ASoC: tas571x: add input channel mixer for TAS5717/19
[cascardo/linux.git] / drivers / scsi / scsi_error.c
1 /*
2  *  scsi_error.c Copyright (C) 1997 Eric Youngdale
3  *
4  *  SCSI error/timeout handling
5  *      Initial versions: Eric Youngdale.  Based upon conversations with
6  *                        Leonard Zubkoff and David Miller at Linux Expo,
7  *                        ideas originating from all over the place.
8  *
9  *      Restructured scsi_unjam_host and associated functions.
10  *      September 04, 2002 Mike Anderson (andmike@us.ibm.com)
11  *
12  *      Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
13  *      minor cleanups.
14  *      September 30, 2002 Mike Anderson (andmike@us.ibm.com)
15  */
16
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/gfp.h>
20 #include <linux/timer.h>
21 #include <linux/string.h>
22 #include <linux/kernel.h>
23 #include <linux/freezer.h>
24 #include <linux/kthread.h>
25 #include <linux/interrupt.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
28 #include <linux/jiffies.h>
29
30 #include <scsi/scsi.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_dbg.h>
33 #include <scsi/scsi_device.h>
34 #include <scsi/scsi_driver.h>
35 #include <scsi/scsi_eh.h>
36 #include <scsi/scsi_common.h>
37 #include <scsi/scsi_transport.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_ioctl.h>
40 #include <scsi/scsi_dh.h>
41 #include <scsi/sg.h>
42
43 #include "scsi_priv.h"
44 #include "scsi_logging.h"
45 #include "scsi_transport_api.h"
46
47 #include <trace/events/scsi.h>
48
49 static void scsi_eh_done(struct scsi_cmnd *scmd);
50
51 /*
52  * These should *probably* be handled by the host itself.
53  * Since it is allowed to sleep, it probably should.
54  */
55 #define BUS_RESET_SETTLE_TIME   (10)
56 #define HOST_RESET_SETTLE_TIME  (10)
57
58 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
59 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
60                                  struct scsi_cmnd *);
61
62 /* called with shost->host_lock held */
63 void scsi_eh_wakeup(struct Scsi_Host *shost)
64 {
65         if (atomic_read(&shost->host_busy) == shost->host_failed) {
66                 trace_scsi_eh_wakeup(shost);
67                 wake_up_process(shost->ehandler);
68                 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
69                         "Waking error handler thread\n"));
70         }
71 }
72
73 /**
74  * scsi_schedule_eh - schedule EH for SCSI host
75  * @shost:      SCSI host to invoke error handling on.
76  *
77  * Schedule SCSI EH without scmd.
78  */
79 void scsi_schedule_eh(struct Scsi_Host *shost)
80 {
81         unsigned long flags;
82
83         spin_lock_irqsave(shost->host_lock, flags);
84
85         if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
86             scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
87                 shost->host_eh_scheduled++;
88                 scsi_eh_wakeup(shost);
89         }
90
91         spin_unlock_irqrestore(shost->host_lock, flags);
92 }
93 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
94
95 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
96 {
97         if (!shost->last_reset || shost->eh_deadline == -1)
98                 return 0;
99
100         /*
101          * 32bit accesses are guaranteed to be atomic
102          * (on all supported architectures), so instead
103          * of using a spinlock we can as well double check
104          * if eh_deadline has been set to 'off' during the
105          * time_before call.
106          */
107         if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
108             shost->eh_deadline > -1)
109                 return 0;
110
111         return 1;
112 }
113
114 /**
115  * scmd_eh_abort_handler - Handle command aborts
116  * @work:       command to be aborted.
117  */
118 void
119 scmd_eh_abort_handler(struct work_struct *work)
120 {
121         struct scsi_cmnd *scmd =
122                 container_of(work, struct scsi_cmnd, abort_work.work);
123         struct scsi_device *sdev = scmd->device;
124         int rtn;
125
126         if (scsi_host_eh_past_deadline(sdev->host)) {
127                 SCSI_LOG_ERROR_RECOVERY(3,
128                         scmd_printk(KERN_INFO, scmd,
129                                     "eh timeout, not aborting\n"));
130         } else {
131                 SCSI_LOG_ERROR_RECOVERY(3,
132                         scmd_printk(KERN_INFO, scmd,
133                                     "aborting command\n"));
134                 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
135                 if (rtn == SUCCESS) {
136                         set_host_byte(scmd, DID_TIME_OUT);
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 retrying "
141                                                     "aborted command\n"));
142                         } else if (!scsi_noretry_cmd(scmd) &&
143                             (++scmd->retries <= scmd->allowed)) {
144                                 SCSI_LOG_ERROR_RECOVERY(3,
145                                         scmd_printk(KERN_WARNING, scmd,
146                                                     "retry aborted command\n"));
147                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
148                                 return;
149                         } else {
150                                 SCSI_LOG_ERROR_RECOVERY(3,
151                                         scmd_printk(KERN_WARNING, scmd,
152                                                     "finish aborted command\n"));
153                                 scsi_finish_command(scmd);
154                                 return;
155                         }
156                 } else {
157                         SCSI_LOG_ERROR_RECOVERY(3,
158                                 scmd_printk(KERN_INFO, scmd,
159                                             "cmd abort %s\n",
160                                             (rtn == FAST_IO_FAIL) ?
161                                             "not send" : "failed"));
162                 }
163         }
164
165         if (!scsi_eh_scmd_add(scmd, 0)) {
166                 SCSI_LOG_ERROR_RECOVERY(3,
167                         scmd_printk(KERN_WARNING, scmd,
168                                     "terminate aborted command\n"));
169                 set_host_byte(scmd, DID_TIME_OUT);
170                 scsi_finish_command(scmd);
171         }
172 }
173
174 /**
175  * scsi_abort_command - schedule a command abort
176  * @scmd:       scmd to abort.
177  *
178  * We only need to abort commands after a command timeout
179  */
180 static int
181 scsi_abort_command(struct scsi_cmnd *scmd)
182 {
183         struct scsi_device *sdev = scmd->device;
184         struct Scsi_Host *shost = sdev->host;
185         unsigned long flags;
186
187         if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
188                 /*
189                  * Retry after abort failed, escalate to next level.
190                  */
191                 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
192                 SCSI_LOG_ERROR_RECOVERY(3,
193                         scmd_printk(KERN_INFO, scmd,
194                                     "previous abort failed\n"));
195                 BUG_ON(delayed_work_pending(&scmd->abort_work));
196                 return FAILED;
197         }
198
199         /*
200          * Do not try a command abort if
201          * SCSI EH has already started.
202          */
203         spin_lock_irqsave(shost->host_lock, flags);
204         if (scsi_host_in_recovery(shost)) {
205                 spin_unlock_irqrestore(shost->host_lock, flags);
206                 SCSI_LOG_ERROR_RECOVERY(3,
207                         scmd_printk(KERN_INFO, scmd,
208                                     "not aborting, host in recovery\n"));
209                 return FAILED;
210         }
211
212         if (shost->eh_deadline != -1 && !shost->last_reset)
213                 shost->last_reset = jiffies;
214         spin_unlock_irqrestore(shost->host_lock, flags);
215
216         scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
217         SCSI_LOG_ERROR_RECOVERY(3,
218                 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
219         queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
220         return SUCCESS;
221 }
222
223 /**
224  * scsi_eh_scmd_add - add scsi cmd to error handling.
225  * @scmd:       scmd to run eh on.
226  * @eh_flag:    optional SCSI_EH flag.
227  *
228  * Return value:
229  *      0 on failure.
230  */
231 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
232 {
233         struct Scsi_Host *shost = scmd->device->host;
234         unsigned long flags;
235         int ret = 0;
236
237         if (!shost->ehandler)
238                 return 0;
239
240         spin_lock_irqsave(shost->host_lock, flags);
241         if (scsi_host_set_state(shost, SHOST_RECOVERY))
242                 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
243                         goto out_unlock;
244
245         if (shost->eh_deadline != -1 && !shost->last_reset)
246                 shost->last_reset = jiffies;
247
248         ret = 1;
249         if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
250                 eh_flag &= ~SCSI_EH_CANCEL_CMD;
251         scmd->eh_eflags |= eh_flag;
252         list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
253         shost->host_failed++;
254         scsi_eh_wakeup(shost);
255  out_unlock:
256         spin_unlock_irqrestore(shost->host_lock, flags);
257         return ret;
258 }
259
260 /**
261  * scsi_times_out - Timeout function for normal scsi commands.
262  * @req:        request that is timing out.
263  *
264  * Notes:
265  *     We do not need to lock this.  There is the potential for a race
266  *     only in that the normal completion handling might run, but if the
267  *     normal completion function determines that the timer has already
268  *     fired, then it mustn't do anything.
269  */
270 enum blk_eh_timer_return scsi_times_out(struct request *req)
271 {
272         struct scsi_cmnd *scmd = req->special;
273         enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
274         struct Scsi_Host *host = scmd->device->host;
275
276         trace_scsi_dispatch_cmd_timeout(scmd);
277         scsi_log_completion(scmd, TIMEOUT_ERROR);
278
279         if (host->eh_deadline != -1 && !host->last_reset)
280                 host->last_reset = jiffies;
281
282         if (host->transportt->eh_timed_out)
283                 rtn = host->transportt->eh_timed_out(scmd);
284         else if (host->hostt->eh_timed_out)
285                 rtn = host->hostt->eh_timed_out(scmd);
286
287         if (rtn == BLK_EH_NOT_HANDLED) {
288                 if (!host->hostt->no_async_abort &&
289                     scsi_abort_command(scmd) == SUCCESS)
290                         return BLK_EH_NOT_HANDLED;
291
292                 set_host_byte(scmd, DID_TIME_OUT);
293                 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))
294                         rtn = BLK_EH_HANDLED;
295         }
296
297         return rtn;
298 }
299
300 /**
301  * scsi_block_when_processing_errors - Prevent cmds from being queued.
302  * @sdev:       Device on which we are performing recovery.
303  *
304  * Description:
305  *     We block until the host is out of error recovery, and then check to
306  *     see whether the host or the device is offline.
