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/timer.h>
20 #include <linux/string.h>
21 #include <linux/slab.h>
22 #include <linux/kernel.h>
23 #include <linux/interrupt.h>
24 #include <linux/blkdev.h>
25 #include <linux/delay.h>
26
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_eh.h>
31 #include <scsi/scsi_host.h>
32 #include <scsi/scsi_ioctl.h>
33 #include <scsi/scsi_request.h>
34
35 #include "scsi_priv.h"
36 #include "scsi_logging.h"
37
38 #define SENSE_TIMEOUT (10*HZ)
39 #define START_UNIT_TIMEOUT (30*HZ)
40
41 /*
42 * These should *probably* be handled by the host itself.
43 * Since it is allowed to sleep, it probably should.
44 */
45 #define BUS_RESET_SETTLE_TIME (10)
46 #define HOST_RESET_SETTLE_TIME (10)
47
48 /* called with shost->host_lock held */
49 void scsi_eh_wakeup(struct Scsi_Host *shost)
50 {
51 if (shost->host_busy == shost->host_failed) {
52 up(shost->eh_wait);
53 SCSI_LOG_ERROR_RECOVERY(5,
54 printk("Waking error handler thread\n"));
55 }
56 }
57
58 /**
59 * scsi_eh_scmd_add - add scsi cmd to error handling.
60 * @scmd: scmd to run eh on.
61 * @eh_flag: optional SCSI_EH flag.
62 *
63 * Return value:
64 * 0 on failure.
65 **/
66 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
67 {
68 struct Scsi_Host *shost = scmd->device->host;
69 unsigned long flags;
70
71 if (shost->eh_wait == NULL)
72 return 0;
73
74 spin_lock_irqsave(shost->host_lock, flags);
75
76 scsi_eh_eflags_set(scmd, eh_flag);
77 /*
78 * FIXME: Can we stop setting owner and state.
79 */
80 scmd->owner = SCSI_OWNER_ERROR_HANDLER;
81 scmd->state = SCSI_STATE_FAILED;
82 /*
83 * Set the serial_number_at_timeout to the current
84 * serial_number
85 */
86 scmd->serial_number_at_timeout = scmd->serial_number;
87 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
88 set_bit(SHOST_RECOVERY, &shost->shost_state);
89 shost->host_failed++;
90 scsi_eh_wakeup(shost);
91 spin_unlock_irqrestore(shost->host_lock, flags);
92 return 1;
93 }
94
95 /**
96 * scsi_add_timer - Start timeout timer for a single scsi command.
97 * @scmd: scsi command that is about to start running.
98 * @timeout: amount of time to allow this command to run.
99 * @complete: timeout function to call if timer isn't canceled.
100 *
101 * Notes:
102 * This should be turned into an inline function. Each scsi command
103 * has its own timer, and as it is added to the queue, we set up the
104 * timer. When the command completes, we cancel the timer.
105 **/
106 void scsi_add_timer(struct scsi_cmnd *scmd, int timeout,
107 void (*complete)(struct scsi_cmnd *))
108 {
109
110 /*
111 * If the clock was already running for this command, then
112 * first delete the timer. The timer handling code gets rather
113 * confused if we don't do this.
114 */
115 if (scmd->eh_timeout.function)
116 del_timer(&scmd->eh_timeout);
117
118 scmd->eh_timeout.data = (unsigned long)scmd;
119 scmd->eh_timeout.expires = jiffies + timeout;
120 scmd->eh_timeout.function = (void (*)(unsigned long)) complete;
121
122 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
123 " %d, (%p)\n", __FUNCTION__,
124 scmd, timeout, complete));
125
126 add_timer(&scmd->eh_timeout);
127 }
128 EXPORT_SYMBOL(scsi_add_timer);
129
130 /**
131 * scsi_delete_timer - Delete/cancel timer for a given function.
132 * @scmd: Cmd that we are canceling timer for
133 *
134 * Notes:
135 * This should be turned into an inline function.
136 *
137 * Return value:
138 * 1 if we were able to detach the timer. 0 if we blew it, and the
139 * timer function has already started to run.
