEEDS_RETRY;
1393
1394                 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1395                         rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1396
1397                 if (rtn == SUCCESS)
1398                         return 0;
1399         }
1400
1401         return 1;
1402 }
1403
1404  /**
1405  * scsi_eh_stu - send START_UNIT if needed
1406  * @shost:      &scsi host being recovered.
1407  * @work_q:     &list_head for pending commands.
1408  * @done_q:     &list_head for processed commands.
1409  *
1410  * Notes:
1411  *    If commands are failing due to not ready, initializing command required,
1412  *      try revalidating the device, which will end up sending a start unit.
1413  */
1414 static int scsi_eh_stu(struct Scsi_Host *shost,
1415                               struct list_head *work_q,
1416                               struct list_head *done_q)
1417 {
1418         struct scsi_cmnd *scmd, *stu_scmd, *next;
1419         struct scsi_device *sdev;
1420
1421         shost_for_each_device(sdev, shost) {
1422                 if (scsi_host_eh_past_deadline(shost)) {
1423                         SCSI_LOG_ERROR_RECOVERY(3,
1424                                 sdev_printk(KERN_INFO, sdev,
1425                                             "%s: skip START_UNIT, past eh deadline\n",
1426                                             current->comm));
1427                         scsi_device_put(sdev);
1428                         break;
1429                 }
1430                 stu_scmd = NULL;
1431                 list_for_each_entry(scmd, work_q, eh_entry)
1432                         if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1433                             scsi_check_sense(scmd) == FAILED ) {
1434                                 stu_scmd = scmd;
1435                                 break;
1436                         }
1437
1438                 if (!stu_scmd)
1439                         continue;
1440
1441                 SCSI_LOG_ERROR_RECOVERY(3,
1442                         sdev_printk(KERN_INFO, sdev,
1443                                      "%s: Sending START_UNIT\n",
1444                                     current->comm));
1445
1446                 if (!scsi_eh_try_stu(stu_scmd)) {
1447                         if (!scsi_device_online(sdev) ||
1448                             !scsi_eh_tur(stu_scmd)) {
1449                                 list_for_each_entry_safe(scmd, next,
1450                                                           work_q, eh_entry) {
1451                                         if (scmd->device == sdev &&
1452                                             scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1453                                                 scsi_eh_finish_cmd(scmd, done_q);
1454                                 }
1455                         }
1456                 } else {
1457                         SCSI_LOG_ERROR_RECOVERY(3,
1458                                 sdev_printk(KERN_INFO, sdev,
1459                                             "%s: START_UNIT failed\n",
1460                                             current->comm));
1461                 }
1462         }
1463
1464         return list_empty(work_q);
1465 }
1466
1467
1468 /**
1469  * scsi_eh_bus_device_reset - send bdr if needed
1470  * @shost:      scsi host being recovered.
1471  * @work_q:     &list_head for pending commands.
1472  * @done_q:     &list_head for processed commands.
1473  *
1474  * Notes:
1475  *    Try a bus device reset.  Still, look to see whether we have multiple
1476  *    devices that are jammed or not - if we have multiple devices, it
1477  *    makes no sense to try bus_device_reset - we really would need to try
1478  *    a bus_reset instead.
1479  */
1480 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1481                                     struct list_head *work_q,
1482                                     struct list_head *done_q)
1483 {
1484         struct scsi_cmnd *scmd, *bdr_scmd, *next;
1485         struct scsi_device *sdev;
1486         enum scsi_disposition rtn;
1487
1488         shost_for_each_device(sdev, shost) {
1489                 if (scsi_host_eh_past_deadline(shost)) {
1490                         SCSI_LOG_ERROR_RECOVERY(3,
1491                                 sdev_printk(KERN_INFO, sdev,
1492                                             "%s: skip BDR, past eh deadline\n",
1493                                              current->comm));
1494                         scsi_device_put(sdev);
1495                         break;
1496                 }
1497                 bdr_scmd = NULL;
1498                 list_for_each_entry(scmd, work_q, eh_entry)
1499                         if (scmd->device == sdev) {
1500                                 bdr_scmd = scmd;
1501                                 break;
1502                         }
1503
1504                 if (!bdr_scmd)
1505                         continue;
1506
1507                 SCSI_LOG_ERROR_RECOVERY(3,
1508                         sdev_printk(KERN_INFO, sdev,
1509                                      "%s: Sending BDR\n", current->comm));
1510                 rtn = scsi_try_bus_device_reset(bdr_scmd);
1511                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1512                         if (!scsi_device_online(sdev) ||
1513                             rtn == FAST_IO_FAIL ||
1514                             !scsi_eh_tur(bdr_scmd)) {
1515                                 list_for_each_entry_safe(scmd, next,
1516                                                          work_q, eh_entry) {
1517                                         if (scmd->device == sdev &&
1518                                             scsi_eh_action(scmd, rtn) != FAILED)
1519                                                 scsi_eh_finish_cmd(scmd,
1520                                                                    done_q);
1521                                 }
1522                         }
1523                 } else {
1524                         SCSI_LOG_ERROR_RECOVERY(3,
1525                                 sdev_printk(KERN_INFO, sdev,
1526                                             "%s: BDR failed\n", current->comm));
1527                 }
1528         }
1529
1530         return list_empty(work_q);
1531 }
1532
1533 /**
1534  * scsi_eh_target_reset - send target reset if needed
1535  * @shost:      scsi host being recovered.
