[SCSI] Fix erratic device offline during EH
[deliverable/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_transport.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_ioctl.h>
39
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
42 #include "scsi_transport_api.h"
43
44 #include <trace/events/scsi.h>
45
46 static void scsi_eh_done(struct scsi_cmnd *scmd);
47
48 /*
49 * These should *probably* be handled by the host itself.
50 * Since it is allowed to sleep, it probably should.
51 */
52 #define BUS_RESET_SETTLE_TIME (10)
53 #define HOST_RESET_SETTLE_TIME (10)
54
55 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
56
57 /* called with shost->host_lock held */
58 void scsi_eh_wakeup(struct Scsi_Host *shost)
59 {
60 if (shost->host_busy == shost->host_failed) {
61 trace_scsi_eh_wakeup(shost);
62 wake_up_process(shost->ehandler);
63 SCSI_LOG_ERROR_RECOVERY(5,
64 printk("Waking error handler thread\n"));
65 }
66 }
67
68 /**
69 * scsi_schedule_eh - schedule EH for SCSI host
70 * @shost: SCSI host to invoke error handling on.
71 *
72 * Schedule SCSI EH without scmd.
73 */
74 void scsi_schedule_eh(struct Scsi_Host *shost)
75 {
76 unsigned long flags;
77
78 spin_lock_irqsave(shost->host_lock, flags);
79
80 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
81 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
82 shost->host_eh_scheduled++;
83 scsi_eh_wakeup(shost);
84 }
85
86 spin_unlock_irqrestore(shost->host_lock, flags);
87 }
88 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
89
90 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
91 {
92 if (!shost->last_reset || !shost->eh_deadline)
93 return 0;
94
95 if (time_before(jiffies,
96 shost->last_reset + shost->eh_deadline))
97 return 0;
98
99 return 1;
100 }
101
102 /**
103 * scsi_eh_scmd_add - add scsi cmd to error handling.
104 * @scmd: scmd to run eh on.
105 * @eh_flag: optional SCSI_EH flag.
106 *
107 * Return value:
108 * 0 on failure.
109 */
110 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
111 {
112 struct Scsi_Host *shost = scmd->device->host;
113 unsigned long flags;
114 int ret = 0;
115
116 if (!shost->ehandler)
117 return 0;
118
119 spin_lock_irqsave(shost->host_lock, flags);
120 if (scsi_host_set_state(shost, SHOST_RECOVERY))
121 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
122 goto out_unlock;
123
124 if (shost->eh_deadline && !shost->last_reset)
125 shost->last_reset = jiffies;
126
127 ret = 1;
128 scmd->eh_eflags |= eh_flag;
129 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
130 shost->host_failed++;
131 scsi_eh_wakeup(shost);
132 out_unlock:
133 spin_unlock_irqrestore(shost->host_lock, flags);
134 return ret;
135 }
136
137 /**
138 * scsi_times_out - Timeout function for normal scsi commands.
139 * @req: request that is timing out.
140 *
141 * Notes:
142 * We do not need to lock this. There is the potential for a race
143 * only in that the normal completion handling might run, but if the
144 * normal completion function determines that the timer has already
145 * fired, then it mustn't do anything.
146 */
147 enum blk_eh_timer_return scsi_times_out(struct request *req)
148 {
149 struct scsi_cmnd *scmd = req->special;
150 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
151 struct Scsi_Host *host = scmd->device->host;
152
153 trace_scsi_dispatch_cmd_timeout(scmd);
154 scsi_log_completion(scmd, TIMEOUT_ERROR);
155
156 if (host->eh_deadline && !host->last_reset)
157 host->last_reset = jiffies;
158
159 if (host->transportt->eh_timed_out)
160 rtn = host->transportt->eh_timed_out(scmd);
161 else if (host->hostt->eh_timed_out)
162 rtn = host->hostt->eh_timed_out(scmd);
163
164 scmd->result |= DID_TIME_OUT << 16;
165
166 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
167 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
168 rtn = BLK_EH_HANDLED;
169
170 return rtn;
171 }
172
173 /**
174 * scsi_block_when_processing_errors - Prevent cmds from being queued.
175 * @sdev: Device on which we are performing recovery.
176 *
177 * Description:
178 * We block until the host is out of error recovery, and then check to
179 * see whether the host or the device is offline.
180 *
181 * Return value:
182 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
183 */
184 int scsi_block_when_processing_errors(struct scsi_device *sdev)
185 {
186 int online;
187
188 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
189
190 online = scsi_device_online(sdev);
191
192 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
193 online));
194
195 return online;
196 }
197 EXPORT_SYMBOL(scsi_block_when_processing_errors);
198
199 #ifdef CONFIG_SCSI_LOGGING
200 /**
201 * scsi_eh_prt_fail_stats - Log info on failures.
202 * @shost: scsi host being recovered.
203 * @work_q: Queue of scsi cmds to process.
204 */
205 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
206 struct list_head *work_q)
207 {
208 struct scsi_cmnd *scmd;
209 struct scsi_device *sdev;
210 int total_failures = 0;
211 int cmd_failed = 0;
212 int cmd_cancel = 0;
213 int devices_failed = 0;
214
215 shost_for_each_device(sdev, shost) {
216 list_for_each_entry(scmd, work_q, eh_entry) {
217 if (scmd->device == sdev) {
218 ++total_failures;
219 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
220 ++cmd_cancel;
221 else
222 ++cmd_failed;
223 }
224 }
225
226 if (cmd_cancel || cmd_failed) {
227 SCSI_LOG_ERROR_RECOVERY(3,
228 sdev_printk(KERN_INFO, sdev,
229 "%s: cmds failed: %d, cancel: %d\n",
230 __func__, cmd_failed,
231 cmd_cancel));
232 cmd_cancel = 0;
233 cmd_failed = 0;
234 ++devices_failed;
235 }
236 }
237
238 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
239 " devices require eh work\n",
240 total_failures, devices_failed));
241 }
242 #endif
243
244 /**
245 * scsi_report_lun_change - Set flag on all *other* devices on the same target
246 * to indicate that a UNIT ATTENTION is expected.
247 * @sdev: Device reporting the UNIT ATTENTION
248 */
249 static void scsi_report_lun_change(struct scsi_device *sdev)
250 {
251 sdev->sdev_target->expecting_lun_change = 1;
252 }
253
254 /**
255 * scsi_report_sense - Examine scsi sense information and log messages for
256 * certain conditions, also issue uevents for some of them.
257 * @sdev: Device reporting the sense code
258 * @sshdr: sshdr to be examined
259 */
260 static void scsi_report_sense(struct scsi_device *sdev,
261 struct scsi_sense_hdr *sshdr)
262 {
263 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
264
265 if (sshdr->sense_key == UNIT_ATTENTION) {
266 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
267 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
268 sdev_printk(KERN_WARNING, sdev,
269 "Inquiry data has changed");
270 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
271 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
272 scsi_report_lun_change(sdev);
273 sdev_printk(KERN_WARNING, sdev,
274 "Warning! Received an indication that the "
275 "LUN assignments on this target have "
276 "changed. The Linux SCSI layer does not "
277 "automatically remap LUN assignments.\n");
278 } else if (sshdr->asc == 0x3f)
279 sdev_printk(KERN_WARNING, sdev,
280 "Warning! Received an indication that the "
281 "operating parameters on this target have "
282 "changed. The Linux SCSI layer does not "
283 "automatically adjust these parameters.\n");
284
285 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
286 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
287 sdev_printk(KERN_WARNING, sdev,
288 "Warning! Received an indication that the "
289 "LUN reached a thin provisioning soft "
290 "threshold.\n");
291 }
292
293 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
294 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
295 sdev_printk(KERN_WARNING, sdev,
296 "Mode parameters changed");
297 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
298 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
299 sdev_printk(KERN_WARNING, sdev,
300 "Capacity data has changed");
301 } else if (sshdr->asc == 0x2a)
302 sdev_printk(KERN_WARNING, sdev,
303 "Parameters changed");
304 }
305
306 if (evt_type != SDEV_EVT_MAXBITS) {
307 set_bit(evt_type, sdev->pending_events);
308 schedule_work(&sdev->event_work);
309 }
310 }
311
312 /**
313 * scsi_check_sense - Examine scsi cmd sense
314 * @scmd: Cmd to have sense checked.
315 *
316 * Return value:
317 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
318 *
319 * Notes:
320 * When a deferred error is detected the current command has
321 * not been executed and needs retrying.
322 */
323 static int scsi_check_sense(struct scsi_cmnd *scmd)
324 {
325 struct scsi_device *sdev = scmd->device;
326 struct scsi_sense_hdr sshdr;
327
328 if (! scsi_command_normalize_sense(scmd, &sshdr))
329 return FAILED; /* no valid sense data */
330
331 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
332 /*
333 * nasty: for mid-layer issued TURs, we need to return the
334 * actual sense data without any recovery attempt. For eh
335 * issued ones, we need to try to recover and interpret
336 */
337 return SUCCESS;
338
339 scsi_report_sense(sdev, &sshdr);
340
341 if (scsi_sense_is_deferred(&sshdr))
342 return NEEDS_RETRY;
343
344 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
345 sdev->scsi_dh_data->scsi_dh->check_sense) {
346 int rc;
347
348 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
349 if (rc != SCSI_RETURN_NOT_HANDLED)
350 return rc;
351 /* handler does not care. Drop down to default handling */
352 }
353
354 /*
355 * Previous logic looked for FILEMARK, EOM or ILI which are
356 * mainly associated with tapes and returned SUCCESS.
