f490994374f6900370358d9921fac51f34fd63f8
[linux-2.6-microblaze.git] / drivers / scsi / scsi_error.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  scsi_error.c Copyright (C) 1997 Eric Youngdale
4  *
5  *  SCSI error/timeout handling
6  *      Initial versions: Eric Youngdale.  Based upon conversations with
7  *                        Leonard Zubkoff and David Miller at Linux Expo,
8  *                        ideas originating from all over the place.
9  *
10  *      Restructured scsi_unjam_host and associated functions.
11  *      September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12  *
13  *      Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14  *      minor cleanups.
15  *      September 30, 2002 Mike Anderson (andmike@us.ibm.com)
16  */
17
18 #include <linux/module.h>
19 #include <linux/sched.h>
20 #include <linux/gfp.h>
21 #include <linux/timer.h>
22 #include <linux/string.h>
23 #include <linux/kernel.h>
24 #include <linux/freezer.h>
25 #include <linux/kthread.h>
26 #include <linux/interrupt.h>
27 #include <linux/blkdev.h>
28 #include <linux/delay.h>
29 #include <linux/jiffies.h>
30
31 #include <scsi/scsi.h>
32 #include <scsi/scsi_cmnd.h>
33 #include <scsi/scsi_dbg.h>
34 #include <scsi/scsi_device.h>
35 #include <scsi/scsi_driver.h>
36 #include <scsi/scsi_eh.h>
37 #include <scsi/scsi_common.h>
38 #include <scsi/scsi_transport.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_ioctl.h>
41 #include <scsi/scsi_dh.h>
42 #include <scsi/scsi_devinfo.h>
43 #include <scsi/sg.h>
44
45 #include "scsi_priv.h"
46 #include "scsi_logging.h"
47 #include "scsi_transport_api.h"
48
49 #include <trace/events/scsi.h>
50
51 #include <asm/unaligned.h>
52
53 static void scsi_eh_done(struct scsi_cmnd *scmd);
54
55 /*
56  * These should *probably* be handled by the host itself.
57  * Since it is allowed to sleep, it probably should.
58  */
59 #define BUS_RESET_SETTLE_TIME   (10)
60 #define HOST_RESET_SETTLE_TIME  (10)
61
62 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
63 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
64                                  struct scsi_cmnd *);
65
66 void scsi_eh_wakeup(struct Scsi_Host *shost)
67 {
68         lockdep_assert_held(shost->host_lock);
69
70         if (scsi_host_busy(shost) == shost->host_failed) {
71                 trace_scsi_eh_wakeup(shost);
72                 wake_up_process(shost->ehandler);
73                 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
74                         "Waking error handler thread\n"));
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);
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 /**
120  * scmd_eh_abort_handler - Handle command aborts
121  * @work:       command to be aborted.
122  *
123  * Note: this function must be called only for a command that has timed out.
124  * Because the block layer marks a request as complete before it calls
125  * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
126  * timed out do not have any effect. Hence it is safe to call
127  * scsi_finish_command() from this function.
128  */
129 void
130 scmd_eh_abort_handler(struct work_struct *work)
131 {
132         struct scsi_cmnd *scmd =
133                 container_of(work, struct scsi_cmnd, abort_work.work);
134         struct scsi_device *sdev = scmd->device;
135         int rtn;
136
137         if (scsi_host_eh_past_deadline(sdev->host)) {
138                 SCSI_LOG_ERROR_RECOVERY(3,
139                         scmd_printk(KERN_INFO, scmd,
140                                     "eh timeout, not aborting\n"));
141         } else {
142                 SCSI_LOG_ERROR_RECOVERY(3,
143                         scmd_printk(KERN_INFO, scmd,
144                                     "aborting command\n"));
145                 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
146                 if (rtn == SUCCESS) {
147                         set_host_byte(scmd, DID_TIME_OUT);
148                         if (scsi_host_eh_past_deadline(sdev->host)) {
149                                 SCSI_LOG_ERROR_RECOVERY(3,
150                                         scmd_printk(KERN_INFO, scmd,
151                                                     "eh timeout, not retrying "
152                                                     "aborted command\n"));
153                         } else if (!scsi_noretry_cmd(scmd) &&
154                             (++scmd->retries <= scmd->allowed)) {
155                                 SCSI_LOG_ERROR_RECOVERY(3,
156                                         scmd_printk(KERN_WARNING, scmd,
157                                                     "retry aborted command\n"));
158                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
159                                 return;
160                         } else {
161                                 SCSI_LOG_ERROR_RECOVERY(3,
162                                         scmd_printk(KERN_WARNING, scmd,
163                                                     "finish aborted command\n"));
164                                 scsi_finish_command(scmd);
165                                 return;
166                         }
167                 } else {
168                         SCSI_LOG_ERROR_RECOVERY(3,
169                                 scmd_printk(KERN_INFO, scmd,
170                                             "cmd abort %s\n",
171                                             (rtn == FAST_IO_FAIL) ?
172                                             "not send" : "failed"));
173                 }
174         }
175
176         scsi_eh_scmd_add(scmd);
177 }
178
179 /**
180  * scsi_abort_command - schedule a command abort
181  * @scmd:       scmd to abort.
182  *
183  * We only need to abort commands after a command timeout
184  */
185 static int
186 scsi_abort_command(struct scsi_cmnd *scmd)
187 {
188         struct scsi_device *sdev = scmd->device;
189         struct Scsi_Host *shost = sdev->host;
190         unsigned long flags;
191
192         if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
193                 /*
194                  * Retry after abort failed, escalate to next level.
195                  */
196                 SCSI_LOG_ERROR_RECOVERY(3,
197                         scmd_printk(KERN_INFO, scmd,
198                                     "previous abort failed\n"));
199                 BUG_ON(delayed_work_pending(&scmd->abort_work));
200                 return FAILED;
201         }
202
203         spin_lock_irqsave(shost->host_lock, flags);
204         if (shost->eh_deadline != -1 && !shost->last_reset)
205                 shost->last_reset = jiffies;
206         spin_unlock_irqrestore(shost->host_lock, flags);
207
208         scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
209         SCSI_LOG_ERROR_RECOVERY(3,
210                 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
211         queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
212         return SUCCESS;
213 }
214
215 /**
216  * scsi_eh_reset - call into ->eh_action to reset internal counters
217  * @scmd:       scmd to run eh on.
218  *
219  * The scsi driver might be carrying internal state about the
220  * devices, so we need to call into the driver to reset the
221  * internal state once the error handler is started.
222  */
223 static void scsi_eh_reset(struct scsi_cmnd *scmd)
224 {
225         if (!blk_rq_is_passthrough(scmd->request)) {
226                 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
227                 if (sdrv->eh_reset)
228                         sdrv->eh_reset(scmd);
229         }
230 }
231
232 static void scsi_eh_inc_host_failed(struct rcu_head *head)
233 {
234         struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
235         struct Scsi_Host *shost = scmd->device->host;
236         unsigned long flags;
237
238         spin_lock_irqsave(shost->host_lock, flags);
239         shost->host_failed++;
240         scsi_eh_wakeup(shost);
241         spin_unlock_irqrestore(shost->host_lock, flags);
242 }
243
244 /**
245  * scsi_eh_scmd_add - add scsi cmd to error handling.
246  * @scmd:       scmd to run eh on.
247  */
248 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
249 {
250         struct Scsi_Host *shost = scmd->device->host;
251         unsigned long flags;
252         int ret;
253
254         WARN_ON_ONCE(!shost->ehandler);
255
256         spin_lock_irqsave(shost->host_lock, flags);
257         if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
258                 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
259                 WARN_ON_ONCE(ret);
260         }
261         if (shost->eh_deadline != -1 && !shost->last_reset)
262                 shost->last_reset = jiffies;
263
264         scsi_eh_reset(scmd);
265         list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
266         spin_unlock_irqrestore(shost->host_lock, flags);
267         /*
268          * Ensure that all tasks observe the host state change before the
269          * host_failed change.
270          */
271         call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
272 }
273
274 /**
275  * scsi_times_out - Timeout function for normal scsi commands.
276  * @req:        request that is timing out.
277  *
278  * Notes:
279  *     We do not need to lock this.  There is the potential for a race
280  *     only in that the normal completion handling might run, but if the
281  *     normal completion function determines that the timer has already
282  *     fired, then it mustn't do anything.
283  */
284 enum blk_eh_timer_return scsi_times_out(struct request *req)
285 {
286         struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
287         enum blk_eh_timer_return rtn = BLK_EH_DONE;
288         struct Scsi_Host *host = scmd->device->host;
289
290         trace_scsi_dispatch_cmd_timeout(scmd);
291         scsi_log_completion(scmd, TIMEOUT_ERROR);
292
293         if (host->eh_deadline != -1 && !host->last_reset)
294                 host->last_reset = jiffies;
295
296         if (host->hostt->eh_timed_out)
297                 rtn = host->hostt->eh_timed_out(scmd);
298
299         if (rtn == BLK_EH_DONE) {
300                 /*
301                  * Set the command to complete first in order to prevent a real
302                  * completion from releasing the command while error handling
303                  * is using it. If the command was already completed, then the
304                  * lower level driver beat the timeout handler, and it is safe
305                  * to return without escalating error recovery.
306                  *
307                  * If timeout handling lost the race to a real completion, the
308                  * block layer may ignore that due to a fake timeout injection,
309                  * so return RESET_TIMER to allow error handling another shot
310                  * at this command.
311                  */
312                 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
313                         return BLK_EH_RESET_TIMER;
314                 if (scsi_abort_command(scmd) != SUCCESS) {
315                         set_host_byte(scmd, DID_TIME_OUT);
316                         scsi_eh_scmd_add(scmd);
317                 }
318         }
319
320         return rtn;
321 }
322
323 /**
324  * scsi_block_when_processing_errors - Prevent cmds from being queued.
325  * @sdev:       Device on which we are performing recovery.
326  *
327  * Description:
328  *     We block until the host is out of error recovery, and then check to
329  *     see whether the host or the device is offline.
