Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6
[sfrench/cifs-2.6.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
29 #include <scsi/scsi.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_transport.h>
35 #include <scsi/scsi_host.h>
36 #include <scsi/scsi_ioctl.h>
37
38 #include "scsi_priv.h"
39 #include "scsi_logging.h"
40 #include "scsi_transport_api.h"
41
42 #include <trace/events/scsi.h>
43
44 #define SENSE_TIMEOUT           (10*HZ)
45
46 /*
47  * These should *probably* be handled by the host itself.
48  * Since it is allowed to sleep, it probably should.
49  */
50 #define BUS_RESET_SETTLE_TIME   (10)
51 #define HOST_RESET_SETTLE_TIME  (10)
52
53 /* called with shost->host_lock held */
54 void scsi_eh_wakeup(struct Scsi_Host *shost)
55 {
56         if (shost->host_busy == shost->host_failed) {
57                 trace_scsi_eh_wakeup(shost);
58                 wake_up_process(shost->ehandler);
59                 SCSI_LOG_ERROR_RECOVERY(5,
60                                 printk("Waking error handler thread\n"));
61         }
62 }
63
64 /**
65  * scsi_schedule_eh - schedule EH for SCSI host
66  * @shost:      SCSI host to invoke error handling on.
67  *
68  * Schedule SCSI EH without scmd.
69  */
70 void scsi_schedule_eh(struct Scsi_Host *shost)
71 {
72         unsigned long flags;
73
74         spin_lock_irqsave(shost->host_lock, flags);
75
76         if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
77             scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
78                 shost->host_eh_scheduled++;
79                 scsi_eh_wakeup(shost);
80         }
81
82         spin_unlock_irqrestore(shost->host_lock, flags);
83 }
84 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
85
86 /**
87  * scsi_eh_scmd_add - add scsi cmd to error handling.
88  * @scmd:       scmd to run eh on.
89  * @eh_flag:    optional SCSI_EH flag.
90  *
91  * Return value:
92  *      0 on failure.
93  */
94 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
95 {
96         struct Scsi_Host *shost = scmd->device->host;
97         unsigned long flags;
98         int ret = 0;
99
100         if (!shost->ehandler)
101                 return 0;
102
103         spin_lock_irqsave(shost->host_lock, flags);
104         if (scsi_host_set_state(shost, SHOST_RECOVERY))
105                 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
106                         goto out_unlock;
107
108         ret = 1;
109         scmd->eh_eflags |= eh_flag;
110         list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
111         shost->host_failed++;
112         scsi_eh_wakeup(shost);
113  out_unlock:
114         spin_unlock_irqrestore(shost->host_lock, flags);
115         return ret;
116 }
117
118 /**
119  * scsi_times_out - Timeout function for normal scsi commands.
120  * @req:        request that is timing out.
121  *
122  * Notes:
123  *     We do not need to lock this.  There is the potential for a race
124  *     only in that the normal completion handling might run, but if the
125  *     normal completion function determines that the timer has already
126  *     fired, then it mustn't do anything.
127  */
128 enum blk_eh_timer_return scsi_times_out(struct request *req)
129 {
130         struct scsi_cmnd *scmd = req->special;
131         enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
132
133         trace_scsi_dispatch_cmd_timeout(scmd);
134         scsi_log_completion(scmd, TIMEOUT_ERROR);
135
136         if (scmd->device->host->transportt->eh_timed_out)
137                 rtn = scmd->device->host->transportt->eh_timed_out(scmd);
138         else if (scmd->device->host->hostt->eh_timed_out)
139                 rtn = scmd->device->host->hostt->eh_timed_out(scmd);
140
141         if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
142                      !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
143                 scmd->result |= DID_TIME_OUT << 16;
144                 rtn = BLK_EH_HANDLED;
145         }
146
147         return rtn;
148 }
149
150 /**
151  * scsi_block_when_processing_errors - Prevent cmds from being queued.
152  * @sdev:       Device on which we are performing recovery.
153  *
154  * Description:
155  *     We block until the host is out of error recovery, and then check to
156  *     see whether the host or the device is offline.
157  *
158  * Return value:
159  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
160  */
161 int scsi_block_when_processing_errors(struct scsi_device *sdev)
162 {
163         int online;
164
165         wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
166
167         online = scsi_device_online(sdev);
168
169         SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
170                                           online));
171
172         return online;
173 }
174 EXPORT_SYMBOL(scsi_block_when_processing_errors);
175
176 #ifdef CONFIG_SCSI_LOGGING
177 /**
178  * scsi_eh_prt_fail_stats - Log info on failures.
179  * @shost:      scsi host being recovered.
180  * @work_q:     Queue of scsi cmds to process.
181  */
182 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
183                                           struct list_head *work_q)
184 {
185         struct scsi_cmnd *scmd;
186         struct scsi_device *sdev;
187         int total_failures = 0;
188         int cmd_failed = 0;
189         int cmd_cancel = 0;
190         int devices_failed = 0;
191
192         shost_for_each_device(sdev, shost) {
193                 list_for_each_entry(scmd, work_q, eh_entry) {
194                         if (scmd->device == sdev) {
195                                 ++total_failures;
196                                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
197                                         ++cmd_cancel;
198                                 else 
199                                         ++cmd_failed;
200                         }
201                 }
202
203                 if (cmd_cancel || cmd_failed) {
204                         SCSI_LOG_ERROR_RECOVERY(3,
205                                 sdev_printk(KERN_INFO, sdev,
206                                             "%s: cmds failed: %d, cancel: %d\n",
207                                             __func__, cmd_failed,
208                                             cmd_cancel));
209                         cmd_cancel = 0;
210                         cmd_failed = 0;
211                         ++devices_failed;
212                 }
213         }
214
215         SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
216                                           " devices require eh work\n",
217                                   total_failures, devices_failed));
218 }
219 #endif
220
221 /**
222  * scsi_check_sense - Examine scsi cmd sense
223  * @scmd:       Cmd to have sense checked.
224  *
225  * Return value:
226  *      SUCCESS or FAILED or NEEDS_RETRY
227  *
228  * Notes:
229  *      When a deferred error is detected the current command has
230  *      not been executed and needs retrying.
231  */
232 static int scsi_check_sense(struct scsi_cmnd *scmd)
233 {
234         struct scsi_device *sdev = scmd->device;
235         struct scsi_sense_hdr sshdr;
236
237         if (! scsi_command_normalize_sense(scmd, &sshdr))
238                 return FAILED;  /* no valid sense data */
239
240         if (scsi_sense_is_deferred(&sshdr))
241                 return NEEDS_RETRY;
242
243         if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
244                         sdev->scsi_dh_data->scsi_dh->check_sense) {
245                 int rc;
246
247                 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
248                 if (rc != SCSI_RETURN_NOT_HANDLED)
249                         return rc;
250                 /* handler does not care. Drop down to default handling */
251         }
252
253         /*
254          * Previous logic looked for FILEMARK, EOM or ILI which are
255          * mainly associated with tapes and returned SUCCESS.
256          */
257         if (sshdr.response_code == 0x70) {
258                 /* fixed format */
259                 if (scmd->sense_buffer[2] & 0xe0)
260                         return SUCCESS;
261         } else {
262                 /*
263                  * descriptor format: look for "stream commands sense data
264                  * descriptor" (see SSC-3). Assume single sense data
265                  * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
266                  */
267                 if ((sshdr.additional_length > 3) &&
268                     (scmd->sense_buffer[8] == 0x4) &&
269                     (scmd->sense_buffer[11] & 0xe0))
270                         return SUCCESS;
271         }
272
273         switch (sshdr.sense_key) {
274         case NO_SENSE:
275                 return SUCCESS;
276         case RECOVERED_ERROR:
277                 return /* soft_error */ SUCCESS;
278
279         case ABORTED_COMMAND:
280                 if (sshdr.asc == 0x10) /* DIF */
281                         return SUCCESS;
282
283                 return NEEDS_RETRY;
284         case NOT_READY:
285         case UNIT_ATTENTION:
286                 /*
287                  * if we are expecting a cc/ua because of a bus reset that we
288                  * performed, treat this just as a retry.  otherwise this is
289                  * information that we should pass up to the upper-level driver
290                  * so that we can deal with it there.
291                  */
292                 if (scmd->device->expecting_cc_ua) {
293                         scmd->device->expecting_cc_ua = 0;
294                         return NEEDS_RETRY;
295                 }
296                 /*
297                  * if the device is in the process of becoming ready, we 
298                  * should retry.