357 */
358 if (sshdr.response_code == 0x70) {
359 /* fixed format */
360 if (scmd->sense_buffer[2] & 0xe0)
361 return SUCCESS;
362 } else {
363 /*
364 * descriptor format: look for "stream commands sense data
365 * descriptor" (see SSC-3). Assume single sense data
366 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
367 */
368 if ((sshdr.additional_length > 3) &&
369 (scmd->sense_buffer[8] == 0x4) &&
370 (scmd->sense_buffer[11] & 0xe0))
371 return SUCCESS;
372 }
373
374 switch (sshdr.sense_key) {
375 case NO_SENSE:
376 return SUCCESS;
377 case RECOVERED_ERROR:
378 return /* soft_error */ SUCCESS;
379
380 case ABORTED_COMMAND:
381 if (sshdr.asc == 0x10) /* DIF */
382 return SUCCESS;
383
384 return NEEDS_RETRY;
385 case NOT_READY:
386 case UNIT_ATTENTION:
387 /*
388 * if we are expecting a cc/ua because of a bus reset that we
389 * performed, treat this just as a retry. otherwise this is
390 * information that we should pass up to the upper-level driver
391 * so that we can deal with it there.
392 */
393 if (scmd->device->expecting_cc_ua) {
394 /*
395 * Because some device does not queue unit
396 * attentions correctly, we carefully check
397 * additional sense code and qualifier so as
398 * not to squash media change unit attention.
399 */
400 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
401 scmd->device->expecting_cc_ua = 0;
402 return NEEDS_RETRY;
403 }
404 }
405 /*
406 * we might also expect a cc/ua if another LUN on the target
407 * reported a UA with an ASC/ASCQ of 3F 0E -
408 * REPORTED LUNS DATA HAS CHANGED.
409 */
410 if (scmd->device->sdev_target->expecting_lun_change &&
411 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
412 return NEEDS_RETRY;
413 /*
414 * if the device is in the process of becoming ready, we
415 * should retry.
416 */
417 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
418 return NEEDS_RETRY;
419 /*
420 * if the device is not started, we need to wake
421 * the error handler to start the motor
422 */
423 if (scmd->device->allow_restart &&
424 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
425 return FAILED;
426 /*
427 * Pass the UA upwards for a determination in the completion
428 * functions.
429 */
430 return SUCCESS;
431
432 /* these are not supported */
433 case DATA_PROTECT:
434 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
435 /* Thin provisioning hard threshold reached */
436 set_host_byte(scmd, DID_ALLOC_FAILURE);
437 return SUCCESS;
438 }
439 case COPY_ABORTED:
440 case VOLUME_OVERFLOW:
441 case MISCOMPARE:
442 case BLANK_CHECK:
443 set_host_byte(scmd, DID_TARGET_FAILURE);
444 return SUCCESS;
445
446 case MEDIUM_ERROR:
447 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
448 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
449 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
450 set_host_byte(scmd, DID_MEDIUM_ERROR);
451 return SUCCESS;
452 }
453 return NEEDS_RETRY;
454
455 case HARDWARE_ERROR:
456 if (scmd->device->retry_hwerror)
457 return ADD_TO_MLQUEUE;
458 else
459 set_host_byte(scmd, DID_TARGET_FAILURE);
460
461 case ILLEGAL_REQUEST:
462 if (sshdr.asc == 0x20 || /* Invalid command operation code */
463 sshdr.asc == 0x21 || /* Logical block address out of range */
464 sshdr.asc == 0x24 || /* Invalid field in cdb */
465 sshdr.asc == 0x26) { /* Parameter value invalid */
466 set_host_byte(scmd, DID_TARGET_FAILURE);
467 }
468 return SUCCESS;
469
470 default:
471 return SUCCESS;
472 }
473 }
474
475 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
476 {
477 struct scsi_host_template *sht = sdev->host->hostt;
478 struct scsi_device *tmp_sdev;
479
480 if (!sht->change_queue_depth ||
481 sdev->queue_depth >= sdev->max_queue_depth)
482 return;
483
484 if (time_before(jiffies,
485 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
486 return;
487
488 if (time_before(jiffies,
489 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
490 return;
491
492 /*
493 * Walk all devices of a target and do
494 * ramp up on them.
495 */
496 shost_for_each_device(tmp_sdev, sdev->host) {
497 if (tmp_sdev->channel != sdev->channel ||
498 tmp_sdev->id != sdev->id ||
499 tmp_sdev->queue_depth == sdev->max_queue_depth)
500 continue;
501 /*
502 * call back into LLD to increase queue_depth by one
503 * with ramp up reason code.
504 */
505 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
506 SCSI_QDEPTH_RAMP_UP);
507 sdev->last_queue_ramp_up = jiffies;
508 }
509 }
510
511 static void scsi_handle_queue_full(struct scsi_device *sdev)
512 {
513 struct scsi_host_template *sht = sdev->host->hostt;
514 struct scsi_device *tmp_sdev;
515
516 if (!sht->change_queue_depth)
517 return;
518
519 shost_for_each_device(tmp_sdev, sdev->host) {
520 if (tmp_sdev->channel != sdev->channel ||
521 tmp_sdev->id != sdev->id)
522 continue;
523 /*
524 * We do not know the number of commands that were at
525 * the device when we got the queue full so we start
526 * from the highest possible value and work our way down.
527 */
528 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
529 SCSI_QDEPTH_QFULL);
530 }
531 }
532
533 /**
534 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
535 * @scmd: SCSI cmd to examine.
536 *
537 * Notes:
538 * This is *only* called when we are examining the status of commands
539 * queued during error recovery. the main difference here is that we
540 * don't allow for the possibility of retries here, and we are a lot
541 * more restrictive about what we consider acceptable.
542 */
543 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
544 {
545 /*
546 * first check the host byte, to see if there is anything in there
547 * that would indicate what we need to do.
548 */
549 if (host_byte(scmd->result) == DID_RESET) {
550 /*
551 * rats. we are already in the error handler, so we now
552 * get to try and figure out what to do next. if the sense
553 * is valid, we have a pretty good idea of what to do.
554 * if not, we mark it as FAILED.
555 */
556 return scsi_check_sense(scmd);
557 }
558 if (host_byte(scmd->result) != DID_OK)
559 return FAILED;
560
561 /*
562 * next, check the message byte.
563 */
564 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
565 return FAILED;
566
567 /*
568 * now, check the status byte to see if this indicates
569 * anything special.
570 */
571 switch (status_byte(scmd->result)) {
572 case GOOD:
573 scsi_handle_queue_ramp_up(scmd->device);
574 case COMMAND_TERMINATED:
575 return SUCCESS;
576 case CHECK_CONDITION:
577 return scsi_check_sense(scmd);
578 case CONDITION_GOOD:
579 case INTERMEDIATE_GOOD:
580 case INTERMEDIATE_C_GOOD:
581 /*
582 * who knows? FIXME(eric)
583 */
584 return SUCCESS;
585 case RESERVATION_CONFLICT:
586 if (scmd->cmnd[0] == TEST_UNIT_READY)
587 /* it is a success, we probed the device and
588 * found it */
589 return SUCCESS;
590 /* otherwise, we failed to send the command */
591 return FAILED;
592 case QUEUE_FULL:
593 scsi_handle_queue_full(scmd->device);
594 /* fall through */
595 case BUSY:
596 return NEEDS_RETRY;
597 default:
598 return FAILED;
599 }
600 return FAILED;
601 }
602
603 /**
604 * scsi_eh_done - Completion function for error handling.
605 * @scmd: Cmd that is done.
606 */
607 static void scsi_eh_done(struct scsi_cmnd *scmd)
608 {
609 struct completion *eh_action;
610
611 SCSI_LOG_ERROR_RECOVERY(3,
612 printk("%s scmd: %p result: %x\n",
613 __func__, scmd, scmd->result));
614
615 eh_action = scmd->device->host->eh_action;
616 if (eh_action)
617 complete(eh_action);
618 }
619
620 /**
621 * scsi_try_host_reset - ask host adapter to reset itself
622 * @scmd: SCSI cmd to send host reset.
623 */
624 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
625 {
626 unsigned long flags;
627 int rtn;
628 struct Scsi_Host *host = scmd->device->host;
629 struct scsi_host_template *hostt = host->hostt;
630
631 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
632 __func__));
633
634 if (!hostt->eh_host_reset_handler)
635 return FAILED;
636
637 rtn = hostt->eh_host_reset_handler(scmd);
638
639 if (rtn == SUCCESS) {
640 if (!hostt->skip_settle_delay)
641 ssleep(HOST_RESET_SETTLE_TIME);
642 spin_lock_irqsave(host->host_lock, flags);
643 scsi_report_bus_reset(host, scmd_channel(scmd));
644 spin_unlock_irqrestore(host->host_lock, flags);
645 }
646
647 return rtn;
648 }
649
650 /**
651 * scsi_try_bus_reset - ask host to perform a bus reset
652 * @scmd: SCSI cmd to send bus reset.