307  *
308  * Return value:
309  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
310  */
311 int scsi_block_when_processing_errors(struct scsi_device *sdev)
312 {
313         int online;
314
315         wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
316
317         online = scsi_device_online(sdev);
318
319         SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
320                 "%s: rtn: %d\n", __func__, online));
321
322         return online;
323 }
324 EXPORT_SYMBOL(scsi_block_when_processing_errors);
325
326 #ifdef CONFIG_SCSI_LOGGING
327 /**
328  * scsi_eh_prt_fail_stats - Log info on failures.
329  * @shost:      scsi host being recovered.
330  * @work_q:     Queue of scsi cmds to process.
331  */
332 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
333                                           struct list_head *work_q)
334 {
335         struct scsi_cmnd *scmd;
336         struct scsi_device *sdev;
337         int total_failures = 0;
338         int cmd_failed = 0;
339         int cmd_cancel = 0;
340         int devices_failed = 0;
341
342         shost_for_each_device(sdev, shost) {
343                 list_for_each_entry(scmd, work_q, eh_entry) {
344                         if (scmd->device == sdev) {
345                                 ++total_failures;
346                                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
347                                         ++cmd_cancel;
348                                 else
349                                         ++cmd_failed;
350                         }
351                 }
352
353                 if (cmd_cancel || cmd_failed) {
354                         SCSI_LOG_ERROR_RECOVERY(3,
355                                 shost_printk(KERN_INFO, shost,
356                                             "%s: cmds failed: %d, cancel: %d\n",
357                                             __func__, cmd_failed,
358                                             cmd_cancel));
359                         cmd_cancel = 0;
360                         cmd_failed = 0;
361                         ++devices_failed;
362                 }
363         }
364
365         SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
366                                    "Total of %d commands on %d"
367                                    " devices require eh work\n",
368                                    total_failures, devices_failed));
369 }
370 #endif
371
372  /**
373  * scsi_report_lun_change - Set flag on all *other* devices on the same target
374  *                          to indicate that a UNIT ATTENTION is expected.
375  * @sdev:       Device reporting the UNIT ATTENTION
376  */
377 static void scsi_report_lun_change(struct scsi_device *sdev)
378 {
379         sdev->sdev_target->expecting_lun_change = 1;
380 }
381
382 /**
383  * scsi_report_sense - Examine scsi sense information and log messages for
384  *                     certain conditions, also issue uevents for some of them.
385  * @sdev:       Device reporting the sense code
386  * @sshdr:      sshdr to be examined
387  */
388 static void scsi_report_sense(struct scsi_device *sdev,
389                               struct scsi_sense_hdr *sshdr)
390 {
391         enum scsi_device_event evt_type = SDEV_EVT_MAXBITS;     /* i.e. none */
392
393         if (sshdr->sense_key == UNIT_ATTENTION) {
394                 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
395                         evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
396                         sdev_printk(KERN_WARNING, sdev,
397                                     "Inquiry data has changed");
398                 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
399                         evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
400                         scsi_report_lun_change(sdev);
401                         sdev_printk(KERN_WARNING, sdev,
402                                     "Warning! Received an indication that the "
403                                     "LUN assignments on this target have "
404                                     "changed. The Linux SCSI layer does not "
405                                     "automatically remap LUN assignments.\n");
406                 } else if (sshdr->asc == 0x3f)
407                         sdev_printk(KERN_WARNING, sdev,
408                                     "Warning! Received an indication that the "
409                                     "operating parameters on this target have "
410                                     "changed. The Linux SCSI layer does not "
411                                     "automatically adjust these parameters.\n");
412
413                 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
414                         evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
415                         sdev_printk(KERN_WARNING, sdev,
416                                     "Warning! Received an indication that the "
417                                     "LUN reached a thin provisioning soft "
418                                     "threshold.\n");
419                 }
420
421                 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
422                         evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
423                         sdev_printk(KERN_WARNING, sdev,
424                                     "Mode parameters changed");
425                 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
426                         evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
427                         sdev_printk(KERN_WARNING, sdev,
428                                     "Asymmetric access state changed");
429                 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
430                         evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
431                         sdev_printk(KERN_WARNING, sdev,
432                                     "Capacity data has changed");
433                 } else if (sshdr->asc == 0x2a)
434                         sdev_printk(KERN_WARNING, sdev,
435                                     "Parameters changed");
436         }
437
438         if (evt_type != SDEV_EVT_MAXBITS) {
439                 set_bit(evt_type, sdev->pending_events);
440                 schedule_work(&sdev->event_work);
441         }
442 }
443
444 /**
445  * scsi_check_sense - Examine scsi cmd sense
446  * @scmd:       Cmd to have sense checked.
447  *
448  * Return value:
449  *      SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
450  *
451  * Notes:
452  *      When a deferred error is detected the current command has
453  *      not been executed and needs retrying.
454  */
455 int scsi_check_sense(struct scsi_cmnd *scmd)
456 {
457         struct scsi_device *sdev = scmd->device;
458         struct scsi_sense_hdr sshdr;
459
460         if (! scsi_command_normalize_sense(scmd, &sshdr))
461                 return FAILED;  /* no valid sense data */
462
463         scsi_report_sense(sdev, &sshdr);
464
465         if (scsi_sense_is_deferred(&sshdr))
466                 return NEEDS_RETRY;
467
468         if (sdev->handler && sdev->handler->check_sense) {
469                 int rc;
470
471                 rc = sdev->handler->check_sense(sdev, &sshdr);
472                 if (rc != SCSI_RETURN_NOT_HANDLED)
473                         return rc;
474                 /* handler does not care. Drop down to default handling */
475         }
476
477         if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
478                 /*
479                  * nasty: for mid-layer issued TURs, we need to return the
480                  * actual sense data without any recovery attempt.  For eh
481                  * issued ones, we need to try to recover and interpret
482                  */
483                 return SUCCESS;
484
485         /*
486          * Previous logic looked for FILEMARK, EOM or ILI which are
487          * mainly associated with tapes and returned SUCCESS.
488          */
489         if (sshdr.response_code == 0x70) {
490                 /* fixed format */
491                 if (scmd->sense_buffer[2] & 0xe0)
492                         return SUCCESS;
493         } else {
494                 /*
495                  * descriptor format: look for "stream commands sense data
496                  * descriptor" (see SSC-3). Assume single sense data
497                  * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
498                  */
499                 if ((sshdr.additional_length > 3) &&
500                     (scmd->sense_buffer[8] == 0x4) &&
501                     (scmd->sense_buffer[11] & 0xe0))
502                         return SUCCESS;
503         }
504
505         switch (sshdr.sense_key) {
506         case NO_SENSE:
507                 return SUCCESS;
508         case RECOVERED_ERROR:
509                 return /* soft_error */ SUCCESS;
510
511         case ABORTED_COMMAND:
512                 if (sshdr.asc == 0x10) /* DIF */
513                         return SUCCESS;
514
515                 return NEEDS_RETRY;
516         case NOT_READY:
517         case UNIT_ATTENTION:
518                 /*
519                  * if we are expecting a cc/ua because of a bus reset that we
520                  * performed, treat this just as a retry.  otherwise this is
521                  * information that we should pass up to the upper-level driver
522                  * so that we can deal with it there.
523                  */
524                 if (scmd->device->expecting_cc_ua) {
525                         /*
526                          * Because some device does not queue unit
527                          * attentions correctly, we carefully check
528                          * additional sense code and qualifier so as
529                          * not to squash media change unit attention.
530                          */
531                         if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
532                                 scmd->device->expecting_cc_ua = 0;
533                                 return NEEDS_RETRY;
534                         }
535                 }
536                 /*
537                  * we might also expect a cc/ua if another LUN on the target
538                  * reported a UA with an ASC/ASCQ of 3F 0E -
539                  * REPORTED LUNS DATA HAS CHANGED.
540                  */
541                 if (scmd->device->sdev_target->expecting_lun_change &&
542                     sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
543                         return NEEDS_RETRY;
544                 /*
545                  * if the device is in the process of becoming ready, we
546                  * should retry.
547                  */
548                 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
549                         return NEEDS_RETRY;
550                 /*
551                  * if the device is not started, we need to wake
552                  * the error handler to start the motor
553                  */
554                 if (scmd->device->allow_restart &&
555                     (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
556                         return FAILED;
557                 /*
558                  * Pass the UA upwards for a determination in the completion
559                  * functions.
560                  */
561                 return SUCCESS;
562
563                 /* these are not supported */
564         case DATA_PROTECT:
565                 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
566                         /* Thin provisioning hard threshold reached */
567                         set_host_byte(scmd, DID_ALLOC_FAILURE);
568                         return SUCCESS;
569                 }
570         case COPY_ABORTED:
571         case VOLUME_OVERFLOW:
572         case MISCOMPARE:
573         case BLANK_CHECK:
574                 set_host_byte(scmd, DID_TARGET_FAILURE);
575                 return SUCCESS;
576
577         case MEDIUM_ERROR:
578                 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
579                     sshdr.asc == 0x13 || /* AMNF DATA FIELD */
580                     sshdr.asc == 0x14) { /* RECORD NOT FOUND */
581                         set_host_byte(scmd, DID_MEDIUM_ERROR);
582                         return SUCCESS;
583                 }
584                 return NEEDS_RETRY;
585
586         case HARDWARE_ERROR:
587                 if (scmd->device->retry_hwerror)
588                         return ADD_TO_MLQUEUE;
589                 else
590                         set_host_byte(scmd, DID_TARGET_FAILURE);
591
592         case ILLEGAL_REQUEST:
593                 if (sshdr.asc == 0x20 || /* Invalid command operation code */
594                     sshdr.asc == 0x21 || /* Logical block address out of range */
595                     sshdr.asc == 0x24 || /* Invalid field in cdb */
596                     sshdr.asc == 0x26) { /* Parameter value invalid */
597                         set_host_byte(scmd, DID_TARGET_FAILURE);
598                 }
599                 return SUCCESS;
600
601         default:
602                 return SUCCESS;
603         }
604 }
605 EXPORT_SYMBOL_GPL(scsi_check_sense);
606
607 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
608 {
609         struct scsi_host_template *sht = sdev->host->hostt;
610         struct scsi_device *tmp_sdev;
611
612         if (!sht->track_queue_depth ||
613             sdev->queue_depth >= sdev->max_queue_depth)
614                 return;
615
616         if (time_before(jiffies,
617             sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
618                 return;
619
620         if (time_before(jiffies,
621             sdev->last_queue_full_time + sdev->queue_ramp_up_period))
622                 return;
623
624         /*
625          * Walk all devices of a target and do
626          * ramp up on them.