299                  */
300                 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
301                         return NEEDS_RETRY;
302                 /*
303                  * if the device is not started, we need to wake
304                  * the error handler to start the motor
305                  */
306                 if (scmd->device->allow_restart &&
307                     (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
308                         return FAILED;
309
310                 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
311                         scmd_printk(KERN_WARNING, scmd,
312                                     "Warning! Received an indication that the "
313                                     "LUN assignments on this target have "
314                                     "changed. The Linux SCSI layer does not "
315                                     "automatically remap LUN assignments.\n");
316                 else if (sshdr.asc == 0x3f)
317                         scmd_printk(KERN_WARNING, scmd,
318                                     "Warning! Received an indication that the "
319                                     "operating parameters on this target have "
320                                     "changed. The Linux SCSI layer does not "
321                                     "automatically adjust these parameters.\n");
322
323                 if (scmd->request->cmd_flags & REQ_HARDBARRIER)
324                         /*
325                          * barrier requests should always retry on UA
326                          * otherwise block will get a spurious error
327                          */
328                         return NEEDS_RETRY;
329                 else
330                         /*
331                          * for normal (non barrier) commands, pass the
332                          * UA upwards for a determination in the
333                          * completion functions
334                          */
335                         return SUCCESS;
336
337                 /* these three are not supported */
338         case COPY_ABORTED:
339         case VOLUME_OVERFLOW:
340         case MISCOMPARE:
341                 return SUCCESS;
342
343         case MEDIUM_ERROR:
344                 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
345                     sshdr.asc == 0x13 || /* AMNF DATA FIELD */
346                     sshdr.asc == 0x14) { /* RECORD NOT FOUND */
347                         return SUCCESS;
348                 }
349                 return NEEDS_RETRY;
350
351         case HARDWARE_ERROR:
352                 if (scmd->device->retry_hwerror)
353                         return ADD_TO_MLQUEUE;
354                 else
355                         return SUCCESS;
356
357         case ILLEGAL_REQUEST:
358         case BLANK_CHECK:
359         case DATA_PROTECT:
360         default:
361                 return SUCCESS;
362         }
363 }
364
365 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
366 {
367         struct scsi_host_template *sht = sdev->host->hostt;
368         struct scsi_device *tmp_sdev;
369
370         if (!sht->change_queue_depth ||
371             sdev->queue_depth >= sdev->max_queue_depth)
372                 return;
373
374         if (time_before(jiffies,
375             sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
376                 return;
377
378         if (time_before(jiffies,
379             sdev->last_queue_full_time + sdev->queue_ramp_up_period))
380                 return;
381
382         /*
383          * Walk all devices of a target and do
384          * ramp up on them.
385          */
386         shost_for_each_device(tmp_sdev, sdev->host) {
387                 if (tmp_sdev->channel != sdev->channel ||
388                     tmp_sdev->id != sdev->id ||
389                     tmp_sdev->queue_depth == sdev->max_queue_depth)
390                         continue;
391                 /*
392                  * call back into LLD to increase queue_depth by one
393                  * with ramp up reason code.
394                  */
395                 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
396                                         SCSI_QDEPTH_RAMP_UP);
397                 sdev->last_queue_ramp_up = jiffies;
398         }
399 }
400
401 static void scsi_handle_queue_full(struct scsi_device *sdev)
402 {
403         struct scsi_host_template *sht = sdev->host->hostt;
404         struct scsi_device *tmp_sdev;
405
406         if (!sht->change_queue_depth)
407                 return;
408
409         shost_for_each_device(tmp_sdev, sdev->host) {
410                 if (tmp_sdev->channel != sdev->channel ||
411                     tmp_sdev->id != sdev->id)
412                         continue;
413                 /*
414                  * We do not know the number of commands that were at
415                  * the device when we got the queue full so we start
416                  * from the highest possible value and work our way down.
417                  */
418                 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
419                                         SCSI_QDEPTH_QFULL);
420         }
421 }
422
423 /**
424  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
425  * @scmd:       SCSI cmd to examine.
426  *
427  * Notes:
428  *    This is *only* called when we are examining the status of commands
429  *    queued during error recovery.  the main difference here is that we
430  *    don't allow for the possibility of retries here, and we are a lot
431  *    more restrictive about what we consider acceptable.
432  */
433 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
434 {
435         /*
436          * first check the host byte, to see if there is anything in there
437          * that would indicate what we need to do.
438          */
439         if (host_byte(scmd->result) == DID_RESET) {
440                 /*
441                  * rats.  we are already in the error handler, so we now
442                  * get to try and figure out what to do next.  if the sense
443                  * is valid, we have a pretty good idea of what to do.
444                  * if not, we mark it as FAILED.
445                  */
446                 return scsi_check_sense(scmd);
447         }
448         if (host_byte(scmd->result) != DID_OK)
449                 return FAILED;
450
451         /*
452          * next, check the message byte.
453          */
454         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
455                 return FAILED;
456
457         /*
458          * now, check the status byte to see if this indicates
459          * anything special.
460          */
461         switch (status_byte(scmd->result)) {
462         case GOOD:
463                 scsi_handle_queue_ramp_up(scmd->device);
464         case COMMAND_TERMINATED:
465                 return SUCCESS;
466         case CHECK_CONDITION:
467                 return scsi_check_sense(scmd);
468         case CONDITION_GOOD:
469         case INTERMEDIATE_GOOD:
470         case INTERMEDIATE_C_GOOD:
471                 /*
472                  * who knows?  FIXME(eric)
473                  */
474                 return SUCCESS;
475         case RESERVATION_CONFLICT:
476                 if (scmd->cmnd[0] == TEST_UNIT_READY)
477                         /* it is a success, we probed the device and
478                          * found it */
479                         return SUCCESS;
480                 /* otherwise, we failed to send the command */
481                 return FAILED;
482         case QUEUE_FULL:
483                 scsi_handle_queue_full(scmd->device);
484                 /* fall through */
485         case BUSY:
486                 return NEEDS_RETRY;
487         default:
488                 return FAILED;
489         }
490         return FAILED;
491 }
492
493 /**
494  * scsi_eh_done - Completion function for error handling.
495  * @scmd:       Cmd that is done.
496  */
497 static void scsi_eh_done(struct scsi_cmnd *scmd)
498 {
499         struct completion     *eh_action;
500
501         SCSI_LOG_ERROR_RECOVERY(3,
502                 printk("%s scmd: %p result: %x\n",
503                         __func__, scmd, scmd->result));
504
505         eh_action = scmd->device->host->eh_action;
506         if (eh_action)
507                 complete(eh_action);
508 }
509
510 /**
511  * scsi_try_host_reset - ask host adapter to reset itself
512  * @scmd:       SCSI cmd to send hsot reset.
513  */
514 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
515 {
516         unsigned long flags;
517         int rtn;
518
519         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
520                                           __func__));
521
522         if (!scmd->device->host->hostt->eh_host_reset_handler)
523                 return FAILED;
524
525         rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
526
527         if (rtn == SUCCESS) {
528                 if (!scmd->device->host->hostt->skip_settle_delay)
529                         ssleep(HOST_RESET_SETTLE_TIME);
530                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
531                 scsi_report_bus_reset(scmd->device->host,
532                                       scmd_channel(scmd));
533                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
534         }
535
536         return rtn;
537 }
538
539 /**
540  * scsi_try_bus_reset - ask host to perform a bus reset
541  * @scmd:       SCSI cmd to send bus reset.
542  */
543 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
544 {
545         unsigned long flags;
546         int rtn;
547
548         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
549                                           __func__));
550
551         if (!scmd->device->host->hostt->eh_bus_reset_handler)
552                 return FAILED;
553
554         rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
555
556         if (rtn == SUCCESS) {
557                 if (!scmd->device->host->hostt->skip_settle_delay)
558                         ssleep(BUS_RESET_SETTLE_TIME);
559                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
560                 scsi_report_bus_reset(scmd->device->host,
561                                       scmd_channel(scmd));
562                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
563         }
564
565         return rtn;
566 }
567
568 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
569 {
570         sdev->was_reset = 1;
571         sdev->expecting_cc_ua = 1;
572 }
573
574 /**
575  * scsi_try_target_reset - Ask host to perform a target reset
576  * @scmd:       SCSI cmd used to send a target reset
577  *
578  * Notes:
579  *    There is no timeout for this operation.  if this operation is
580  *    unreliable for a given host, then the host itself needs to put a
581  *    timer on it, and set the host back to a consistent state prior to
582  *    returning.