330  *
331  * Return value:
332  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
333  */
334 int scsi_block_when_processing_errors(struct scsi_device *sdev)
335 {
336         int online;
337
338         wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
339
340         online = scsi_device_online(sdev);
341
342         return online;
343 }
344 EXPORT_SYMBOL(scsi_block_when_processing_errors);
345
346 #ifdef CONFIG_SCSI_LOGGING
347 /**
348  * scsi_eh_prt_fail_stats - Log info on failures.
349  * @shost:      scsi host being recovered.
350  * @work_q:     Queue of scsi cmds to process.
351  */
352 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
353                                           struct list_head *work_q)
354 {
355         struct scsi_cmnd *scmd;
356         struct scsi_device *sdev;
357         int total_failures = 0;
358         int cmd_failed = 0;
359         int cmd_cancel = 0;
360         int devices_failed = 0;
361
362         shost_for_each_device(sdev, shost) {
363                 list_for_each_entry(scmd, work_q, eh_entry) {
364                         if (scmd->device == sdev) {
365                                 ++total_failures;
366                                 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
367                                         ++cmd_cancel;
368                                 else
369                                         ++cmd_failed;
370                         }
371                 }
372
373                 if (cmd_cancel || cmd_failed) {
374                         SCSI_LOG_ERROR_RECOVERY(3,
375                                 shost_printk(KERN_INFO, shost,
376                                             "%s: cmds failed: %d, cancel: %d\n",
377                                             __func__, cmd_failed,
378                                             cmd_cancel));
379                         cmd_cancel = 0;
380                         cmd_failed = 0;
381                         ++devices_failed;
382                 }
383         }
384
385         SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
386                                    "Total of %d commands on %d"
387                                    " devices require eh work\n",
388                                    total_failures, devices_failed));
389 }
390 #endif
391
392  /**
393  * scsi_report_lun_change - Set flag on all *other* devices on the same target
394  *                          to indicate that a UNIT ATTENTION is expected.
395  * @sdev:       Device reporting the UNIT ATTENTION
396  */
397 static void scsi_report_lun_change(struct scsi_device *sdev)
398 {
399         sdev->sdev_target->expecting_lun_change = 1;
400 }
401
402 /**
403  * scsi_report_sense - Examine scsi sense information and log messages for
404  *                     certain conditions, also issue uevents for some of them.
405  * @sdev:       Device reporting the sense code
406  * @sshdr:      sshdr to be examined
407  */
408 static void scsi_report_sense(struct scsi_device *sdev,
409                               struct scsi_sense_hdr *sshdr)
410 {
411         enum scsi_device_event evt_type = SDEV_EVT_MAXBITS;     /* i.e. none */
412
413         if (sshdr->sense_key == UNIT_ATTENTION) {
414                 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
415                         evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
416                         sdev_printk(KERN_WARNING, sdev,
417                                     "Inquiry data has changed");
418                 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
419                         evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
420                         scsi_report_lun_change(sdev);
421                         sdev_printk(KERN_WARNING, sdev,
422                                     "Warning! Received an indication that the "
423                                     "LUN assignments on this target have "
424                                     "changed. The Linux SCSI layer does not "
425                                     "automatically remap LUN assignments.\n");
426                 } else if (sshdr->asc == 0x3f)
427                         sdev_printk(KERN_WARNING, sdev,
428                                     "Warning! Received an indication that the "
429                                     "operating parameters on this target have "
430                                     "changed. The Linux SCSI layer does not "
431                                     "automatically adjust these parameters.\n");
432
433                 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
434                         evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
435                         sdev_printk(KERN_WARNING, sdev,
436                                     "Warning! Received an indication that the "
437                                     "LUN reached a thin provisioning soft "
438                                     "threshold.\n");
439                 }
440
441                 if (sshdr->asc == 0x29) {
442                         evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
443                         sdev_printk(KERN_WARNING, sdev,
444                                     "Power-on or device reset occurred\n");
445                 }
446
447                 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
448                         evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
449                         sdev_printk(KERN_WARNING, sdev,
450                                     "Mode parameters changed");
451                 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
452                         evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
453                         sdev_printk(KERN_WARNING, sdev,
454                                     "Asymmetric access state changed");
455                 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
456                         evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
457                         sdev_printk(KERN_WARNING, sdev,
458                                     "Capacity data has changed");
459                 } else if (sshdr->asc == 0x2a)
460                         sdev_printk(KERN_WARNING, sdev,
461                                     "Parameters changed");
462         }
463
464         if (evt_type != SDEV_EVT_MAXBITS) {
465                 set_bit(evt_type, sdev->pending_events);
466                 schedule_work(&sdev->event_work);
467         }
468 }
469
470 /**
471  * scsi_check_sense - Examine scsi cmd sense
472  * @scmd:       Cmd to have sense checked.
473  *
474  * Return value:
475  *      SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
476  *
477  * Notes:
478  *      When a deferred error is detected the current command has
479  *      not been executed and needs retrying.
480  */
481 int scsi_check_sense(struct scsi_cmnd *scmd)
482 {
483         struct scsi_device *sdev = scmd->device;
484         struct scsi_sense_hdr sshdr;
485
486         if (! scsi_command_normalize_sense(scmd, &sshdr))
487                 return FAILED;  /* no valid sense data */
488
489         scsi_report_sense(sdev, &sshdr);
490
491         if (scsi_sense_is_deferred(&sshdr))
492                 return NEEDS_RETRY;
493
494         if (sdev->handler && sdev->handler->check_sense) {
495                 int rc;
496
497                 rc = sdev->handler->check_sense(sdev, &sshdr);
498                 if (rc != SCSI_RETURN_NOT_HANDLED)
499                         return rc;
500                 /* handler does not care. Drop down to default handling */
501         }
502
503         if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
504                 /*
505                  * nasty: for mid-layer issued TURs, we need to return the
506                  * actual sense data without any recovery attempt.  For eh
507                  * issued ones, we need to try to recover and interpret
508                  */
509                 return SUCCESS;
510
511         /*
512          * Previous logic looked for FILEMARK, EOM or ILI which are
513          * mainly associated with tapes and returned SUCCESS.
514          */
515         if (sshdr.response_code == 0x70) {
516                 /* fixed format */
517                 if (scmd->sense_buffer[2] & 0xe0)
518                         return SUCCESS;
519         } else {
520                 /*
521                  * descriptor format: look for "stream commands sense data
522                  * descriptor" (see SSC-3). Assume single sense data
523                  * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
524                  */
525                 if ((sshdr.additional_length > 3) &&
526                     (scmd->sense_buffer[8] == 0x4) &&
527                     (scmd->sense_buffer[11] & 0xe0))
528                         return SUCCESS;
529         }
530
531         switch (sshdr.sense_key) {
532         case NO_SENSE:
533                 return SUCCESS;
534         case RECOVERED_ERROR:
535                 return /* soft_error */ SUCCESS;
536
537         case ABORTED_COMMAND:
538                 if (sshdr.asc == 0x10) /* DIF */
539                         return SUCCESS;
540
541                 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
542                         return ADD_TO_MLQUEUE;
543                 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
544                     sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
545                         return ADD_TO_MLQUEUE;
546
547                 return NEEDS_RETRY;
548         case NOT_READY:
549         case UNIT_ATTENTION:
550                 /*
551                  * if we are expecting a cc/ua because of a bus reset that we
552                  * performed, treat this just as a retry.  otherwise this is
553                  * information that we should pass up to the upper-level driver
554                  * so that we can deal with it there.
555                  */
556                 if (scmd->device->expecting_cc_ua) {
557                         /*
558                          * Because some device does not queue unit
559                          * attentions correctly, we carefully check
560                          * additional sense code and qualifier so as
561                          * not to squash media change unit attention.
562                          */
563                         if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
564                                 scmd->device->expecting_cc_ua = 0;
565                                 return NEEDS_RETRY;
566                         }
567                 }
568                 /*
569                  * we might also expect a cc/ua if another LUN on the target
570                  * reported a UA with an ASC/ASCQ of 3F 0E -
571                  * REPORTED LUNS DATA HAS CHANGED.
572                  */
573                 if (scmd->device->sdev_target->expecting_lun_change &&
574                     sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
575                         return NEEDS_RETRY;
576                 /*
577                  * if the device is in the process of becoming ready, we
578                  * should retry.
579                  */
580                 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
581                         return NEEDS_RETRY;
582                 /*
583                  * if the device is not started, we need to wake
584                  * the error handler to start the motor
585                  */
586                 if (scmd->device->allow_restart &&
587                     (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
588                         return FAILED;
589                 /*
590                  * Pass the UA upwards for a determination in the completion
591                  * functions.
592                  */
593                 return SUCCESS;
594
595                 /* these are not supported */
596         case DATA_PROTECT:
597                 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
598                         /* Thin provisioning hard threshold reached */
599                         set_host_byte(scmd, DID_ALLOC_FAILURE);
600                         return SUCCESS;
601                 }
602                 /* FALLTHROUGH */
603         case COPY_ABORTED:
604         case VOLUME_OVERFLOW:
605         case MISCOMPARE:
606         case BLANK_CHECK:
607                 set_host_byte(scmd, DID_TARGET_FAILURE);
608                 return SUCCESS;
609
610         case MEDIUM_ERROR:
611                 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
612                     sshdr.asc == 0x13 || /* AMNF DATA FIELD */
613                     sshdr.asc == 0x14) { /* RECORD NOT FOUND */
614                         set_host_byte(scmd, DID_MEDIUM_ERROR);
615                         return SUCCESS;
616                 }
617                 return NEEDS_RETRY;
618
619         case HARDWARE_ERROR:
620                 if (scmd->device->retry_hwerror)
621                         return ADD_TO_MLQUEUE;
622                 else
623                         set_host_byte(scmd, DID_TARGET_FAILURE);
624                 /* FALLTHROUGH */
625
626         case ILLEGAL_REQUEST:
627                 if (sshdr.asc == 0x20 || /* Invalid command operation code */
628                     sshdr.asc == 0x21 || /* Logical block address out of range */
629                     sshdr.asc == 0x22 || /* Invalid function */
630                     sshdr.asc == 0x24 || /* Invalid field in cdb */
631                     sshdr.asc == 0x26 || /* Parameter value invalid */
632                     sshdr.asc == 0x27) { /* Write protected */
633                         set_host_byte(scmd, DID_TARGET_FAILURE);
634                 }
635                 return SUCCESS;
636
637         default:
638                 return SUCCESS;
639         }
640 }
641 EXPORT_SYMBOL_GPL(scsi_check_sense);
642
643 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
644 {
645         struct scsi_host_template *sht = sdev->host->hostt;
646         struct scsi_device *tmp_sdev;
647
648         if (!sht->track_queue_depth ||
649             sdev->queue_depth >= sdev->max_queue_depth)
650                 return;
651
652         if (time_before(jiffies,
653             sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
654                 return;
655
656         if (time_before(jiffies,
657             sdev->last_queue_full_time + sdev->queue_ramp_up_period))
658                 return;
659
660         /*
661          * Walk all devices of a target and do
662          * ramp up on them.