140 **/
141 int scsi_delete_timer(struct scsi_cmnd *scmd)
142 {
143 int rtn;
144
145 rtn = del_timer(&scmd->eh_timeout);
146
147 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
148 " rtn: %d\n", __FUNCTION__,
149 scmd, rtn));
150
151 scmd->eh_timeout.data = (unsigned long)NULL;
152 scmd->eh_timeout.function = NULL;
153
154 return rtn;
155 }
156 EXPORT_SYMBOL(scsi_delete_timer);
157
158 /**
159 * scsi_times_out - Timeout function for normal scsi commands.
160 * @scmd: Cmd that is timing out.
161 *
162 * Notes:
163 * We do not need to lock this. There is the potential for a race
164 * only in that the normal completion handling might run, but if the
165 * normal completion function determines that the timer has already
166 * fired, then it mustn't do anything.
167 **/
168 void scsi_times_out(struct scsi_cmnd *scmd)
169 {
170 scsi_log_completion(scmd, TIMEOUT_ERROR);
171
172 if (scmd->device->host->hostt->eh_timed_out)
173 switch (scmd->device->host->hostt->eh_timed_out(scmd)) {
174 case EH_HANDLED:
175 __scsi_done(scmd);
176 return;
177 case EH_RESET_TIMER:
178 /* This allows a single retry even of a command
179 * with allowed == 0 */
180 if (scmd->retries++ > scmd->allowed)
181 break;
182 scsi_add_timer(scmd, scmd->timeout_per_command,
183 scsi_times_out);
184 return;
185 case EH_NOT_HANDLED:
186 break;
187 }
188
189 if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
190 panic("Error handler thread not present at %p %p %s %d",
191 scmd, scmd->device->host, __FILE__, __LINE__);
192 }
193 }
194
195 /**
196 * scsi_block_when_processing_errors - Prevent cmds from being queued.
197 * @sdev: Device on which we are performing recovery.
198 *
199 * Description:
200 * We block until the host is out of error recovery, and then check to
201 * see whether the host or the device is offline.
202 *
203 * Return value:
204 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
205 **/
206 int scsi_block_when_processing_errors(struct scsi_device *sdev)
207 {
208 int online;
209
210 wait_event(sdev->host->host_wait, (!test_bit(SHOST_RECOVERY, &sdev->host->shost_state)));
211
212 online = scsi_device_online(sdev);
213
214 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__,
215 online));
216
217 return online;
218 }
219 EXPORT_SYMBOL(scsi_block_when_processing_errors);
220
221 #ifdef CONFIG_SCSI_LOGGING
222 /**
223 * scsi_eh_prt_fail_stats - Log info on failures.
224 * @shost: scsi host being recovered.
225 * @work_q: Queue of scsi cmds to process.
226 **/
227 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
228 struct list_head *work_q)
229 {
230 struct scsi_cmnd *scmd;
231 struct scsi_device *sdev;
232 int total_failures = 0;
233 int cmd_failed = 0;
234 int cmd_cancel = 0;
235 int devices_failed = 0;
236
237 shost_for_each_device(sdev, shost) {
238 list_for_each_entry(scmd, work_q, eh_entry) {
239 if (scmd->device == sdev) {
240 ++total_failures;
241 if (scsi_eh_eflags_chk(scmd,
242 SCSI_EH_CANCEL_CMD))
243 ++cmd_cancel;
244 else
245 ++cmd_failed;
246 }
247 }
248
249 if (cmd_cancel || cmd_failed) {
250 SCSI_LOG_ERROR_RECOVERY(3,
251 printk("%s: %d:%d:%d:%d cmds failed: %d,"
252 " cancel: %d\n",
253 __FUNCTION__, shost->host_no,
254 sdev->channel, sdev->id, sdev->lun,
255 cmd_failed, cmd_cancel));
256 cmd_cancel = 0;
257 cmd_failed = 0;
258 ++devices_failed;
259 }
260 }
261
262 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
263 " devices require eh work\n",
264 total_failures, devices_failed));
265 }
266 #endif
267
268 /**
269 * scsi_check_sense - Examine scsi cmd sense
270 * @scmd: Cmd to have sense checked.
271 *
272 * Return value:
273 * SUCCESS or FAILED or NEEDS_RETRY
274 *
275 * Notes:
276 * When a deferred error is detected the current command has
277 * not been executed and needs retrying.