583  */
584 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
585 {
586         unsigned long flags;
587         int rtn;
588
589         if (!scmd->device->host->hostt->eh_target_reset_handler)
590                 return FAILED;
591
592         rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
593         if (rtn == SUCCESS) {
594                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
595                 __starget_for_each_device(scsi_target(scmd->device), NULL,
596                                           __scsi_report_device_reset);
597                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
598         }
599
600         return rtn;
601 }
602
603 /**
604  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
605  * @scmd:       SCSI cmd used to send BDR
606  *
607  * Notes:
608  *    There is no timeout for this operation.  if this operation is
609  *    unreliable for a given host, then the host itself needs to put a
610  *    timer on it, and set the host back to a consistent state prior to
611  *    returning.
612  */
613 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
614 {
615         int rtn;
616
617         if (!scmd->device->host->hostt->eh_device_reset_handler)
618                 return FAILED;
619
620         rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
621         if (rtn == SUCCESS)
622                 __scsi_report_device_reset(scmd->device, NULL);
623         return rtn;
624 }
625
626 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
627 {
628         if (!scmd->device->host->hostt->eh_abort_handler)
629                 return FAILED;
630
631         return scmd->device->host->hostt->eh_abort_handler(scmd);
632 }
633
634 /**
635  * scsi_try_to_abort_cmd - Ask host to abort a running command.
636  * @scmd:       SCSI cmd to abort from Lower Level.
637  *
638  * Notes:
639  *    This function will not return until the user's completion function
640  *    has been called.  there is no timeout on this operation.  if the
641  *    author of the low-level driver wishes this operation to be timed,
642  *    they can provide this facility themselves.  helper functions in
643  *    scsi_error.c can be supplied to make this easier to do.
644  */
645 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
646 {
647         /*
648          * scsi_done was called just after the command timed out and before
649          * we had a chance to process it. (db)
650          */
651         if (scmd->serial_number == 0)
652                 return SUCCESS;
653         return __scsi_try_to_abort_cmd(scmd);
654 }
655
656 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
657 {
658         if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
659                 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
660                         if (scsi_try_target_reset(scmd) != SUCCESS)
661                                 if (scsi_try_bus_reset(scmd) != SUCCESS)
662                                         scsi_try_host_reset(scmd);
663 }
664
665 /**
666  * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recory
667  * @scmd:       SCSI command structure to hijack
668  * @ses:        structure to save restore information
669  * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
670  * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
671  * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
672  *
673  * This function is used to save a scsi command information before re-execution
674  * as part of the error recovery process.  If @sense_bytes is 0 the command
675  * sent must be one that does not transfer any data.  If @sense_bytes != 0
676  * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
677  * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
678  */
679 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
680                         unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
681 {
682         struct scsi_device *sdev = scmd->device;
683
684         /*
685          * We need saved copies of a number of fields - this is because
686          * error handling may need to overwrite these with different values
687          * to run different commands, and once error handling is complete,
688          * we will need to restore these values prior to running the actual
689          * command.
690          */
691         ses->cmd_len = scmd->cmd_len;
692         ses->cmnd = scmd->cmnd;
693         ses->data_direction = scmd->sc_data_direction;
694         ses->sdb = scmd->sdb;
695         ses->next_rq = scmd->request->next_rq;
696         ses->result = scmd->result;
697         ses->underflow = scmd->underflow;
698         ses->prot_op = scmd->prot_op;
699
700         scmd->prot_op = SCSI_PROT_NORMAL;
701         scmd->cmnd = ses->eh_cmnd;
702         memset(scmd->cmnd, 0, BLK_MAX_CDB);
703         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
704         scmd->request->next_rq = NULL;
705
706         if (sense_bytes) {
707                 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
708                                          sense_bytes);
709                 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
710                             scmd->sdb.length);
711                 scmd->sdb.table.sgl = &ses->sense_sgl;
712                 scmd->sc_data_direction = DMA_FROM_DEVICE;
713                 scmd->sdb.table.nents = 1;
714                 scmd->cmnd[0] = REQUEST_SENSE;
715                 scmd->cmnd[4] = scmd->sdb.length;
716                 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
717         } else {
718                 scmd->sc_data_direction = DMA_NONE;
719                 if (cmnd) {
720                         BUG_ON(cmnd_size > BLK_MAX_CDB);
721                         memcpy(scmd->cmnd, cmnd, cmnd_size);
722                         scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
723                 }
724         }
725
726         scmd->underflow = 0;
727
728         if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
729                 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
730                         (sdev->lun << 5 & 0xe0);
731
732         /*
733          * Zero the sense buffer.  The scsi spec mandates that any
734          * untransferred sense data should be interpreted as being zero.
735          */
736         memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
737 }
738 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
739
740 /**
741  * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recory
742  * @scmd:       SCSI command structure to restore
743  * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
744  *
745  * Undo any damage done by above scsi_eh_prep_cmnd().
746  */
747 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
748 {
749         /*
750          * Restore original data
751          */
752         scmd->cmd_len = ses->cmd_len;
753         scmd->cmnd = ses->cmnd;
754         scmd->sc_data_direction = ses->data_direction;
755         scmd->sdb = ses->sdb;
756         scmd->request->next_rq = ses->next_rq;
757         scmd->result = ses->result;
758         scmd->underflow = ses->underflow;
759         scmd->prot_op = ses->prot_op;
760 }
761 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
762
763 /**
764  * scsi_send_eh_cmnd  - submit a scsi command as part of error recory
765  * @scmd:       SCSI command structure to hijack
766  * @cmnd:       CDB to send
767  * @cmnd_size:  size in bytes of @cmnd
768  * @timeout:    timeout for this request
769  * @sense_bytes: size of sense data to copy or 0
770  *
771  * This function is used to send a scsi command down to a target device
772  * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
773  *
774  * Return value:
775  *    SUCCESS or FAILED or NEEDS_RETRY
776  */
777 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
778                              int cmnd_size, int timeout, unsigned sense_bytes)
779 {
780         struct scsi_device *sdev = scmd->device;
781         struct Scsi_Host *shost = sdev->host;
782         DECLARE_COMPLETION_ONSTACK(done);
783         unsigned long timeleft;
784         unsigned long flags;
785         struct scsi_eh_save ses;
786         int rtn;
787
788         scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
789         shost->eh_action = &done;
790
791         spin_lock_irqsave(shost->host_lock, flags);
792         scsi_log_send(scmd);
793         shost->hostt->queuecommand(scmd, scsi_eh_done);
794         spin_unlock_irqrestore(shost->host_lock, flags);
795
796         timeleft = wait_for_completion_timeout(&done, timeout);
797
798         shost->eh_action = NULL;
799
800         scsi_log_completion(scmd, SUCCESS);
801
802         SCSI_LOG_ERROR_RECOVERY(3,
803                 printk("%s: scmd: %p, timeleft: %ld\n",
804                         __func__, scmd, timeleft));
805
806         /*
807          * If there is time left scsi_eh_done got called, and we will
808          * examine the actual status codes to see whether the command
809          * actually did complete normally, else tell the host to forget
810          * about this command.
811          */
812         if (timeleft) {
813                 rtn = scsi_eh_completed_normally(scmd);
814                 SCSI_LOG_ERROR_RECOVERY(3,
815                         printk("%s: scsi_eh_completed_normally %x\n",
816                                __func__, rtn));
817
818                 switch (rtn) {
819                 case SUCCESS:
820                 case NEEDS_RETRY:
821                 case FAILED:
822                         break;
823                 case ADD_TO_MLQUEUE:
824                         rtn = NEEDS_RETRY;
825                         break;
826                 default:
827                         rtn = FAILED;
828                         break;
829                 }
830         } else {
831                 scsi_abort_eh_cmnd(scmd);
832                 rtn = FAILED;
833         }
834
835         scsi_eh_restore_cmnd(scmd, &ses);
836         return rtn;
837 }
838
839 /**
840  * scsi_request_sense - Request sense data from a particular target.
841  * @scmd:       SCSI cmd for request sense.
842  *
843  * Notes:
844  *    Some hosts automatically obtain this information, others require
845  *    that we obtain it on our own. This function will *not* return until
846  *    the command either times out, or it completes.
847  */
848 static int scsi_request_sense(struct scsi_cmnd *scmd)
849 {
850         return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
851 }
852
853 /**
854  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
855  * @scmd:       Original SCSI cmd that eh has finished.
856  * @done_q:     Queue for processed commands.
857  *
858  * Notes:
859  *    We don't want to use the normal command completion while we are are
860  *    still handling errors - it may cause other commands to be queued,
861  *    and that would disturb what we are doing.  Thus we really want to
862  *    keep a list of pending commands for final completion, and once we
863  *    are ready to leave error handling we handle completion for real.