1536  * @work_q:     &list_head for pending commands.
1537  * @done_q:     &list_head for processed commands.
1538  *
1539  * Notes:
1540  *    Try a target reset.
1541  */
1542 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1543                                 struct list_head *work_q,
1544                                 struct list_head *done_q)
1545 {
1546         LIST_HEAD(tmp_list);
1547         LIST_HEAD(check_list);
1548
1549         list_splice_init(work_q, &tmp_list);
1550
1551         while (!list_empty(&tmp_list)) {
1552                 struct scsi_cmnd *next, *scmd;
1553                 enum scsi_disposition rtn;
1554                 unsigned int id;
1555
1556                 if (scsi_host_eh_past_deadline(shost)) {
1557                         /* push back on work queue for further processing */
1558                         list_splice_init(&check_list, work_q);
1559                         list_splice_init(&tmp_list, work_q);
1560                         SCSI_LOG_ERROR_RECOVERY(3,
1561                                 shost_printk(KERN_INFO, shost,
1562                                             "%s: Skip target reset, past eh deadline\n",
1563                                              current->comm));
1564                         return list_empty(work_q);
1565                 }
1566
1567                 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1568                 id = scmd_id(scmd);
1569
1570                 SCSI_LOG_ERROR_RECOVERY(3,
1571                         shost_printk(KERN_INFO, shost,
1572                                      "%s: Sending target reset to target %d\n",
1573                                      current->comm, id));
1574                 rtn = scsi_try_target_reset(scmd);
1575                 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1576                         SCSI_LOG_ERROR_RECOVERY(3,
1577                                 shost_printk(KERN_INFO, shost,
1578                                              "%s: Target reset failed"
1579                                              " target: %d\n",
1580                                              current->comm, id));
1581                 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1582                         if (scmd_id(scmd) != id)
1583                                 continue;
1584
1585                         if (rtn == SUCCESS)
1586                                 list_move_tail(&scmd->eh_entry, &check_list);
1587                         else if (rtn == FAST_IO_FAIL)
1588                                 scsi_eh_finish_cmd(scmd, done_q);
1589                         else
1590                                 /* push back on work queue for further processing */
1591                                 list_move(&scmd->eh_entry, work_q);
1592                 }
1593         }
1594
1595         return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1596 }
1597
1598 /**
1599  * scsi_eh_bus_reset - send a bus reset
1600  * @shost:      &scsi host being recovered.
1601  * @work_q:     &list_head for pending commands.
1602  * @done_q:     &list_head for processed commands.
1603  */
1604 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1605                              struct list_head *work_q,
1606                              struct list_head *done_q)
1607 {
1608         struct scsi_cmnd *scmd, *chan_scmd, *next;
1609         LIST_HEAD(check_list);
1610         unsigned int channel;
1611         enum scsi_disposition rtn;
1612
1613         /*
1614          * we really want to loop over the various channels, and do this on
1615          * a channel by channel basis.  we should also check to see if any
1616          * of the failed commands are on soft_reset devices, and if so, skip
1617          * the reset.
1618          */
1619
1620         for (channel = 0; channel <= shost->max_channel; channel++) {
1621                 if (scsi_host_eh_past_deadline(shost)) {
1622                         list_splice_init(&check_list, work_q);
1623                         SCSI_LOG_ERROR_RECOVERY(3,
1624                                 shost_printk(KERN_INFO, shost,
1625                                             "%s: skip BRST, past eh deadline\n",
1626                                              current->comm));
1627                         return list_empty(work_q);
1628                 }
1629
1630                 chan_scmd = NULL;
1631                 list_for_each_entry(scmd, work_q, eh_entry) {
1632                         if (channel == scmd_channel(scmd)) {
1633                                 chan_scmd = scmd;
1634                                 break;
1635                                 /*
1636                                  * FIXME add back in some support for
1637                                  * soft_reset devices.