627          */
628         shost_for_each_device(tmp_sdev, sdev->host) {
629                 if (tmp_sdev->channel != sdev->channel ||
630                     tmp_sdev->id != sdev->id ||
631                     tmp_sdev->queue_depth == sdev->max_queue_depth)
632                         continue;
633
634                 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
635                 sdev->last_queue_ramp_up = jiffies;
636         }
637 }
638
639 static void scsi_handle_queue_full(struct scsi_device *sdev)
640 {
641         struct scsi_host_template *sht = sdev->host->hostt;
642         struct scsi_device *tmp_sdev;
643
644         if (!sht->track_queue_depth)
645                 return;
646
647         shost_for_each_device(tmp_sdev, sdev->host) {
648                 if (tmp_sdev->channel != sdev->channel ||
649                     tmp_sdev->id != sdev->id)
650                         continue;
651                 /*
652                  * We do not know the number of commands that were at
653                  * the device when we got the queue full so we start
654                  * from the highest possible value and work our way down.
655                  */
656                 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
657         }
658 }
659
660 /**
661  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
662  * @scmd:       SCSI cmd to examine.
663  *
664  * Notes:
665  *    This is *only* called when we are examining the status of commands
666  *    queued during error recovery.  the main difference here is that we
667  *    don't allow for the possibility of retries here, and we are a lot
668  *    more restrictive about what we consider acceptable.
669  */
670 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
671 {
672         /*
673          * first check the host byte, to see if there is anything in there
674          * that would indicate what we need to do.
675          */
676         if (host_byte(scmd->result) == DID_RESET) {
677                 /*
678                  * rats.  we are already in the error handler, so we now
679                  * get to try and figure out what to do next.  if the sense
680                  * is valid, we have a pretty good idea of what to do.
681                  * if not, we mark it as FAILED.
682                  */
683                 return scsi_check_sense(scmd);
684         }
685         if (host_byte(scmd->result) != DID_OK)
686                 return FAILED;
687
688         /*
689          * next, check the message byte.
690          */
691         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
692                 return FAILED;
693
694         /*
695          * now, check the status byte to see if this indicates
696          * anything special.
697          */
698         switch (status_byte(scmd->result)) {
699         case GOOD:
700                 scsi_handle_queue_ramp_up(scmd->device);
701         case COMMAND_TERMINATED:
702                 return SUCCESS;
703         case CHECK_CONDITION:
704                 return scsi_check_sense(scmd);
705         case CONDITION_GOOD:
706         case INTERMEDIATE_GOOD:
707         case INTERMEDIATE_C_GOOD:
708                 /*
709                  * who knows?  FIXME(eric)
710                  */
711                 return SUCCESS;
712         case RESERVATION_CONFLICT:
713                 if (scmd->cmnd[0] == TEST_UNIT_READY)
714                         /* it is a success, we probed the device and
715                          * found it */
716                         return SUCCESS;
717                 /* otherwise, we failed to send the command */
718                 return FAILED;
719         case QUEUE_FULL:
720                 scsi_handle_queue_full(scmd->device);
721                 /* fall through */
722         case BUSY:
723                 return NEEDS_RETRY;
724         default:
725                 return FAILED;
726         }
727         return FAILED;
728 }
729
730 /**
731  * scsi_eh_done - Completion function for error handling.
732  * @scmd:       Cmd that is done.
733  */
734 static void scsi_eh_done(struct scsi_cmnd *scmd)
735 {
736         struct completion *eh_action;
737
738         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
739                         "%s result: %x\n", __func__, scmd->result));
740
741         eh_action = scmd->device->host->eh_action;
742         if (eh_action)
743                 complete(eh_action);
744 }
745
746 /**
747  * scsi_try_host_reset - ask host adapter to reset itself
748  * @scmd:       SCSI cmd to send host reset.
749  */
750 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
751 {
752         unsigned long flags;
753         int rtn;
754         struct Scsi_Host *host = scmd->device->host;
755         struct scsi_host_template *hostt = host->hostt;
756
757         SCSI_LOG_ERROR_RECOVERY(3,
758                 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
759
760         if (!hostt->eh_host_reset_handler)
761                 return FAILED;
762
763         rtn = hostt->eh_host_reset_handler(scmd);
764
765         if (rtn == SUCCESS) {
766                 if (!hostt->skip_settle_delay)
767                         ssleep(HOST_RESET_SETTLE_TIME);
768                 spin_lock_irqsave(host->host_lock, flags);
769                 scsi_report_bus_reset(host, scmd_channel(scmd));
770                 spin_unlock_irqrestore(host->host_lock, flags);
771         }
772
773         return rtn;
774 }
775
776 /**
777  * scsi_try_bus_reset - ask host to perform a bus reset
778  * @scmd:       SCSI cmd to send bus reset.
779  */
780 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
781 {
782         unsigned long flags;
783         int rtn;
784         struct Scsi_Host *host = scmd->device->host;
785         struct scsi_host_template *hostt = host->hostt;
786
787         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
788                 "%s: Snd Bus RST\n", __func__));
789
790         if (!hostt->eh_bus_reset_handler)
791                 return FAILED;
792
793         rtn = hostt->eh_bus_reset_handler(scmd);
794
795         if (rtn == SUCCESS) {
796                 if (!hostt->skip_settle_delay)
797                         ssleep(BUS_RESET_SETTLE_TIME);
798                 spin_lock_irqsave(host->host_lock, flags);
799                 scsi_report_bus_reset(host, scmd_channel(scmd));
800                 spin_unlock_irqrestore(host->host_lock, flags);
801         }
802
803         return rtn;
804 }
805
806 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
807 {
808         sdev->was_reset = 1;
809         sdev->expecting_cc_ua = 1;
810 }
811
812 /**
813  * scsi_try_target_reset - Ask host to perform a target reset
814  * @scmd:       SCSI cmd used to send a target reset
815  *
816  * Notes:
817  *    There is no timeout for this operation.  if this operation is
818  *    unreliable for a given host, then the host itself needs to put a
819  *    timer on it, and set the host back to a consistent state prior to
820  *    returning.
821  */
822 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
823 {
824         unsigned long flags;
825         int rtn;
826         struct Scsi_Host *host = scmd->device->host;
827         struct scsi_host_template *hostt = host->hostt;
828
829         if (!hostt->eh_target_reset_handler)
830                 return FAILED;
831
832         rtn = hostt->eh_target_reset_handler(scmd);
833         if (rtn == SUCCESS) {
834                 spin_lock_irqsave(host->host_lock, flags);
835                 __starget_for_each_device(scsi_target(scmd->device), NULL,
836                                           __scsi_report_device_reset);
837                 spin_unlock_irqrestore(host->host_lock, flags);
838         }
839
840         return rtn;
841 }
842
843 /**
844  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
845  * @scmd:       SCSI cmd used to send BDR
846  *
847  * Notes:
848  *    There is no timeout for this operation.  if this operation is
849  *    unreliable for a given host, then the host itself needs to put a
850  *    timer on it, and set the host back to a consistent state prior to
851  *    returning.
852  */
853 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
854 {
855         int rtn;
856         struct scsi_host_template *hostt = scmd->device->host->hostt;
857
858         if (!hostt->eh_device_reset_handler)
859                 return FAILED;
860
861         rtn = hostt->eh_device_reset_handler(scmd);
862         if (rtn == SUCCESS)
863                 __scsi_report_device_reset(scmd->device, NULL);
864         return rtn;
865 }
866
867 /**
868  * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
869  * @hostt:      SCSI driver host template
870  * @scmd:       SCSI cmd used to send a target reset
871  *
872  * Return value:
873  *      SUCCESS, FAILED, or FAST_IO_FAIL
874  *
875  * Notes:
876  *    SUCCESS does not necessarily indicate that the command
877  *    has been aborted; it only indicates that the LLDDs
878  *    has cleared all references to that command.
879  *    LLDDs should return FAILED only if an abort was required
880  *    but could not be executed. LLDDs should return FAST_IO_FAIL
881  *    if the device is temporarily unavailable (eg due to a
882  *    link down on FibreChannel)
883  */
884 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
885                                  struct scsi_cmnd *scmd)
886 {
887         if (!hostt->eh_abort_handler)
888                 return FAILED;
889
890         return hostt->eh_abort_handler(scmd);
891 }
892
893 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
894 {
895         if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
896                 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
897                         if (scsi_try_target_reset(scmd) != SUCCESS)
898                                 if (scsi_try_bus_reset(scmd) != SUCCESS)
899                                         scsi_try_host_reset(scmd);
900 }
901
902 /**
903  * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recovery
904  * @scmd:       SCSI command structure to hijack
905  * @ses:        structure to save restore information
906  * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
907  * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
908  * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
909  *
910  * This function is used to save a scsi command information before re-execution
911  * as part of the error recovery process.  If @sense_bytes is 0 the command
912  * sent must be one that does not transfer any data.  If @sense_bytes != 0
913  * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
914  * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
915  */
916 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
917                         unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
918 {
919         struct scsi_device *sdev = scmd->device;
920
921         /*
922          * We need saved copies of a number of fields - this is because
923          * error handling may need to overwrite these with different values
924          * to run different commands, and once error handling is complete,
925          * we will need to restore these values prior to running the actual
926          * command.
927          */
928         ses->cmd_len = scmd->cmd_len;
929         ses->cmnd = scmd->cmnd;
930         ses->data_direction = scmd->sc_data_direction;
931         ses->sdb = scmd->sdb;
932         ses->next_rq = scmd->request->next_rq;
933         ses->result = scmd->result;
934         ses->underflow = scmd->underflow;
935         ses->prot_op = scmd->prot_op;
936
937         scmd->prot_op = SCSI_PROT_NORMAL;
938         scmd->eh_eflags = 0;
939         scmd->cmnd = ses->eh_cmnd;
940         memset(scmd->cmnd, 0, BLK_MAX_CDB);
941         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
942         scmd->request->next_rq = NULL;
943         scmd->result = 0;
944
945         if (sense_bytes) {
946                 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
947                                          sense_bytes);
948                 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
949                             scmd->sdb.length);
950                 scmd->sdb.table.sgl = &ses->sense_sgl;
951                 scmd->sc_data_direction = DMA_FROM_DEVICE;
952                 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
953                 scmd->cmnd[0] = REQUEST_SENSE;
954                 scmd->cmnd[4] = scmd->sdb.length;
955                 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
956         } else {
957                 scmd->sc_data_direction = DMA_NONE;
958                 if (cmnd) {
959                         BUG_ON(cmnd_size > BLK_MAX_CDB);
960                         memcpy(scmd->cmnd, cmnd, cmnd_size);
961                         scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
962                 }
963         }
964
965         scmd->underflow = 0;
966
967         if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
968                 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
969                         (sdev->lun << 5 & 0xe0);
970
971         /*
972          * Zero the sense buffer.  The scsi spec mandates that any
973          * untransferred sense data should be interpreted as being zero.