864  */
865 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
866 {
867         scmd->device->host->host_failed--;
868         scmd->eh_eflags = 0;
869         list_move_tail(&scmd->eh_entry, done_q);
870 }
871 EXPORT_SYMBOL(scsi_eh_finish_cmd);
872
873 /**
874  * scsi_eh_get_sense - Get device sense data.
875  * @work_q:     Queue of commands to process.
876  * @done_q:     Queue of processed commands.
877  *
878  * Description:
879  *    See if we need to request sense information.  if so, then get it
880  *    now, so we have a better idea of what to do.  
881  *
882  * Notes:
883  *    This has the unfortunate side effect that if a shost adapter does
884  *    not automatically request sense information, we end up shutting
885  *    it down before we request it.
886  *
887  *    All drivers should request sense information internally these days,
888  *    so for now all I have to say is tough noogies if you end up in here.
889  *
890  *    XXX: Long term this code should go away, but that needs an audit of
891  *         all LLDDs first.
892  */
893 int scsi_eh_get_sense(struct list_head *work_q,
894                       struct list_head *done_q)
895 {
896         struct scsi_cmnd *scmd, *next;
897         int rtn;
898
899         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
900                 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
901                     SCSI_SENSE_VALID(scmd))
902                         continue;
903
904                 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
905                                                   "%s: requesting sense\n",
906                                                   current->comm));
907                 rtn = scsi_request_sense(scmd);
908                 if (rtn != SUCCESS)
909                         continue;
910
911                 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
912                                                   " result %x\n", scmd,
913                                                   scmd->result));
914                 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
915
916                 rtn = scsi_decide_disposition(scmd);
917
918                 /*
919                  * if the result was normal, then just pass it along to the
920                  * upper level.
921                  */
922                 if (rtn == SUCCESS)
923                         /* we don't want this command reissued, just
924                          * finished with the sense data, so set
925                          * retries to the max allowed to ensure it
926                          * won't get reissued */
927                         scmd->retries = scmd->allowed;
928                 else if (rtn != NEEDS_RETRY)
929                         continue;
930
931                 scsi_eh_finish_cmd(scmd, done_q);
932         }
933
934         return list_empty(work_q);
935 }
936 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
937
938 /**
939  * scsi_eh_tur - Send TUR to device.
940  * @scmd:       &scsi_cmnd to send TUR
941  *
942  * Return value:
943  *    0 - Device is ready. 1 - Device NOT ready.
944  */
945 static int scsi_eh_tur(struct scsi_cmnd *scmd)
946 {
947         static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
948         int retry_cnt = 1, rtn;
949
950 retry_tur:
951         rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
952
953         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
954                 __func__, scmd, rtn));
955
956         switch (rtn) {
957         case NEEDS_RETRY:
958                 if (retry_cnt--)
959                         goto retry_tur;
960                 /*FALLTHRU*/
961         case SUCCESS:
962                 return 0;
963         default:
964                 return 1;
965         }
966 }
967
968 /**
969  * scsi_eh_abort_cmds - abort pending commands.
970  * @work_q:     &list_head for pending commands.
971  * @done_q:     &list_head for processed commands.
972  *
973  * Decription:
974  *    Try and see whether or not it makes sense to try and abort the
975  *    running command.  This only works out to be the case if we have one
976  *    command that has timed out.  If the command simply failed, it makes
977  *    no sense to try and abort the command, since as far as the shost
978  *    adapter is concerned, it isn't running.
979  */
980 static int scsi_eh_abort_cmds(struct list_head *work_q,
981                               struct list_head *done_q)
982 {
983         struct scsi_cmnd *scmd, *next;
984         int rtn;
985
986         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
987                 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
988                         continue;
989                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
990                                                   "0x%p\n", current->comm,
991                                                   scmd));
992                 rtn = scsi_try_to_abort_cmd(scmd);
993                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
994                         scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
995                         if (!scsi_device_online(scmd->device) ||
996                             rtn == FAST_IO_FAIL ||
997                             !scsi_eh_tur(scmd)) {
998                                 scsi_eh_finish_cmd(scmd, done_q);
999                         }
1000                                 
1001                 } else
1002                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1003                                                           " cmd failed:"
1004                                                           "0x%p\n",
1005                                                           current->comm,
1006                                                           scmd));
1007         }
1008
1009         return list_empty(work_q);
1010 }
1011
1012 /**
1013  * scsi_eh_try_stu - Send START_UNIT to device.
1014  * @scmd:       &scsi_cmnd to send START_UNIT
1015  *
1016  * Return value:
1017  *    0 - Device is ready. 1 - Device NOT ready.
1018  */
1019 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1020 {
1021         static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1022
1023         if (scmd->device->allow_restart) {
1024                 int i, rtn = NEEDS_RETRY;
1025
1026                 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1027                         rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1028
1029                 if (rtn == SUCCESS)
1030                         return 0;
1031         }
1032
1033         return 1;
1034 }
1035
1036  /**
1037  * scsi_eh_stu - send START_UNIT if needed
1038  * @shost:      &scsi host being recovered.
1039  * @work_q:     &list_head for pending commands.
1040  * @done_q:     &list_head for processed commands.
1041  *
1042  * Notes:
1043  *    If commands are failing due to not ready, initializing command required,
1044  *      try revalidating the device, which will end up sending a start unit. 
1045  */
1046 static int scsi_eh_stu(struct Scsi_Host *shost,
1047                               struct list_head *work_q,
1048                               struct list_head *done_q)
1049 {
1050         struct scsi_cmnd *scmd, *stu_scmd, *next;
1051         struct scsi_device *sdev;
1052
1053         shost_for_each_device(sdev, shost) {
1054                 stu_scmd = NULL;
1055                 list_for_each_entry(scmd, work_q, eh_entry)
1056                         if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1057                             scsi_check_sense(scmd) == FAILED ) {
1058                                 stu_scmd = scmd;
1059                                 break;
1060                         }
1061
1062                 if (!stu_scmd)
1063                         continue;
1064
1065                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1066                                                   " 0x%p\n", current->comm, sdev));
1067
1068                 if (!scsi_eh_try_stu(stu_scmd)) {
1069                         if (!scsi_device_online(sdev) ||
1070                             !scsi_eh_tur(stu_scmd)) {
1071                                 list_for_each_entry_safe(scmd, next,
1072                                                           work_q, eh_entry) {
1073                                         if (scmd->device == sdev)
1074                                                 scsi_eh_finish_cmd(scmd, done_q);
1075                                 }
1076                         }
1077                 } else {
1078                         SCSI_LOG_ERROR_RECOVERY(3,
1079                                                 printk("%s: START_UNIT failed to sdev:"
1080                                                        " 0x%p\n", current->comm, sdev));
1081                 }
1082         }
1083
1084         return list_empty(work_q);
1085 }
1086
1087
1088 /**
1089  * scsi_eh_bus_device_reset - send bdr if needed
1090  * @shost:      scsi host being recovered.
1091  * @work_q:     &list_head for pending commands.
1092  * @done_q:     &list_head for processed commands.
1093  *
1094  * Notes:
1095  *    Try a bus device reset.  Still, look to see whether we have multiple
1096  *    devices that are jammed or not - if we have multiple devices, it
1097  *    makes no sense to try bus_device_reset - we really would need to try
1098  *    a bus_reset instead. 
1099  */
1100 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1101                                     struct list_head *work_q,
1102                                     struct list_head *done_q)
1103 {
1104         struct scsi_cmnd *scmd, *bdr_scmd, *next;
1105         struct scsi_device *sdev;
1106         int rtn;
1107
1108         shost_for_each_device(sdev, shost) {
1109                 bdr_scmd = NULL;
1110                 list_for_each_entry(scmd, work_q, eh_entry)
1111                         if (scmd->device == sdev) {
1112                                 bdr_scmd = scmd;
1113                                 break;
1114                         }
1115
1116                 if (!bdr_scmd)
1117                         continue;
1118
1119                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1120                                                   " 0x%p\n", current->comm,
1121                                                   sdev));
1122                 rtn = scsi_try_bus_device_reset(bdr_scmd);
1123                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1124                         if (!scsi_device_online(sdev) ||
1125                             rtn == FAST_IO_FAIL ||
1126                             !scsi_eh_tur(bdr_scmd)) {
1127                                 list_for_each_entry_safe(scmd, next,
1128                                                          work_q, eh_entry) {
1129                                         if (scmd->device == sdev)
1130                                                 scsi_eh_finish_cmd(scmd,
1131                                                                    done_q);
1132                                 }
1133                         }
1134                 } else {
1135                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1136                                                           " failed sdev:"
1137                                                           "0x%p\n",
1138                                                           current->comm,
1139                                                            sdev));
1140                 }
1141         }
1142
1143         return list_empty(work_q);
1144 }
1145
1146 /**
1147  * scsi_eh_target_reset - send target reset if needed
1148  * @shost:      scsi host being recovered.