1638                                  */
1639                         }
1640                 }
1641
1642                 if (!chan_scmd)
1643                         continue;
1644                 SCSI_LOG_ERROR_RECOVERY(3,
1645                         shost_printk(KERN_INFO, shost,
1646                                      "%s: Sending BRST chan: %d\n",
1647                                      current->comm, channel));
1648                 rtn = scsi_try_bus_reset(chan_scmd);
1649                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1650                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1651                                 if (channel == scmd_channel(scmd)) {
1652                                         if (rtn == FAST_IO_FAIL)
1653                                                 scsi_eh_finish_cmd(scmd,
1654                                                                    done_q);
1655                                         else
1656                                                 list_move_tail(&scmd->eh_entry,
1657                                                                &check_list);
1658                                 }
1659                         }
1660                 } else {
1661                         SCSI_LOG_ERROR_RECOVERY(3,
1662                                 shost_printk(KERN_INFO, shost,
1663                                              "%s: BRST failed chan: %d\n",
1664                                              current->comm, channel));
1665                 }
1666         }
1667         return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1668 }
1669
1670 /**
1671  * scsi_eh_host_reset - send a host reset
1672  * @shost:      host to be reset.
1673  * @work_q:     &list_head for pending commands.
1674  * @done_q:     &list_head for processed commands.
1675  */
1676 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1677                               struct list_head *work_q,
1678                               struct list_head *done_q)
1679 {
1680         struct scsi_cmnd *scmd, *next;
1681         LIST_HEAD(check_list);
1682         enum scsi_disposition rtn;
1683
1684         if (!list_empty(work_q)) {
1685                 scmd = list_entry(work_q->next,
1686                                   struct scsi_cmnd, eh_entry);
1687
1688                 SCSI_LOG_ERROR_RECOVERY(3,
1689                         shost_printk(KERN_INFO, shost,
1690                                      "%s: Sending HRST\n",
1691                                      current->comm));
1692
1693                 rtn = scsi_try_host_reset(scmd);
1694                 if (rtn == SUCCESS) {
1695                         list_splice_init(work_q, &check_list);
1696                 } else if (rtn == FAST_IO_FAIL) {
1697                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1698                                         scsi_eh_finish_cmd(scmd, done_q);
1699                         }
1700                 } else {
1701                         SCSI_LOG_ERROR_RECOVERY(3,
1702                                 shost_printk(KERN_INFO, shost,
1703                                              "%s: HRST failed\n",
1704                                              current->comm));
1705                 }
1706         }
1707         return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1708 }
1709
1710 /**
1711  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1712  * @work_q:     &list_head for pending commands.
1713  * @done_q:     &list_head for processed commands.
1714  */
1715 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1716                                   struct list_head *done_q)
1717 {
1718         struct scsi_cmnd *scmd, *next;
1719         struct scsi_device *sdev;
1720
1721         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1722                 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1723                             "not ready after error recovery\n");
1724                 sdev = scmd->device;
1725
1726                 mutex_lock(&sdev->state_mutex);
1727                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1728                 mutex_unlock(&sdev->state_mutex);
1729
1730                 scsi_eh_finish_cmd(scmd, done_q);
1731         }
1732         return;
1733 }
1734
1735 /**
1736  * scsi_noretry_cmd - determine if command should be failed fast
1737  * @scmd:       SCSI cmd to examine.
1738  */
1739 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1740 {
1741         switch (host_byte(scmd->result)) {
1742         case DID_OK:
1743                 break;
1744         case DID_TIME_OUT:
1745                 goto check_type;
1746         case DID_BUS_BUSY:
1747                 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1748         case DID_PARITY:
1749                 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1750         case DID_ERROR:
1751                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1752                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1753                         return 0;
1754                 fallthrough;
1755         case DID_SOFT_ERROR:
1756                 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1757         }
1758
1759         if (status_byte(scmd->result) != CHECK_CONDITION)
1760                 return 0;
1761
1762 check_type:
1763         /*
1764          * assume caller has checked sense and determined
1765          * the check condition was retryable.
1766          */
1767         if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1768             blk_rq_is_passthrough(scmd->request))
1769                 return 1;
1770
1771         return 0;
1772 }
1773
1774 /**
1775  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1776  * @scmd:       SCSI cmd to examine.
1777  *
1778  * Notes:
1779  *    This is *only* called when we are examining the status after sending
1780  *    out the actual data command.  any commands that are queued for error
1781  *    recovery (e.g. test_unit_ready) do *not* come through here.
1782  *
1783  *    When this routine returns failed, it means the error handler thread
1784  *    is woken.  In cases where the error code indicates an error that
1785  *    doesn't require the error handler read (i.e. we don't need to
1786  *    abort/reset), this function should return SUCCESS.
1787  */
1788 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
1789 {
1790         enum scsi_disposition rtn;
1791
1792         /*
1793          * if the device is offline, then we clearly just pass the result back
1794          * up to the top level.