663          */
664         shost_for_each_device(tmp_sdev, sdev->host) {
665                 if (tmp_sdev->channel != sdev->channel ||
666                     tmp_sdev->id != sdev->id ||
667                     tmp_sdev->queue_depth == sdev->max_queue_depth)
668                         continue;
669
670                 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
671                 sdev->last_queue_ramp_up = jiffies;
672         }
673 }
674
675 static void scsi_handle_queue_full(struct scsi_device *sdev)
676 {
677         struct scsi_host_template *sht = sdev->host->hostt;
678         struct scsi_device *tmp_sdev;
679
680         if (!sht->track_queue_depth)
681                 return;
682
683         shost_for_each_device(tmp_sdev, sdev->host) {
684                 if (tmp_sdev->channel != sdev->channel ||
685                     tmp_sdev->id != sdev->id)
686                         continue;
687                 /*
688                  * We do not know the number of commands that were at
689                  * the device when we got the queue full so we start
690                  * from the highest possible value and work our way down.
691                  */
692                 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
693         }
694 }
695
696 /**
697  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
698  * @scmd:       SCSI cmd to examine.
699  *
700  * Notes:
701  *    This is *only* called when we are examining the status of commands
702  *    queued during error recovery.  the main difference here is that we
703  *    don't allow for the possibility of retries here, and we are a lot
704  *    more restrictive about what we consider acceptable.
705  */
706 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
707 {
708         /*
709          * first check the host byte, to see if there is anything in there
710          * that would indicate what we need to do.
711          */
712         if (host_byte(scmd->result) == DID_RESET) {
713                 /*
714                  * rats.  we are already in the error handler, so we now
715                  * get to try and figure out what to do next.  if the sense
716                  * is valid, we have a pretty good idea of what to do.
717                  * if not, we mark it as FAILED.
718                  */
719                 return scsi_check_sense(scmd);
720         }
721         if (host_byte(scmd->result) != DID_OK)
722                 return FAILED;
723
724         /*
725          * next, check the message byte.
726          */
727         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
728                 return FAILED;
729
730         /*
731          * now, check the status byte to see if this indicates
732          * anything special.
733          */
734         switch (status_byte(scmd->result)) {
735         case GOOD:
736                 scsi_handle_queue_ramp_up(scmd->device);
737                 /* FALLTHROUGH */
738         case COMMAND_TERMINATED:
739                 return SUCCESS;
740         case CHECK_CONDITION:
741                 return scsi_check_sense(scmd);
742         case CONDITION_GOOD:
743         case INTERMEDIATE_GOOD:
744         case INTERMEDIATE_C_GOOD:
745                 /*
746                  * who knows?  FIXME(eric)
747                  */
748                 return SUCCESS;
749         case RESERVATION_CONFLICT:
750                 if (scmd->cmnd[0] == TEST_UNIT_READY)
751                         /* it is a success, we probed the device and
752                          * found it */
753                         return SUCCESS;
754                 /* otherwise, we failed to send the command */
755                 return FAILED;
756         case QUEUE_FULL:
757                 scsi_handle_queue_full(scmd->device);
758                 /* fall through */
759         case BUSY:
760                 return NEEDS_RETRY;
761         default:
762                 return FAILED;
763         }
764         return FAILED;
765 }
766
767 /**
768  * scsi_eh_done - Completion function for error handling.
769  * @scmd:       Cmd that is done.
770  */
771 static void scsi_eh_done(struct scsi_cmnd *scmd)
772 {
773         struct completion *eh_action;
774
775         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
776                         "%s result: %x\n", __func__, scmd->result));
777
778         eh_action = scmd->device->host->eh_action;
779         if (eh_action)
780                 complete(eh_action);
781 }
782
783 /**
784  * scsi_try_host_reset - ask host adapter to reset itself
785  * @scmd:       SCSI cmd to send host reset.
786  */
787 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
788 {
789         unsigned long flags;
790         int rtn;
791         struct Scsi_Host *host = scmd->device->host;
792         struct scsi_host_template *hostt = host->hostt;
793
794         SCSI_LOG_ERROR_RECOVERY(3,
795                 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
796
797         if (!hostt->eh_host_reset_handler)
798                 return FAILED;
799
800         rtn = hostt->eh_host_reset_handler(scmd);
801
802         if (rtn == SUCCESS) {
803                 if (!hostt->skip_settle_delay)
804                         ssleep(HOST_RESET_SETTLE_TIME);
805                 spin_lock_irqsave(host->host_lock, flags);
806                 scsi_report_bus_reset(host, scmd_channel(scmd));
807                 spin_unlock_irqrestore(host->host_lock, flags);
808         }
809
810         return rtn;
811 }
812
813 /**
814  * scsi_try_bus_reset - ask host to perform a bus reset
815  * @scmd:       SCSI cmd to send bus reset.
816  */
817 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
818 {
819         unsigned long flags;
820         int rtn;
821         struct Scsi_Host *host = scmd->device->host;
822         struct scsi_host_template *hostt = host->hostt;
823
824         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
825                 "%s: Snd Bus RST\n", __func__));
826
827         if (!hostt->eh_bus_reset_handler)
828                 return FAILED;
829
830         rtn = hostt->eh_bus_reset_handler(scmd);
831
832         if (rtn == SUCCESS) {
833                 if (!hostt->skip_settle_delay)
834                         ssleep(BUS_RESET_SETTLE_TIME);
835                 spin_lock_irqsave(host->host_lock, flags);
836                 scsi_report_bus_reset(host, scmd_channel(scmd));
837                 spin_unlock_irqrestore(host->host_lock, flags);
838         }
839
840         return rtn;
841 }
842
843 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
844 {
845         sdev->was_reset = 1;
846         sdev->expecting_cc_ua = 1;
847 }
848
849 /**
850  * scsi_try_target_reset - Ask host to perform a target reset
851  * @scmd:       SCSI cmd used to send a target reset
852  *
853  * Notes:
854  *    There is no timeout for this operation.  if this operation is
855  *    unreliable for a given host, then the host itself needs to put a
856  *    timer on it, and set the host back to a consistent state prior to
857  *    returning.
858  */
859 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
860 {
861         unsigned long flags;
862         int rtn;
863         struct Scsi_Host *host = scmd->device->host;
864         struct scsi_host_template *hostt = host->hostt;
865
866         if (!hostt->eh_target_reset_handler)
867                 return FAILED;
868
869         rtn = hostt->eh_target_reset_handler(scmd);
870         if (rtn == SUCCESS) {
871                 spin_lock_irqsave(host->host_lock, flags);
872                 __starget_for_each_device(scsi_target(scmd->device), NULL,
873                                           __scsi_report_device_reset);
874                 spin_unlock_irqrestore(host->host_lock, flags);
875         }
876
877         return rtn;
878 }
879
880 /**
881  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
882  * @scmd:       SCSI cmd used to send BDR
883  *
884  * Notes:
885  *    There is no timeout for this operation.  if this operation is
886  *    unreliable for a given host, then the host itself needs to put a
887  *    timer on it, and set the host back to a consistent state prior to
888  *    returning.
889  */
890 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
891 {
892         int rtn;
893         struct scsi_host_template *hostt = scmd->device->host->hostt;
894
895         if (!hostt->eh_device_reset_handler)
896                 return FAILED;
897
898         rtn = hostt->eh_device_reset_handler(scmd);
899         if (rtn == SUCCESS)
900                 __scsi_report_device_reset(scmd->device, NULL);
901         return rtn;
902 }
903
904 /**
905  * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
906  * @hostt:      SCSI driver host template
907  * @scmd:       SCSI cmd used to send a target reset
908  *
909  * Return value:
910  *      SUCCESS, FAILED, or FAST_IO_FAIL
911  *
912  * Notes:
913  *    SUCCESS does not necessarily indicate that the command
914  *    has been aborted; it only indicates that the LLDDs
915  *    has cleared all references to that command.
916  *    LLDDs should return FAILED only if an abort was required
917  *    but could not be executed. LLDDs should return FAST_IO_FAIL
918  *    if the device is temporarily unavailable (eg due to a
919  *    link down on FibreChannel)
920  */
921 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
922                                  struct scsi_cmnd *scmd)
923 {
924         if (!hostt->eh_abort_handler)
925                 return FAILED;
926
927         return hostt->eh_abort_handler(scmd);
928 }
929
930 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
931 {
932         if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
933                 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
934                         if (scsi_try_target_reset(scmd) != SUCCESS)
935                                 if (scsi_try_bus_reset(scmd) != SUCCESS)
936                                         scsi_try_host_reset(scmd);
937 }
938
939 /**
940  * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recovery
941  * @scmd:       SCSI command structure to hijack
942  * @ses:        structure to save restore information
943  * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
944  * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
945  * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
946  *
947  * This function is used to save a scsi command information before re-execution
948  * as part of the error recovery process.  If @sense_bytes is 0 the command
949  * sent must be one that does not transfer any data.  If @sense_bytes != 0
950  * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
951  * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
952  */
953 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
954                         unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
955 {
956         struct scsi_device *sdev = scmd->device;
957
958         /*
959          * We need saved copies of a number of fields - this is because
960          * error handling may need to overwrite these with different values
961          * to run different commands, and once error handling is complete,
962          * we will need to restore these values prior to running the actual
963          * command.