1149  * @work_q:     &list_head for pending commands.
1150  * @done_q:     &list_head for processed commands.
1151  *
1152  * Notes:
1153  *    Try a target reset.
1154  */
1155 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1156                                 struct list_head *work_q,
1157                                 struct list_head *done_q)
1158 {
1159         struct scsi_cmnd *scmd, *tgtr_scmd, *next;
1160         unsigned int id = 0;
1161         int rtn;
1162
1163         do {
1164                 tgtr_scmd = NULL;
1165                 list_for_each_entry(scmd, work_q, eh_entry) {
1166                         if (id == scmd_id(scmd)) {
1167                                 tgtr_scmd = scmd;
1168                                 break;
1169                         }
1170                 }
1171                 if (!tgtr_scmd) {
1172                         /* not one exactly equal; find the next highest */
1173                         list_for_each_entry(scmd, work_q, eh_entry) {
1174                                 if (scmd_id(scmd) > id &&
1175                                     (!tgtr_scmd ||
1176                                      scmd_id(tgtr_scmd) > scmd_id(scmd)))
1177                                                 tgtr_scmd = scmd;
1178                         }
1179                 }
1180                 if (!tgtr_scmd)
1181                         /* no more commands, that's it */
1182                         break;
1183
1184                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1185                                                   "to target %d\n",
1186                                                   current->comm, id));
1187                 rtn = scsi_try_target_reset(tgtr_scmd);
1188                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1189                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1190                                 if (id == scmd_id(scmd))
1191                                         if (!scsi_device_online(scmd->device) ||
1192                                             rtn == FAST_IO_FAIL ||
1193                                             !scsi_eh_tur(tgtr_scmd))
1194                                                 scsi_eh_finish_cmd(scmd,
1195                                                                    done_q);
1196                         }
1197                 } else
1198                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1199                                                           " failed target: "
1200                                                           "%d\n",
1201                                                           current->comm, id));
1202                 id++;
1203         } while(id != 0);
1204
1205         return list_empty(work_q);
1206 }
1207
1208 /**
1209  * scsi_eh_bus_reset - send a bus reset 
1210  * @shost:      &scsi host being recovered.
1211  * @work_q:     &list_head for pending commands.
1212  * @done_q:     &list_head for processed commands.
1213  */
1214 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1215                              struct list_head *work_q,
1216                              struct list_head *done_q)
1217 {
1218         struct scsi_cmnd *scmd, *chan_scmd, *next;
1219         unsigned int channel;
1220         int rtn;
1221
1222         /*
1223          * we really want to loop over the various channels, and do this on
1224          * a channel by channel basis.  we should also check to see if any
1225          * of the failed commands are on soft_reset devices, and if so, skip
1226          * the reset.  
1227          */
1228
1229         for (channel = 0; channel <= shost->max_channel; channel++) {
1230                 chan_scmd = NULL;
1231                 list_for_each_entry(scmd, work_q, eh_entry) {
1232                         if (channel == scmd_channel(scmd)) {
1233                                 chan_scmd = scmd;
1234                                 break;
1235                                 /*
1236                                  * FIXME add back in some support for
1237                                  * soft_reset devices.
1238                                  */
1239                         }
1240                 }
1241
1242                 if (!chan_scmd)
1243                         continue;
1244                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1245                                                   " %d\n", current->comm,
1246                                                   channel));
1247                 rtn = scsi_try_bus_reset(chan_scmd);
1248                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1249                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1250                                 if (channel == scmd_channel(scmd))
1251                                         if (!scsi_device_online(scmd->device) ||
1252                                             rtn == FAST_IO_FAIL ||
1253                                             !scsi_eh_tur(scmd))
1254                                                 scsi_eh_finish_cmd(scmd,
1255                                                                    done_q);
1256                         }
1257                 } else {
1258                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1259                                                           " failed chan: %d\n",
1260                                                           current->comm,
1261                                                           channel));
1262                 }
1263         }
1264         return list_empty(work_q);
1265 }
1266
1267 /**
1268  * scsi_eh_host_reset - send a host reset 
1269  * @work_q:     list_head for processed commands.
1270  * @done_q:     list_head for processed commands.
1271  */
1272 static int scsi_eh_host_reset(struct list_head *work_q,
1273                               struct list_head *done_q)
1274 {
1275         struct scsi_cmnd *scmd, *next;
1276         int rtn;
1277
1278         if (!list_empty(work_q)) {
1279                 scmd = list_entry(work_q->next,
1280                                   struct scsi_cmnd, eh_entry);
1281
1282                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1283                                                   , current->comm));
1284
1285                 rtn = scsi_try_host_reset(scmd);
1286                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1287                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1288                                 if (!scsi_device_online(scmd->device) ||
1289                                     rtn == FAST_IO_FAIL ||
1290                                     (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1291                                     !scsi_eh_tur(scmd))
1292                                         scsi_eh_finish_cmd(scmd, done_q);
1293                         }
1294                 } else {
1295                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1296                                                           " failed\n",
1297                                                           current->comm));
1298                 }
1299         }
1300         return list_empty(work_q);
1301 }
1302
1303 /**
1304  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1305  * @work_q:     list_head for processed commands.
1306  * @done_q:     list_head for processed commands.
1307  */
1308 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1309                                   struct list_head *done_q)
1310 {
1311         struct scsi_cmnd *scmd, *next;
1312
1313         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1314                 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1315                             "not ready after error recovery\n");
1316                 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1317                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1318                         /*
1319                          * FIXME: Handle lost cmds.
1320                          */
1321                 }
1322                 scsi_eh_finish_cmd(scmd, done_q);
1323         }
1324         return;
1325 }
1326
1327 /**
1328  * scsi_noretry_cmd - determinte if command should be failed fast
1329  * @scmd:       SCSI cmd to examine.
1330  */
1331 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1332 {
1333         switch (host_byte(scmd->result)) {
1334         case DID_OK:
1335                 break;
1336         case DID_BUS_BUSY:
1337                 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1338         case DID_PARITY:
1339                 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1340         case DID_ERROR:
1341                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1342                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1343                         return 0;
1344                 /* fall through */
1345         case DID_SOFT_ERROR:
1346                 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1347         }
1348
1349         switch (status_byte(scmd->result)) {
1350         case CHECK_CONDITION:
1351                 /*
1352                  * assume caller has checked sense and determinted
1353                  * the check condition was retryable.
1354                  */
1355                 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1356                     scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1357                         return 1;
1358         }
1359
1360         return 0;
1361 }
1362
1363 /**
1364  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1365  * @scmd:       SCSI cmd to examine.
1366  *
1367  * Notes:
1368  *    This is *only* called when we are examining the status after sending
1369  *    out the actual data command.  any commands that are queued for error
1370  *    recovery (e.g. test_unit_ready) do *not* come through here.
1371  *
1372  *    When this routine returns failed, it means the error handler thread
1373  *    is woken.  In cases where the error code indicates an error that
1374  *    doesn't require the error handler read (i.e. we don't need to
1375  *    abort/reset), this function should return SUCCESS.
1376  */
1377 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1378 {
1379         int rtn;
1380
1381         /*
1382          * if the device is offline, then we clearly just pass the result back
1383          * up to the top level.
1384          */
1385         if (!scsi_device_online(scmd->device)) {
1386                 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1387                                                   " as SUCCESS\n",
1388                                                   __func__));
1389                 return SUCCESS;
1390         }
1391
1392         /*
1393          * first check the host byte, to see if there is anything in there
1394          * that would indicate what we need to do.
1395          */
1396         switch (host_byte(scmd->result)) {
1397         case DID_PASSTHROUGH:
1398                 /*
1399                  * no matter what, pass this through to the upper layer.
1400                  * nuke this special code so that it looks like we are saying
1401                  * did_ok.
1402                  */
1403                 scmd->result &= 0xff00ffff;
1404                 return SUCCESS;
1405         case DID_OK:
1406                 /*
1407                  * looks good.  drop through, and check the next byte.