278 **/
279 static int scsi_check_sense(struct scsi_cmnd *scmd)
280 {
281 struct scsi_sense_hdr sshdr;
282
283 if (! scsi_command_normalize_sense(scmd, &sshdr))
284 return FAILED; /* no valid sense data */
285
286 if (scsi_sense_is_deferred(&sshdr))
287 return NEEDS_RETRY;
288
289 /*
290 * Previous logic looked for FILEMARK, EOM or ILI which are
291 * mainly associated with tapes and returned SUCCESS.
292 */
293 if (sshdr.response_code == 0x70) {
294 /* fixed format */
295 if (scmd->sense_buffer[2] & 0xe0)
296 return SUCCESS;
297 } else {
298 /*
299 * descriptor format: look for "stream commands sense data
300 * descriptor" (see SSC-3). Assume single sense data
301 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
302 */
303 if ((sshdr.additional_length > 3) &&
304 (scmd->sense_buffer[8] == 0x4) &&
305 (scmd->sense_buffer[11] & 0xe0))
306 return SUCCESS;
307 }
308
309 switch (sshdr.sense_key) {
310 case NO_SENSE:
311 return SUCCESS;
312 case RECOVERED_ERROR:
313 return /* soft_error */ SUCCESS;
314
315 case ABORTED_COMMAND:
316 return NEEDS_RETRY;
317 case NOT_READY:
318 case UNIT_ATTENTION:
319 /*
320 * if we are expecting a cc/ua because of a bus reset that we
321 * performed, treat this just as a retry. otherwise this is
322 * information that we should pass up to the upper-level driver
323 * so that we can deal with it there.
324 */
325 if (scmd->device->expecting_cc_ua) {
326 scmd->device->expecting_cc_ua = 0;
327 return NEEDS_RETRY;
328 }
329 /*
330 * if the device is in the process of becoming ready, we
331 * should retry.
332 */
333 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
334 return NEEDS_RETRY;
335 /*
336 * if the device is not started, we need to wake
337 * the error handler to start the motor
338 */
339 if (scmd->device->allow_restart &&
340 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
341 return FAILED;
342 return SUCCESS;
343
344 /* these three are not supported */
345 case COPY_ABORTED:
346 case VOLUME_OVERFLOW:
347 case MISCOMPARE:
348 return SUCCESS;
349
350 case MEDIUM_ERROR:
351 return NEEDS_RETRY;
352
353 case ILLEGAL_REQUEST:
354 case BLANK_CHECK:
355 case DATA_PROTECT:
356 case HARDWARE_ERROR:
357 default:
358 return SUCCESS;
359 }
360 }
361
362 /**
363 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
364 * @scmd: SCSI cmd to examine.
365 *
366 * Notes:
367 * This is *only* called when we are examining the status of commands
368 * queued during error recovery. the main difference here is that we
369 * don't allow for the possibility of retries here, and we are a lot
370 * more restrictive about what we consider acceptable.
371 **/
372 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
373 {
374 /*
375 * first check the host byte, to see if there is anything in there
376 * that would indicate what we need to do.
377 */
378 if (host_byte(scmd->result) == DID_RESET) {
379 /*
380 * rats. we are already in the error handler, so we now
381 * get to try and figure out what to do next. if the sense
382 * is valid, we have a pretty good idea of what to do.
383 * if not, we mark it as FAILED.
384 */
385 return scsi_check_sense(scmd);
386 }
387 if (host_byte(scmd->result) != DID_OK)
388 return FAILED;
389
390 /*
391 * next, check the message byte.
392 */
393 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
394 return FAILED;
395
396 /*
397 * now, check the status byte to see if this indicates
398 * anything special.
399 */
400 switch (status_byte(scmd->result)) {
401 case GOOD:
402 case COMMAND_TERMINATED:
403 return SUCCESS;
404 case CHECK_CONDITION:
405 return scsi_check_sense(scmd);
406 case CONDITION_GOOD:
407 case INTERMEDIATE_GOOD:
408 case INTERMEDIATE_C_GOOD:
409 /*
410 * who knows? FIXME(eric)
411 */
412 return SUCCESS;
413 case BUSY:
414 case QUEUE_FULL:
415 case RESERVATION_CONFLICT:
416 default:
417 return FAILED;
418 }
419 return FAILED;
420 }
421
422 /**
423 * scsi_eh_times_out - timeout function for error handling.