964          */
965         ses->cmd_len = scmd->cmd_len;
966         ses->cmnd = scmd->cmnd;
967         ses->data_direction = scmd->sc_data_direction;
968         ses->sdb = scmd->sdb;
969         ses->result = scmd->result;
970         ses->underflow = scmd->underflow;
971         ses->prot_op = scmd->prot_op;
972         ses->eh_eflags = scmd->eh_eflags;
973
974         scmd->prot_op = SCSI_PROT_NORMAL;
975         scmd->eh_eflags = 0;
976         scmd->cmnd = ses->eh_cmnd;
977         memset(scmd->cmnd, 0, BLK_MAX_CDB);
978         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
979         scmd->result = 0;
980
981         if (sense_bytes) {
982                 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
983                                          sense_bytes);
984                 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
985                             scmd->sdb.length);
986                 scmd->sdb.table.sgl = &ses->sense_sgl;
987                 scmd->sc_data_direction = DMA_FROM_DEVICE;
988                 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
989                 scmd->cmnd[0] = REQUEST_SENSE;
990                 scmd->cmnd[4] = scmd->sdb.length;
991                 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
992         } else {
993                 scmd->sc_data_direction = DMA_NONE;
994                 if (cmnd) {
995                         BUG_ON(cmnd_size > BLK_MAX_CDB);
996                         memcpy(scmd->cmnd, cmnd, cmnd_size);
997                         scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
998                 }
999         }
1000
1001         scmd->underflow = 0;
1002
1003         if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1004                 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1005                         (sdev->lun << 5 & 0xe0);
1006
1007         /*
1008          * Zero the sense buffer.  The scsi spec mandates that any
1009          * untransferred sense data should be interpreted as being zero.
1010          */
1011         memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1012 }
1013 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1014
1015 /**
1016  * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recovery
1017  * @scmd:       SCSI command structure to restore
1018  * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
1019  *
1020  * Undo any damage done by above scsi_eh_prep_cmnd().
1021  */
1022 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1023 {
1024         /*
1025          * Restore original data
1026          */
1027         scmd->cmd_len = ses->cmd_len;
1028         scmd->cmnd = ses->cmnd;
1029         scmd->sc_data_direction = ses->data_direction;
1030         scmd->sdb = ses->sdb;
1031         scmd->result = ses->result;
1032         scmd->underflow = ses->underflow;
1033         scmd->prot_op = ses->prot_op;
1034         scmd->eh_eflags = ses->eh_eflags;
1035 }
1036 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1037
1038 /**
1039  * scsi_send_eh_cmnd  - submit a scsi command as part of error recovery
1040  * @scmd:       SCSI command structure to hijack
1041  * @cmnd:       CDB to send
1042  * @cmnd_size:  size in bytes of @cmnd
1043  * @timeout:    timeout for this request
1044  * @sense_bytes: size of sense data to copy or 0
1045  *
1046  * This function is used to send a scsi command down to a target device
1047  * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1048  *
1049  * Return value:
1050  *    SUCCESS or FAILED or NEEDS_RETRY
1051  */
1052 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1053                              int cmnd_size, int timeout, unsigned sense_bytes)
1054 {
1055         struct scsi_device *sdev = scmd->device;
1056         struct Scsi_Host *shost = sdev->host;
1057         DECLARE_COMPLETION_ONSTACK(done);
1058         unsigned long timeleft = timeout;
1059         struct scsi_eh_save ses;
1060         const unsigned long stall_for = msecs_to_jiffies(100);
1061         int rtn;
1062
1063 retry:
1064         scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1065         shost->eh_action = &done;
1066
1067         scsi_log_send(scmd);
1068         scmd->scsi_done = scsi_eh_done;
1069         rtn = shost->hostt->queuecommand(shost, scmd);
1070         if (rtn) {
1071                 if (timeleft > stall_for) {
1072                         scsi_eh_restore_cmnd(scmd, &ses);
1073                         timeleft -= stall_for;
1074                         msleep(jiffies_to_msecs(stall_for));
1075                         goto retry;
1076                 }
1077                 /* signal not to enter either branch of the if () below */
1078                 timeleft = 0;
1079                 rtn = FAILED;
1080         } else {
1081                 timeleft = wait_for_completion_timeout(&done, timeout);
1082                 rtn = SUCCESS;
1083         }
1084
1085         shost->eh_action = NULL;
1086
1087         scsi_log_completion(scmd, rtn);
1088
1089         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1090                         "%s timeleft: %ld\n",
1091                         __func__, timeleft));
1092
1093         /*
1094          * If there is time left scsi_eh_done got called, and we will examine
1095          * the actual status codes to see whether the command actually did
1096          * complete normally, else if we have a zero return and no time left,
1097          * the command must still be pending, so abort it and return FAILED.
1098          * If we never actually managed to issue the command, because
1099          * ->queuecommand() kept returning non zero, use the rtn = FAILED
1100          * value above (so don't execute either branch of the if)
1101          */
1102         if (timeleft) {
1103                 rtn = scsi_eh_completed_normally(scmd);
1104                 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1105                         "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1106
1107                 switch (rtn) {
1108                 case SUCCESS:
1109                 case NEEDS_RETRY:
1110                 case FAILED:
1111                         break;
1112                 case ADD_TO_MLQUEUE:
1113                         rtn = NEEDS_RETRY;
1114                         break;
1115                 default:
1116                         rtn = FAILED;
1117                         break;
1118                 }
1119         } else if (rtn != FAILED) {
1120                 scsi_abort_eh_cmnd(scmd);
1121                 rtn = FAILED;
1122         }
1123
1124         scsi_eh_restore_cmnd(scmd, &ses);
1125
1126         return rtn;
1127 }
1128
1129 /**
1130  * scsi_request_sense - Request sense data from a particular target.
1131  * @scmd:       SCSI cmd for request sense.
1132  *
1133  * Notes:
1134  *    Some hosts automatically obtain this information, others require
1135  *    that we obtain it on our own. This function will *not* return until
1136  *    the command either times out, or it completes.
1137  */
1138 static int scsi_request_sense(struct scsi_cmnd *scmd)
1139 {
1140         return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1141 }
1142
1143 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1144 {
1145         if (!blk_rq_is_passthrough(scmd->request)) {
1146                 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1147                 if (sdrv->eh_action)
1148                         rtn = sdrv->eh_action(scmd, rtn);
1149         }
1150         return rtn;
1151 }
1152
1153 /**
1154  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1155  * @scmd:       Original SCSI cmd that eh has finished.
1156  * @done_q:     Queue for processed commands.
1157  *
1158  * Notes:
1159  *    We don't want to use the normal command completion while we are are
1160  *    still handling errors - it may cause other commands to be queued,
1161  *    and that would disturb what we are doing.  Thus we really want to
1162  *    keep a list of pending commands for final completion, and once we
1163  *    are ready to leave error handling we handle completion for real.
1164  */
1165 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1166 {
1167         list_move_tail(&scmd->eh_entry, done_q);
1168 }
1169 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1170
1171 /**
1172  * scsi_eh_get_sense - Get device sense data.
1173  * @work_q:     Queue of commands to process.
1174  * @done_q:     Queue of processed commands.
1175  *
1176  * Description:
1177  *    See if we need to request sense information.  if so, then get it
1178  *    now, so we have a better idea of what to do.
1179  *
1180  * Notes:
1181  *    This has the unfortunate side effect that if a shost adapter does
1182  *    not automatically request sense information, we end up shutting
1183  *    it down before we request it.
1184  *
1185  *    All drivers should request sense information internally these days,
1186  *    so for now all I have to say is tough noogies if you end up in here.
1187  *
1188  *    XXX: Long term this code should go away, but that needs an audit of
1189  *         all LLDDs first.
1190  */
1191 int scsi_eh_get_sense(struct list_head *work_q,
1192                       struct list_head *done_q)
1193 {
1194         struct scsi_cmnd *scmd, *next;
1195         struct Scsi_Host *shost;
1196         int rtn;
1197
1198         /*
1199          * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1200          * should not get sense.
1201          */
1202         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1203                 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1204                     SCSI_SENSE_VALID(scmd))
1205                         continue;
1206
1207                 shost = scmd->device->host;
1208                 if (scsi_host_eh_past_deadline(shost)) {
1209                         SCSI_LOG_ERROR_RECOVERY(3,
1210                                 scmd_printk(KERN_INFO, scmd,
1211                                             "%s: skip request sense, past eh deadline\n",
1212                                              current->comm));
1213                         break;
1214                 }
1215                 if (status_byte(scmd->result) != CHECK_CONDITION)
1216                         /*
1217                          * don't request sense if there's no check condition
1218                          * status because the error we're processing isn't one
1219                          * that has a sense code (and some devices get
1220                          * confused by sense requests out of the blue)
1221                          */
1222                         continue;
1223
1224                 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1225                                                   "%s: requesting sense\n",
1226                                                   current->comm));
1227                 rtn = scsi_request_sense(scmd);
1228                 if (rtn != SUCCESS)
1229                         continue;
1230
1231                 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1232                         "sense requested, result %x\n", scmd->result));
1233                 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1234
1235                 rtn = scsi_decide_disposition(scmd);
1236
1237                 /*
1238                  * if the result was normal, then just pass it along to the
1239                  * upper level.
1240                  */
1241                 if (rtn == SUCCESS)
1242                         /* we don't want this command reissued, just
1243                          * finished with the sense data, so set
1244                          * retries to the max allowed to ensure it
1245                          * won't get reissued */
1246                         scmd->retries = scmd->allowed;
1247                 else if (rtn != NEEDS_RETRY)
1248                         continue;
1249
1250                 scsi_eh_finish_cmd(scmd, done_q);
1251         }
1252
1253         return list_empty(work_q);
1254 }
1255 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1256
1257 /**
1258  * scsi_eh_tur - Send TUR to device.
1259  * @scmd:       &scsi_cmnd to send TUR
1260  *
1261  * Return value:
1262  *    0 - Device is ready. 1 - Device NOT ready.