1408                  */
1409                 break;
1410         case DID_NO_CONNECT:
1411         case DID_BAD_TARGET:
1412         case DID_ABORT:
1413                 /*
1414                  * note - this means that we just report the status back
1415                  * to the top level driver, not that we actually think
1416                  * that it indicates SUCCESS.
1417                  */
1418                 return SUCCESS;
1419                 /*
1420                  * when the low level driver returns did_soft_error,
1421                  * it is responsible for keeping an internal retry counter 
1422                  * in order to avoid endless loops (db)
1423                  *
1424                  * actually this is a bug in this function here.  we should
1425                  * be mindful of the maximum number of retries specified
1426                  * and not get stuck in a loop.
1427                  */
1428         case DID_SOFT_ERROR:
1429                 goto maybe_retry;
1430         case DID_IMM_RETRY:
1431                 return NEEDS_RETRY;
1432
1433         case DID_REQUEUE:
1434                 return ADD_TO_MLQUEUE;
1435         case DID_TRANSPORT_DISRUPTED:
1436                 /*
1437                  * LLD/transport was disrupted during processing of the IO.
1438                  * The transport class is now blocked/blocking,
1439                  * and the transport will decide what to do with the IO
1440                  * based on its timers and recovery capablilities if
1441                  * there are enough retries.
1442                  */
1443                 goto maybe_retry;
1444         case DID_TRANSPORT_FAILFAST:
1445                 /*
1446                  * The transport decided to failfast the IO (most likely
1447                  * the fast io fail tmo fired), so send IO directly upwards.
1448                  */
1449                 return SUCCESS;
1450         case DID_ERROR:
1451                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1452                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1453                         /*
1454                          * execute reservation conflict processing code
1455                          * lower down
1456                          */
1457                         break;
1458                 /* fallthrough */
1459
1460         case DID_BUS_BUSY:
1461         case DID_PARITY:
1462                 goto maybe_retry;
1463         case DID_TIME_OUT:
1464                 /*
1465                  * when we scan the bus, we get timeout messages for
1466                  * these commands if there is no device available.
1467                  * other hosts report did_no_connect for the same thing.
1468                  */
1469                 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1470                      scmd->cmnd[0] == INQUIRY)) {
1471                         return SUCCESS;
1472                 } else {
1473                         return FAILED;
1474                 }
1475         case DID_RESET:
1476                 return SUCCESS;
1477         default:
1478                 return FAILED;
1479         }
1480
1481         /*
1482          * next, check the message byte.
1483          */
1484         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1485                 return FAILED;
1486
1487         /*
1488          * check the status byte to see if this indicates anything special.
1489          */
1490         switch (status_byte(scmd->result)) {
1491         case QUEUE_FULL:
1492                 scsi_handle_queue_full(scmd->device);
1493                 /*
1494                  * the case of trying to send too many commands to a
1495                  * tagged queueing device.
1496                  */
1497         case BUSY:
1498                 /*
1499                  * device can't talk to us at the moment.  Should only
1500                  * occur (SAM-3) when the task queue is empty, so will cause
1501                  * the empty queue handling to trigger a stall in the
1502                  * device.
1503                  */
1504                 return ADD_TO_MLQUEUE;
1505         case GOOD:
1506                 scsi_handle_queue_ramp_up(scmd->device);
1507         case COMMAND_TERMINATED:
1508                 return SUCCESS;
1509         case TASK_ABORTED:
1510                 goto maybe_retry;
1511         case CHECK_CONDITION:
1512                 rtn = scsi_check_sense(scmd);
1513                 if (rtn == NEEDS_RETRY)
1514                         goto maybe_retry;
1515                 /* if rtn == FAILED, we have no sense information;
1516                  * returning FAILED will wake the error handler thread
1517                  * to collect the sense and redo the decide
1518                  * disposition */
1519                 return rtn;
1520         case CONDITION_GOOD:
1521         case INTERMEDIATE_GOOD:
1522         case INTERMEDIATE_C_GOOD:
1523         case ACA_ACTIVE:
1524                 /*
1525                  * who knows?  FIXME(eric)
1526                  */
1527                 return SUCCESS;
1528
1529         case RESERVATION_CONFLICT:
1530                 sdev_printk(KERN_INFO, scmd->device,
1531                             "reservation conflict\n");
1532                 return SUCCESS; /* causes immediate i/o error */
1533         default:
1534                 return FAILED;
1535         }
1536         return FAILED;
1537
1538       maybe_retry:
1539
1540         /* we requeue for retry because the error was retryable, and
1541          * the request was not marked fast fail.  Note that above,
1542          * even if the request is marked fast fail, we still requeue
1543          * for queue congestion conditions (QUEUE_FULL or BUSY) */
1544         if ((++scmd->retries) <= scmd->allowed
1545             && !scsi_noretry_cmd(scmd)) {
1546                 return NEEDS_RETRY;
1547         } else {
1548                 /*
1549                  * no more retries - report this one back to upper level.
1550                  */
1551                 return SUCCESS;
1552         }
1553 }
1554
1555 static void eh_lock_door_done(struct request *req, int uptodate)
1556 {
1557         __blk_put_request(req->q, req);
1558 }
1559
1560 /**
1561  * scsi_eh_lock_door - Prevent medium removal for the specified device
1562  * @sdev:       SCSI device to prevent medium removal
1563  *
1564  * Locking:
1565  *      We must be called from process context.
1566  *
1567  * Notes:
1568  *      We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1569  *      head of the devices request queue, and continue.
1570  */
1571 static void scsi_eh_lock_door(struct scsi_device *sdev)
1572 {
1573         struct request *req;
1574
1575         /*
1576          * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1577          * request becomes available
1578          */
1579         req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1580
1581         req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1582         req->cmd[1] = 0;
1583         req->cmd[2] = 0;
1584         req->cmd[3] = 0;
1585         req->cmd[4] = SCSI_REMOVAL_PREVENT;
1586         req->cmd[5] = 0;
1587
1588         req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1589
1590         req->cmd_type = REQ_TYPE_BLOCK_PC;
1591         req->cmd_flags |= REQ_QUIET;
1592         req->timeout = 10 * HZ;
1593         req->retries = 5;
1594
1595         blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1596 }
1597
1598 /**
1599  * scsi_restart_operations - restart io operations to the specified host.
1600  * @shost:      Host we are restarting.
1601  *
1602  * Notes:
1603  *    When we entered the error handler, we blocked all further i/o to
1604  *    this device.  we need to 'reverse' this process.
1605  */
1606 static void scsi_restart_operations(struct Scsi_Host *shost)
1607 {
1608         struct scsi_device *sdev;
1609         unsigned long flags;
1610
1611         /*
1612          * If the door was locked, we need to insert a door lock request
1613          * onto the head of the SCSI request queue for the device.  There
1614          * is no point trying to lock the door of an off-line device.
1615          */
1616         shost_for_each_device(sdev, shost) {
1617                 if (scsi_device_online(sdev) && sdev->locked)
1618                         scsi_eh_lock_door(sdev);
1619         }
1620
1621         /*
1622          * next free up anything directly waiting upon the host.  this
1623          * will be requests for character device operations, and also for
1624          * ioctls to queued block devices.
1625          */
1626         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1627                                           __func__));
1628
1629         spin_lock_irqsave(shost->host_lock, flags);
1630         if (scsi_host_set_state(shost, SHOST_RUNNING))
1631                 if (scsi_host_set_state(shost, SHOST_CANCEL))
1632                         BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1633         spin_unlock_irqrestore(shost->host_lock, flags);
1634
1635         wake_up(&shost->host_wait);
1636
1637         /*
1638          * finally we need to re-initiate requests that may be pending.  we will
1639          * have had everything blocked while error handling is taking place, and
1640          * now that error recovery is done, we will need to ensure that these
1641          * requests are started.
1642          */
1643         scsi_run_host_queues(shost);
1644 }
1645
1646 /**
1647  * scsi_eh_ready_devs - check device ready state and recover if not.
1648  * @shost:      host to be recovered.
1649  * @work_q:     &list_head for pending commands.
1650  * @done_q:     &list_head for processed commands.
1651  */
1652 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1653                         struct list_head *work_q,
1654                         struct list_head *done_q)
1655 {
1656         if (!scsi_eh_stu(shost, work_q, done_q))
1657                 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1658                         if (!scsi_eh_target_reset(shost, work_q, done_q))
1659                                 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1660                                         if (!scsi_eh_host_reset(work_q, done_q))
1661                                                 scsi_eh_offline_sdevs(work_q,
1662                                                                       done_q);
1663 }
1664 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1665
1666 /**
1667  * scsi_eh_flush_done_q - finish processed commands or retry them.