653 */
654 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
655 {
656 unsigned long flags;
657 int rtn;
658 struct Scsi_Host *host = scmd->device->host;
659 struct scsi_host_template *hostt = host->hostt;
660
661 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
662 __func__));
663
664 if (!hostt->eh_bus_reset_handler)
665 return FAILED;
666
667 rtn = hostt->eh_bus_reset_handler(scmd);
668
669 if (rtn == SUCCESS) {
670 if (!hostt->skip_settle_delay)
671 ssleep(BUS_RESET_SETTLE_TIME);
672 spin_lock_irqsave(host->host_lock, flags);
673 scsi_report_bus_reset(host, scmd_channel(scmd));
674 spin_unlock_irqrestore(host->host_lock, flags);
675 }
676
677 return rtn;
678 }
679
680 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
681 {
682 sdev->was_reset = 1;
683 sdev->expecting_cc_ua = 1;
684 }
685
686 /**
687 * scsi_try_target_reset - Ask host to perform a target reset
688 * @scmd: SCSI cmd used to send a target reset
689 *
690 * Notes:
691 * There is no timeout for this operation. if this operation is
692 * unreliable for a given host, then the host itself needs to put a
693 * timer on it, and set the host back to a consistent state prior to
694 * returning.
695 */
696 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
697 {
698 unsigned long flags;
699 int rtn;
700 struct Scsi_Host *host = scmd->device->host;
701 struct scsi_host_template *hostt = host->hostt;
702
703 if (!hostt->eh_target_reset_handler)
704 return FAILED;
705
706 rtn = hostt->eh_target_reset_handler(scmd);
707 if (rtn == SUCCESS) {
708 spin_lock_irqsave(host->host_lock, flags);
709 __starget_for_each_device(scsi_target(scmd->device), NULL,
710 __scsi_report_device_reset);
711 spin_unlock_irqrestore(host->host_lock, flags);
712 }
713
714 return rtn;
715 }
716
717 /**
718 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
719 * @scmd: SCSI cmd used to send BDR
720 *
721 * Notes:
722 * There is no timeout for this operation. if this operation is
723 * unreliable for a given host, then the host itself needs to put a
724 * timer on it, and set the host back to a consistent state prior to
725 * returning.
726 */
727 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
728 {
729 int rtn;
730 struct scsi_host_template *hostt = scmd->device->host->hostt;
731
732 if (!hostt->eh_device_reset_handler)
733 return FAILED;
734
735 rtn = hostt->eh_device_reset_handler(scmd);
736 if (rtn == SUCCESS)
737 __scsi_report_device_reset(scmd->device, NULL);
738 return rtn;
739 }
740
741 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
742 {
743 if (!hostt->eh_abort_handler)
744 return FAILED;
745
746 return hostt->eh_abort_handler(scmd);
747 }
748
749 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
750 {
751 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
752 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
753 if (scsi_try_target_reset(scmd) != SUCCESS)
754 if (scsi_try_bus_reset(scmd) != SUCCESS)
755 scsi_try_host_reset(scmd);
756 }
757
758 /**
759 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
760 * @scmd: SCSI command structure to hijack
761 * @ses: structure to save restore information
762 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
763 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
764 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
765 *
766 * This function is used to save a scsi command information before re-execution
767 * as part of the error recovery process. If @sense_bytes is 0 the command
768 * sent must be one that does not transfer any data. If @sense_bytes != 0
769 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
770 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
771 */
772 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
773 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
774 {
775 struct scsi_device *sdev = scmd->device;
776
777 /*
778 * We need saved copies of a number of fields - this is because
779 * error handling may need to overwrite these with different values
780 * to run different commands, and once error handling is complete,
781 * we will need to restore these values prior to running the actual
782 * command.
783 */
784 ses->cmd_len = scmd->cmd_len;
785 ses->cmnd = scmd->cmnd;
786 ses->data_direction = scmd->sc_data_direction;
787 ses->sdb = scmd->sdb;
788 ses->next_rq = scmd->request->next_rq;
789 ses->result = scmd->result;
790 ses->underflow = scmd->underflow;
791 ses->prot_op = scmd->prot_op;
792
793 scmd->prot_op = SCSI_PROT_NORMAL;
794 scmd->cmnd = ses->eh_cmnd;
795 memset(scmd->cmnd, 0, BLK_MAX_CDB);
796 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
797 scmd->request->next_rq = NULL;
798
799 if (sense_bytes) {
800 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
801 sense_bytes);
802 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
803 scmd->sdb.length);
804 scmd->sdb.table.sgl = &ses->sense_sgl;
805 scmd->sc_data_direction = DMA_FROM_DEVICE;
806 scmd->sdb.table.nents = 1;
807 scmd->cmnd[0] = REQUEST_SENSE;
808 scmd->cmnd[4] = scmd->sdb.length;
809 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
810 } else {
811 scmd->sc_data_direction = DMA_NONE;
812 if (cmnd) {
813 BUG_ON(cmnd_size > BLK_MAX_CDB);
814 memcpy(scmd->cmnd, cmnd, cmnd_size);
815 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
816 }
817 }
818
819 scmd->underflow = 0;
820
821 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
822 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
823 (sdev->lun << 5 & 0xe0);
824
825 /*
826 * Zero the sense buffer. The scsi spec mandates that any
827 * untransferred sense data should be interpreted as being zero.
828 */
829 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
830 }
831 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
832
833 /**
834 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
835 * @scmd: SCSI command structure to restore
836 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
837 *
838 * Undo any damage done by above scsi_eh_prep_cmnd().
839 */
840 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
841 {
842 /*
843 * Restore original data
844 */
845 scmd->cmd_len = ses->cmd_len;
846 scmd->cmnd = ses->cmnd;
847 scmd->sc_data_direction = ses->data_direction;
848 scmd->sdb = ses->sdb;
849 scmd->request->next_rq = ses->next_rq;
850 scmd->result = ses->result;
851 scmd->underflow = ses->underflow;
852 scmd->prot_op = ses->prot_op;
853 }
854 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
855
856 /**
857 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
858 * @scmd: SCSI command structure to hijack
859 * @cmnd: CDB to send
860 * @cmnd_size: size in bytes of @cmnd
861 * @timeout: timeout for this request
862 * @sense_bytes: size of sense data to copy or 0
863 *
864 * This function is used to send a scsi command down to a target device
865 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
866 *
867 * Return value:
868 * SUCCESS or FAILED or NEEDS_RETRY
869 */
870 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
871 int cmnd_size, int timeout, unsigned sense_bytes)
872 {
873 struct scsi_device *sdev = scmd->device;
874 struct Scsi_Host *shost = sdev->host;
875 DECLARE_COMPLETION_ONSTACK(done);
876 unsigned long timeleft = timeout;
877 struct scsi_eh_save ses;
878 const unsigned long stall_for = msecs_to_jiffies(100);
879 int rtn;
880
881 retry:
882 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
883 shost->eh_action = &done;
884
885 scsi_log_send(scmd);
886 scmd->scsi_done = scsi_eh_done;
887 rtn = shost->hostt->queuecommand(shost, scmd);
888 if (rtn) {
889 if (timeleft > stall_for) {
890 scsi_eh_restore_cmnd(scmd, &ses);
891 timeleft -= stall_for;
892 msleep(jiffies_to_msecs(stall_for));
893 goto retry;
894 }
895 /* signal not to enter either branch of the if () below */
896 timeleft = 0;
897 rtn = NEEDS_RETRY;
898 } else {
899 timeleft = wait_for_completion_timeout(&done, timeout);
900 }
901
902 shost->eh_action = NULL;
903
904 scsi_log_completion(scmd, rtn);
905
906 SCSI_LOG_ERROR_RECOVERY(3,
907 printk("%s: scmd: %p, timeleft: %ld\n",
908 __func__, scmd, timeleft));
909
910 /*
911 * If there is time left scsi_eh_done got called, and we will examine
912 * the actual status codes to see whether the command actually did
913 * complete normally, else if we have a zero return and no time left,
914 * the command must still be pending, so abort it and return FAILED.
915 * If we never actually managed to issue the command, because
916 * ->queuecommand() kept returning non zero, use the rtn = FAILED
917 * value above (so don't execute either branch of the if)
918 */
919 if (timeleft) {
920 rtn = scsi_eh_completed_normally(scmd);
921 SCSI_LOG_ERROR_RECOVERY(3,
922 printk("%s: scsi_eh_completed_normally %x\n",
923 __func__, rtn));
924
925 switch (rtn) {
926 case SUCCESS:
927 case NEEDS_RETRY:
928 case FAILED:
929 break;
930 case ADD_TO_MLQUEUE:
931 rtn = NEEDS_RETRY;
932 break;
933 default:
934 rtn = FAILED;
935 break;
936 }
937 } else if (!rtn) {
938 scsi_abort_eh_cmnd(scmd);
939 rtn = FAILED;
940 }
941
942 scsi_eh_restore_cmnd(scmd, &ses);
943
944 return rtn;
945 }
946
947 /**
948 * scsi_request_sense - Request sense data from a particular target.
949 * @scmd: SCSI cmd for request sense.
950 *
951 * Notes:
952 * Some hosts automatically obtain this information, others require
953 * that we obtain it on our own. This function will *not* return until
954 * the command either times out, or it completes.
955 */
956 static int scsi_request_sense(struct scsi_cmnd *scmd)
957 {
958 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
959 }
960
961 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
962 {
963 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
964 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
965 if (sdrv->eh_action)
966 rtn = sdrv->eh_action(scmd, rtn);
967 }
968 return rtn;
969 }
970
971 /**
972 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
973 * @scmd: Original SCSI cmd that eh has finished.
974 * @done_q: Queue for processed commands.