1795          */
1796         if (!scsi_device_online(scmd->device)) {
1797                 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1798                         "%s: device offline - report as SUCCESS\n", __func__));
1799                 return SUCCESS;
1800         }
1801
1802         /*
1803          * first check the host byte, to see if there is anything in there
1804          * that would indicate what we need to do.
1805          */
1806         switch (host_byte(scmd->result)) {
1807         case DID_PASSTHROUGH:
1808                 /*
1809                  * no matter what, pass this through to the upper layer.
1810                  * nuke this special code so that it looks like we are saying
1811                  * did_ok.
1812                  */
1813                 scmd->result &= 0xff00ffff;
1814                 return SUCCESS;
1815         case DID_OK:
1816                 /*
1817                  * looks good.  drop through, and check the next byte.
1818                  */
1819                 break;
1820         case DID_ABORT:
1821                 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1822                         set_host_byte(scmd, DID_TIME_OUT);
1823                         return SUCCESS;
1824                 }
1825                 fallthrough;
1826         case DID_NO_CONNECT:
1827         case DID_BAD_TARGET:
1828                 /*
1829                  * note - this means that we just report the status back
1830                  * to the top level driver, not that we actually think
1831                  * that it indicates SUCCESS.
1832                  */
1833                 return SUCCESS;
1834         case DID_SOFT_ERROR:
1835                 /*
1836                  * when the low level driver returns did_soft_error,
1837                  * it is responsible for keeping an internal retry counter
1838                  * in order to avoid endless loops (db)
1839                  */
1840                 goto maybe_retry;
1841         case DID_IMM_RETRY:
1842                 return NEEDS_RETRY;
1843
1844         case DID_REQUEUE:
1845                 return ADD_TO_MLQUEUE;
1846         case DID_TRANSPORT_DISRUPTED:
1847                 /*
1848                  * LLD/transport was disrupted during processing of the IO.
1849                  * The transport class is now blocked/blocking,
1850                  * and the transport will decide what to do with the IO
1851                  * based on its timers and recovery capablilities if
1852                  * there are enough retries.
1853                  */
1854                 goto maybe_retry;
1855         case DID_TRANSPORT_FAILFAST:
1856                 /*
1857                  * The transport decided to failfast the IO (most likely
1858                  * the fast io fail tmo fired), so send IO directly upwards.
1859                  */
1860                 return SUCCESS;
1861         case DID_ERROR:
1862                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1863                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1864                         /*
1865                          * execute reservation conflict processing code
1866                          * lower down
1867                          */
1868                         break;
1869                 fallthrough;
1870         case DID_BUS_BUSY:
1871         case DID_PARITY:
1872                 goto maybe_retry;
1873         case DID_TIME_OUT:
1874                 /*
1875                  * when we scan the bus, we get timeout messages for
1876                  * these commands if there is no device available.
1877                  * other hosts report did_no_connect for the same thing.
1878                  */
1879                 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1880                      scmd->cmnd[0] == INQUIRY)) {
1881                         return SUCCESS;
1882                 } else {
1883                         return FAILED;
1884                 }
1885         case DID_RESET:
1886                 return SUCCESS;
1887         default:
1888                 return FAILED;
1889         }
1890
1891         /*
1892          * next, check the message byte.
1893          */
1894         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1895                 return FAILED;
1896
1897         /*
1898          * check the status byte to see if this indicates anything special.
1899          */
1900         switch (status_byte(scmd->result)) {
1901         case QUEUE_FULL:
1902                 scsi_handle_queue_full(scmd->device);
1903                 /*
1904                  * the case of trying to send too many commands to a
1905                  * tagged queueing device.
1906                  */
1907                 fallthrough;
1908         case BUSY:
1909                 /*
1910                  * device can't talk to us at the moment.  Should only
1911                  * occur (SAM-3) when the task queue is empty, so will cause
1912                  * the empty queue handling to trigger a stall in the
1913                  * device.