1263  */
1264 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1265 {
1266         static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1267         int retry_cnt = 1, rtn;
1268
1269 retry_tur:
1270         rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1271                                 scmd->device->eh_timeout, 0);
1272
1273         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1274                 "%s return: %x\n", __func__, rtn));
1275
1276         switch (rtn) {
1277         case NEEDS_RETRY:
1278                 if (retry_cnt--)
1279                         goto retry_tur;
1280                 /*FALLTHRU*/
1281         case SUCCESS:
1282                 return 0;
1283         default:
1284                 return 1;
1285         }
1286 }
1287
1288 /**
1289  * scsi_eh_test_devices - check if devices are responding from error recovery.
1290  * @cmd_list:   scsi commands in error recovery.
1291  * @work_q:     queue for commands which still need more error recovery
1292  * @done_q:     queue for commands which are finished
1293  * @try_stu:    boolean on if a STU command should be tried in addition to TUR.
1294  *
1295  * Decription:
1296  *    Tests if devices are in a working state.  Commands to devices now in
1297  *    a working state are sent to the done_q while commands to devices which
1298  *    are still failing to respond are returned to the work_q for more
1299  *    processing.
1300  **/
1301 static int scsi_eh_test_devices(struct list_head *cmd_list,
1302                                 struct list_head *work_q,
1303                                 struct list_head *done_q, int try_stu)
1304 {
1305         struct scsi_cmnd *scmd, *next;
1306         struct scsi_device *sdev;
1307         int finish_cmds;
1308
1309         while (!list_empty(cmd_list)) {
1310                 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1311                 sdev = scmd->device;
1312
1313                 if (!try_stu) {
1314                         if (scsi_host_eh_past_deadline(sdev->host)) {
1315                                 /* Push items back onto work_q */
1316                                 list_splice_init(cmd_list, work_q);
1317                                 SCSI_LOG_ERROR_RECOVERY(3,
1318                                         sdev_printk(KERN_INFO, sdev,
1319                                                     "%s: skip test device, past eh deadline",
1320                                                     current->comm));
1321                                 break;
1322                         }
1323                 }
1324
1325                 finish_cmds = !scsi_device_online(scmd->device) ||
1326                         (try_stu && !scsi_eh_try_stu(scmd) &&
1327                          !scsi_eh_tur(scmd)) ||
1328                         !scsi_eh_tur(scmd);
1329
1330                 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1331                         if (scmd->device == sdev) {
1332                                 if (finish_cmds &&
1333                                     (try_stu ||
1334                                      scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1335                                         scsi_eh_finish_cmd(scmd, done_q);
1336                                 else
1337                                         list_move_tail(&scmd->eh_entry, work_q);
1338                         }
1339         }
1340         return list_empty(work_q);
1341 }
1342
1343 /**
1344  * scsi_eh_try_stu - Send START_UNIT to device.
1345  * @scmd:       &scsi_cmnd to send START_UNIT
1346  *
1347  * Return value:
1348  *    0 - Device is ready. 1 - Device NOT ready.
1349  */
1350 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1351 {
1352         static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1353
1354         if (scmd->device->allow_restart) {
1355                 int i, rtn = NEEDS_RETRY;
1356
1357                 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1358                         rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1359
1360                 if (rtn == SUCCESS)
1361                         return 0;
1362         }
1363
1364         return 1;
1365 }
1366
1367  /**
1368  * scsi_eh_stu - send START_UNIT if needed
1369  * @shost:      &scsi host being recovered.
1370  * @work_q:     &list_head for pending commands.
1371  * @done_q:     &list_head for processed commands.
1372  *
1373  * Notes:
1374  *    If commands are failing due to not ready, initializing command required,
1375  *      try revalidating the device, which will end up sending a start unit.
1376  */
1377 static int scsi_eh_stu(struct Scsi_Host *shost,
1378                               struct list_head *work_q,
1379                               struct list_head *done_q)
1380 {
1381         struct scsi_cmnd *scmd, *stu_scmd, *next;
1382         struct scsi_device *sdev;
1383
1384         shost_for_each_device(sdev, shost) {
1385                 if (scsi_host_eh_past_deadline(shost)) {
1386                         SCSI_LOG_ERROR_RECOVERY(3,
1387                                 sdev_printk(KERN_INFO, sdev,
1388                                             "%s: skip START_UNIT, past eh deadline\n",
1389                                             current->comm));
1390                         break;
1391                 }
1392                 stu_scmd = NULL;
1393                 list_for_each_entry(scmd, work_q, eh_entry)
1394                         if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1395                             scsi_check_sense(scmd) == FAILED ) {
1396                                 stu_scmd = scmd;
1397                                 break;
1398                         }
1399
1400                 if (!stu_scmd)
1401                         continue;
1402
1403                 SCSI_LOG_ERROR_RECOVERY(3,
1404                         sdev_printk(KERN_INFO, sdev,
1405                                      "%s: Sending START_UNIT\n",
1406                                     current->comm));
1407
1408                 if (!scsi_eh_try_stu(stu_scmd)) {
1409                         if (!scsi_device_online(sdev) ||
1410                             !scsi_eh_tur(stu_scmd)) {
1411                                 list_for_each_entry_safe(scmd, next,
1412                                                           work_q, eh_entry) {
1413                                         if (scmd->device == sdev &&
1414                                             scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1415                                                 scsi_eh_finish_cmd(scmd, done_q);
1416                                 }
1417                         }
1418                 } else {
1419                         SCSI_LOG_ERROR_RECOVERY(3,
1420                                 sdev_printk(KERN_INFO, sdev,
1421                                             "%s: START_UNIT failed\n",
1422                                             current->comm));
1423                 }
1424         }
1425
1426         return list_empty(work_q);
1427 }
1428
1429
1430 /**
1431  * scsi_eh_bus_device_reset - send bdr if needed
1432  * @shost:      scsi host being recovered.
1433  * @work_q:     &list_head for pending commands.
1434  * @done_q:     &list_head for processed commands.
1435  *
1436  * Notes:
1437  *    Try a bus device reset.  Still, look to see whether we have multiple
1438  *    devices that are jammed or not - if we have multiple devices, it
1439  *    makes no sense to try bus_device_reset - we really would need to try
1440  *    a bus_reset instead.
1441  */
1442 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1443                                     struct list_head *work_q,
1444                                     struct list_head *done_q)
1445 {
1446         struct scsi_cmnd *scmd, *bdr_scmd, *next;
1447         struct scsi_device *sdev;
1448         int rtn;
1449
1450         shost_for_each_device(sdev, shost) {
1451                 if (scsi_host_eh_past_deadline(shost)) {
1452                         SCSI_LOG_ERROR_RECOVERY(3,
1453                                 sdev_printk(KERN_INFO, sdev,
1454                                             "%s: skip BDR, past eh deadline\n",
1455                                              current->comm));
1456                         break;
1457                 }
1458                 bdr_scmd = NULL;
1459                 list_for_each_entry(scmd, work_q, eh_entry)
1460                         if (scmd->device == sdev) {
1461                                 bdr_scmd = scmd;
1462                                 break;
1463                         }
1464
1465                 if (!bdr_scmd)
1466                         continue;
1467
1468                 SCSI_LOG_ERROR_RECOVERY(3,
1469                         sdev_printk(KERN_INFO, sdev,
1470                                      "%s: Sending BDR\n", current->comm));
1471                 rtn = scsi_try_bus_device_reset(bdr_scmd);
1472                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1473                         if (!scsi_device_online(sdev) ||
1474                             rtn == FAST_IO_FAIL ||
1475                             !scsi_eh_tur(bdr_scmd)) {
1476                                 list_for_each_entry_safe(scmd, next,
1477                                                          work_q, eh_entry) {
1478                                         if (scmd->device == sdev &&
1479                                             scsi_eh_action(scmd, rtn) != FAILED)
1480                                                 scsi_eh_finish_cmd(scmd,
1481                                                                    done_q);
1482                                 }
1483                         }
1484                 } else {
1485                         SCSI_LOG_ERROR_RECOVERY(3,
1486                                 sdev_printk(KERN_INFO, sdev,
1487                                             "%s: BDR failed\n", current->comm));
1488                 }
1489         }
1490
1491         return list_empty(work_q);
1492 }
1493
1494 /**
1495  * scsi_eh_target_reset - send target reset if needed
1496  * @shost:      scsi host being recovered.
1497  * @work_q:     &list_head for pending commands.
1498  * @done_q:     &list_head for processed commands.
1499  *
1500  * Notes:
1501  *    Try a target reset.
1502  */
1503 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1504                                 struct list_head *work_q,
1505                                 struct list_head *done_q)
1506 {
1507         LIST_HEAD(tmp_list);
1508         LIST_HEAD(check_list);
1509
1510         list_splice_init(work_q, &tmp_list);
1511
1512         while (!list_empty(&tmp_list)) {
1513                 struct scsi_cmnd *next, *scmd;
1514                 int rtn;
1515                 unsigned int id;
1516
1517                 if (scsi_host_eh_past_deadline(shost)) {
1518                         /* push back on work queue for further processing */
1519                         list_splice_init(&check_list, work_q);
1520                         list_splice_init(&tmp_list, work_q);
1521                         SCSI_LOG_ERROR_RECOVERY(3,
1522                                 shost_printk(KERN_INFO, shost,
1523                                             "%s: Skip target reset, past eh deadline\n",
1524                                              current->comm));
1525                         return list_empty(work_q);
1526                 }
1527
1528                 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1529                 id = scmd_id(scmd);
1530
1531                 SCSI_LOG_ERROR_RECOVERY(3,
1532                         shost_printk(KERN_INFO, shost,
1533                                      "%s: Sending target reset to target %d\n",
1534                                      current->comm, id));
1535                 rtn = scsi_try_target_reset(scmd);
1536                 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1537                         SCSI_LOG_ERROR_RECOVERY(3,
1538                                 shost_printk(KERN_INFO, shost,
1539                                              "%s: Target reset failed"
1540                                              " target: %d\n",
1541                                              current->comm, id));
1542                 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1543                         if (scmd_id(scmd) != id)
1544                                 continue;
1545
1546                         if (rtn == SUCCESS)
1547                                 list_move_tail(&scmd->eh_entry, &check_list);
1548                         else if (rtn == FAST_IO_FAIL)
1549                                 scsi_eh_finish_cmd(scmd, done_q);
1550                         else
1551                                 /* push back on work queue for further processing */
1552                                 list_move(&scmd->eh_entry, work_q);
1553                 }
1554         }
1555
1556         return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1557 }
1558
1559 /**
1560  * scsi_eh_bus_reset - send a bus reset
1561  * @shost:      &scsi host being recovered.