1668  * @done_q:     list_head of processed commands.
1669  */
1670 void scsi_eh_flush_done_q(struct list_head *done_q)
1671 {
1672         struct scsi_cmnd *scmd, *next;
1673
1674         list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1675                 list_del_init(&scmd->eh_entry);
1676                 if (scsi_device_online(scmd->device) &&
1677                     !scsi_noretry_cmd(scmd) &&
1678                     (++scmd->retries <= scmd->allowed)) {
1679                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1680                                                           " retry cmd: %p\n",
1681                                                           current->comm,
1682                                                           scmd));
1683                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1684                 } else {
1685                         /*
1686                          * If just we got sense for the device (called
1687                          * scsi_eh_get_sense), scmd->result is already
1688                          * set, do not set DRIVER_TIMEOUT.
1689                          */
1690                         if (!scmd->result)
1691                                 scmd->result |= (DRIVER_TIMEOUT << 24);
1692                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1693                                                         " cmd: %p\n",
1694                                                         current->comm, scmd));
1695                         scsi_finish_command(scmd);
1696                 }
1697         }
1698 }
1699 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1700
1701 /**
1702  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1703  * @shost:      Host to unjam.
1704  *
1705  * Notes:
1706  *    When we come in here, we *know* that all commands on the bus have
1707  *    either completed, failed or timed out.  we also know that no further
1708  *    commands are being sent to the host, so things are relatively quiet
1709  *    and we have freedom to fiddle with things as we wish.
1710  *
1711  *    This is only the *default* implementation.  it is possible for
1712  *    individual drivers to supply their own version of this function, and
1713  *    if the maintainer wishes to do this, it is strongly suggested that
1714  *    this function be taken as a template and modified.  this function
1715  *    was designed to correctly handle problems for about 95% of the
1716  *    different cases out there, and it should always provide at least a
1717  *    reasonable amount of error recovery.
1718  *
1719  *    Any command marked 'failed' or 'timeout' must eventually have
1720  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
1721  *    here, so when we restart the host after we return it should have an
1722  *    empty queue.
1723  */
1724 static void scsi_unjam_host(struct Scsi_Host *shost)
1725 {
1726         unsigned long flags;
1727         LIST_HEAD(eh_work_q);
1728         LIST_HEAD(eh_done_q);
1729
1730         spin_lock_irqsave(shost->host_lock, flags);
1731         list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1732         spin_unlock_irqrestore(shost->host_lock, flags);
1733
1734         SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1735
1736         if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1737                 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1738                         scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1739
1740         scsi_eh_flush_done_q(&eh_done_q);
1741 }
1742
1743 /**
1744  * scsi_error_handler - SCSI error handler thread
1745  * @data:       Host for which we are running.
1746  *
1747  * Notes:
1748  *    This is the main error handling loop.  This is run as a kernel thread
1749  *    for every SCSI host and handles all error handling activity.
1750  */
1751 int scsi_error_handler(void *data)
1752 {
1753         struct Scsi_Host *shost = data;
1754
1755         /*
1756          * We use TASK_INTERRUPTIBLE so that the thread is not
1757          * counted against the load average as a running process.
1758          * We never actually get interrupted because kthread_run
1759          * disables signal delivery for the created thread.
1760          */
1761         set_current_state(TASK_INTERRUPTIBLE);
1762         while (!kthread_should_stop()) {
1763                 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1764                     shost->host_failed != shost->host_busy) {
1765                         SCSI_LOG_ERROR_RECOVERY(1,
1766                                 printk("Error handler scsi_eh_%d sleeping\n",
1767                                         shost->host_no));
1768                         schedule();
1769                         set_current_state(TASK_INTERRUPTIBLE);
1770                         continue;
1771                 }
1772
1773                 __set_current_state(TASK_RUNNING);
1774                 SCSI_LOG_ERROR_RECOVERY(1,
1775                         printk("Error handler scsi_eh_%d waking up\n",
1776                                 shost->host_no));
1777
1778                 /*
1779                  * We have a host that is failing for some reason.  Figure out
1780                  * what we need to do to get it up and online again (if we can).
1781                  * If we fail, we end up taking the thing offline.
1782                  */
1783                 if (scsi_autopm_get_host(shost) != 0) {
1784                         SCSI_LOG_ERROR_RECOVERY(1,
1785                                 printk(KERN_ERR "Error handler scsi_eh_%d "
1786                                                 "unable to autoresume\n",
1787                                                 shost->host_no));
1788                         continue;
1789                 }
1790
1791                 if (shost->transportt->eh_strategy_handler)
1792                         shost->transportt->eh_strategy_handler(shost);
1793                 else
1794                         scsi_unjam_host(shost);
1795
1796                 /*
1797                  * Note - if the above fails completely, the action is to take
1798                  * individual devices offline and flush the queue of any
1799                  * outstanding requests that may have been pending.  When we
1800                  * restart, we restart any I/O to any other devices on the bus
1801                  * which are still online.
1802                  */
1803                 scsi_restart_operations(shost);
1804                 scsi_autopm_put_host(shost);
1805                 set_current_state(TASK_INTERRUPTIBLE);
1806         }
1807         __set_current_state(TASK_RUNNING);
1808
1809         SCSI_LOG_ERROR_RECOVERY(1,
1810                 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1811         shost->ehandler = NULL;
1812         return 0;
1813 }
1814
1815 /*
1816  * Function:    scsi_report_bus_reset()
1817  *
1818  * Purpose:     Utility function used by low-level drivers to report that
1819  *              they have observed a bus reset on the bus being handled.
1820  *
1821  * Arguments:   shost       - Host in question
1822  *              channel     - channel on which reset was observed.
1823  *
1824  * Returns:     Nothing
1825  *
1826  * Lock status: Host lock must be held.
1827  *
1828  * Notes:       This only needs to be called if the reset is one which
1829  *              originates from an unknown location.  Resets originated
1830  *              by the mid-level itself don't need to call this, but there
1831  *              should be no harm.
1832  *
1833  *              The main purpose of this is to make sure that a CHECK_CONDITION
1834  *              is properly treated.
1835  */
1836 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1837 {
1838         struct scsi_device *sdev;
1839
1840         __shost_for_each_device(sdev, shost) {
1841                 if (channel == sdev_channel(sdev))
1842                         __scsi_report_device_reset(sdev, NULL);
1843         }
1844 }
1845 EXPORT_SYMBOL(scsi_report_bus_reset);
1846
1847 /*
1848  * Function:    scsi_report_device_reset()
1849  *
1850  * Purpose:     Utility function used by low-level drivers to report that
1851  *              they have observed a device reset on the device being handled.
1852  *
1853  * Arguments:   shost       - Host in question
1854  *              channel     - channel on which reset was observed
1855  *              target      - target on which reset was observed
1856  *
1857  * Returns:     Nothing
1858  *
1859  * Lock status: Host lock must be held
1860  *
1861  * Notes:       This only needs to be called if the reset is one which
1862  *              originates from an unknown location.  Resets originated
1863  *              by the mid-level itself don't need to call this, but there
1864  *              should be no harm.
1865  *
1866  *              The main purpose of this is to make sure that a CHECK_CONDITION
1867  *              is properly treated.
1868  */
1869 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1870 {
1871         struct scsi_device *sdev;
1872
1873         __shost_for_each_device(sdev, shost) {
1874                 if (channel == sdev_channel(sdev) &&
1875                     target == sdev_id(sdev))
1876                         __scsi_report_device_reset(sdev, NULL);
1877         }
1878 }
1879 EXPORT_SYMBOL(scsi_report_device_reset);
1880
1881 static void
1882 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1883 {
1884 }
1885
1886 /*
1887  * Function:    scsi_reset_provider
1888  *
1889  * Purpose:     Send requested reset to a bus or device at any phase.
1890  *
1891  * Arguments:   device  - device to send reset to
1892  *              flag - reset type (see scsi.h)
1893  *
1894  * Returns:     SUCCESS/FAILURE.
1895  *
1896  * Notes:       This is used by the SCSI Generic driver to provide
1897  *              Bus/Device reset capability.