1914                  */
1915                 return ADD_TO_MLQUEUE;
1916         case GOOD:
1917                 if (scmd->cmnd[0] == REPORT_LUNS)
1918                         scmd->device->sdev_target->expecting_lun_change = 0;
1919                 scsi_handle_queue_ramp_up(scmd->device);
1920                 fallthrough;
1921         case COMMAND_TERMINATED:
1922                 return SUCCESS;
1923         case TASK_ABORTED:
1924                 goto maybe_retry;
1925         case CHECK_CONDITION:
1926                 rtn = scsi_check_sense(scmd);
1927                 if (rtn == NEEDS_RETRY)
1928                         goto maybe_retry;
1929                 /* if rtn == FAILED, we have no sense information;
1930                  * returning FAILED will wake the error handler thread
1931                  * to collect the sense and redo the decide
1932                  * disposition */
1933                 return rtn;
1934         case CONDITION_GOOD:
1935         case INTERMEDIATE_GOOD:
1936         case INTERMEDIATE_C_GOOD:
1937         case ACA_ACTIVE:
1938                 /*
1939                  * who knows?  FIXME(eric)
1940                  */
1941                 return SUCCESS;
1942
1943         case RESERVATION_CONFLICT:
1944                 sdev_printk(KERN_INFO, scmd->device,
1945                             "reservation conflict\n");
1946                 set_host_byte(scmd, DID_NEXUS_FAILURE);
1947                 return SUCCESS; /* causes immediate i/o error */
1948         default:
1949                 return FAILED;
1950         }
1951         return FAILED;
1952
1953 maybe_retry:
1954
1955         /* we requeue for retry because the error was retryable, and
1956          * the request was not marked fast fail.  Note that above,
1957          * even if the request is marked fast fail, we still requeue
1958          * for queue congestion conditions (QUEUE_FULL or BUSY) */
1959         if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
1960                 return NEEDS_RETRY;
1961         } else {
1962                 /*
1963                  * no more retries - report this one back to upper level.
1964                  */
1965                 return SUCCESS;
1966         }
1967 }
1968
1969 static void eh_lock_door_done(struct request *req, blk_status_t status)
1970 {
1971         blk_put_request(req);
1972 }
1973
1974 /**
1975  * scsi_eh_lock_door - Prevent medium removal for the specified device
1976  * @sdev:       SCSI device to prevent medium removal
1977  *
1978  * Locking:
1979  *      We must be called from process context.
1980  *
1981  * Notes:
1982  *      We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1983  *      head of the devices request queue, and continue.
1984  */
1985 static void scsi_eh_lock_door(struct scsi_device *sdev)
1986 {
1987         struct request *req;
1988         struct scsi_request *rq;
1989
1990         req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, 0);
1991         if (IS_ERR(req))
1992                 return;
1993         rq = scsi_req(req);
1994
1995         rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1996         rq->cmd[1] = 0;
1997         rq->cmd[2] = 0;
1998         rq->cmd[3] = 0;
1999         rq->cmd[4] = SCSI_REMOVAL_PREVENT;
2000         rq->cmd[5] = 0;
2001         rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
2002
2003         req->rq_flags |= RQF_QUIET;
2004         req->timeout = 10 * HZ;
2005         rq->retries = 5;
2006
2007         blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
2008 }
2009
2010 /**
2011  * scsi_restart_operations - restart io operations to the specified host.
2012  * @shost:      Host we are restarting.
2013  *
2014  * Notes:
2015  *    When we entered the error handler, we blocked all further i/o to
2016  *    this device.  we need to 'reverse' this process.
2017  */
2018 static void scsi_restart_operations(struct Scsi_Host *shost)
2019 {
2020         struct scsi_device *sdev;
2021         unsigned long flags;
2022
2023         /*
2024          * If the door was locked, we need to insert a door lock request
2025          * onto the head of the SCSI request queue for the device.  There
2026          * is no point trying to lock the door of an off-line device.
2027          */
2028         shost_for_each_device(sdev, shost) {
2029                 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2030                         scsi_eh_lock_door(sdev);
2031                         sdev->was_reset = 0;
2032                 }
2033         }
2034
2035         /*
2036          * next free up anything directly waiting upon the host.  this
2037          * will be requests for character device operations, and also for
2038          * ioctls to queued block devices.
2039          */
2040         SCSI_LOG_ERROR_RECOVERY(3,
2041                 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2042
2043         spin_lock_irqsave(shost->host_lock, flags);
2044         if (scsi_host_set_state(shost, SHOST_RUNNING))
2045                 if (scsi_host_set_state(shost, SHOST_CANCEL))
2046                         BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2047         spin_unlock_irqrestore(shost->host_lock, flags);
2048
2049         wake_up(&shost->host_wait);
2050
2051         /*
2052          * finally we need to re-initiate requests that may be pending.  we will
2053          * have had everything blocked while error handling is taking place, and
2054          * now that error recovery is done, we will need to ensure that these
2055          * requests are started.
2056          */
2057         scsi_run_host_queues(shost);
2058
2059         /*
2060          * if eh is active and host_eh_scheduled is pending we need to re-run
2061          * recovery.  we do this check after scsi_run_host_queues() to allow
2062          * everything pent up since the last eh run a chance to make forward
2063          * progress before we sync again.  Either we'll immediately re-run
2064          * recovery or scsi_device_unbusy() will wake us again when these
2065          * pending commands complete.
2066          */
2067         spin_lock_irqsave(shost->host_lock, flags);
2068         if (shost->host_eh_scheduled)
2069                 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2070                         WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2071         spin_unlock_irqrestore(shost->host_lock, flags);
2072 }
2073
2074 /**
2075  * scsi_eh_ready_devs - check device ready state and recover if not.