975 *
976 * Notes:
977 * We don't want to use the normal command completion while we are are
978 * still handling errors - it may cause other commands to be queued,
979 * and that would disturb what we are doing. Thus we really want to
980 * keep a list of pending commands for final completion, and once we
981 * are ready to leave error handling we handle completion for real.
982 */
983 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
984 {
985 scmd->device->host->host_failed--;
986 scmd->eh_eflags = 0;
987 list_move_tail(&scmd->eh_entry, done_q);
988 }
989 EXPORT_SYMBOL(scsi_eh_finish_cmd);
990
991 /**
992 * scsi_eh_get_sense - Get device sense data.
993 * @work_q: Queue of commands to process.
994 * @done_q: Queue of processed commands.
995 *
996 * Description:
997 * See if we need to request sense information. if so, then get it
998 * now, so we have a better idea of what to do.
999 *
1000 * Notes:
1001 * This has the unfortunate side effect that if a shost adapter does
1002 * not automatically request sense information, we end up shutting
1003 * it down before we request it.
1004 *
1005 * All drivers should request sense information internally these days,
1006 * so for now all I have to say is tough noogies if you end up in here.
1007 *
1008 * XXX: Long term this code should go away, but that needs an audit of
1009 * all LLDDs first.
1010 */
1011 int scsi_eh_get_sense(struct list_head *work_q,
1012 struct list_head *done_q)
1013 {
1014 struct scsi_cmnd *scmd, *next;
1015 struct Scsi_Host *shost;
1016 int rtn;
1017 unsigned long flags;
1018
1019 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1020 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1021 SCSI_SENSE_VALID(scmd))
1022 continue;
1023
1024 shost = scmd->device->host;
1025 spin_lock_irqsave(shost->host_lock, flags);
1026 if (scsi_host_eh_past_deadline(shost)) {
1027 spin_unlock_irqrestore(shost->host_lock, flags);
1028 SCSI_LOG_ERROR_RECOVERY(3,
1029 shost_printk(KERN_INFO, shost,
1030 "skip %s, past eh deadline\n",
1031 __func__));
1032 break;
1033 }
1034 spin_unlock_irqrestore(shost->host_lock, flags);
1035 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1036 "%s: requesting sense\n",
1037 current->comm));
1038 rtn = scsi_request_sense(scmd);
1039 if (rtn != SUCCESS)
1040 continue;
1041
1042 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
1043 " result %x\n", scmd,
1044 scmd->result));
1045 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
1046
1047 rtn = scsi_decide_disposition(scmd);
1048
1049 /*
1050 * if the result was normal, then just pass it along to the
1051 * upper level.
1052 */
1053 if (rtn == SUCCESS)
1054 /* we don't want this command reissued, just
1055 * finished with the sense data, so set
1056 * retries to the max allowed to ensure it
1057 * won't get reissued */
1058 scmd->retries = scmd->allowed;
1059 else if (rtn != NEEDS_RETRY)
1060 continue;
1061
1062 scsi_eh_finish_cmd(scmd, done_q);
1063 }
1064
1065 return list_empty(work_q);
1066 }
1067 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1068
1069 /**
1070 * scsi_eh_tur - Send TUR to device.
1071 * @scmd: &scsi_cmnd to send TUR
1072 *
1073 * Return value:
1074 * 0 - Device is ready. 1 - Device NOT ready.
1075 */
1076 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1077 {
1078 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1079 int retry_cnt = 1, rtn;
1080
1081 retry_tur:
1082 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1083 scmd->device->eh_timeout, 0);
1084
1085 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
1086 __func__, scmd, rtn));
1087
1088 switch (rtn) {
1089 case NEEDS_RETRY:
1090 if (retry_cnt--)
1091 goto retry_tur;
1092 /*FALLTHRU*/
1093 case SUCCESS:
1094 return 0;
1095 default:
1096 return 1;
1097 }
1098 }
1099
1100 /**
1101 * scsi_eh_test_devices - check if devices are responding from error recovery.
1102 * @cmd_list: scsi commands in error recovery.
1103 * @work_q: queue for commands which still need more error recovery
1104 * @done_q: queue for commands which are finished
1105 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1106 *
1107 * Decription:
1108 * Tests if devices are in a working state. Commands to devices now in
1109 * a working state are sent to the done_q while commands to devices which
1110 * are still failing to respond are returned to the work_q for more
1111 * processing.
1112 **/
1113 static int scsi_eh_test_devices(struct list_head *cmd_list,
1114 struct list_head *work_q,
1115 struct list_head *done_q, int try_stu)
1116 {
1117 struct scsi_cmnd *scmd, *next;
1118 struct scsi_device *sdev;
1119 int finish_cmds;
1120 unsigned long flags;
1121
1122 while (!list_empty(cmd_list)) {
1123 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1124 sdev = scmd->device;
1125
1126 if (!try_stu) {
1127 spin_lock_irqsave(sdev->host->host_lock, flags);
1128 if (scsi_host_eh_past_deadline(sdev->host)) {
1129 /* Push items back onto work_q */
1130 list_splice_init(cmd_list, work_q);
1131 spin_unlock_irqrestore(sdev->host->host_lock,
1132 flags);
1133 SCSI_LOG_ERROR_RECOVERY(3,
1134 shost_printk(KERN_INFO, sdev->host,
1135 "skip %s, past eh deadline",
1136 __func__));
1137 break;
1138 }
1139 spin_unlock_irqrestore(sdev->host->host_lock, flags);
1140 }
1141
1142 finish_cmds = !scsi_device_online(scmd->device) ||
1143 (try_stu && !scsi_eh_try_stu(scmd) &&
1144 !scsi_eh_tur(scmd)) ||
1145 !scsi_eh_tur(scmd);
1146
1147 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1148 if (scmd->device == sdev) {
1149 if (finish_cmds &&
1150 (try_stu ||
1151 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1152 scsi_eh_finish_cmd(scmd, done_q);
1153 else
1154 list_move_tail(&scmd->eh_entry, work_q);
1155 }
1156 }
1157 return list_empty(work_q);
1158 }
1159
1160
1161 /**
1162 * scsi_eh_abort_cmds - abort pending commands.
1163 * @work_q: &list_head for pending commands.
1164 * @done_q: &list_head for processed commands.
1165 *
1166 * Decription:
1167 * Try and see whether or not it makes sense to try and abort the
1168 * running command. This only works out to be the case if we have one
1169 * command that has timed out. If the command simply failed, it makes
1170 * no sense to try and abort the command, since as far as the shost
1171 * adapter is concerned, it isn't running.
1172 */
1173 static int scsi_eh_abort_cmds(struct list_head *work_q,
1174 struct list_head *done_q)
1175 {
1176 struct scsi_cmnd *scmd, *next;
1177 LIST_HEAD(check_list);
1178 int rtn;
1179 struct Scsi_Host *shost;
1180 unsigned long flags;
1181
1182 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1183 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1184 continue;
1185 shost = scmd->device->host;
1186 spin_lock_irqsave(shost->host_lock, flags);
1187 if (scsi_host_eh_past_deadline(shost)) {
1188 spin_unlock_irqrestore(shost->host_lock, flags);
1189 list_splice_init(&check_list, work_q);
1190 SCSI_LOG_ERROR_RECOVERY(3,
1191 shost_printk(KERN_INFO, shost,
1192 "skip %s, past eh deadline\n",
1193 __func__));
1194 return list_empty(work_q);
1195 }
1196 spin_unlock_irqrestore(shost->host_lock, flags);
1197 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1198 "0x%p\n", current->comm,
1199 scmd));
1200 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
1201 if (rtn == FAILED) {
1202 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1203 " cmd failed:"
1204 "0x%p\n",
1205 current->comm,
1206 scmd));
1207 list_splice_init(&check_list, work_q);
1208 return list_empty(work_q);
1209 }
1210 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1211 if (rtn == FAST_IO_FAIL)
1212 scsi_eh_finish_cmd(scmd, done_q);
1213 else
1214 list_move_tail(&scmd->eh_entry, &check_list);
1215 }
1216
1217 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1218 }
1219
1220 /**
1221 * scsi_eh_try_stu - Send START_UNIT to device.
1222 * @scmd: &scsi_cmnd to send START_UNIT
1223 *
1224 * Return value:
1225 * 0 - Device is ready. 1 - Device NOT ready.
1226 */
1227 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1228 {
1229 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1230
1231 if (scmd->device->allow_restart) {
1232 int i, rtn = NEEDS_RETRY;
1233
1234 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1235 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1236
1237 if (rtn == SUCCESS)
1238 return 0;
1239 }
1240
1241 return 1;
1242 }
1243
1244 /**
1245 * scsi_eh_stu - send START_UNIT if needed
1246 * @shost: &scsi host being recovered.
1247 * @work_q: &list_head for pending commands.
1248 * @done_q: &list_head for processed commands.
1249 *
1250 * Notes:
1251 * If commands are failing due to not ready, initializing command required,
1252 * try revalidating the device, which will end up sending a start unit.