974          */
975         memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
976 }
977 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
978
979 /**
980  * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recovery
981  * @scmd:       SCSI command structure to restore
982  * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
983  *
984  * Undo any damage done by above scsi_eh_prep_cmnd().
985  */
986 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
987 {
988         /*
989          * Restore original data
990          */
991         scmd->cmd_len = ses->cmd_len;
992         scmd->cmnd = ses->cmnd;
993         scmd->sc_data_direction = ses->data_direction;
994         scmd->sdb = ses->sdb;
995         scmd->request->next_rq = ses->next_rq;
996         scmd->result = ses->result;
997         scmd->underflow = ses->underflow;
998         scmd->prot_op = ses->prot_op;
999 }
1000 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1001
1002 /**
1003  * scsi_send_eh_cmnd  - submit a scsi command as part of error recovery
1004  * @scmd:       SCSI command structure to hijack
1005  * @cmnd:       CDB to send
1006  * @cmnd_size:  size in bytes of @cmnd
1007  * @timeout:    timeout for this request
1008  * @sense_bytes: size of sense data to copy or 0
1009  *
1010  * This function is used to send a scsi command down to a target device
1011  * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1012  *
1013  * Return value:
1014  *    SUCCESS or FAILED or NEEDS_RETRY
1015  */
1016 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1017                              int cmnd_size, int timeout, unsigned sense_bytes)
1018 {
1019         struct scsi_device *sdev = scmd->device;
1020         struct Scsi_Host *shost = sdev->host;
1021         DECLARE_COMPLETION_ONSTACK(done);
1022         unsigned long timeleft = timeout;
1023         struct scsi_eh_save ses;
1024         const unsigned long stall_for = msecs_to_jiffies(100);
1025         int rtn;
1026
1027 retry:
1028         scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1029         shost->eh_action = &done;
1030
1031         scsi_log_send(scmd);
1032         scmd->scsi_done = scsi_eh_done;
1033         rtn = shost->hostt->queuecommand(shost, scmd);
1034         if (rtn) {
1035                 if (timeleft > stall_for) {
1036                         scsi_eh_restore_cmnd(scmd, &ses);
1037                         timeleft -= stall_for;
1038                         msleep(jiffies_to_msecs(stall_for));
1039                         goto retry;
1040                 }
1041                 /* signal not to enter either branch of the if () below */
1042                 timeleft = 0;
1043                 rtn = FAILED;
1044         } else {
1045                 timeleft = wait_for_completion_timeout(&done, timeout);
1046                 rtn = SUCCESS;
1047         }
1048
1049         shost->eh_action = NULL;
1050
1051         scsi_log_completion(scmd, rtn);
1052
1053         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1054                         "%s timeleft: %ld\n",
1055                         __func__, timeleft));
1056
1057         /*
1058          * If there is time left scsi_eh_done got called, and we will examine
1059          * the actual status codes to see whether the command actually did
1060          * complete normally, else if we have a zero return and no time left,
1061          * the command must still be pending, so abort it and return FAILED.
1062          * If we never actually managed to issue the command, because
1063          * ->queuecommand() kept returning non zero, use the rtn = FAILED
1064          * value above (so don't execute either branch of the if)
1065          */
1066         if (timeleft) {
1067                 rtn = scsi_eh_completed_normally(scmd);
1068                 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1069                         "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1070
1071                 switch (rtn) {
1072                 case SUCCESS:
1073                 case NEEDS_RETRY:
1074                 case FAILED:
1075                         break;
1076                 case ADD_TO_MLQUEUE:
1077                         rtn = NEEDS_RETRY;
1078                         break;
1079                 default:
1080                         rtn = FAILED;
1081                         break;
1082                 }
1083         } else if (rtn != FAILED) {
1084                 scsi_abort_eh_cmnd(scmd);
1085                 rtn = FAILED;
1086         }
1087
1088         scsi_eh_restore_cmnd(scmd, &ses);
1089
1090         return rtn;
1091 }
1092
1093 /**
1094  * scsi_request_sense - Request sense data from a particular target.
1095  * @scmd:       SCSI cmd for request sense.
1096  *
1097  * Notes:
1098  *    Some hosts automatically obtain this information, others require
1099  *    that we obtain it on our own. This function will *not* return until
1100  *    the command either times out, or it completes.
1101  */
1102 static int scsi_request_sense(struct scsi_cmnd *scmd)
1103 {
1104         return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1105 }
1106
1107 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1108 {
1109         if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1110                 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1111                 if (sdrv->eh_action)
1112                         rtn = sdrv->eh_action(scmd, rtn);
1113         }
1114         return rtn;
1115 }
1116
1117 /**
1118  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1119  * @scmd:       Original SCSI cmd that eh has finished.
1120  * @done_q:     Queue for processed commands.
1121  *
1122  * Notes:
1123  *    We don't want to use the normal command completion while we are are
1124  *    still handling errors - it may cause other commands to be queued,
1125  *    and that would disturb what we are doing.  Thus we really want to
1126  *    keep a list of pending commands for final completion, and once we
1127  *    are ready to leave error handling we handle completion for real.
1128  */
1129 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1130 {
1131         scmd->device->host->host_failed--;
1132         scmd->eh_eflags = 0;
1133         list_move_tail(&scmd->eh_entry, done_q);
1134 }
1135 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1136
1137 /**
1138  * scsi_eh_get_sense - Get device sense data.
1139  * @work_q:     Queue of commands to process.
1140  * @done_q:     Queue of processed commands.
1141  *
1142  * Description:
1143  *    See if we need to request sense information.  if so, then get it
1144  *    now, so we have a better idea of what to do.
1145  *
1146  * Notes:
1147  *    This has the unfortunate side effect that if a shost adapter does
1148  *    not automatically request sense information, we end up shutting
1149  *    it down before we request it.
1150  *
1151  *    All drivers should request sense information internally these days,
1152  *    so for now all I have to say is tough noogies if you end up in here.
1153  *
1154  *    XXX: Long term this code should go away, but that needs an audit of
1155  *         all LLDDs first.
1156  */
1157 int scsi_eh_get_sense(struct list_head *work_q,
1158                       struct list_head *done_q)
1159 {
1160         struct scsi_cmnd *scmd, *next;
1161         struct Scsi_Host *shost;
1162         int rtn;
1163
1164         /*
1165          * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1166          * should not get sense.
1167          */
1168         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1169                 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1170                     (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1171                     SCSI_SENSE_VALID(scmd))
1172                         continue;
1173
1174                 shost = scmd->device->host;
1175                 if (scsi_host_eh_past_deadline(shost)) {
1176                         SCSI_LOG_ERROR_RECOVERY(3,
1177                                 scmd_printk(KERN_INFO, scmd,
1178                                             "%s: skip request sense, past eh deadline\n",
1179                                              current->comm));
1180                         break;
1181                 }
1182                 if (status_byte(scmd->result) != CHECK_CONDITION)
1183                         /*
1184                          * don't request sense if there's no check condition
1185                          * status because the error we're processing isn't one
1186                          * that has a sense code (and some devices get
1187                          * confused by sense requests out of the blue)
1188                          */
1189                         continue;
1190
1191                 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1192                                                   "%s: requesting sense\n",
1193                                                   current->comm));
1194                 rtn = scsi_request_sense(scmd);
1195                 if (rtn != SUCCESS)
1196                         continue;
1197
1198                 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1199                         "sense requested, result %x\n", scmd->result));
1200                 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1201
1202                 rtn = scsi_decide_disposition(scmd);
1203
1204                 /*
1205                  * if the result was normal, then just pass it along to the
1206                  * upper level.
1207                  */
1208                 if (rtn == SUCCESS)
1209                         /* we don't want this command reissued, just
1210                          * finished with the sense data, so set
1211                          * retries to the max allowed to ensure it
1212                          * won't get reissued */
1213                         scmd->retries = scmd->allowed;
1214                 else if (rtn != NEEDS_RETRY)
1215                         continue;
1216
1217                 scsi_eh_finish_cmd(scmd, done_q);
1218         }
1219
1220         return list_empty(work_q);
1221 }
1222 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1223
1224 /**
1225  * scsi_eh_tur - Send TUR to device.
1226  * @scmd:       &scsi_cmnd to send TUR
1227  *
1228  * Return value:
1229  *    0 - Device is ready. 1 - Device NOT ready.
1230  */
1231 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1232 {
1233         static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1234         int retry_cnt = 1, rtn;
1235
1236 retry_tur:
1237         rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1238                                 scmd->device->eh_timeout, 0);
1239
1240         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1241                 "%s return: %x\n", __func__, rtn));
1242
1243         switch (rtn) {
1244         case NEEDS_RETRY:
1245                 if (retry_cnt--)
1246                         goto retry_tur;
1247                 /*FALLTHRU*/
1248         case SUCCESS:
1249                 return 0;
1250         default:
1251                 return 1;
1252         }
1253 }
1254
1255 /**
1256  * scsi_eh_test_devices - check if devices are responding from error recovery.
1257  * @cmd_list:   scsi commands in error recovery.
1258  * @work_q:     queue for commands which still need more error recovery
1259  * @done_q:     queue for commands which are finished
1260  * @try_stu:    boolean on if a STU command should be tried in addition to TUR.
1261  *
1262  * Decription:
1263  *    Tests if devices are in a working state.  Commands to devices now in
1264  *    a working state are sent to the done_q while commands to devices which
1265  *    are still failing to respond are returned to the work_q for more
1266  *    processing.