424 * @scmd: Cmd that is timing out.
425 *
426 * Notes:
427 * During error handling, the kernel thread will be sleeping waiting
428 * for some action to complete on the device. our only job is to
429 * record that it timed out, and to wake up the thread.
430 **/
431 static void scsi_eh_times_out(struct scsi_cmnd *scmd)
432 {
433 scsi_eh_eflags_set(scmd, SCSI_EH_REC_TIMEOUT);
434 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd:%p\n", __FUNCTION__,
435 scmd));
436
437 if (scmd->device->host->eh_action)
438 up(scmd->device->host->eh_action);
439 }
440
441 /**
442 * scsi_eh_done - Completion function for error handling.
443 * @scmd: Cmd that is done.
444 **/
445 static void scsi_eh_done(struct scsi_cmnd *scmd)
446 {
447 /*
448 * if the timeout handler is already running, then just set the
449 * flag which says we finished late, and return. we have no
450 * way of stopping the timeout handler from running, so we must
451 * always defer to it.
452 */
453 if (del_timer(&scmd->eh_timeout)) {
454 scmd->request->rq_status = RQ_SCSI_DONE;
455 scmd->owner = SCSI_OWNER_ERROR_HANDLER;
456
457 SCSI_LOG_ERROR_RECOVERY(3, printk("%s scmd: %p result: %x\n",
458 __FUNCTION__, scmd, scmd->result));
459
460 if (scmd->device->host->eh_action)
461 up(scmd->device->host->eh_action);
462 }
463 }
464
465 /**
466 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery.
467 * @scmd: SCSI Cmd to send.
468 * @timeout: Timeout for cmd.
469 *
470 * Notes:
471 * The initialization of the structures is quite a bit different in
472 * this case, and furthermore, there is a different completion handler
473 * vs scsi_dispatch_cmd.
474 * Return value:
475 * SUCCESS or FAILED or NEEDS_RETRY
476 **/
477 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout)
478 {
479 struct Scsi_Host *host = scmd->device->host;
480 DECLARE_MUTEX_LOCKED(sem);
481 unsigned long flags;
482 int rtn = SUCCESS;
483
484 /*
485 * we will use a queued command if possible, otherwise we will
486 * emulate the queuing and calling of completion function ourselves.
487 */
488 scmd->owner = SCSI_OWNER_LOWLEVEL;
489
490 if (scmd->device->scsi_level <= SCSI_2)
491 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
492 (scmd->device->lun << 5 & 0xe0);
493
494 scsi_add_timer(scmd, timeout, scsi_eh_times_out);
495
496 /*
497 * set up the semaphore so we wait for the command to complete.
498 */
499 scmd->device->host->eh_action = &sem;
500 scmd->request->rq_status = RQ_SCSI_BUSY;
501
502 spin_lock_irqsave(scmd->device->host->host_lock, flags);
503 scsi_log_send(scmd);
504 host->hostt->queuecommand(scmd, scsi_eh_done);
505 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
506
507 down(&sem);
508 scsi_log_completion(scmd, SUCCESS);
509
510 scmd->device->host->eh_action = NULL;
511
512 /*
513 * see if timeout. if so, tell the host to forget about it.
514 * in other words, we don't want a callback any more.
515 */
516 if (scsi_eh_eflags_chk(scmd, SCSI_EH_REC_TIMEOUT)) {
517 scsi_eh_eflags_clr(scmd, SCSI_EH_REC_TIMEOUT);
518 scmd->owner = SCSI_OWNER_LOWLEVEL;
519
520 /*
521 * as far as the low level driver is
522 * concerned, this command is still active, so
523 * we must give the low level driver a chance
524 * to abort it. (db)
525 *
526 * FIXME(eric) - we are not tracking whether we could
527 * abort a timed out command or not. not sure how
528 * we should treat them differently anyways.