1562  * @work_q:     &list_head for pending commands.
1563  * @done_q:     &list_head for processed commands.
1564  */
1565 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1566                              struct list_head *work_q,
1567                              struct list_head *done_q)
1568 {
1569         struct scsi_cmnd *scmd, *chan_scmd, *next;
1570         LIST_HEAD(check_list);
1571         unsigned int channel;
1572         int rtn;
1573
1574         /*
1575          * we really want to loop over the various channels, and do this on
1576          * a channel by channel basis.  we should also check to see if any
1577          * of the failed commands are on soft_reset devices, and if so, skip
1578          * the reset.
1579          */
1580
1581         for (channel = 0; channel <= shost->max_channel; channel++) {
1582                 if (scsi_host_eh_past_deadline(shost)) {
1583                         list_splice_init(&check_list, work_q);
1584                         SCSI_LOG_ERROR_RECOVERY(3,
1585                                 shost_printk(KERN_INFO, shost,
1586                                             "%s: skip BRST, past eh deadline\n",
1587                                              current->comm));
1588                         return list_empty(work_q);
1589                 }
1590
1591                 chan_scmd = NULL;
1592                 list_for_each_entry(scmd, work_q, eh_entry) {
1593                         if (channel == scmd_channel(scmd)) {
1594                                 chan_scmd = scmd;
1595                                 break;
1596                                 /*
1597                                  * FIXME add back in some support for
1598                                  * soft_reset devices.
1599                                  */
1600                         }
1601                 }
1602
1603                 if (!chan_scmd)
1604                         continue;
1605                 SCSI_LOG_ERROR_RECOVERY(3,
1606                         shost_printk(KERN_INFO, shost,
1607                                      "%s: Sending BRST chan: %d\n",
1608                                      current->comm, channel));
1609                 rtn = scsi_try_bus_reset(chan_scmd);
1610                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1611                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1612                                 if (channel == scmd_channel(scmd)) {
1613                                         if (rtn == FAST_IO_FAIL)
1614                                                 scsi_eh_finish_cmd(scmd,
1615                                                                    done_q);
1616                                         else
1617                                                 list_move_tail(&scmd->eh_entry,
1618                                                                &check_list);
1619                                 }
1620                         }
1621                 } else {
1622                         SCSI_LOG_ERROR_RECOVERY(3,
1623                                 shost_printk(KERN_INFO, shost,
1624                                              "%s: BRST failed chan: %d\n",
1625                                              current->comm, channel));
1626                 }
1627         }
1628         return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1629 }
1630
1631 /**
1632  * scsi_eh_host_reset - send a host reset
1633  * @shost:      host to be reset.
1634  * @work_q:     &list_head for pending commands.
1635  * @done_q:     &list_head for processed commands.
1636  */
1637 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1638                               struct list_head *work_q,
1639                               struct list_head *done_q)
1640 {
1641         struct scsi_cmnd *scmd, *next;
1642         LIST_HEAD(check_list);
1643         int rtn;
1644
1645         if (!list_empty(work_q)) {
1646                 scmd = list_entry(work_q->next,
1647                                   struct scsi_cmnd, eh_entry);
1648
1649                 SCSI_LOG_ERROR_RECOVERY(3,
1650                         shost_printk(KERN_INFO, shost,
1651                                      "%s: Sending HRST\n",
1652                                      current->comm));
1653
1654                 rtn = scsi_try_host_reset(scmd);
1655                 if (rtn == SUCCESS) {
1656                         list_splice_init(work_q, &check_list);
1657                 } else if (rtn == FAST_IO_FAIL) {
1658                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1659                                         scsi_eh_finish_cmd(scmd, done_q);
1660                         }
1661                 } else {
1662                         SCSI_LOG_ERROR_RECOVERY(3,
1663                                 shost_printk(KERN_INFO, shost,
1664                                              "%s: HRST failed\n",
1665                                              current->comm));
1666                 }
1667         }
1668         return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1669 }
1670
1671 /**
1672  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1673  * @work_q:     &list_head for pending commands.
1674  * @done_q:     &list_head for processed commands.
1675  */
1676 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1677                                   struct list_head *done_q)
1678 {
1679         struct scsi_cmnd *scmd, *next;
1680         struct scsi_device *sdev;
1681
1682         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1683                 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1684                             "not ready after error recovery\n");
1685                 sdev = scmd->device;
1686
1687                 mutex_lock(&sdev->state_mutex);
1688                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1689                 mutex_unlock(&sdev->state_mutex);
1690
1691                 scsi_eh_finish_cmd(scmd, done_q);
1692         }
1693         return;
1694 }
1695
1696 /**
1697  * scsi_noretry_cmd - determine if command should be failed fast
1698  * @scmd:       SCSI cmd to examine.
1699  */
1700 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1701 {
1702         switch (host_byte(scmd->result)) {
1703         case DID_OK:
1704                 break;
1705         case DID_TIME_OUT:
1706                 goto check_type;
1707         case DID_BUS_BUSY:
1708                 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1709         case DID_PARITY:
1710                 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1711         case DID_ERROR:
1712                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1713                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1714                         return 0;
1715                 /* fall through */
1716         case DID_SOFT_ERROR:
1717                 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1718         }
1719
1720         if (status_byte(scmd->result) != CHECK_CONDITION)
1721                 return 0;
1722
1723 check_type:
1724         /*
1725          * assume caller has checked sense and determined
1726          * the check condition was retryable.
1727          */
1728         if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1729             blk_rq_is_passthrough(scmd->request))
1730                 return 1;
1731         else
1732                 return 0;
1733 }
1734
1735 /**
1736  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1737  * @scmd:       SCSI cmd to examine.
1738  *
1739  * Notes:
1740  *    This is *only* called when we are examining the status after sending
1741  *    out the actual data command.  any commands that are queued for error
1742  *    recovery (e.g. test_unit_ready) do *not* come through here.
1743  *
1744  *    When this routine returns failed, it means the error handler thread
1745  *    is woken.  In cases where the error code indicates an error that
1746  *    doesn't require the error handler read (i.e. we don't need to
1747  *    abort/reset), this function should return SUCCESS.
1748  */
1749 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1750 {
1751         int rtn;
1752
1753         /*
1754          * if the device is offline, then we clearly just pass the result back
1755          * up to the top level.
1756          */
1757         if (!scsi_device_online(scmd->device)) {
1758                 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1759                         "%s: device offline - report as SUCCESS\n", __func__));
1760                 return SUCCESS;
1761         }
1762
1763         /*
1764          * first check the host byte, to see if there is anything in there
1765          * that would indicate what we need to do.
1766          */
1767         switch (host_byte(scmd->result)) {
1768         case DID_PASSTHROUGH:
1769                 /*
1770                  * no matter what, pass this through to the upper layer.
1771                  * nuke this special code so that it looks like we are saying
1772                  * did_ok.
1773                  */
1774                 scmd->result &= 0xff00ffff;
1775                 return SUCCESS;
1776         case DID_OK:
1777                 /*
1778                  * looks good.  drop through, and check the next byte.
1779                  */
1780                 break;
1781         case DID_ABORT:
1782                 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1783                         set_host_byte(scmd, DID_TIME_OUT);
1784                         return SUCCESS;
1785                 }
1786                 /* FALLTHROUGH */
1787         case DID_NO_CONNECT:
1788         case DID_BAD_TARGET:
1789                 /*
1790                  * note - this means that we just report the status back
1791                  * to the top level driver, not that we actually think
1792                  * that it indicates SUCCESS.
1793                  */
1794                 return SUCCESS;
1795         case DID_SOFT_ERROR:
1796                 /*
1797                  * when the low level driver returns did_soft_error,
1798                  * it is responsible for keeping an internal retry counter
1799                  * in order to avoid endless loops (db)
1800                  */
1801                 goto maybe_retry;
1802         case DID_IMM_RETRY:
1803                 return NEEDS_RETRY;
1804
1805         case DID_REQUEUE:
1806                 return ADD_TO_MLQUEUE;
1807         case DID_TRANSPORT_DISRUPTED:
1808                 /*
1809                  * LLD/transport was disrupted during processing of the IO.
1810                  * The transport class is now blocked/blocking,
1811                  * and the transport will decide what to do with the IO
1812                  * based on its timers and recovery capablilities if
1813                  * there are enough retries.
1814                  */
1815                 goto maybe_retry;
1816         case DID_TRANSPORT_FAILFAST:
1817                 /*
1818                  * The transport decided to failfast the IO (most likely
1819                  * the fast io fail tmo fired), so send IO directly upwards.
1820                  */
1821                 return SUCCESS;
1822         case DID_ERROR:
1823                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1824                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1825                         /*
1826                          * execute reservation conflict processing code
1827                          * lower down
1828                          */
1829                         break;
1830                 /* fallthrough */
1831         case DID_BUS_BUSY:
1832         case DID_PARITY:
1833                 goto maybe_retry;
1834         case DID_TIME_OUT:
1835                 /*
1836                  * when we scan the bus, we get timeout messages for
1837                  * these commands if there is no device available.
1838                  * other hosts report did_no_connect for the same thing.
1839                  */
1840                 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1841                      scmd->cmnd[0] == INQUIRY)) {
1842                         return SUCCESS;
1843                 } else {
1844                         return FAILED;
1845                 }
1846         case DID_RESET:
1847                 return SUCCESS;
1848         default:
1849                 return FAILED;
1850         }
1851
1852         /*
1853          * next, check the message byte.
1854          */
1855         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1856                 return FAILED;
1857
1858         /*
1859          * check the status byte to see if this indicates anything special.
1860          */
1861         switch (status_byte(scmd->result)) {
1862         case QUEUE_FULL:
1863                 scsi_handle_queue_full(scmd->device);
1864                 /*
1865                  * the case of trying to send too many commands to a
1866                  * tagged queueing device.