1267  **/
1268 static int scsi_eh_test_devices(struct list_head *cmd_list,
1269                                 struct list_head *work_q,
1270                                 struct list_head *done_q, int try_stu)
1271 {
1272         struct scsi_cmnd *scmd, *next;
1273         struct scsi_device *sdev;
1274         int finish_cmds;
1275
1276         while (!list_empty(cmd_list)) {
1277                 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1278                 sdev = scmd->device;
1279
1280                 if (!try_stu) {
1281                         if (scsi_host_eh_past_deadline(sdev->host)) {
1282                                 /* Push items back onto work_q */
1283                                 list_splice_init(cmd_list, work_q);
1284                                 SCSI_LOG_ERROR_RECOVERY(3,
1285                                         sdev_printk(KERN_INFO, sdev,
1286                                                     "%s: skip test device, past eh deadline",
1287                                                     current->comm));
1288                                 break;
1289                         }
1290                 }
1291
1292                 finish_cmds = !scsi_device_online(scmd->device) ||
1293                         (try_stu && !scsi_eh_try_stu(scmd) &&
1294                          !scsi_eh_tur(scmd)) ||
1295                         !scsi_eh_tur(scmd);
1296
1297                 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1298                         if (scmd->device == sdev) {
1299                                 if (finish_cmds &&
1300                                     (try_stu ||
1301                                      scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1302                                         scsi_eh_finish_cmd(scmd, done_q);
1303                                 else
1304                                         list_move_tail(&scmd->eh_entry, work_q);
1305                         }
1306         }
1307         return list_empty(work_q);
1308 }
1309
1310
1311 /**
1312  * scsi_eh_abort_cmds - abort pending commands.
1313  * @work_q:     &list_head for pending commands.
1314  * @done_q:     &list_head for processed commands.
1315  *
1316  * Decription:
1317  *    Try and see whether or not it makes sense to try and abort the
1318  *    running command.  This only works out to be the case if we have one
1319  *    command that has timed out.  If the command simply failed, it makes
1320  *    no sense to try and abort the command, since as far as the shost
1321  *    adapter is concerned, it isn't running.
1322  */
1323 static int scsi_eh_abort_cmds(struct list_head *work_q,
1324                               struct list_head *done_q)
1325 {
1326         struct scsi_cmnd *scmd, *next;
1327         LIST_HEAD(check_list);
1328         int rtn;
1329         struct Scsi_Host *shost;
1330
1331         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1332                 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1333                         continue;
1334                 shost = scmd->device->host;
1335                 if (scsi_host_eh_past_deadline(shost)) {
1336                         list_splice_init(&check_list, work_q);
1337                         SCSI_LOG_ERROR_RECOVERY(3,
1338                                 scmd_printk(KERN_INFO, scmd,
1339                                             "%s: skip aborting cmd, past eh deadline\n",
1340                                             current->comm));
1341                         return list_empty(work_q);
1342                 }
1343                 SCSI_LOG_ERROR_RECOVERY(3,
1344                         scmd_printk(KERN_INFO, scmd,
1345                                      "%s: aborting cmd\n", current->comm));
1346                 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
1347                 if (rtn == FAILED) {
1348                         SCSI_LOG_ERROR_RECOVERY(3,
1349                                 scmd_printk(KERN_INFO, scmd,
1350                                             "%s: aborting cmd failed\n",
1351                                              current->comm));
1352                         list_splice_init(&check_list, work_q);
1353                         return list_empty(work_q);
1354                 }
1355                 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1356                 if (rtn == FAST_IO_FAIL)
1357                         scsi_eh_finish_cmd(scmd, done_q);
1358                 else
1359                         list_move_tail(&scmd->eh_entry, &check_list);
1360         }
1361
1362         return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1363 }
1364
1365 /**
1366  * scsi_eh_try_stu - Send START_UNIT to device.
1367  * @scmd:       &scsi_cmnd to send START_UNIT
1368  *
1369  * Return value:
1370  *    0 - Device is ready. 1 - Device NOT ready.
1371  */
1372 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1373 {
1374         static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1375
1376         if (scmd->device->allow_restart) {
1377                 int i, rtn = NEEDS_RETRY;
1378
1379                 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1380                         rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1381
1382                 if (rtn == SUCCESS)
1383                         return 0;
1384         }
1385
1386         return 1;
1387 }
1388
1389  /**
1390  * scsi_eh_stu - send START_UNIT if needed
1391  * @shost:      &scsi host being recovered.
1392  * @work_q:     &list_head for pending commands.
1393  * @done_q:     &list_head for processed commands.
1394  *
1395  * Notes:
1396  *    If commands are failing due to not ready, initializing command required,
1397  *      try revalidating the device, which will end up sending a start unit.
1398  */
1399 static int scsi_eh_stu(struct Scsi_Host *shost,
1400                               struct list_head *work_q,
1401                               struct list_head *done_q)
1402 {
1403         struct scsi_cmnd *scmd, *stu_scmd, *next;
1404         struct scsi_device *sdev;
1405
1406         shost_for_each_device(sdev, shost) {
1407                 if (scsi_host_eh_past_deadline(shost)) {
1408                         SCSI_LOG_ERROR_RECOVERY(3,
1409                                 sdev_printk(KERN_INFO, sdev,
1410                                             "%s: skip START_UNIT, past eh deadline\n",
1411                                             current->comm));
1412                         break;
1413                 }
1414                 stu_scmd = NULL;
1415                 list_for_each_entry(scmd, work_q, eh_entry)
1416                         if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1417                             scsi_check_sense(scmd) == FAILED ) {
1418                                 stu_scmd = scmd;
1419                                 break;
1420                         }
1421
1422                 if (!stu_scmd)
1423                         continue;
1424
1425                 SCSI_LOG_ERROR_RECOVERY(3,
1426                         sdev_printk(KERN_INFO, sdev,
1427                                      "%s: Sending START_UNIT\n",
1428                                     current->comm));
1429
1430                 if (!scsi_eh_try_stu(stu_scmd)) {
1431                         if (!scsi_device_online(sdev) ||
1432                             !scsi_eh_tur(stu_scmd)) {
1433                                 list_for_each_entry_safe(scmd, next,
1434                                                           work_q, eh_entry) {
1435                                         if (scmd->device == sdev &&
1436                                             scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1437                                                 scsi_eh_finish_cmd(scmd, done_q);
1438                                 }
1439                         }
1440                 } else {
1441                         SCSI_LOG_ERROR_RECOVERY(3,
1442                                 sdev_printk(KERN_INFO, sdev,
1443                                             "%s: START_UNIT failed\n",
1444                                             current->comm));
1445                 }
1446         }
1447
1448         return list_empty(work_q);
1449 }
1450
1451
1452 /**
1453  * scsi_eh_bus_device_reset - send bdr if needed
1454  * @shost:      scsi host being recovered.
1455  * @work_q:     &list_head for pending commands.
1456  * @done_q:     &list_head for processed commands.
1457  *
1458  * Notes:
1459  *    Try a bus device reset.  Still, look to see whether we have multiple
1460  *    devices that are jammed or not - if we have multiple devices, it
1461  *    makes no sense to try bus_device_reset - we really would need to try
1462  *    a bus_reset instead.
1463  */
1464 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1465                                     struct list_head *work_q,
1466                                     struct list_head *done_q)
1467 {
1468         struct scsi_cmnd *scmd, *bdr_scmd, *next;
1469         struct scsi_device *sdev;
1470         int rtn;
1471
1472         shost_for_each_device(sdev, shost) {
1473                 if (scsi_host_eh_past_deadline(shost)) {
1474                         SCSI_LOG_ERROR_RECOVERY(3,
1475                                 sdev_printk(KERN_INFO, sdev,
1476                                             "%s: skip BDR, past eh deadline\n",
1477                                              current->comm));
1478                         break;
1479                 }
1480                 bdr_scmd = NULL;
1481                 list_for_each_entry(scmd, work_q, eh_entry)
1482                         if (scmd->device == sdev) {
1483                                 bdr_scmd = scmd;
1484                                 break;
1485                         }
1486
1487                 if (!bdr_scmd)
1488                         continue;
1489
1490                 SCSI_LOG_ERROR_RECOVERY(3,
1491                         sdev_printk(KERN_INFO, sdev,
1492                                      "%s: Sending BDR\n", current->comm));
1493                 rtn = scsi_try_bus_device_reset(bdr_scmd);
1494                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1495                         if (!scsi_device_online(sdev) ||
1496                             rtn == FAST_IO_FAIL ||
1497                             !scsi_eh_tur(bdr_scmd)) {
1498                                 list_for_each_entry_safe(scmd, next,
1499                                                          work_q, eh_entry) {
1500                                         if (scmd->device == sdev &&
1501                                             scsi_eh_action(scmd, rtn) != FAILED)
1502                                                 scsi_eh_finish_cmd(scmd,
1503                                                                    done_q);
1504                                 }
1505                         }
1506                 } else {
1507                         SCSI_LOG_ERROR_RECOVERY(3,
1508                                 sdev_printk(KERN_INFO, sdev,
1509                                             "%s: BDR failed\n", current->comm));
1510                 }
1511         }
1512
1513         return list_empty(work_q);
1514 }
1515
1516 /**
1517  * scsi_eh_target_reset - send target reset if needed
1518  * @shost:      scsi host being recovered.
1519  * @work_q:     &list_head for pending commands.
1520  * @done_q:     &list_head for processed commands.
1521  *
1522  * Notes:
1523  *    Try a target reset.
1524  */
1525 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1526                                 struct list_head *work_q,
1527                                 struct list_head *done_q)
1528 {
1529         LIST_HEAD(tmp_list);
1530         LIST_HEAD(check_list);
1531
1532         list_splice_init(work_q, &tmp_list);
1533
1534         while (!list_empty(&tmp_list)) {
1535                 struct scsi_cmnd *next, *scmd;
1536                 int rtn;
1537                 unsigned int id;
1538
1539                 if (scsi_host_eh_past_deadline(shost)) {
1540                         /* push back on work queue for further processing */
1541                         list_splice_init(&check_list, work_q);
1542                         list_splice_init(&tmp_list, work_q);
1543                         SCSI_LOG_ERROR_RECOVERY(3,
1544                                 shost_printk(KERN_INFO, shost,
1545                                             "%s: Skip target reset, past eh deadline\n",
1546                                              current->comm));
1547                         return list_empty(work_q);
1548                 }
1549
1550                 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1551                 id = scmd_id(scmd);
1552
1553                 SCSI_LOG_ERROR_RECOVERY(3,
1554                         shost_printk(KERN_INFO, shost,
1555                                      "%s: Sending target reset to target %d\n",
1556                                      current->comm, id));
1557                 rtn = scsi_try_target_reset(scmd);
1558                 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1559                         SCSI_LOG_ERROR_RECOVERY(3,
1560                                 shost_printk(KERN_INFO, shost,
1561                                              "%s: Target reset failed"
1562                                              " target: %d\n",
1563                                              current->comm, id));
1564                 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1565                         if (scmd_id(scmd) != id)
1566                                 continue;
1567
1568                         if (rtn == SUCCESS)
1569                                 list_move_tail(&scmd->eh_entry, &check_list);
1570                         else if (rtn == FAST_IO_FAIL)
1571                                 scsi_eh_finish_cmd(scmd, done_q);
1572                         else
1573                                 /* push back on work queue for further processing */
1574                                 list_move(&scmd->eh_entry, work_q);
1575                 }
1576         }
1577
1578         return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1579 }
1580
1581 /**
1582  * scsi_eh_bus_reset - send a bus reset
1583  * @shost:      &scsi host being recovered.