529 */
530 spin_lock_irqsave(scmd->device->host->host_lock, flags);
531 if (scmd->device->host->hostt->eh_abort_handler)
532 scmd->device->host->hostt->eh_abort_handler(scmd);
533 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
534
535 scmd->request->rq_status = RQ_SCSI_DONE;
536 scmd->owner = SCSI_OWNER_ERROR_HANDLER;
537
538 rtn = FAILED;
539 }
540
541 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd: %p, rtn:%x\n",
542 __FUNCTION__, scmd, rtn));
543
544 /*
545 * now examine the actual status codes to see whether the command
546 * actually did complete normally.
547 */
548 if (rtn == SUCCESS) {
549 rtn = scsi_eh_completed_normally(scmd);
550 SCSI_LOG_ERROR_RECOVERY(3,
551 printk("%s: scsi_eh_completed_normally %x\n",
552 __FUNCTION__, rtn));
553 switch (rtn) {
554 case SUCCESS:
555 case NEEDS_RETRY:
556 case FAILED:
557 break;
558 default:
559 rtn = FAILED;
560 break;
561 }
562 }
563
564 return rtn;
565 }
566
567 /**
568 * scsi_request_sense - Request sense data from a particular target.
569 * @scmd: SCSI cmd for request sense.
570 *
571 * Notes:
572 * Some hosts automatically obtain this information, others require
573 * that we obtain it on our own. This function will *not* return until
574 * the command either times out, or it completes.
575 **/
576 static int scsi_request_sense(struct scsi_cmnd *scmd)
577 {
578 static unsigned char generic_sense[6] =
579 {REQUEST_SENSE, 0, 0, 0, 252, 0};
580 unsigned char *scsi_result;
581 int saved_result;
582 int rtn;
583
584 memcpy(scmd->cmnd, generic_sense, sizeof(generic_sense));
585
586 scsi_result = kmalloc(252, GFP_ATOMIC | (scmd->device->host->hostt->unchecked_isa_dma) ? __GFP_DMA : 0);
587
588
589 if (unlikely(!scsi_result)) {
590 printk(KERN_ERR "%s: cannot allocate scsi_result.\n",
591 __FUNCTION__);
592 return FAILED;
593 }
594
595 /*
596 * zero the sense buffer. some host adapters automatically always
597 * request sense, so it is not a good idea that
598 * scmd->request_buffer and scmd->sense_buffer point to the same
599 * address (db). 0 is not a valid sense code.
600 */
601 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
602 memset(scsi_result, 0, 252);
603
604 saved_result = scmd->result;
605 scmd->request_buffer = scsi_result;
606 scmd->request_bufflen = 252;
607 scmd->use_sg = 0;
608 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
609 scmd->sc_data_direction = DMA_FROM_DEVICE;
610 scmd->underflow = 0;
611
612 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT);
613
614 /* last chance to have valid sense data */
615 if(!SCSI_SENSE_VALID(scmd)) {
616 memcpy(scmd->sense_buffer, scmd->request_buffer,
617 sizeof(scmd->sense_buffer));
618 }
619
620 kfree(scsi_result);
621
622 /*
623 * when we eventually call scsi_finish, we really wish to complete
624 * the original request, so let's restore the original data. (db)
625 */
626 scsi_setup_cmd_retry(scmd);
627 scmd->result = saved_result;
628 return rtn;
629 }
630
631 /**
632 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
633 * @scmd: Original SCSI cmd that eh has finished.
634 * @done_q: Queue for processed commands.
635 *
636 * Notes:
637 * We don't want to use the normal command completion while we are are
638 * still handling errors - it may cause other commands to be queued,
639 * and that would disturb what we are doing. thus we really want to
640 * keep a list of pending commands for final completion, and once we
641 * are ready to leave error handling we handle completion for real.
642 **/
643 static void scsi_eh_finish_cmd(struct scsi_cmnd *scmd,
644 struct list_head *done_q)
645 {
646 scmd->device->host->host_failed--;
647 scmd->state = SCSI_STATE_BHQUEUE;
648
649 scsi_eh_eflags_clr_all(scmd);
650
651 /*
652 * set this back so that the upper level can correctly free up
653 * things.
654 */
655 scsi_setup_cmd_retry(scmd);
656 list_move_tail(&scmd->eh_entry, done_q);
657 }
658
659 /**
660 * scsi_eh_get_sense - Get device sense data.