2076  * @shost:      host to be recovered.
2077  * @work_q:     &list_head for pending commands.
2078  * @done_q:     &list_head for processed commands.
2079  */
2080 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2081                         struct list_head *work_q,
2082                         struct list_head *done_q)
2083 {
2084         if (!scsi_eh_stu(shost, work_q, done_q))
2085                 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2086                         if (!scsi_eh_target_reset(shost, work_q, done_q))
2087                                 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2088                                         if (!scsi_eh_host_reset(shost, work_q, done_q))
2089                                                 scsi_eh_offline_sdevs(work_q,
2090                                                                       done_q);
2091 }
2092 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2093
2094 /**
2095  * scsi_eh_flush_done_q - finish processed commands or retry them.
2096  * @done_q:     list_head of processed commands.
2097  */
2098 void scsi_eh_flush_done_q(struct list_head *done_q)
2099 {
2100         struct scsi_cmnd *scmd, *next;
2101
2102         list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2103                 list_del_init(&scmd->eh_entry);
2104                 if (scsi_device_online(scmd->device) &&
2105                     !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd)) {
2106                         SCSI_LOG_ERROR_RECOVERY(3,
2107                                 scmd_printk(KERN_INFO, scmd,
2108                                              "%s: flush retry cmd\n",
2109                                              current->comm));
2110                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2111                 } else {
2112                         /*
2113                          * If just we got sense for the device (called
2114                          * scsi_eh_get_sense), scmd->result is already
2115                          * set, do not set DRIVER_TIMEOUT.
2116                          */
2117                         if (!scmd->result)
2118                                 scmd->result |= (DRIVER_TIMEOUT << 24);
2119                         SCSI_LOG_ERROR_RECOVERY(3,
2120                                 scmd_printk(KERN_INFO, scmd,
2121                                              "%s: flush finish cmd\n",
2122                                              current->comm));
2123                         scsi_finish_command(scmd);
2124                 }
2125         }
2126 }
2127 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2128
2129 /**
2130  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2131  * @shost:      Host to unjam.
2132  *
2133  * Notes:
2134  *    When we come in here, we *know* that all commands on the bus have
2135  *    either completed, failed or timed out.  we also know that no further
2136  *    commands are being sent to the host, so things are relatively quiet
2137  *    and we have freedom to fiddle with things as we wish.
2138  *
2139  *    This is only the *default* implementation.  it is possible for
2140  *    individual drivers to supply their own version of this function, and
2141  *    if the maintainer wishes to do this, it is strongly suggested that
2142  *    this function be taken as a template and modified.  this function
2143  *    was designed to correctly handle problems for about 95% of the
2144  *    different cases out there, and it should always provide at least a
2145  *    reasonable amount of error recovery.
2146  *
2147  *    Any command marked 'failed' or 'timeout' must eventually have
2148  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
2149  *    here, so when we restart the host after we return it should have an
2150  *    empty queue.
2151  */
2152 static void scsi_unjam_host(struct Scsi_Host *shost)
2153 {
2154         unsigned long flags;
2155         LIST_HEAD(eh_work_q);
2156         LIST_HEAD(eh_done_q);
2157
2158         spin_lock_irqsave(shost->host_lock, flags);
2159         list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2160         spin_unlock_irqrestore(shost->host_lock, flags);
2161
2162         SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2163
2164         if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2165                 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2166
2167         spin_lock_irqsave(shost->host_lock, flags);
2168         if (shost->eh_deadline != -1)
2169                 shost->last_reset = 0;
2170         spin_unlock_irqrestore(shost->host_lock, flags);
2171         scsi_eh_flush_done_q(&eh_done_q);
2172 }
2173
2174 /**
2175  * scsi_error_handler - SCSI error handler thread
2176  * @data:       Host for which we are running.
2177  *
2178  * Notes:
2179  *    This is the main error handling loop.  This is run as a kernel thread
2180  *    for every SCSI host and handles all error handling activity.
2181  */
2182 int scsi_error_handler(void *data)
2183 {
2184         struct Scsi_Host *shost = data;
2185
2186         /*
2187          * We use TASK_INTERRUPTIBLE so that the thread is not
2188          * counted against the load average as a running process.
2189          * We never actually get interrupted because kthread_run
2190          * disables signal delivery for the created thread.