1253 */
1254 static int scsi_eh_stu(struct Scsi_Host *shost,
1255 struct list_head *work_q,
1256 struct list_head *done_q)
1257 {
1258 struct scsi_cmnd *scmd, *stu_scmd, *next;
1259 struct scsi_device *sdev;
1260 unsigned long flags;
1261
1262 shost_for_each_device(sdev, shost) {
1263 spin_lock_irqsave(shost->host_lock, flags);
1264 if (scsi_host_eh_past_deadline(shost)) {
1265 spin_unlock_irqrestore(shost->host_lock, flags);
1266 SCSI_LOG_ERROR_RECOVERY(3,
1267 shost_printk(KERN_INFO, shost,
1268 "skip %s, past eh deadline\n",
1269 __func__));
1270 break;
1271 }
1272 spin_unlock_irqrestore(shost->host_lock, flags);
1273 stu_scmd = NULL;
1274 list_for_each_entry(scmd, work_q, eh_entry)
1275 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1276 scsi_check_sense(scmd) == FAILED ) {
1277 stu_scmd = scmd;
1278 break;
1279 }
1280
1281 if (!stu_scmd)
1282 continue;
1283
1284 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1285 " 0x%p\n", current->comm, sdev));
1286
1287 if (!scsi_eh_try_stu(stu_scmd)) {
1288 if (!scsi_device_online(sdev) ||
1289 !scsi_eh_tur(stu_scmd)) {
1290 list_for_each_entry_safe(scmd, next,
1291 work_q, eh_entry) {
1292 if (scmd->device == sdev &&
1293 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1294 scsi_eh_finish_cmd(scmd, done_q);
1295 }
1296 }
1297 } else {
1298 SCSI_LOG_ERROR_RECOVERY(3,
1299 printk("%s: START_UNIT failed to sdev:"
1300 " 0x%p\n", current->comm, sdev));
1301 }
1302 }
1303
1304 return list_empty(work_q);
1305 }
1306
1307
1308 /**
1309 * scsi_eh_bus_device_reset - send bdr if needed
1310 * @shost: scsi host being recovered.
1311 * @work_q: &list_head for pending commands.
1312 * @done_q: &list_head for processed commands.
1313 *
1314 * Notes:
1315 * Try a bus device reset. Still, look to see whether we have multiple
1316 * devices that are jammed or not - if we have multiple devices, it
1317 * makes no sense to try bus_device_reset - we really would need to try
1318 * a bus_reset instead.
1319 */
1320 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1321 struct list_head *work_q,
1322 struct list_head *done_q)
1323 {
1324 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1325 struct scsi_device *sdev;
1326 unsigned long flags;
1327 int rtn;
1328
1329 shost_for_each_device(sdev, shost) {
1330 spin_lock_irqsave(shost->host_lock, flags);
1331 if (scsi_host_eh_past_deadline(shost)) {
1332 spin_unlock_irqrestore(shost->host_lock, flags);
1333 SCSI_LOG_ERROR_RECOVERY(3,
1334 shost_printk(KERN_INFO, shost,
1335 "skip %s, past eh deadline\n",
1336 __func__));
1337 break;
1338 }
1339 spin_unlock_irqrestore(shost->host_lock, flags);
1340 bdr_scmd = NULL;
1341 list_for_each_entry(scmd, work_q, eh_entry)
1342 if (scmd->device == sdev) {
1343 bdr_scmd = scmd;
1344 break;
1345 }
1346
1347 if (!bdr_scmd)
1348 continue;
1349
1350 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1351 " 0x%p\n", current->comm,
1352 sdev));
1353 rtn = scsi_try_bus_device_reset(bdr_scmd);
1354 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1355 if (!scsi_device_online(sdev) ||
1356 rtn == FAST_IO_FAIL ||
1357 !scsi_eh_tur(bdr_scmd)) {
1358 list_for_each_entry_safe(scmd, next,
1359 work_q, eh_entry) {
1360 if (scmd->device == sdev &&
1361 scsi_eh_action(scmd, rtn) != FAILED)
1362 scsi_eh_finish_cmd(scmd,
1363 done_q);
1364 }
1365 }
1366 } else {
1367 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1368 " failed sdev:"
1369 "0x%p\n",
1370 current->comm,
1371 sdev));
1372 }
1373 }
1374
1375 return list_empty(work_q);
1376 }
1377
1378 /**
1379 * scsi_eh_target_reset - send target reset if needed
1380 * @shost: scsi host being recovered.
1381 * @work_q: &list_head for pending commands.
1382 * @done_q: &list_head for processed commands.
1383 *
1384 * Notes:
1385 * Try a target reset.
1386 */
1387 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1388 struct list_head *work_q,
1389 struct list_head *done_q)
1390 {
1391 LIST_HEAD(tmp_list);
1392 LIST_HEAD(check_list);
1393
1394 list_splice_init(work_q, &tmp_list);
1395
1396 while (!list_empty(&tmp_list)) {
1397 struct scsi_cmnd *next, *scmd;
1398 int rtn;
1399 unsigned int id;
1400 unsigned long flags;
1401
1402 spin_lock_irqsave(shost->host_lock, flags);
1403 if (scsi_host_eh_past_deadline(shost)) {
1404 spin_unlock_irqrestore(shost->host_lock, flags);
1405 /* push back on work queue for further processing */
1406 list_splice_init(&check_list, work_q);
1407 list_splice_init(&tmp_list, work_q);
1408 SCSI_LOG_ERROR_RECOVERY(3,
1409 shost_printk(KERN_INFO, shost,
1410 "skip %s, past eh deadline\n",
1411 __func__));
1412 return list_empty(work_q);
1413 }
1414 spin_unlock_irqrestore(shost->host_lock, flags);
1415
1416 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1417 id = scmd_id(scmd);
1418
1419 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1420 "to target %d\n",
1421 current->comm, id));
1422 rtn = scsi_try_target_reset(scmd);
1423 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1424 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1425 " failed target: "
1426 "%d\n",
1427 current->comm, id));
1428 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1429 if (scmd_id(scmd) != id)
1430 continue;
1431
1432 if (rtn == SUCCESS)
1433 list_move_tail(&scmd->eh_entry, &check_list);
1434 else if (rtn == FAST_IO_FAIL)
1435 scsi_eh_finish_cmd(scmd, done_q);
1436 else
1437 /* push back on work queue for further processing */
1438 list_move(&scmd->eh_entry, work_q);
1439 }
1440 }
1441
1442 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1443 }
1444
1445 /**
1446 * scsi_eh_bus_reset - send a bus reset
1447 * @shost: &scsi host being recovered.
1448 * @work_q: &list_head for pending commands.
1449 * @done_q: &list_head for processed commands.
1450 */
1451 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1452 struct list_head *work_q,
1453 struct list_head *done_q)
1454 {
1455 struct scsi_cmnd *scmd, *chan_scmd, *next;
1456 LIST_HEAD(check_list);
1457 unsigned int channel;
1458 int rtn;
1459 unsigned long flags;
1460
1461 /*
1462 * we really want to loop over the various channels, and do this on
1463 * a channel by channel basis. we should also check to see if any
1464 * of the failed commands are on soft_reset devices, and if so, skip
1465 * the reset.
1466 */
1467
1468 for (channel = 0; channel <= shost->max_channel; channel++) {
1469 spin_lock_irqsave(shost->host_lock, flags);
1470 if (scsi_host_eh_past_deadline(shost)) {
1471 spin_unlock_irqrestore(shost->host_lock, flags);
1472 list_splice_init(&check_list, work_q);
1473 SCSI_LOG_ERROR_RECOVERY(3,
1474 shost_printk(KERN_INFO, shost,
1475 "skip %s, past eh deadline\n",
1476 __func__));
1477 return list_empty(work_q);
1478 }
1479 spin_unlock_irqrestore(shost->host_lock, flags);
1480
1481 chan_scmd = NULL;
1482 list_for_each_entry(scmd, work_q, eh_entry) {
1483 if (channel == scmd_channel(scmd)) {
1484 chan_scmd = scmd;
1485 break;
1486 /*
1487 * FIXME add back in some support for
1488 * soft_reset devices.
1489 */
1490 }
1491 }
1492
1493 if (!chan_scmd)
1494 continue;
1495 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1496 " %d\n", current->comm,
1497 channel));
1498 rtn = scsi_try_bus_reset(chan_scmd);
1499 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1500 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1501 if (channel == scmd_channel(scmd)) {
1502 if (rtn == FAST_IO_FAIL)
1503 scsi_eh_finish_cmd(scmd,
1504 done_q);
1505 else
1506 list_move_tail(&scmd->eh_entry,
1507 &check_list);
1508 }
1509 }
1510 } else {
1511 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1512 " failed chan: %d\n",
1513 current->comm,
1514 channel));
1515 }
1516 }
1517 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1518 }
1519
1520 /**
1521 * scsi_eh_host_reset - send a host reset
1522 * @work_q: list_head for processed commands.
1523 * @done_q: list_head for processed commands.
1524 */
1525 static int scsi_eh_host_reset(struct list_head *work_q,
1526 struct list_head *done_q)
1527 {
1528 struct scsi_cmnd *scmd, *next;
1529 LIST_HEAD(check_list);
1530 int rtn;
1531
1532 if (!list_empty(work_q)) {
1533 scmd = list_entry(work_q->next,
1534 struct scsi_cmnd, eh_entry);
1535
1536 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1537 , current->comm));
1538
1539 rtn = scsi_try_host_reset(scmd);
1540 if (rtn == SUCCESS) {
1541 list_splice_init(work_q, &check_list);
1542 } else if (rtn == FAST_IO_FAIL) {
1543 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1544 scsi_eh_finish_cmd(scmd, done_q);
1545 }
1546 } else {
1547 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1548 " failed\n",
1549 current->comm));
1550 }
1551 }
1552 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1553 }
1554
1555 /**
1556 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1557 * @work_q: list_head for processed commands.
1558 * @done_q: list_head for processed commands.