661 * @work_q: Queue of commands to process.
662 * @done_q: Queue of proccessed commands..
663 *
664 * Description:
665 * See if we need to request sense information. if so, then get it
666 * now, so we have a better idea of what to do.
667 *
668 * Notes:
669 * This has the unfortunate side effect that if a shost adapter does
670 * not automatically request sense information, that we end up shutting
671 * it down before we request it.
672 *
673 * All drivers should request sense information internally these days,
674 * so for now all I have to say is tough noogies if you end up in here.
675 *
676 * XXX: Long term this code should go away, but that needs an audit of
677 * all LLDDs first.
678 **/
679 static int scsi_eh_get_sense(struct list_head *work_q,
680 struct list_head *done_q)
681 {
682 struct list_head *lh, *lh_sf;
683 struct scsi_cmnd *scmd;
684 int rtn;
685
686 list_for_each_safe(lh, lh_sf, work_q) {
687 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
688 if (scsi_eh_eflags_chk(scmd, SCSI_EH_CANCEL_CMD) ||
689 SCSI_SENSE_VALID(scmd))
690 continue;
691
692 SCSI_LOG_ERROR_RECOVERY(2, printk("%s: requesting sense"
693 " for id: %d\n",
694 current->comm,
695 scmd->device->id));
696 rtn = scsi_request_sense(scmd);
697 if (rtn != SUCCESS)
698 continue;
699
700 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
701 " result %x\n", scmd,
702 scmd->result));
703 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
704
705 rtn = scsi_decide_disposition(scmd);
706
707 /*
708 * if the result was normal, then just pass it along to the
709 * upper level.
710 */
711 if (rtn == SUCCESS)
712 /* we don't want this command reissued, just
713 * finished with the sense data, so set
714 * retries to the max allowed to ensure it
715 * won't get reissued */
716 scmd->retries = scmd->allowed;
717 else if (rtn != NEEDS_RETRY)
718 continue;
719
720 scsi_eh_finish_cmd(scmd, done_q);
721 }
722
723 return list_empty(work_q);
724 }
725
726 /**
727 * scsi_try_to_abort_cmd - Ask host to abort a running command.
728 * @scmd: SCSI cmd to abort from Lower Level.
729 *
730 * Notes:
731 * This function will not return until the user's completion function
732 * has been called. there is no timeout on this operation. if the
733 * author of the low-level driver wishes this operation to be timed,
734 * they can provide this facility themselves. helper functions in
735 * scsi_error.c can be supplied to make this easier to do.
736 **/
737 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
738 {
739 unsigned long flags;
740 int rtn = FAILED;
741
742 if (!scmd->device->host->hostt->eh_abort_handler)
743 return rtn;
744
745 /*
746 * scsi_done was called just after the command timed out and before
747 * we had a chance to process it. (db)
748 */
749 if (scmd->serial_number == 0)
750 return SUCCESS;
751
752 scmd->owner = SCSI_OWNER_LOWLEVEL;
753
754 spin_lock_irqsave(scmd->device->host->host_lock, flags);
755 rtn = scmd->device->host->hostt->eh_abort_handler(scmd);
756 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
757
758 return rtn;
759 }
760
761 /**
762 * scsi_eh_tur - Send TUR to device.
763 * @scmd: Scsi cmd to send TUR
764 *
765 * Return value:
766 * 0 - Device is ready. 1 - Device NOT ready.
767 **/
768 static int scsi_eh_tur(struct scsi_cmnd *scmd)
769 {
770 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
771 int retry_cnt = 1, rtn;
772
773 retry_tur:
774 memcpy(scmd->cmnd, tur_command, sizeof(tur_command));
775
776 /*
777 * zero the sense buffer. the scsi spec mandates that any
778 * untransferred sense data should be interpreted as being zero.
779 */
780 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
781
782 scmd->request_buffer = NULL;
783 scmd->request_bufflen = 0;
784 scmd->use_sg = 0;
785 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
786 scmd->underflow = 0;
787 scmd->sc_data_direction = DMA_NONE;
788
789 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT);
790
791 /*
792 * when we eventually call scsi_finish, we really wish to complete
793 * the original request, so let's restore the original data. (db)
794 */
795 scsi_setup_cmd_retry(scmd);
796
797 /*
798 * hey, we are done. let's look to see what happened.