1584  * @work_q:     &list_head for pending commands.
1585  * @done_q:     &list_head for processed commands.
1586  */
1587 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1588                              struct list_head *work_q,
1589                              struct list_head *done_q)
1590 {
1591         struct scsi_cmnd *scmd, *chan_scmd, *next;
1592         LIST_HEAD(check_list);
1593         unsigned int channel;
1594         int rtn;
1595
1596         /*
1597          * we really want to loop over the various channels, and do this on
1598          * a channel by channel basis.  we should also check to see if any
1599          * of the failed commands are on soft_reset devices, and if so, skip
1600          * the reset.
1601          */
1602
1603         for (channel = 0; channel <= shost->max_channel; channel++) {
1604                 if (scsi_host_eh_past_deadline(shost)) {
1605                         list_splice_init(&check_list, work_q);
1606                         SCSI_LOG_ERROR_RECOVERY(3,
1607                                 shost_printk(KERN_INFO, shost,
1608                                             "%s: skip BRST, past eh deadline\n",
1609                                              current->comm));
1610                         return list_empty(work_q);
1611                 }
1612
1613                 chan_scmd = NULL;
1614                 list_for_each_entry(scmd, work_q, eh_entry) {
1615                         if (channel == scmd_channel(scmd)) {
1616                                 chan_scmd = scmd;
1617                                 break;
1618                                 /*
1619                                  * FIXME add back in some support for
1620                                  * soft_reset devices.
1621                                  */
1622                         }
1623                 }
1624
1625                 if (!chan_scmd)
1626                         continue;
1627                 SCSI_LOG_ERROR_RECOVERY(3,
1628                         shost_printk(KERN_INFO, shost,
1629                                      "%s: Sending BRST chan: %d\n",
1630                                      current->comm, channel));
1631                 rtn = scsi_try_bus_reset(chan_scmd);
1632                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1633                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1634                                 if (channel == scmd_channel(scmd)) {
1635                                         if (rtn == FAST_IO_FAIL)
1636                                                 scsi_eh_finish_cmd(scmd,
1637                                                                    done_q);
1638                                         else
1639                                                 list_move_tail(&scmd->eh_entry,
1640                                                                &check_list);
1641                                 }
1642                         }
1643                 } else {
1644                         SCSI_LOG_ERROR_RECOVERY(3,
1645                                 shost_printk(KERN_INFO, shost,
1646                                              "%s: BRST failed chan: %d\n",
1647                                              current->comm, channel));
1648                 }
1649         }
1650         return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1651 }
1652
1653 /**
1654  * scsi_eh_host_reset - send a host reset
1655  * @shost:      host to be reset.
1656  * @work_q:     &list_head for pending commands.
1657  * @done_q:     &list_head for processed commands.
1658  */
1659 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1660                               struct list_head *work_q,
1661                               struct list_head *done_q)
1662 {
1663         struct scsi_cmnd *scmd, *next;
1664         LIST_HEAD(check_list);
1665         int rtn;
1666
1667         if (!list_empty(work_q)) {
1668                 scmd = list_entry(work_q->next,
1669                                   struct scsi_cmnd, eh_entry);
1670
1671                 SCSI_LOG_ERROR_RECOVERY(3,
1672                         shost_printk(KERN_INFO, shost,
1673                                      "%s: Sending HRST\n",
1674                                      current->comm));
1675
1676                 rtn = scsi_try_host_reset(scmd);
1677                 if (rtn == SUCCESS) {
1678                         list_splice_init(work_q, &check_list);
1679                 } else if (rtn == FAST_IO_FAIL) {
1680                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1681                                         scsi_eh_finish_cmd(scmd, done_q);
1682                         }
1683                 } else {
1684                         SCSI_LOG_ERROR_RECOVERY(3,
1685                                 shost_printk(KERN_INFO, shost,
1686                                              "%s: HRST failed\n",
1687                                              current->comm));
1688                 }
1689         }
1690         return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1691 }
1692
1693 /**
1694  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1695  * @work_q:     &list_head for pending commands.
1696  * @done_q:     &list_head for processed commands.
1697  */
1698 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1699                                   struct list_head *done_q)
1700 {
1701         struct scsi_cmnd *scmd, *next;
1702
1703         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1704                 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1705                             "not ready after error recovery\n");
1706                 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1707                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1708                         /*
1709                          * FIXME: Handle lost cmds.
1710                          */
1711                 }
1712                 scsi_eh_finish_cmd(scmd, done_q);
1713         }
1714         return;
1715 }
1716
1717 /**
1718  * scsi_noretry_cmd - determine if command should be failed fast
1719  * @scmd:       SCSI cmd to examine.
1720  */
1721 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1722 {
1723         switch (host_byte(scmd->result)) {
1724         case DID_OK:
1725                 break;
1726         case DID_TIME_OUT:
1727                 goto check_type;
1728         case DID_BUS_BUSY:
1729                 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1730         case DID_PARITY:
1731                 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1732         case DID_ERROR:
1733                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1734                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1735                         return 0;
1736                 /* fall through */
1737         case DID_SOFT_ERROR:
1738                 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1739         }
1740
1741         if (status_byte(scmd->result) != CHECK_CONDITION)
1742                 return 0;
1743
1744 check_type:
1745         /*
1746          * assume caller has checked sense and determined
1747          * the check condition was retryable.
1748          */
1749         if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1750             scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1751                 return 1;
1752         else
1753                 return 0;
1754 }
1755
1756 /**
1757  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1758  * @scmd:       SCSI cmd to examine.
1759  *
1760  * Notes:
1761  *    This is *only* called when we are examining the status after sending
1762  *    out the actual data command.  any commands that are queued for error
1763  *    recovery (e.g. test_unit_ready) do *not* come through here.
1764  *
1765  *    When this routine returns failed, it means the error handler thread
1766  *    is woken.  In cases where the error code indicates an error that
1767  *    doesn't require the error handler read (i.e. we don't need to
1768  *    abort/reset), this function should return SUCCESS.
1769  */
1770 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1771 {
1772         int rtn;
1773
1774         /*
1775          * if the device is offline, then we clearly just pass the result back
1776          * up to the top level.
1777          */
1778         if (!scsi_device_online(scmd->device)) {
1779                 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1780                         "%s: device offline - report as SUCCESS\n", __func__));
1781                 return SUCCESS;
1782         }
1783
1784         /*
1785          * first check the host byte, to see if there is anything in there
1786          * that would indicate what we need to do.
1787          */
1788         switch (host_byte(scmd->result)) {
1789         case DID_PASSTHROUGH:
1790                 /*
1791                  * no matter what, pass this through to the upper layer.
1792                  * nuke this special code so that it looks like we are saying
1793                  * did_ok.
1794                  */
1795                 scmd->result &= 0xff00ffff;
1796                 return SUCCESS;
1797         case DID_OK:
1798                 /*
1799                  * looks good.  drop through, and check the next byte.
1800                  */
1801                 break;
1802         case DID_ABORT:
1803                 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1804                         set_host_byte(scmd, DID_TIME_OUT);
1805                         return SUCCESS;
1806                 }
1807         case DID_NO_CONNECT:
1808         case DID_BAD_TARGET:
1809                 /*
1810                  * note - this means that we just report the status back
1811                  * to the top level driver, not that we actually think
1812                  * that it indicates SUCCESS.
1813                  */
1814                 return SUCCESS;
1815                 /*
1816                  * when the low level driver returns did_soft_error,
1817                  * it is responsible for keeping an internal retry counter
1818                  * in order to avoid endless loops (db)
1819                  *
1820                  * actually this is a bug in this function here.  we should
1821                  * be mindful of the maximum number of retries specified
1822                  * and not get stuck in a loop.
1823                  */
1824         case DID_SOFT_ERROR:
1825                 goto maybe_retry;
1826         case DID_IMM_RETRY:
1827                 return NEEDS_RETRY;
1828
1829         case DID_REQUEUE:
1830                 return ADD_TO_MLQUEUE;
1831         case DID_TRANSPORT_DISRUPTED:
1832                 /*
1833                  * LLD/transport was disrupted during processing of the IO.
1834                  * The transport class is now blocked/blocking,
1835                  * and the transport will decide what to do with the IO
1836                  * based on its timers and recovery capablilities if
1837                  * there are enough retries.
1838                  */
1839                 goto maybe_retry;
1840         case DID_TRANSPORT_FAILFAST:
1841                 /*
1842                  * The transport decided to failfast the IO (most likely
1843                  * the fast io fail tmo fired), so send IO directly upwards.
1844                  */
1845                 return SUCCESS;
1846         case DID_ERROR:
1847                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1848                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1849                         /*
1850                          * execute reservation conflict processing code
1851                          * lower down
1852                          */
1853                         break;
1854                 /* fallthrough */
1855         case DID_BUS_BUSY:
1856         case DID_PARITY:
1857                 goto maybe_retry;
1858         case DID_TIME_OUT:
1859                 /*
1860                  * when we scan the bus, we get timeout messages for
1861                  * these commands if there is no device available.
1862                  * other hosts report did_no_connect for the same thing.
1863                  */
1864                 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1865                      scmd->cmnd[0] == INQUIRY)) {
1866                         return SUCCESS;
1867                 } else {
1868                         return FAILED;
1869                 }
1870         case DID_RESET:
1871                 return SUCCESS;
1872         default:
1873                 return FAILED;
1874         }
1875
1876         /*
1877          * next, check the message byte.