1559 */
1560 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1561 struct list_head *done_q)
1562 {
1563 struct scsi_cmnd *scmd, *next;
1564
1565 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1566 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1567 "not ready after error recovery\n");
1568 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1569 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1570 /*
1571 * FIXME: Handle lost cmds.
1572 */
1573 }
1574 scsi_eh_finish_cmd(scmd, done_q);
1575 }
1576 return;
1577 }
1578
1579 /**
1580 * scsi_noretry_cmd - determinte if command should be failed fast
1581 * @scmd: SCSI cmd to examine.
1582 */
1583 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1584 {
1585 switch (host_byte(scmd->result)) {
1586 case DID_OK:
1587 break;
1588 case DID_BUS_BUSY:
1589 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1590 case DID_PARITY:
1591 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1592 case DID_ERROR:
1593 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1594 status_byte(scmd->result) == RESERVATION_CONFLICT)
1595 return 0;
1596 /* fall through */
1597 case DID_SOFT_ERROR:
1598 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1599 }
1600
1601 switch (status_byte(scmd->result)) {
1602 case CHECK_CONDITION:
1603 /*
1604 * assume caller has checked sense and determinted
1605 * the check condition was retryable.
1606 */
1607 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1608 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1609 return 1;
1610 }
1611
1612 return 0;
1613 }
1614
1615 /**
1616 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1617 * @scmd: SCSI cmd to examine.
1618 *
1619 * Notes:
1620 * This is *only* called when we are examining the status after sending
1621 * out the actual data command. any commands that are queued for error
1622 * recovery (e.g. test_unit_ready) do *not* come through here.
1623 *
1624 * When this routine returns failed, it means the error handler thread
1625 * is woken. In cases where the error code indicates an error that
1626 * doesn't require the error handler read (i.e. we don't need to
1627 * abort/reset), this function should return SUCCESS.
1628 */
1629 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1630 {
1631 int rtn;
1632
1633 /*
1634 * if the device is offline, then we clearly just pass the result back
1635 * up to the top level.
1636 */
1637 if (!scsi_device_online(scmd->device)) {
1638 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1639 " as SUCCESS\n",
1640 __func__));
1641 return SUCCESS;
1642 }
1643
1644 /*
1645 * first check the host byte, to see if there is anything in there
1646 * that would indicate what we need to do.
1647 */
1648 switch (host_byte(scmd->result)) {
1649 case DID_PASSTHROUGH:
1650 /*
1651 * no matter what, pass this through to the upper layer.
1652 * nuke this special code so that it looks like we are saying
1653 * did_ok.
1654 */
1655 scmd->result &= 0xff00ffff;
1656 return SUCCESS;
1657 case DID_OK:
1658 /*
1659 * looks good. drop through, and check the next byte.
1660 */
1661 break;
1662 case DID_NO_CONNECT:
1663 case DID_BAD_TARGET:
1664 case DID_ABORT:
1665 /*
1666 * note - this means that we just report the status back
1667 * to the top level driver, not that we actually think
1668 * that it indicates SUCCESS.
1669 */
1670 return SUCCESS;
1671 /*
1672 * when the low level driver returns did_soft_error,
1673 * it is responsible for keeping an internal retry counter
1674 * in order to avoid endless loops (db)
1675 *
1676 * actually this is a bug in this function here. we should
1677 * be mindful of the maximum number of retries specified
1678 * and not get stuck in a loop.
1679 */
1680 case DID_SOFT_ERROR:
1681 goto maybe_retry;
1682 case DID_IMM_RETRY:
1683 return NEEDS_RETRY;
1684
1685 case DID_REQUEUE:
1686 return ADD_TO_MLQUEUE;
1687 case DID_TRANSPORT_DISRUPTED:
1688 /*
1689 * LLD/transport was disrupted during processing of the IO.
1690 * The transport class is now blocked/blocking,
1691 * and the transport will decide what to do with the IO
1692 * based on its timers and recovery capablilities if
1693 * there are enough retries.
1694 */
1695 goto maybe_retry;
1696 case DID_TRANSPORT_FAILFAST:
1697 /*
1698 * The transport decided to failfast the IO (most likely
1699 * the fast io fail tmo fired), so send IO directly upwards.
1700 */
1701 return SUCCESS;
1702 case DID_ERROR:
1703 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1704 status_byte(scmd->result) == RESERVATION_CONFLICT)
1705 /*
1706 * execute reservation conflict processing code
1707 * lower down
1708 */
1709 break;
1710 /* fallthrough */
1711 case DID_BUS_BUSY:
1712 case DID_PARITY:
1713 goto maybe_retry;
1714 case DID_TIME_OUT:
1715 /*
1716 * when we scan the bus, we get timeout messages for
1717 * these commands if there is no device available.
1718 * other hosts report did_no_connect for the same thing.
1719 */
1720 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1721 scmd->cmnd[0] == INQUIRY)) {
1722 return SUCCESS;
1723 } else {
1724 return FAILED;
1725 }
1726 case DID_RESET:
1727 return SUCCESS;
1728 default:
1729 return FAILED;
1730 }
1731
1732 /*
1733 * next, check the message byte.
1734 */
1735 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1736 return FAILED;
1737
1738 /*
1739 * check the status byte to see if this indicates anything special.
1740 */
1741 switch (status_byte(scmd->result)) {
1742 case QUEUE_FULL:
1743 scsi_handle_queue_full(scmd->device);
1744 /*
1745 * the case of trying to send too many commands to a
1746 * tagged queueing device.
1747 */
1748 case BUSY:
1749 /*
1750 * device can't talk to us at the moment. Should only
1751 * occur (SAM-3) when the task queue is empty, so will cause
1752 * the empty queue handling to trigger a stall in the
1753 * device.
1754 */
1755 return ADD_TO_MLQUEUE;
1756 case GOOD:
1757 if (scmd->cmnd[0] == REPORT_LUNS)
1758 scmd->device->sdev_target->expecting_lun_change = 0;
1759 scsi_handle_queue_ramp_up(scmd->device);
1760 case COMMAND_TERMINATED:
1761 return SUCCESS;
1762 case TASK_ABORTED:
1763 goto maybe_retry;
1764 case CHECK_CONDITION:
1765 rtn = scsi_check_sense(scmd);
1766 if (rtn == NEEDS_RETRY)
1767 goto maybe_retry;
1768 /* if rtn == FAILED, we have no sense information;
1769 * returning FAILED will wake the error handler thread
1770 * to collect the sense and redo the decide
1771 * disposition */
1772 return rtn;
1773 case CONDITION_GOOD:
1774 case INTERMEDIATE_GOOD:
1775 case INTERMEDIATE_C_GOOD:
1776 case ACA_ACTIVE:
1777 /*
1778 * who knows? FIXME(eric)
1779 */
1780 return SUCCESS;
1781
1782 case RESERVATION_CONFLICT:
1783 sdev_printk(KERN_INFO, scmd->device,
1784 "reservation conflict\n");
1785 set_host_byte(scmd, DID_NEXUS_FAILURE);
1786 return SUCCESS; /* causes immediate i/o error */
1787 default:
1788 return FAILED;
1789 }
1790 return FAILED;
1791
1792 maybe_retry:
1793
1794 /* we requeue for retry because the error was retryable, and
1795 * the request was not marked fast fail. Note that above,
1796 * even if the request is marked fast fail, we still requeue
1797 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1798 if ((++scmd->retries) <= scmd->allowed
1799 && !scsi_noretry_cmd(scmd)) {
1800 return NEEDS_RETRY;
1801 } else {
1802 /*
1803 * no more retries - report this one back to upper level.
1804 */
1805 return SUCCESS;
1806 }
1807 }
1808
1809 static void eh_lock_door_done(struct request *req, int uptodate)
1810 {
1811 __blk_put_request(req->q, req);
1812 }
1813
1814 /**
1815 * scsi_eh_lock_door - Prevent medium removal for the specified device
1816 * @sdev: SCSI device to prevent medium removal
1817 *
1818 * Locking:
1819 * We must be called from process context.
1820 *
1821 * Notes:
1822 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1823 * head of the devices request queue, and continue.
1824 */
1825 static void scsi_eh_lock_door(struct scsi_device *sdev)
1826 {
1827 struct request *req;
1828
1829 /*
1830 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1831 * request becomes available
1832 */
1833 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1834
1835 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1836 req->cmd[1] = 0;
1837 req->cmd[2] = 0;
1838 req->cmd[3] = 0;
1839 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1840 req->cmd[5] = 0;
1841
1842 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1843
1844 req->cmd_type = REQ_TYPE_BLOCK_PC;
1845 req->cmd_flags |= REQ_QUIET;
1846 req->timeout = 10 * HZ;
1847 req->retries = 5;
1848
1849 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1850 }
1851
1852 /**
1853 * scsi_restart_operations - restart io operations to the specified host.
1854 * @shost: Host we are restarting.
1855 *
1856 * Notes:
1857 * When we entered the error handler, we blocked all further i/o to
1858 * this device. we need to 'reverse' this process.
1859 */
1860 static void scsi_restart_operations(struct Scsi_Host *shost)
1861 {
1862 struct scsi_device *sdev;
1863 unsigned long flags;
1864
1865 /*
1866 * If the door was locked, we need to insert a door lock request
1867 * onto the head of the SCSI request queue for the device. There
1868 * is no point trying to lock the door of an off-line device.