1867                  */
1868                 /* FALLTHROUGH */
1869         case BUSY:
1870                 /*
1871                  * device can't talk to us at the moment.  Should only
1872                  * occur (SAM-3) when the task queue is empty, so will cause
1873                  * the empty queue handling to trigger a stall in the
1874                  * device.
1875                  */
1876                 return ADD_TO_MLQUEUE;
1877         case GOOD:
1878                 if (scmd->cmnd[0] == REPORT_LUNS)
1879                         scmd->device->sdev_target->expecting_lun_change = 0;
1880                 scsi_handle_queue_ramp_up(scmd->device);
1881                 /* FALLTHROUGH */
1882         case COMMAND_TERMINATED:
1883                 return SUCCESS;
1884         case TASK_ABORTED:
1885                 goto maybe_retry;
1886         case CHECK_CONDITION:
1887                 rtn = scsi_check_sense(scmd);
1888                 if (rtn == NEEDS_RETRY)
1889                         goto maybe_retry;
1890                 /* if rtn == FAILED, we have no sense information;
1891                  * returning FAILED will wake the error handler thread
1892                  * to collect the sense and redo the decide
1893                  * disposition */
1894                 return rtn;
1895         case CONDITION_GOOD:
1896         case INTERMEDIATE_GOOD:
1897         case INTERMEDIATE_C_GOOD:
1898         case ACA_ACTIVE:
1899                 /*
1900                  * who knows?  FIXME(eric)
1901                  */
1902                 return SUCCESS;
1903
1904         case RESERVATION_CONFLICT:
1905                 sdev_printk(KERN_INFO, scmd->device,
1906                             "reservation conflict\n");
1907                 set_host_byte(scmd, DID_NEXUS_FAILURE);
1908                 return SUCCESS; /* causes immediate i/o error */
1909         default:
1910                 return FAILED;
1911         }
1912         return FAILED;
1913
1914 maybe_retry:
1915
1916         /* we requeue for retry because the error was retryable, and
1917          * the request was not marked fast fail.  Note that above,
1918          * even if the request is marked fast fail, we still requeue
1919          * for queue congestion conditions (QUEUE_FULL or BUSY) */
1920         if ((++scmd->retries) <= scmd->allowed
1921             && !scsi_noretry_cmd(scmd)) {
1922                 return NEEDS_RETRY;
1923         } else {
1924                 /*
1925                  * no more retries - report this one back to upper level.
1926                  */
1927                 return SUCCESS;
1928         }
1929 }
1930
1931 static void eh_lock_door_done(struct request *req, blk_status_t status)
1932 {
1933         blk_put_request(req);
1934 }
1935
1936 /**
1937  * scsi_eh_lock_door - Prevent medium removal for the specified device
1938  * @sdev:       SCSI device to prevent medium removal
1939  *
1940  * Locking:
1941  *      We must be called from process context.
1942  *
1943  * Notes:
1944  *      We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1945  *      head of the devices request queue, and continue.
1946  */
1947 static void scsi_eh_lock_door(struct scsi_device *sdev)
1948 {
1949         struct request *req;
1950         struct scsi_request *rq;
1951
1952         req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, 0);
1953         if (IS_ERR(req))
1954                 return;
1955         rq = scsi_req(req);
1956
1957         rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1958         rq->cmd[1] = 0;
1959         rq->cmd[2] = 0;
1960         rq->cmd[3] = 0;
1961         rq->cmd[4] = SCSI_REMOVAL_PREVENT;
1962         rq->cmd[5] = 0;
1963         rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
1964
1965         req->rq_flags |= RQF_QUIET;
1966         req->timeout = 10 * HZ;
1967         rq->retries = 5;
1968
1969         blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1970 }
1971
1972 /**
1973  * scsi_restart_operations - restart io operations to the specified host.
1974  * @shost:      Host we are restarting.
1975  *
1976  * Notes:
1977  *    When we entered the error handler, we blocked all further i/o to
1978  *    this device.  we need to 'reverse' this process.
1979  */
1980 static void scsi_restart_operations(struct Scsi_Host *shost)
1981 {
1982         struct scsi_device *sdev;
1983         unsigned long flags;
1984
1985         /*
1986          * If the door was locked, we need to insert a door lock request
1987          * onto the head of the SCSI request queue for the device.  There
1988          * is no point trying to lock the door of an off-line device.
1989          */
1990         shost_for_each_device(sdev, shost) {
1991                 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
1992                         scsi_eh_lock_door(sdev);
1993                         sdev->was_reset = 0;
1994                 }
1995         }
1996
1997         /*
1998          * next free up anything directly waiting upon the host.  this
1999          * will be requests for character device operations, and also for
2000          * ioctls to queued block devices.
2001          */
2002         SCSI_LOG_ERROR_RECOVERY(3,
2003                 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2004
2005         spin_lock_irqsave(shost->host_lock, flags);
2006         if (scsi_host_set_state(shost, SHOST_RUNNING))
2007                 if (scsi_host_set_state(shost, SHOST_CANCEL))
2008                         BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2009         spin_unlock_irqrestore(shost->host_lock, flags);
2010
2011         wake_up(&shost->host_wait);
2012
2013         /*
2014          * finally we need to re-initiate requests that may be pending.  we will
2015          * have had everything blocked while error handling is taking place, and
2016          * now that error recovery is done, we will need to ensure that these
2017          * requests are started.
2018          */
2019         scsi_run_host_queues(shost);
2020
2021         /*
2022          * if eh is active and host_eh_scheduled is pending we need to re-run
2023          * recovery.  we do this check after scsi_run_host_queues() to allow
2024          * everything pent up since the last eh run a chance to make forward
2025          * progress before we sync again.  Either we'll immediately re-run
2026          * recovery or scsi_device_unbusy() will wake us again when these
2027          * pending commands complete.
2028          */
2029         spin_lock_irqsave(shost->host_lock, flags);
2030         if (shost->host_eh_scheduled)
2031                 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2032                         WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2033         spin_unlock_irqrestore(shost->host_lock, flags);
2034 }
2035
2036 /**
2037  * scsi_eh_ready_devs - check device ready state and recover if not.
2038  * @shost:      host to be recovered.
2039  * @work_q:     &list_head for pending commands.
2040  * @done_q:     &list_head for processed commands.
2041  */
2042 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2043                         struct list_head *work_q,
2044                         struct list_head *done_q)
2045 {
2046         if (!scsi_eh_stu(shost, work_q, done_q))
2047                 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2048                         if (!scsi_eh_target_reset(shost, work_q, done_q))
2049                                 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2050                                         if (!scsi_eh_host_reset(shost, work_q, done_q))
2051                                                 scsi_eh_offline_sdevs(work_q,
2052                                                                       done_q);
2053 }
2054 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2055
2056 /**
2057  * scsi_eh_flush_done_q - finish processed commands or retry them.
2058  * @done_q:     list_head of processed commands.
2059  */
2060 void scsi_eh_flush_done_q(struct list_head *done_q)
2061 {
2062         struct scsi_cmnd *scmd, *next;
2063
2064         list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2065                 list_del_init(&scmd->eh_entry);
2066                 if (scsi_device_online(scmd->device) &&
2067                     !scsi_noretry_cmd(scmd) &&
2068                     (++scmd->retries <= scmd->allowed)) {
2069                         SCSI_LOG_ERROR_RECOVERY(3,
2070                                 scmd_printk(KERN_INFO, scmd,
2071                                              "%s: flush retry cmd\n",
2072                                              current->comm));
2073                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2074                 } else {
2075                         /*
2076                          * If just we got sense for the device (called
2077                          * scsi_eh_get_sense), scmd->result is already
2078                          * set, do not set DRIVER_TIMEOUT.
2079                          */
2080                         if (!scmd->result)
2081                                 scmd->result |= (DRIVER_TIMEOUT << 24);
2082                         SCSI_LOG_ERROR_RECOVERY(3,
2083                                 scmd_printk(KERN_INFO, scmd,
2084                                              "%s: flush finish cmd\n",
2085                                              current->comm));
2086                         scsi_finish_command(scmd);
2087                 }
2088         }
2089 }
2090 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2091
2092 /**
2093  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2094  * @shost:      Host to unjam.
2095  *
2096  * Notes:
2097  *    When we come in here, we *know* that all commands on the bus have
2098  *    either completed, failed or timed out.  we also know that no further
2099  *    commands are being sent to the host, so things are relatively quiet
2100  *    and we have freedom to fiddle with things as we wish.
2101  *
2102  *    This is only the *default* implementation.  it is possible for
2103  *    individual drivers to supply their own version of this function, and
2104  *    if the maintainer wishes to do this, it is strongly suggested that
2105  *    this function be taken as a template and modified.  this function
2106  *    was designed to correctly handle problems for about 95% of the
2107  *    different cases out there, and it should always provide at least a
2108  *    reasonable amount of error recovery.
2109  *
2110  *    Any command marked 'failed' or 'timeout' must eventually have
2111  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
2112  *    here, so when we restart the host after we return it should have an
2113  *    empty queue.
2114  */
2115 static void scsi_unjam_host(struct Scsi_Host *shost)
2116 {
2117         unsigned long flags;
2118         LIST_HEAD(eh_work_q);
2119         LIST_HEAD(eh_done_q);
2120
2121         spin_lock_irqsave(shost->host_lock, flags);
2122         list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2123         spin_unlock_irqrestore(shost->host_lock, flags);
2124
2125         SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2126
2127         if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2128                 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2129
2130         spin_lock_irqsave(shost->host_lock, flags);
2131         if (shost->eh_deadline != -1)
2132                 shost->last_reset = 0;
2133         spin_unlock_irqrestore(shost->host_lock, flags);
2134         scsi_eh_flush_done_q(&eh_done_q);
2135 }
2136
2137 /**
2138  * scsi_error_handler - SCSI error handler thread
2139  * @data:       Host for which we are running.
2140  *
2141  * Notes:
2142  *    This is the main error handling loop.  This is run as a kernel thread
2143  *    for every SCSI host and handles all error handling activity.