1878          */
1879         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1880                 return FAILED;
1881
1882         /*
1883          * check the status byte to see if this indicates anything special.
1884          */
1885         switch (status_byte(scmd->result)) {
1886         case QUEUE_FULL:
1887                 scsi_handle_queue_full(scmd->device);
1888                 /*
1889                  * the case of trying to send too many commands to a
1890                  * tagged queueing device.
1891                  */
1892         case BUSY:
1893                 /*
1894                  * device can't talk to us at the moment.  Should only
1895                  * occur (SAM-3) when the task queue is empty, so will cause
1896                  * the empty queue handling to trigger a stall in the
1897                  * device.
1898                  */
1899                 return ADD_TO_MLQUEUE;
1900         case GOOD:
1901                 if (scmd->cmnd[0] == REPORT_LUNS)
1902                         scmd->device->sdev_target->expecting_lun_change = 0;
1903                 scsi_handle_queue_ramp_up(scmd->device);
1904         case COMMAND_TERMINATED:
1905                 return SUCCESS;
1906         case TASK_ABORTED:
1907                 goto maybe_retry;
1908         case CHECK_CONDITION:
1909                 rtn = scsi_check_sense(scmd);
1910                 if (rtn == NEEDS_RETRY)
1911                         goto maybe_retry;
1912                 /* if rtn == FAILED, we have no sense information;
1913                  * returning FAILED will wake the error handler thread
1914                  * to collect the sense and redo the decide
1915                  * disposition */
1916                 return rtn;
1917         case CONDITION_GOOD:
1918         case INTERMEDIATE_GOOD:
1919         case INTERMEDIATE_C_GOOD:
1920         case ACA_ACTIVE:
1921                 /*
1922                  * who knows?  FIXME(eric)
1923                  */
1924                 return SUCCESS;
1925
1926         case RESERVATION_CONFLICT:
1927                 sdev_printk(KERN_INFO, scmd->device,
1928                             "reservation conflict\n");
1929                 set_host_byte(scmd, DID_NEXUS_FAILURE);
1930                 return SUCCESS; /* causes immediate i/o error */
1931         default:
1932                 return FAILED;
1933         }
1934         return FAILED;
1935
1936       maybe_retry:
1937
1938         /* we requeue for retry because the error was retryable, and
1939          * the request was not marked fast fail.  Note that above,
1940          * even if the request is marked fast fail, we still requeue
1941          * for queue congestion conditions (QUEUE_FULL or BUSY) */
1942         if ((++scmd->retries) <= scmd->allowed
1943             && !scsi_noretry_cmd(scmd)) {
1944                 return NEEDS_RETRY;
1945         } else {
1946                 /*
1947                  * no more retries - report this one back to upper level.
1948                  */
1949                 return SUCCESS;
1950         }
1951 }
1952
1953 static void eh_lock_door_done(struct request *req, int uptodate)
1954 {
1955         __blk_put_request(req->q, req);
1956 }
1957
1958 /**
1959  * scsi_eh_lock_door - Prevent medium removal for the specified device
1960  * @sdev:       SCSI device to prevent medium removal
1961  *
1962  * Locking:
1963  *      We must be called from process context.
1964  *
1965  * Notes:
1966  *      We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1967  *      head of the devices request queue, and continue.
1968  */
1969 static void scsi_eh_lock_door(struct scsi_device *sdev)
1970 {
1971         struct request *req;
1972
1973         /*
1974          * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a
1975          * request becomes available
1976          */
1977         req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1978         if (IS_ERR(req))
1979                 return;
1980
1981         blk_rq_set_block_pc(req);
1982
1983         req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1984         req->cmd[1] = 0;
1985         req->cmd[2] = 0;
1986         req->cmd[3] = 0;
1987         req->cmd[4] = SCSI_REMOVAL_PREVENT;
1988         req->cmd[5] = 0;
1989
1990         req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1991
1992         req->cmd_flags |= REQ_QUIET;
1993         req->timeout = 10 * HZ;
1994         req->retries = 5;
1995
1996         blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1997 }
1998
1999 /**
2000  * scsi_restart_operations - restart io operations to the specified host.
2001  * @shost:      Host we are restarting.
2002  *
2003  * Notes:
2004  *    When we entered the error handler, we blocked all further i/o to
2005  *    this device.  we need to 'reverse' this process.
2006  */
2007 static void scsi_restart_operations(struct Scsi_Host *shost)
2008 {
2009         struct scsi_device *sdev;
2010         unsigned long flags;
2011
2012         /*
2013          * If the door was locked, we need to insert a door lock request
2014          * onto the head of the SCSI request queue for the device.  There
2015          * is no point trying to lock the door of an off-line device.
2016          */
2017         shost_for_each_device(sdev, shost) {
2018                 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2019                         scsi_eh_lock_door(sdev);
2020                         sdev->was_reset = 0;
2021                 }
2022         }
2023
2024         /*
2025          * next free up anything directly waiting upon the host.  this
2026          * will be requests for character device operations, and also for
2027          * ioctls to queued block devices.
2028          */
2029         SCSI_LOG_ERROR_RECOVERY(3,
2030                 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2031
2032         spin_lock_irqsave(shost->host_lock, flags);
2033         if (scsi_host_set_state(shost, SHOST_RUNNING))
2034                 if (scsi_host_set_state(shost, SHOST_CANCEL))
2035                         BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2036         spin_unlock_irqrestore(shost->host_lock, flags);
2037
2038         wake_up(&shost->host_wait);
2039
2040         /*
2041          * finally we need to re-initiate requests that may be pending.  we will
2042          * have had everything blocked while error handling is taking place, and
2043          * now that error recovery is done, we will need to ensure that these
2044          * requests are started.
2045          */
2046         scsi_run_host_queues(shost);
2047
2048         /*
2049          * if eh is active and host_eh_scheduled is pending we need to re-run
2050          * recovery.  we do this check after scsi_run_host_queues() to allow
2051          * everything pent up since the last eh run a chance to make forward
2052          * progress before we sync again.  Either we'll immediately re-run
2053          * recovery or scsi_device_unbusy() will wake us again when these
2054          * pending commands complete.
2055          */
2056         spin_lock_irqsave(shost->host_lock, flags);
2057         if (shost->host_eh_scheduled)
2058                 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2059                         WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2060         spin_unlock_irqrestore(shost->host_lock, flags);
2061 }
2062
2063 /**
2064  * scsi_eh_ready_devs - check device ready state and recover if not.
2065  * @shost:      host to be recovered.
2066  * @work_q:     &list_head for pending commands.
2067  * @done_q:     &list_head for processed commands.
2068  */
2069 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2070                         struct list_head *work_q,
2071                         struct list_head *done_q)
2072 {
2073         if (!scsi_eh_stu(shost, work_q, done_q))
2074                 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2075                         if (!scsi_eh_target_reset(shost, work_q, done_q))
2076                                 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2077                                         if (!scsi_eh_host_reset(shost, work_q, done_q))
2078                                                 scsi_eh_offline_sdevs(work_q,
2079                                                                       done_q);
2080 }
2081 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2082
2083 /**
2084  * scsi_eh_flush_done_q - finish processed commands or retry them.
2085  * @done_q:     list_head of processed commands.
2086  */
2087 void scsi_eh_flush_done_q(struct list_head *done_q)
2088 {
2089         struct scsi_cmnd *scmd, *next;
2090
2091         list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2092                 list_del_init(&scmd->eh_entry);
2093                 if (scsi_device_online(scmd->device) &&
2094                     !scsi_noretry_cmd(scmd) &&
2095                     (++scmd->retries <= scmd->allowed)) {
2096                         SCSI_LOG_ERROR_RECOVERY(3,
2097                                 scmd_printk(KERN_INFO, scmd,
2098                                              "%s: flush retry cmd\n",
2099                                              current->comm));
2100                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2101                 } else {
2102                         /*
2103                          * If just we got sense for the device (called
2104                          * scsi_eh_get_sense), scmd->result is already
2105                          * set, do not set DRIVER_TIMEOUT.
2106                          */
2107                         if (!scmd->result)
2108                                 scmd->result |= (DRIVER_TIMEOUT << 24);
2109                         SCSI_LOG_ERROR_RECOVERY(3,
2110                                 scmd_printk(KERN_INFO, scmd,
2111                                              "%s: flush finish cmd\n",
2112                                              current->comm));
2113                         scsi_finish_command(scmd);
2114                 }
2115         }
2116 }
2117 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2118
2119 /**
2120  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2121  * @shost:      Host to unjam.
2122  *
2123  * Notes:
2124  *    When we come in here, we *know* that all commands on the bus have
2125  *    either completed, failed or timed out.  we also know that no further
2126  *    commands are being sent to the host, so things are relatively quiet
2127  *    and we have freedom to fiddle with things as we wish.
2128  *
2129  *    This is only the *default* implementation.  it is possible for
2130  *    individual drivers to supply their own version of this function, and
2131  *    if the maintainer wishes to do this, it is strongly suggested that
2132  *    this function be taken as a template and modified.  this function
2133  *    was designed to correctly handle problems for about 95% of the
2134  *    different cases out there, and it should always provide at least a
2135  *    reasonable amount of error recovery.
2136  *
2137  *    Any command marked 'failed' or 'timeout' must eventually have
2138  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
2139  *    here, so when we restart the host after we return it should have an
2140  *    empty queue.
2141  */
2142 static void scsi_unjam_host(struct Scsi_Host *shost)
2143 {
2144         unsigned long flags;
2145         LIST_HEAD(eh_work_q);
2146         LIST_HEAD(eh_done_q);
2147
2148         spin_lock_irqsave(shost->host_lock, flags);
2149         list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2150         spin_unlock_irqrestore(shost->host_lock, flags);
2151
2152         SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2153
2154         if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2155                 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2156                         scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2157
2158         spin_lock_irqsave(shost->host_lock, flags);
2159         if (shost->eh_deadline != -1)
2160                 shost->last_reset = 0;
2161         spin_unlock_irqrestore(shost->host_lock, flags);
2162         scsi_eh_flush_done_q(&eh_done_q);
2163 }
2164
2165 /**
2166  * scsi_error_handler - SCSI error handler thread
2167  * @data:       Host for which we are running.
2168  *
2169  * Notes:
2170  *    This is the main error handling loop.  This is run as a kernel thread
2171  *    for every SCSI host and handles all error handling activity.