1869 */
1870 shost_for_each_device(sdev, shost) {
1871 if (scsi_device_online(sdev) && sdev->locked)
1872 scsi_eh_lock_door(sdev);
1873 }
1874
1875 /*
1876 * next free up anything directly waiting upon the host. this
1877 * will be requests for character device operations, and also for
1878 * ioctls to queued block devices.
1879 */
1880 SCSI_LOG_ERROR_RECOVERY(3,
1881 printk("scsi_eh_%d waking up host to restart\n",
1882 shost->host_no));
1883
1884 spin_lock_irqsave(shost->host_lock, flags);
1885 if (scsi_host_set_state(shost, SHOST_RUNNING))
1886 if (scsi_host_set_state(shost, SHOST_CANCEL))
1887 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1888 spin_unlock_irqrestore(shost->host_lock, flags);
1889
1890 wake_up(&shost->host_wait);
1891
1892 /*
1893 * finally we need to re-initiate requests that may be pending. we will
1894 * have had everything blocked while error handling is taking place, and
1895 * now that error recovery is done, we will need to ensure that these
1896 * requests are started.
1897 */
1898 scsi_run_host_queues(shost);
1899
1900 /*
1901 * if eh is active and host_eh_scheduled is pending we need to re-run
1902 * recovery. we do this check after scsi_run_host_queues() to allow
1903 * everything pent up since the last eh run a chance to make forward
1904 * progress before we sync again. Either we'll immediately re-run
1905 * recovery or scsi_device_unbusy() will wake us again when these
1906 * pending commands complete.
1907 */
1908 spin_lock_irqsave(shost->host_lock, flags);
1909 if (shost->host_eh_scheduled)
1910 if (scsi_host_set_state(shost, SHOST_RECOVERY))
1911 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
1912 spin_unlock_irqrestore(shost->host_lock, flags);
1913 }
1914
1915 /**
1916 * scsi_eh_ready_devs - check device ready state and recover if not.
1917 * @shost: host to be recovered.
1918 * @work_q: &list_head for pending commands.
1919 * @done_q: &list_head for processed commands.
1920 */
1921 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1922 struct list_head *work_q,
1923 struct list_head *done_q)
1924 {
1925 if (!scsi_eh_stu(shost, work_q, done_q))
1926 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1927 if (!scsi_eh_target_reset(shost, work_q, done_q))
1928 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1929 if (!scsi_eh_host_reset(work_q, done_q))
1930 scsi_eh_offline_sdevs(work_q,
1931 done_q);
1932 }
1933 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1934
1935 /**
1936 * scsi_eh_flush_done_q - finish processed commands or retry them.
1937 * @done_q: list_head of processed commands.
1938 */
1939 void scsi_eh_flush_done_q(struct list_head *done_q)
1940 {
1941 struct scsi_cmnd *scmd, *next;
1942
1943 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1944 list_del_init(&scmd->eh_entry);
1945 if (scsi_device_online(scmd->device) &&
1946 !scsi_noretry_cmd(scmd) &&
1947 (++scmd->retries <= scmd->allowed)) {
1948 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1949 " retry cmd: %p\n",
1950 current->comm,
1951 scmd));
1952 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1953 } else {
1954 /*
1955 * If just we got sense for the device (called
1956 * scsi_eh_get_sense), scmd->result is already
1957 * set, do not set DRIVER_TIMEOUT.
1958 */
1959 if (!scmd->result)
1960 scmd->result |= (DRIVER_TIMEOUT << 24);
1961 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1962 " cmd: %p\n",
1963 current->comm, scmd));
1964 scsi_finish_command(scmd);
1965 }
1966 }
1967 }
1968 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1969
1970 /**
1971 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1972 * @shost: Host to unjam.
1973 *
1974 * Notes:
1975 * When we come in here, we *know* that all commands on the bus have
1976 * either completed, failed or timed out. we also know that no further
1977 * commands are being sent to the host, so things are relatively quiet
1978 * and we have freedom to fiddle with things as we wish.
1979 *
1980 * This is only the *default* implementation. it is possible for
1981 * individual drivers to supply their own version of this function, and
1982 * if the maintainer wishes to do this, it is strongly suggested that
1983 * this function be taken as a template and modified. this function
1984 * was designed to correctly handle problems for about 95% of the
1985 * different cases out there, and it should always provide at least a
1986 * reasonable amount of error recovery.
1987 *
1988 * Any command marked 'failed' or 'timeout' must eventually have
1989 * scsi_finish_cmd() called for it. we do all of the retry stuff
1990 * here, so when we restart the host after we return it should have an
1991 * empty queue.
1992 */
1993 static void scsi_unjam_host(struct Scsi_Host *shost)
1994 {
1995 unsigned long flags;
1996 LIST_HEAD(eh_work_q);
1997 LIST_HEAD(eh_done_q);
1998
1999 spin_lock_irqsave(shost->host_lock, flags);
2000 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2001 spin_unlock_irqrestore(shost->host_lock, flags);
2002
2003 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2004
2005 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2006 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2007 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2008
2009 spin_lock_irqsave(shost->host_lock, flags);
2010 if (shost->eh_deadline)
2011 shost->last_reset = 0;
2012 spin_unlock_irqrestore(shost->host_lock, flags);
2013 scsi_eh_flush_done_q(&eh_done_q);
2014 }
2015
2016 /**
2017 * scsi_error_handler - SCSI error handler thread
2018 * @data: Host for which we are running.
2019 *
2020 * Notes:
2021 * This is the main error handling loop. This is run as a kernel thread
2022 * for every SCSI host and handles all error handling activity.
2023 */
2024 int scsi_error_handler(void *data)
2025 {
2026 struct Scsi_Host *shost = data;
2027
2028 /*
2029 * We use TASK_INTERRUPTIBLE so that the thread is not
2030 * counted against the load average as a running process.
2031 * We never actually get interrupted because kthread_run
2032 * disables signal delivery for the created thread.
2033 */
2034 while (!kthread_should_stop()) {
2035 set_current_state(TASK_INTERRUPTIBLE);
2036 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2037 shost->host_failed != shost->host_busy) {
2038 SCSI_LOG_ERROR_RECOVERY(1,
2039 printk("scsi_eh_%d: sleeping\n",
2040 shost->host_no));
2041 schedule();
2042 continue;
2043 }
2044
2045 __set_current_state(TASK_RUNNING);
2046 SCSI_LOG_ERROR_RECOVERY(1,
2047 printk("scsi_eh_%d: waking up %d/%d/%d\n",
2048 shost->host_no, shost->host_eh_scheduled,
2049 shost->host_failed, shost->host_busy));
2050
2051 /*
2052 * We have a host that is failing for some reason. Figure out
2053 * what we need to do to get it up and online again (if we can).
2054 * If we fail, we end up taking the thing offline.
2055 */
2056 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2057 SCSI_LOG_ERROR_RECOVERY(1,
2058 printk(KERN_ERR "Error handler scsi_eh_%d "
2059 "unable to autoresume\n",
2060 shost->host_no));
2061 continue;
2062 }
2063
2064 if (shost->transportt->eh_strategy_handler)
2065 shost->transportt->eh_strategy_handler(shost);
2066 else
2067 scsi_unjam_host(shost);
2068
2069 /*
2070 * Note - if the above fails completely, the action is to take
2071 * individual devices offline and flush the queue of any
2072 * outstanding requests that may have been pending. When we
2073 * restart, we restart any I/O to any other devices on the bus
2074 * which are still online.
2075 */
2076 scsi_restart_operations(shost);
2077 if (!shost->eh_noresume)
2078 scsi_autopm_put_host(shost);
2079 }
2080 __set_current_state(TASK_RUNNING);
2081
2082 SCSI_LOG_ERROR_RECOVERY(1,
2083 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
2084 shost->ehandler = NULL;
2085 return 0;
2086 }
2087
2088 /*
2089 * Function: scsi_report_bus_reset()
2090 *
2091 * Purpose: Utility function used by low-level drivers to report that
2092 * they have observed a bus reset on the bus being handled.
2093 *
2094 * Arguments: shost - Host in question
2095 * channel - channel on which reset was observed.
2096 *
2097 * Returns: Nothing
2098 *
2099 * Lock status: Host lock must be held.
2100 *
2101 * Notes: This only needs to be called if the reset is one which
2102 * originates from an unknown location. Resets originated
2103 * by the mid-level itself don't need to call this, but there
2104 * should be no harm.
2105 *
2106 * The main purpose of this is to make sure that a CHECK_CONDITION
2107 * is properly treated.
2108 */
2109 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2110 {
2111 struct scsi_device *sdev;
2112
2113 __shost_for_each_device(sdev, shost) {
2114 if (channel == sdev_channel(sdev))
2115 __scsi_report_device_reset(sdev, NULL);
2116 }
2117 }
2118 EXPORT_SYMBOL(scsi_report_bus_reset);
2119
2120 /*
2121 * Function: scsi_report_device_reset()
2122 *
2123 * Purpose: Utility function used by low-level drivers to report that
2124 * they have observed a device reset on the device being handled.
2125 *
2126 * Arguments: shost - Host in question
2127 * channel - channel on which reset was observed
2128 * target - target on which reset was observed
2129 *
2130 * Returns: Nothing
2131 *
2132 * Lock status: Host lock must be held
2133 *
2134 * Notes: This only needs to be called if the reset is one which
2135 * originates from an unknown location. Resets originated
2136 * by the mid-level itself don't need to call this, but there
2137 * should be no harm.