2191          */
2192         while (true) {
2193                 /*
2194                  * The sequence in kthread_stop() sets the stop flag first
2195                  * then wakes the process.  To avoid missed wakeups, the task
2196                  * should always be in a non running state before the stop
2197                  * flag is checked
2198                  */
2199                 set_current_state(TASK_INTERRUPTIBLE);
2200                 if (kthread_should_stop())
2201                         break;
2202
2203                 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2204                     shost->host_failed != scsi_host_busy(shost)) {
2205                         SCSI_LOG_ERROR_RECOVERY(1,
2206                                 shost_printk(KERN_INFO, shost,
2207                                              "scsi_eh_%d: sleeping\n",
2208                                              shost->host_no));
2209                         schedule();
2210                         continue;
2211                 }
2212
2213                 __set_current_state(TASK_RUNNING);
2214                 SCSI_LOG_ERROR_RECOVERY(1,
2215                         shost_printk(KERN_INFO, shost,
2216                                      "scsi_eh_%d: waking up %d/%d/%d\n",
2217                                      shost->host_no, shost->host_eh_scheduled,
2218                                      shost->host_failed,
2219                                      scsi_host_busy(shost)));
2220
2221                 /*
2222                  * We have a host that is failing for some reason.  Figure out
2223                  * what we need to do to get it up and online again (if we can).
2224                  * If we fail, we end up taking the thing offline.
2225                  */
2226                 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2227                         SCSI_LOG_ERROR_RECOVERY(1,
2228                                 shost_printk(KERN_ERR, shost,
2229                                              "scsi_eh_%d: unable to autoresume\n",
2230                                              shost->host_no));
2231                         continue;
2232                 }
2233
2234                 if (shost->transportt->eh_strategy_handler)
2235                         shost->transportt->eh_strategy_handler(shost);
2236                 else
2237                         scsi_unjam_host(shost);
2238
2239                 /* All scmds have been handled */
2240                 shost->host_failed = 0;
2241
2242                 /*
2243                  * Note - if the above fails completely, the action is to take
2244                  * individual devices offline and flush the queue of any
2245                  * outstanding requests that may have been pending.  When we
2246                  * restart, we restart any I/O to any other devices on the bus
2247                  * which are still online.
2248                  */
2249                 scsi_restart_operations(shost);
2250                 if (!shost->eh_noresume)
2251                         scsi_autopm_put_host(shost);
2252         }
2253         __set_current_state(TASK_RUNNING);
2254
2255         SCSI_LOG_ERROR_RECOVERY(1,
2256                 shost_printk(KERN_INFO, shost,
2257                              "Error handler scsi_eh_%d exiting\n",
2258                              shost->host_no));
2259         shost->ehandler = NULL;
2260         return 0;
2261 }
2262
2263 /*
2264  * Function:    scsi_report_bus_reset()
2265  *
2266  * Purpose:     Utility function used by low-level drivers to report that
2267  *              they have observed a bus reset on the bus being handled.
2268  *
2269  * Arguments:   shost       - Host in question
2270  *              channel     - channel on which reset was observed.
2271  *
2272  * Returns:     Nothing
2273  *
2274  * Lock status: Host lock must be held.
2275  *
2276  * Notes:       This only needs to be called if the reset is one which
2277  *              originates from an unknown location.  Resets originated
2278  *              by the mid-level itself don't need to call this, but there
2279  *              should be no harm.
2280  *
2281  *              The main purpose of this is to make sure that a CHECK_CONDITION
2282  *              is properly treated.
2283  */
2284 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2285 {
2286         struct scsi_device *sdev;
2287
2288         __shost_for_each_device(sdev, shost) {
2289                 if (channel == sdev_channel(sdev))
2290                         __scsi_report_device_reset(sdev, NULL);
2291         }
2292 }
2293 EXPORT_SYMBOL(scsi_report_bus_reset);
2294
2295 /*
2296  * Function:    scsi_report_device_reset()
2297  *
2298  * Purpose:     Utility function used by low-level drivers to report that
2299  *              they have observed a device reset on the device being handled.
2300  *
2301  * Arguments:   shost       - Host in question
2302  *              channel     - channel on which reset was observed
2303  *              target      - target on which reset was observed
2304  *
2305  * Returns:     Nothing
2306  *
2307  * Lock status: Host lock must be held
2308  *
2309  * Notes:       This only needs to be called if the reset is one which
2310  *              originates from an unknown location.  Resets originated
2311  *              by the mid-level itself don't need to call this, but there
2312  *              should be no harm.
2313  *
2314  *              The main purpose of this is to make sure that a CHECK_CONDITION
2315  *              is properly treated.