2172  */
2173 int scsi_error_handler(void *data)
2174 {
2175         struct Scsi_Host *shost = data;
2176
2177         /*
2178          * We use TASK_INTERRUPTIBLE so that the thread is not
2179          * counted against the load average as a running process.
2180          * We never actually get interrupted because kthread_run
2181          * disables signal delivery for the created thread.
2182          */
2183         while (true) {
2184                 /*
2185                  * The sequence in kthread_stop() sets the stop flag first
2186                  * then wakes the process.  To avoid missed wakeups, the task
2187                  * should always be in a non running state before the stop
2188                  * flag is checked
2189                  */
2190                 set_current_state(TASK_INTERRUPTIBLE);
2191                 if (kthread_should_stop())
2192                         break;
2193
2194                 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2195                     shost->host_failed != atomic_read(&shost->host_busy)) {
2196                         SCSI_LOG_ERROR_RECOVERY(1,
2197                                 shost_printk(KERN_INFO, shost,
2198                                              "scsi_eh_%d: sleeping\n",
2199                                              shost->host_no));
2200                         schedule();
2201                         continue;
2202                 }
2203
2204                 __set_current_state(TASK_RUNNING);
2205                 SCSI_LOG_ERROR_RECOVERY(1,
2206                         shost_printk(KERN_INFO, shost,
2207                                      "scsi_eh_%d: waking up %d/%d/%d\n",
2208                                      shost->host_no, shost->host_eh_scheduled,
2209                                      shost->host_failed,
2210                                      atomic_read(&shost->host_busy)));
2211
2212                 /*
2213                  * We have a host that is failing for some reason.  Figure out
2214                  * what we need to do to get it up and online again (if we can).
2215                  * If we fail, we end up taking the thing offline.
2216                  */
2217                 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2218                         SCSI_LOG_ERROR_RECOVERY(1,
2219                                 shost_printk(KERN_ERR, shost,
2220                                              "scsi_eh_%d: unable to autoresume\n",
2221                                              shost->host_no));
2222                         continue;
2223                 }
2224
2225                 if (shost->transportt->eh_strategy_handler)
2226                         shost->transportt->eh_strategy_handler(shost);
2227                 else
2228                         scsi_unjam_host(shost);
2229
2230                 /*
2231                  * Note - if the above fails completely, the action is to take
2232                  * individual devices offline and flush the queue of any
2233                  * outstanding requests that may have been pending.  When we
2234                  * restart, we restart any I/O to any other devices on the bus
2235                  * which are still online.
2236                  */
2237                 scsi_restart_operations(shost);
2238                 if (!shost->eh_noresume)
2239                         scsi_autopm_put_host(shost);
2240         }
2241         __set_current_state(TASK_RUNNING);
2242
2243         SCSI_LOG_ERROR_RECOVERY(1,
2244                 shost_printk(KERN_INFO, shost,
2245                              "Error handler scsi_eh_%d exiting\n",
2246                              shost->host_no));
2247         shost->ehandler = NULL;
2248         return 0;
2249 }
2250
2251 /*
2252  * Function:    scsi_report_bus_reset()
2253  *
2254  * Purpose:     Utility function used by low-level drivers to report that
2255  *              they have observed a bus reset on the bus being handled.
2256  *
2257  * Arguments:   shost       - Host in question
2258  *              channel     - channel on which reset was observed.
2259  *
2260  * Returns:     Nothing
2261  *
2262  * Lock status: Host lock must be held.
2263  *
2264  * Notes:       This only needs to be called if the reset is one which
2265  *              originates from an unknown location.  Resets originated
2266  *              by the mid-level itself don't need to call this, but there
2267  *              should be no harm.
2268  *
2269  *              The main purpose of this is to make sure that a CHECK_CONDITION
2270  *              is properly treated.
2271  */
2272 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2273 {
2274         struct scsi_device *sdev;
2275
2276         __shost_for_each_device(sdev, shost) {
2277                 if (channel == sdev_channel(sdev))
2278                         __scsi_report_device_reset(sdev, NULL);
2279         }
2280 }
2281 EXPORT_SYMBOL(scsi_report_bus_reset);
2282
2283 /*
2284  * Function:    scsi_report_device_reset()
2285  *
2286  * Purpose:     Utility function used by low-level drivers to report that
2287  *              they have observed a device reset on the device being handled.
2288  *
2289  * Arguments:   shost       - Host in question
2290  *              channel     - channel on which reset was observed
2291  *              target      - target on which reset was observed
2292  *
2293  * Returns:     Nothing
2294  *
2295  * Lock status: Host lock must be held
2296  *
2297  * Notes:       This only needs to be called if the reset is one which
2298  *              originates from an unknown location.  Resets originated
2299  *              by the mid-level itself don't need to call this, but there
2300  *              should be no harm.
2301  *
2302  *              The main purpose of this is to make sure that a CHECK_CONDITION
2303  *              is properly treated.
2304  */
2305 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2306 {
2307         struct scsi_device *sdev;
2308
2309         __shost_for_each_device(sdev, shost) {
2310                 if (channel == sdev_channel(sdev) &&
2311                     target == sdev_id(sdev))
2312                         __scsi_report_device_reset(sdev, NULL);
2313         }
2314 }
2315 EXPORT_SYMBOL(scsi_report_device_reset);
2316
2317 static void
2318 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2319 {
2320 }
2321
2322 /**
2323  * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2324  * @dev:        scsi_device to operate on
2325  * @arg:        reset type (see sg.h)
2326  */
2327 int
2328 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2329 {
2330         struct scsi_cmnd *scmd;
2331         struct Scsi_Host *shost = dev->host;
2332         struct request req;
2333         unsigned long flags;
2334         int error = 0, rtn, val;
2335
2336         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2337                 return -EACCES;
2338
2339         error = get_user(val, arg);
2340         if (error)
2341                 return error;
2342
2343         if (scsi_autopm_get_host(shost) < 0)
2344                 return -EIO;
2345
2346         error = -EIO;
2347         scmd = scsi_get_command(dev, GFP_KERNEL);
2348         if (!scmd)
2349                 goto out_put_autopm_host;
2350
2351         blk_rq_init(NULL, &req);
2352         scmd->request = &req;
2353
2354         scmd->cmnd = req.cmd;
2355
2356         scmd->scsi_done         = scsi_reset_provider_done_command;
2357         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2358
2359         scmd->cmd_len                   = 0;
2360
2361         scmd->sc_data_direction         = DMA_BIDIRECTIONAL;
2362
2363         spin_lock_irqsave(shost->host_lock, flags);
2364         shost->tmf_in_progress = 1;
2365         spin_unlock_irqrestore(shost->host_lock, flags);
2366
2367         switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2368         case SG_SCSI_RESET_NOTHING:
2369                 rtn = SUCCESS;
2370                 break;
2371         case SG_SCSI_RESET_DEVICE:
2372                 rtn = scsi_try_bus_device_reset(scmd);
2373                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2374                         break;
2375                 /* FALLTHROUGH */
2376         case SG_SCSI_RESET_TARGET:
2377                 rtn = scsi_try_target_reset(scmd);
2378                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2379                         break;
2380                 /* FALLTHROUGH */
2381         case SG_SCSI_RESET_BUS:
2382                 rtn = scsi_try_bus_reset(scmd);
2383                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2384                         break;
2385                 /* FALLTHROUGH */
2386         case SG_SCSI_RESET_HOST:
2387                 rtn = scsi_try_host_reset(scmd);
2388                 if (rtn == SUCCESS)
2389                         break;
2390         default:
2391                 /* FALLTHROUGH */
2392                 rtn = FAILED;
2393                 break;
2394         }
2395
2396         error = (rtn == SUCCESS) ? 0 : -EIO;
2397
2398         spin_lock_irqsave(shost->host_lock, flags);
2399         shost->tmf_in_progress = 0;
2400         spin_unlock_irqrestore(shost->host_lock, flags);
2401
2402         /*
2403          * be sure to wake up anyone who was sleeping or had their queue
2404          * suspended while we performed the TMF.
2405          */
2406         SCSI_LOG_ERROR_RECOVERY(3,
2407                 shost_printk(KERN_INFO, shost,
2408                              "waking up host to restart after TMF\n"));
2409
2410         wake_up(&shost->host_wait);
2411         scsi_run_host_queues(shost);
2412
2413         scsi_put_command(scmd);
2414
2415 out_put_autopm_host:
2416         scsi_autopm_put_host(shost);
2417         return error;
2418 }
2419 EXPORT_SYMBOL(scsi_ioctl_reset);
2420
2421 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2422                                   struct scsi_sense_hdr *sshdr)
2423 {
2424         return scsi_normalize_sense(cmd->sense_buffer,
2425                         SCSI_SENSE_BUFFERSIZE, sshdr);
2426 }
2427 EXPORT_SYMBOL(scsi_command_normalize_sense);
2428
2429 /**
2430  * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2431  * @sense_buffer:       byte array of sense data
2432  * @sb_len:             number of valid bytes in sense_buffer
2433  * @info_out:           pointer to 64 integer where 8 or 4 byte information
2434  *                      field will be placed if found.
2435  *
2436  * Return value:
2437  *      1 if information field found, 0 if not found.
2438  */
2439 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2440                             u64 * info_out)
2441 {
2442         int j;
2443         const u8 * ucp;
2444         u64 ull;
2445
2446         if (sb_len < 7)
2447                 return 0;
2448         switch (sense_buffer[0] & 0x7f) {
2449         case 0x70:
2450         case 0x71:
2451                 if (sense_buffer[0] & 0x80) {
2452                         *info_out = (sense_buffer[3] << 24) +
2453                                     (sense_buffer[4] << 16) +
2454                                     (sense_buffer[5] << 8) + sense_buffer[6];
2455                         return 1;
2456                 } else
2457                         return 0;
2458         case 0x72:
2459         case 0x73:
2460                 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2461                                            0 /* info desc */);
2462                 if (ucp && (0xa == ucp[1])) {
2463                         ull = 0;
2464                         for (j = 0; j < 8; ++j) {
2465                                 if (j > 0)
2466                                         ull <<= 8;
2467                                 ull |= ucp[4 + j];
2468                         }
2469                         *info_out = ull;
2470                         return 1;
2471                 } else
2472                         return 0;
2473         default:
2474                 return 0;
2475         }
2476 }
2477 EXPORT_SYMBOL(scsi_get_sense_info_fld);