2138 *
2139 * The main purpose of this is to make sure that a CHECK_CONDITION
2140 * is properly treated.
2141 */
2142 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2143 {
2144 struct scsi_device *sdev;
2145
2146 __shost_for_each_device(sdev, shost) {
2147 if (channel == sdev_channel(sdev) &&
2148 target == sdev_id(sdev))
2149 __scsi_report_device_reset(sdev, NULL);
2150 }
2151 }
2152 EXPORT_SYMBOL(scsi_report_device_reset);
2153
2154 static void
2155 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2156 {
2157 }
2158
2159 /*
2160 * Function: scsi_reset_provider
2161 *
2162 * Purpose: Send requested reset to a bus or device at any phase.
2163 *
2164 * Arguments: device - device to send reset to
2165 * flag - reset type (see scsi.h)
2166 *
2167 * Returns: SUCCESS/FAILURE.
2168 *
2169 * Notes: This is used by the SCSI Generic driver to provide
2170 * Bus/Device reset capability.
2171 */
2172 int
2173 scsi_reset_provider(struct scsi_device *dev, int flag)
2174 {
2175 struct scsi_cmnd *scmd;
2176 struct Scsi_Host *shost = dev->host;
2177 struct request req;
2178 unsigned long flags;
2179 int rtn;
2180
2181 if (scsi_autopm_get_host(shost) < 0)
2182 return FAILED;
2183
2184 scmd = scsi_get_command(dev, GFP_KERNEL);
2185 blk_rq_init(NULL, &req);
2186 scmd->request = &req;
2187
2188 scmd->cmnd = req.cmd;
2189
2190 scmd->scsi_done = scsi_reset_provider_done_command;
2191 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2192
2193 scmd->cmd_len = 0;
2194
2195 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2196
2197 spin_lock_irqsave(shost->host_lock, flags);
2198 shost->tmf_in_progress = 1;
2199 spin_unlock_irqrestore(shost->host_lock, flags);
2200
2201 switch (flag) {
2202 case SCSI_TRY_RESET_DEVICE:
2203 rtn = scsi_try_bus_device_reset(scmd);
2204 if (rtn == SUCCESS)
2205 break;
2206 /* FALLTHROUGH */
2207 case SCSI_TRY_RESET_TARGET:
2208 rtn = scsi_try_target_reset(scmd);
2209 if (rtn == SUCCESS)
2210 break;
2211 /* FALLTHROUGH */
2212 case SCSI_TRY_RESET_BUS:
2213 rtn = scsi_try_bus_reset(scmd);
2214 if (rtn == SUCCESS)
2215 break;
2216 /* FALLTHROUGH */
2217 case SCSI_TRY_RESET_HOST:
2218 rtn = scsi_try_host_reset(scmd);
2219 break;
2220 default:
2221 rtn = FAILED;
2222 }
2223
2224 spin_lock_irqsave(shost->host_lock, flags);
2225 shost->tmf_in_progress = 0;
2226 spin_unlock_irqrestore(shost->host_lock, flags);
2227
2228 /*
2229 * be sure to wake up anyone who was sleeping or had their queue
2230 * suspended while we performed the TMF.
2231 */
2232 SCSI_LOG_ERROR_RECOVERY(3,
2233 printk("%s: waking up host to restart after TMF\n",
2234 __func__));
2235
2236 wake_up(&shost->host_wait);
2237
2238 scsi_run_host_queues(shost);
2239
2240 scsi_next_command(scmd);
2241 scsi_autopm_put_host(shost);
2242 return rtn;
2243 }
2244 EXPORT_SYMBOL(scsi_reset_provider);
2245
2246 /**
2247 * scsi_normalize_sense - normalize main elements from either fixed or
2248 * descriptor sense data format into a common format.
2249 *
2250 * @sense_buffer: byte array containing sense data returned by device
2251 * @sb_len: number of valid bytes in sense_buffer
2252 * @sshdr: pointer to instance of structure that common
2253 * elements are written to.
2254 *
2255 * Notes:
2256 * The "main elements" from sense data are: response_code, sense_key,
2257 * asc, ascq and additional_length (only for descriptor format).
2258 *
2259 * Typically this function can be called after a device has
2260 * responded to a SCSI command with the CHECK_CONDITION status.
2261 *
2262 * Return value:
2263 * 1 if valid sense data information found, else 0;
2264 */
2265 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2266 struct scsi_sense_hdr *sshdr)
2267 {
2268 if (!sense_buffer || !sb_len)
2269 return 0;
2270
2271 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2272
2273 sshdr->response_code = (sense_buffer[0] & 0x7f);
2274
2275 if (!scsi_sense_valid(sshdr))
2276 return 0;
2277
2278 if (sshdr->response_code >= 0x72) {
2279 /*
2280 * descriptor format
2281 */
2282 if (sb_len > 1)
2283 sshdr->sense_key = (sense_buffer[1] & 0xf);
2284 if (sb_len > 2)
2285 sshdr->asc = sense_buffer[2];
2286 if (sb_len > 3)
2287 sshdr->ascq = sense_buffer[3];
2288 if (sb_len > 7)
2289 sshdr->additional_length = sense_buffer[7];
2290 } else {
2291 /*
2292 * fixed format
2293 */
2294 if (sb_len > 2)
2295 sshdr->sense_key = (sense_buffer[2] & 0xf);
2296 if (sb_len > 7) {
2297 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2298 sb_len : (sense_buffer[7] + 8);
2299 if (sb_len > 12)
2300 sshdr->asc = sense_buffer[12];
2301 if (sb_len > 13)
2302 sshdr->ascq = sense_buffer[13];
2303 }
2304 }
2305
2306 return 1;
2307 }
2308 EXPORT_SYMBOL(scsi_normalize_sense);
2309
2310 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2311 struct scsi_sense_hdr *sshdr)
2312 {
2313 return scsi_normalize_sense(cmd->sense_buffer,
2314 SCSI_SENSE_BUFFERSIZE, sshdr);
2315 }
2316 EXPORT_SYMBOL(scsi_command_normalize_sense);
2317
2318 /**
2319 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2320 * @sense_buffer: byte array of descriptor format sense data
2321 * @sb_len: number of valid bytes in sense_buffer
2322 * @desc_type: value of descriptor type to find
2323 * (e.g. 0 -> information)
2324 *
2325 * Notes:
2326 * only valid when sense data is in descriptor format
2327 *
2328 * Return value:
2329 * pointer to start of (first) descriptor if found else NULL
2330 */
2331 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2332 int desc_type)
2333 {
2334 int add_sen_len, add_len, desc_len, k;
2335 const u8 * descp;
2336
2337 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2338 return NULL;
2339 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2340 return NULL;
2341 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2342 add_sen_len : (sb_len - 8);
2343 descp = &sense_buffer[8];
2344 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2345 descp += desc_len;
2346 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2347 desc_len = add_len + 2;
2348 if (descp[0] == desc_type)
2349 return descp;
2350 if (add_len < 0) // short descriptor ??
2351 break;
2352 }
2353 return NULL;
2354 }
2355 EXPORT_SYMBOL(scsi_sense_desc_find);
2356
2357 /**
2358 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2359 * @sense_buffer: byte array of sense data
2360 * @sb_len: number of valid bytes in sense_buffer
2361 * @info_out: pointer to 64 integer where 8 or 4 byte information
2362 * field will be placed if found.
2363 *
2364 * Return value:
2365 * 1 if information field found, 0 if not found.
2366 */
2367 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2368 u64 * info_out)
2369 {
2370 int j;
2371 const u8 * ucp;
2372 u64 ull;
2373
2374 if (sb_len < 7)
2375 return 0;
2376 switch (sense_buffer[0] & 0x7f) {
2377 case 0x70:
2378 case 0x71:
2379 if (sense_buffer[0] & 0x80) {
2380 *info_out = (sense_buffer[3] << 24) +
2381 (sense_buffer[4] << 16) +
2382 (sense_buffer[5] << 8) + sense_buffer[6];
2383 return 1;
2384 } else
2385 return 0;
2386 case 0x72:
2387 case 0x73:
2388 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2389 0 /* info desc */);
2390 if (ucp && (0xa == ucp[1])) {
2391 ull = 0;
2392 for (j = 0; j < 8; ++j) {
2393 if (j > 0)
2394 ull <<= 8;
2395 ull |= ucp[4 + j];
2396 }
2397 *info_out = ull;
2398 return 1;
2399 } else
2400 return 0;
2401 default:
2402 return 0;
2403 }
2404 }
2405 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2406
2407 /**
2408 * scsi_build_sense_buffer - build sense data in a buffer
2409 * @desc: Sense format (non zero == descriptor format,
2410 * 0 == fixed format)
2411 * @buf: Where to build sense data
2412 * @key: Sense key
2413 * @asc: Additional sense code
2414 * @ascq: Additional sense code qualifier
2415 *
2416 **/
2417 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2418 {
2419 if (desc) {
2420 buf[0] = 0x72; /* descriptor, current */
2421 buf[1] = key;
2422 buf[2] = asc;
2423 buf[3] = ascq;
2424 buf[7] = 0;
2425 } else {
2426 buf[0] = 0x70; /* fixed, current */
2427 buf[2] = key;
2428 buf[7] = 0xa;
2429 buf[12] = asc;
2430 buf[13] = ascq;
2431 }
2432 }
2433 EXPORT_SYMBOL(scsi_build_sense_buffer);
This page took 0.083025 seconds and 5 git commands to generate.