2144  */
2145 int scsi_error_handler(void *data)
2146 {
2147         struct Scsi_Host *shost = data;
2148
2149         /*
2150          * We use TASK_INTERRUPTIBLE so that the thread is not
2151          * counted against the load average as a running process.
2152          * We never actually get interrupted because kthread_run
2153          * disables signal delivery for the created thread.
2154          */
2155         while (true) {
2156                 /*
2157                  * The sequence in kthread_stop() sets the stop flag first
2158                  * then wakes the process.  To avoid missed wakeups, the task
2159                  * should always be in a non running state before the stop
2160                  * flag is checked
2161                  */
2162                 set_current_state(TASK_INTERRUPTIBLE);
2163                 if (kthread_should_stop())
2164                         break;
2165
2166                 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2167                     shost->host_failed != scsi_host_busy(shost)) {
2168                         SCSI_LOG_ERROR_RECOVERY(1,
2169                                 shost_printk(KERN_INFO, shost,
2170                                              "scsi_eh_%d: sleeping\n",
2171                                              shost->host_no));
2172                         schedule();
2173                         continue;
2174                 }
2175
2176                 __set_current_state(TASK_RUNNING);
2177                 SCSI_LOG_ERROR_RECOVERY(1,
2178                         shost_printk(KERN_INFO, shost,
2179                                      "scsi_eh_%d: waking up %d/%d/%d\n",
2180                                      shost->host_no, shost->host_eh_scheduled,
2181                                      shost->host_failed,
2182                                      scsi_host_busy(shost)));
2183
2184                 /*
2185                  * We have a host that is failing for some reason.  Figure out
2186                  * what we need to do to get it up and online again (if we can).
2187                  * If we fail, we end up taking the thing offline.
2188                  */
2189                 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2190                         SCSI_LOG_ERROR_RECOVERY(1,
2191                                 shost_printk(KERN_ERR, shost,
2192                                              "scsi_eh_%d: unable to autoresume\n",
2193                                              shost->host_no));
2194                         continue;
2195                 }
2196
2197                 if (shost->transportt->eh_strategy_handler)
2198                         shost->transportt->eh_strategy_handler(shost);
2199                 else
2200                         scsi_unjam_host(shost);
2201
2202                 /* All scmds have been handled */
2203                 shost->host_failed = 0;
2204
2205                 /*
2206                  * Note - if the above fails completely, the action is to take
2207                  * individual devices offline and flush the queue of any
2208                  * outstanding requests that may have been pending.  When we
2209                  * restart, we restart any I/O to any other devices on the bus
2210                  * which are still online.
2211                  */
2212                 scsi_restart_operations(shost);
2213                 if (!shost->eh_noresume)
2214                         scsi_autopm_put_host(shost);
2215         }
2216         __set_current_state(TASK_RUNNING);
2217
2218         SCSI_LOG_ERROR_RECOVERY(1,
2219                 shost_printk(KERN_INFO, shost,
2220                              "Error handler scsi_eh_%d exiting\n",
2221                              shost->host_no));
2222         shost->ehandler = NULL;
2223         return 0;
2224 }
2225
2226 /*
2227  * Function:    scsi_report_bus_reset()
2228  *
2229  * Purpose:     Utility function used by low-level drivers to report that
2230  *              they have observed a bus reset on the bus being handled.
2231  *
2232  * Arguments:   shost       - Host in question
2233  *              channel     - channel on which reset was observed.
2234  *
2235  * Returns:     Nothing
2236  *
2237  * Lock status: Host lock must be held.
2238  *
2239  * Notes:       This only needs to be called if the reset is one which
2240  *              originates from an unknown location.  Resets originated
2241  *              by the mid-level itself don't need to call this, but there
2242  *              should be no harm.
2243  *
2244  *              The main purpose of this is to make sure that a CHECK_CONDITION
2245  *              is properly treated.
2246  */
2247 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2248 {
2249         struct scsi_device *sdev;
2250
2251         __shost_for_each_device(sdev, shost) {
2252                 if (channel == sdev_channel(sdev))
2253                         __scsi_report_device_reset(sdev, NULL);
2254         }
2255 }
2256 EXPORT_SYMBOL(scsi_report_bus_reset);
2257
2258 /*
2259  * Function:    scsi_report_device_reset()
2260  *
2261  * Purpose:     Utility function used by low-level drivers to report that
2262  *              they have observed a device reset on the device being handled.
2263  *
2264  * Arguments:   shost       - Host in question
2265  *              channel     - channel on which reset was observed
2266  *              target      - target on which reset was observed
2267  *
2268  * Returns:     Nothing
2269  *
2270  * Lock status: Host lock must be held
2271  *
2272  * Notes:       This only needs to be called if the reset is one which
2273  *              originates from an unknown location.  Resets originated
2274  *              by the mid-level itself don't need to call this, but there
2275  *              should be no harm.
2276  *
2277  *              The main purpose of this is to make sure that a CHECK_CONDITION
2278  *              is properly treated.
2279  */
2280 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2281 {
2282         struct scsi_device *sdev;
2283
2284         __shost_for_each_device(sdev, shost) {
2285                 if (channel == sdev_channel(sdev) &&
2286                     target == sdev_id(sdev))
2287                         __scsi_report_device_reset(sdev, NULL);
2288         }
2289 }
2290 EXPORT_SYMBOL(scsi_report_device_reset);
2291
2292 static void
2293 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2294 {
2295 }
2296
2297 /**
2298  * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2299  * @dev:        scsi_device to operate on
2300  * @arg:        reset type (see sg.h)
2301  */
2302 int
2303 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2304 {
2305         struct scsi_cmnd *scmd;
2306         struct Scsi_Host *shost = dev->host;
2307         struct request *rq;
2308         unsigned long flags;
2309         int error = 0, rtn, val;
2310
2311         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2312                 return -EACCES;
2313
2314         error = get_user(val, arg);
2315         if (error)
2316                 return error;
2317
2318         if (scsi_autopm_get_host(shost) < 0)
2319                 return -EIO;
2320
2321         error = -EIO;
2322         rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2323                         shost->hostt->cmd_size, GFP_KERNEL);
2324         if (!rq)
2325                 goto out_put_autopm_host;
2326         blk_rq_init(NULL, rq);
2327
2328         scmd = (struct scsi_cmnd *)(rq + 1);
2329         scsi_init_command(dev, scmd);
2330         scmd->request = rq;
2331         scmd->cmnd = scsi_req(rq)->cmd;
2332
2333         scmd->scsi_done         = scsi_reset_provider_done_command;
2334         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2335
2336         scmd->cmd_len                   = 0;
2337
2338         scmd->sc_data_direction         = DMA_BIDIRECTIONAL;
2339
2340         spin_lock_irqsave(shost->host_lock, flags);
2341         shost->tmf_in_progress = 1;
2342         spin_unlock_irqrestore(shost->host_lock, flags);
2343
2344         switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2345         case SG_SCSI_RESET_NOTHING:
2346                 rtn = SUCCESS;
2347                 break;
2348         case SG_SCSI_RESET_DEVICE:
2349                 rtn = scsi_try_bus_device_reset(scmd);
2350                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2351                         break;
2352                 /* FALLTHROUGH */
2353         case SG_SCSI_RESET_TARGET:
2354                 rtn = scsi_try_target_reset(scmd);
2355                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2356                         break;
2357                 /* FALLTHROUGH */
2358         case SG_SCSI_RESET_BUS:
2359                 rtn = scsi_try_bus_reset(scmd);
2360                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2361                         break;
2362                 /* FALLTHROUGH */
2363         case SG_SCSI_RESET_HOST:
2364                 rtn = scsi_try_host_reset(scmd);
2365                 if (rtn == SUCCESS)
2366                         break;
2367                 /* FALLTHROUGH */
2368         default:
2369                 rtn = FAILED;
2370                 break;
2371         }
2372
2373         error = (rtn == SUCCESS) ? 0 : -EIO;
2374
2375         spin_lock_irqsave(shost->host_lock, flags);
2376         shost->tmf_in_progress = 0;
2377         spin_unlock_irqrestore(shost->host_lock, flags);
2378
2379         /*
2380          * be sure to wake up anyone who was sleeping or had their queue
2381          * suspended while we performed the TMF.
2382          */
2383         SCSI_LOG_ERROR_RECOVERY(3,
2384                 shost_printk(KERN_INFO, shost,
2385                              "waking up host to restart after TMF\n"));
2386
2387         wake_up(&shost->host_wait);
2388         scsi_run_host_queues(shost);
2389
2390         scsi_put_command(scmd);
2391         kfree(rq);
2392
2393 out_put_autopm_host:
2394         scsi_autopm_put_host(shost);
2395         return error;
2396 }
2397
2398 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2399                                   struct scsi_sense_hdr *sshdr)
2400 {
2401         return scsi_normalize_sense(cmd->sense_buffer,
2402                         SCSI_SENSE_BUFFERSIZE, sshdr);
2403 }
2404 EXPORT_SYMBOL(scsi_command_normalize_sense);
2405
2406 /**
2407  * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2408  * @sense_buffer:       byte array of sense data
2409  * @sb_len:             number of valid bytes in sense_buffer
2410  * @info_out:           pointer to 64 integer where 8 or 4 byte information
2411  *                      field will be placed if found.
2412  *
2413  * Return value:
2414  *      true if information field found, false if not found.
2415  */
2416 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2417                              u64 *info_out)
2418 {
2419         const u8 * ucp;
2420
2421         if (sb_len < 7)
2422                 return false;
2423         switch (sense_buffer[0] & 0x7f) {
2424         case 0x70:
2425         case 0x71:
2426                 if (sense_buffer[0] & 0x80) {
2427                         *info_out = get_unaligned_be32(&sense_buffer[3]);
2428                         return true;
2429                 }
2430                 return false;
2431         case 0x72:
2432         case 0x73:
2433                 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2434                                            0 /* info desc */);
2435                 if (ucp && (0xa == ucp[1])) {
2436                         *info_out = get_unaligned_be64(&ucp[4]);
2437                         return true;
2438                 }
2439                 return false;
2440         default:
2441                 return false;
2442         }
2443 }
2444 EXPORT_SYMBOL(scsi_get_sense_info_fld);