2316  */
2317 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2318 {
2319         struct scsi_device *sdev;
2320
2321         __shost_for_each_device(sdev, shost) {
2322                 if (channel == sdev_channel(sdev) &&
2323                     target == sdev_id(sdev))
2324                         __scsi_report_device_reset(sdev, NULL);
2325         }
2326 }
2327 EXPORT_SYMBOL(scsi_report_device_reset);
2328
2329 static void
2330 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2331 {
2332 }
2333
2334 /**
2335  * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2336  * @dev:        scsi_device to operate on
2337  * @arg:        reset type (see sg.h)
2338  */
2339 int
2340 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2341 {
2342         struct scsi_cmnd *scmd;
2343         struct Scsi_Host *shost = dev->host;
2344         struct request *rq;
2345         unsigned long flags;
2346         int error = 0, val;
2347         enum scsi_disposition rtn;
2348
2349         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2350                 return -EACCES;
2351
2352         error = get_user(val, arg);
2353         if (error)
2354                 return error;
2355
2356         if (scsi_autopm_get_host(shost) < 0)
2357                 return -EIO;
2358
2359         error = -EIO;
2360         rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2361                         shost->hostt->cmd_size, GFP_KERNEL);
2362         if (!rq)
2363                 goto out_put_autopm_host;
2364         blk_rq_init(NULL, rq);
2365
2366         scmd = (struct scsi_cmnd *)(rq + 1);
2367         scsi_init_command(dev, scmd);
2368         scmd->request = rq;
2369         scmd->cmnd = scsi_req(rq)->cmd;
2370         scmd->flags |= SCMD_LAST;
2371
2372         scmd->scsi_done         = scsi_reset_provider_done_command;
2373         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2374
2375         scmd->cmd_len                   = 0;
2376
2377         scmd->sc_data_direction         = DMA_BIDIRECTIONAL;
2378
2379         spin_lock_irqsave(shost->host_lock, flags);
2380         shost->tmf_in_progress = 1;
2381         spin_unlock_irqrestore(shost->host_lock, flags);
2382
2383         switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2384         case SG_SCSI_RESET_NOTHING:
2385                 rtn = SUCCESS;
2386                 break;
2387         case SG_SCSI_RESET_DEVICE:
2388                 rtn = scsi_try_bus_device_reset(scmd);
2389                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2390                         break;
2391                 fallthrough;
2392         case SG_SCSI_RESET_TARGET:
2393                 rtn = scsi_try_target_reset(scmd);
2394                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2395                         break;
2396                 fallthrough;
2397         case SG_SCSI_RESET_BUS:
2398                 rtn = scsi_try_bus_reset(scmd);
2399                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2400                         break;
2401                 fallthrough;
2402         case SG_SCSI_RESET_HOST:
2403                 rtn = scsi_try_host_reset(scmd);
2404                 if (rtn == SUCCESS)
2405                         break;
2406                 fallthrough;
2407         default:
2408                 rtn = FAILED;
2409                 break;
2410         }
2411
2412         error = (rtn == SUCCESS) ? 0 : -EIO;
2413
2414         spin_lock_irqsave(shost->host_lock, flags);
2415         shost->tmf_in_progress = 0;
2416         spin_unlock_irqrestore(shost->host_lock, flags);
2417
2418         /*
2419          * be sure to wake up anyone who was sleeping or had their queue
2420          * suspended while we performed the TMF.
2421          */
2422         SCSI_LOG_ERROR_RECOVERY(3,
2423                 shost_printk(KERN_INFO, shost,
2424                              "waking up host to restart after TMF\n"));
2425
2426         wake_up(&shost->host_wait);
2427         scsi_run_host_queues(shost);
2428
2429         kfree(rq);
2430
2431 out_put_autopm_host:
2432         scsi_autopm_put_host(shost);
2433         return error;
2434 }
2435
2436 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2437                                   struct scsi_sense_hdr *sshdr)
2438 {
2439         return scsi_normalize_sense(cmd->sense_buffer,
2440                         SCSI_SENSE_BUFFERSIZE, sshdr);
2441 }
2442 EXPORT_SYMBOL(scsi_command_normalize_sense);
2443
2444 /**
2445  * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2446  * @sense_buffer:       byte array of sense data
2447  * @sb_len:             number of valid bytes in sense_buffer
2448  * @info_out:           pointer to 64 integer where 8 or 4 byte information
2449  *                      field will be placed if found.
2450  *
2451  * Return value:
2452  *      true if information field found, false if not found.
2453  */
2454 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2455                              u64 *info_out)
2456 {
2457         const u8 * ucp;
2458
2459         if (sb_len < 7)
2460                 return false;
2461         switch (sense_buffer[0] & 0x7f) {
2462         case 0x70:
2463         case 0x71:
2464                 if (sense_buffer[0] & 0x80) {
2465                         *info_out = get_unaligned_be32(&sense_buffer[3]);
2466                         return true;
2467                 }
2468                 return false;
2469         case 0x72:
2470         case 0x73:
2471                 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2472                                            0 /* info desc */);
2473                 if (ucp && (0xa == ucp[1])) {
2474                         *info_out = get_unaligned_be64(&ucp[4]);
2475                         return true;
2476                 }
2477                 return false;
2478         default:
2479                 return false;
2480         }
2481 }
2482 EXPORT_SYMBOL(scsi_get_sense_info_fld);