1898  */
1899 int
1900 scsi_reset_provider(struct scsi_device *dev, int flag)
1901 {
1902         struct scsi_cmnd *scmd;
1903         struct Scsi_Host *shost = dev->host;
1904         struct request req;
1905         unsigned long flags;
1906         int rtn;
1907
1908         if (scsi_autopm_get_host(shost) < 0)
1909                 return FAILED;
1910
1911         scmd = scsi_get_command(dev, GFP_KERNEL);
1912         blk_rq_init(NULL, &req);
1913         scmd->request = &req;
1914
1915         scmd->cmnd = req.cmd;
1916
1917         scmd->scsi_done         = scsi_reset_provider_done_command;
1918         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1919
1920         scmd->cmd_len                   = 0;
1921
1922         scmd->sc_data_direction         = DMA_BIDIRECTIONAL;
1923
1924         spin_lock_irqsave(shost->host_lock, flags);
1925         shost->tmf_in_progress = 1;
1926         spin_unlock_irqrestore(shost->host_lock, flags);
1927
1928         switch (flag) {
1929         case SCSI_TRY_RESET_DEVICE:
1930                 rtn = scsi_try_bus_device_reset(scmd);
1931                 if (rtn == SUCCESS)
1932                         break;
1933                 /* FALLTHROUGH */
1934         case SCSI_TRY_RESET_TARGET:
1935                 rtn = scsi_try_target_reset(scmd);
1936                 if (rtn == SUCCESS)
1937                         break;
1938                 /* FALLTHROUGH */
1939         case SCSI_TRY_RESET_BUS:
1940                 rtn = scsi_try_bus_reset(scmd);
1941                 if (rtn == SUCCESS)
1942                         break;
1943                 /* FALLTHROUGH */
1944         case SCSI_TRY_RESET_HOST:
1945                 rtn = scsi_try_host_reset(scmd);
1946                 break;
1947         default:
1948                 rtn = FAILED;
1949         }
1950
1951         spin_lock_irqsave(shost->host_lock, flags);
1952         shost->tmf_in_progress = 0;
1953         spin_unlock_irqrestore(shost->host_lock, flags);
1954
1955         /*
1956          * be sure to wake up anyone who was sleeping or had their queue
1957          * suspended while we performed the TMF.
1958          */
1959         SCSI_LOG_ERROR_RECOVERY(3,
1960                 printk("%s: waking up host to restart after TMF\n",
1961                 __func__));
1962
1963         wake_up(&shost->host_wait);
1964
1965         scsi_run_host_queues(shost);
1966
1967         scsi_next_command(scmd);
1968         scsi_autopm_put_host(shost);
1969         return rtn;
1970 }
1971 EXPORT_SYMBOL(scsi_reset_provider);
1972
1973 /**
1974  * scsi_normalize_sense - normalize main elements from either fixed or
1975  *                      descriptor sense data format into a common format.
1976  *
1977  * @sense_buffer:       byte array containing sense data returned by device
1978  * @sb_len:             number of valid bytes in sense_buffer
1979  * @sshdr:              pointer to instance of structure that common
1980  *                      elements are written to.
1981  *
1982  * Notes:
1983  *      The "main elements" from sense data are: response_code, sense_key,
1984  *      asc, ascq and additional_length (only for descriptor format).
1985  *
1986  *      Typically this function can be called after a device has
1987  *      responded to a SCSI command with the CHECK_CONDITION status.
1988  *
1989  * Return value:
1990  *      1 if valid sense data information found, else 0;
1991  */
1992 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1993                          struct scsi_sense_hdr *sshdr)
1994 {
1995         if (!sense_buffer || !sb_len)
1996                 return 0;
1997
1998         memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1999
2000         sshdr->response_code = (sense_buffer[0] & 0x7f);
2001
2002         if (!scsi_sense_valid(sshdr))
2003                 return 0;
2004
2005         if (sshdr->response_code >= 0x72) {
2006                 /*
2007                  * descriptor format
2008                  */
2009                 if (sb_len > 1)
2010                         sshdr->sense_key = (sense_buffer[1] & 0xf);
2011                 if (sb_len > 2)
2012                         sshdr->asc = sense_buffer[2];
2013                 if (sb_len > 3)
2014                         sshdr->ascq = sense_buffer[3];
2015                 if (sb_len > 7)
2016                         sshdr->additional_length = sense_buffer[7];
2017         } else {
2018                 /* 
2019                  * fixed format
2020                  */
2021                 if (sb_len > 2)
2022                         sshdr->sense_key = (sense_buffer[2] & 0xf);
2023                 if (sb_len > 7) {
2024                         sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2025                                          sb_len : (sense_buffer[7] + 8);
2026                         if (sb_len > 12)
2027                                 sshdr->asc = sense_buffer[12];
2028                         if (sb_len > 13)
2029                                 sshdr->ascq = sense_buffer[13];
2030                 }
2031         }
2032
2033         return 1;
2034 }
2035 EXPORT_SYMBOL(scsi_normalize_sense);
2036
2037 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2038                                  struct scsi_sense_hdr *sshdr)
2039 {
2040         return scsi_normalize_sense(cmd->sense_buffer,
2041                         SCSI_SENSE_BUFFERSIZE, sshdr);
2042 }
2043 EXPORT_SYMBOL(scsi_command_normalize_sense);
2044
2045 /**
2046  * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2047  * @sense_buffer:       byte array of descriptor format sense data
2048  * @sb_len:             number of valid bytes in sense_buffer
2049  * @desc_type:          value of descriptor type to find
2050  *                      (e.g. 0 -> information)
2051  *
2052  * Notes:
2053  *      only valid when sense data is in descriptor format
2054  *
2055  * Return value:
2056  *      pointer to start of (first) descriptor if found else NULL
2057  */
2058 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2059                                 int desc_type)
2060 {
2061         int add_sen_len, add_len, desc_len, k;
2062         const u8 * descp;
2063
2064         if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2065                 return NULL;
2066         if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2067                 return NULL;
2068         add_sen_len = (add_sen_len < (sb_len - 8)) ?
2069                         add_sen_len : (sb_len - 8);
2070         descp = &sense_buffer[8];
2071         for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2072                 descp += desc_len;
2073                 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2074                 desc_len = add_len + 2;
2075                 if (descp[0] == desc_type)
2076                         return descp;
2077                 if (add_len < 0) // short descriptor ??
2078                         break;
2079         }
2080         return NULL;
2081 }
2082 EXPORT_SYMBOL(scsi_sense_desc_find);
2083
2084 /**
2085  * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2086  * @sense_buffer:       byte array of sense data
2087  * @sb_len:             number of valid bytes in sense_buffer
2088  * @info_out:           pointer to 64 integer where 8 or 4 byte information
2089  *                      field will be placed if found.
2090  *
2091  * Return value:
2092  *      1 if information field found, 0 if not found.
2093  */
2094 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2095                             u64 * info_out)
2096 {
2097         int j;
2098         const u8 * ucp;
2099         u64 ull;
2100
2101         if (sb_len < 7)
2102                 return 0;
2103         switch (sense_buffer[0] & 0x7f) {
2104         case 0x70:
2105         case 0x71:
2106                 if (sense_buffer[0] & 0x80) {
2107                         *info_out = (sense_buffer[3] << 24) +
2108                                     (sense_buffer[4] << 16) +
2109                                     (sense_buffer[5] << 8) + sense_buffer[6];
2110                         return 1;
2111                 } else
2112                         return 0;
2113         case 0x72:
2114         case 0x73:
2115                 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2116                                            0 /* info desc */);
2117                 if (ucp && (0xa == ucp[1])) {
2118                         ull = 0;
2119                         for (j = 0; j < 8; ++j) {
2120                                 if (j > 0)
2121                                         ull <<= 8;
2122                                 ull |= ucp[4 + j];
2123                         }
2124                         *info_out = ull;
2125                         return 1;
2126                 } else
2127                         return 0;
2128         default:
2129                 return 0;
2130         }
2131 }
2132 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2133
2134 /**
2135  * scsi_build_sense_buffer - build sense data in a buffer
2136  * @desc:       Sense format (non zero == descriptor format,
2137  *              0 == fixed format)
2138  * @buf:        Where to build sense data
2139  * @key:        Sense key
2140  * @asc:        Additional sense code
2141  * @ascq:       Additional sense code qualifier
2142  *
2143  **/
2144 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2145 {
2146         if (desc) {
2147                 buf[0] = 0x72;  /* descriptor, current */
2148                 buf[1] = key;
2149                 buf[2] = asc;
2150                 buf[3] = ascq;
2151                 buf[7] = 0;
2152         } else {
2153                 buf[0] = 0x70;  /* fixed, current */
2154                 buf[2] = key;
2155                 buf[7] = 0xa;
2156                 buf[12] = asc;
2157                 buf[13] = ascq;
2158         }
2159 }
2160 EXPORT_SYMBOL(scsi_build_sense_buffer);