799 */
800 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
801 __FUNCTION__, scmd, rtn));
802 if (rtn == SUCCESS)
803 return 0;
804 else if (rtn == NEEDS_RETRY)
805 if (retry_cnt--)
806 goto retry_tur;
807 return 1;
808 }
809
810 /**
811 * scsi_eh_abort_cmds - abort canceled commands.
812 * @shost: scsi host being recovered.
813 * @eh_done_q: list_head for processed commands.
814 *
815 * Decription:
816 * Try and see whether or not it makes sense to try and abort the
817 * running command. this only works out to be the case if we have one
818 * command that has timed out. if the command simply failed, it makes
819 * no sense to try and abort the command, since as far as the shost
820 * adapter is concerned, it isn't running.
821 **/
822 static int scsi_eh_abort_cmds(struct list_head *work_q,
823 struct list_head *done_q)
824 {
825 struct list_head *lh, *lh_sf;
826 struct scsi_cmnd *scmd;
827 int rtn;
828
829 list_for_each_safe(lh, lh_sf, work_q) {
830 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
831 if (!scsi_eh_eflags_chk(scmd, SCSI_EH_CANCEL_CMD))
832 continue;
833 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
834 "0x%p\n", current->comm,
835 scmd));
836 rtn = scsi_try_to_abort_cmd(scmd);
837 if (rtn == SUCCESS) {
838 scsi_eh_eflags_clr(scmd, SCSI_EH_CANCEL_CMD);
839 if (!scsi_device_online(scmd->device) ||
840 !scsi_eh_tur(scmd)) {
841 scsi_eh_finish_cmd(scmd, done_q);
842 }
843
844 } else
845 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
846 " cmd failed:"
847 "0x%p\n",
848 current->comm,
849 scmd));
850 }
851
852 return list_empty(work_q);
853 }
854
855 /**
856 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
857 * @scmd: SCSI cmd used to send BDR
858 *
859 * Notes:
860 * There is no timeout for this operation. if this operation is
861 * unreliable for a given host, then the host itself needs to put a
862 * timer on it, and set the host back to a consistent state prior to
863 * returning.
864 **/
865 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
866 {
867 unsigned long flags;
868 int rtn = FAILED;
869
870 if (!scmd->device->host->hostt->eh_device_reset_handler)
871 return rtn;
872
873 scmd->owner = SCSI_OWNER_LOWLEVEL;
874
875 spin_lock_irqsave(scmd->device->host->host_lock, flags);
876 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
877 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
878
879 if (rtn == SUCCESS) {
880 scmd->device->was_reset = 1;
881 scmd->device->expecting_cc_ua = 1;
882 }
883
884 return rtn;
885 }
886
887 /**
888 * scsi_eh_try_stu - Send START_UNIT to device.
889 * @scmd: Scsi cmd to send START_UNIT
890 *
891 * Return value:
892 * 0 - Device is ready. 1 - Device NOT ready.
893 **/
894 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
895 {
896 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
897 int rtn;
898
899 if (!scmd->device->allow_restart)
900 return 1;
901
902 memcpy(scmd->cmnd, stu_command, sizeof(stu_command));
903
904 /*
905 * zero the sense buffer. the scsi spec mandates that any
906 * untransferred sense data should be interpreted as being zero.
907 */
908 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
909
910 scmd->request_buffer = NULL;
911 scmd->request_bufflen = 0;
912 scmd->use_sg = 0;
913 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
914 scmd->underflow = 0;
915 scmd->sc_data_direction = DMA_NONE;
916
917 rtn = scsi_send_eh_cmnd(scmd, START_UNIT_TIMEOUT);
918
919 /*
920 * when we eventually call scsi_finish, we really wish to complete
921 * the original request, so let's restore the original data. (db)
922 */
923 scsi_setup_cmd_retry(scmd);
924
925 /*
926 * hey, we are done. let's look to see what happened.
927 */
928 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
929 __FUNCTION__, scmd, rtn));
930 if (rtn == SUCCESS)
931 return 0;
932 return 1;
933 }
934
935 /**
936 * scsi_eh_stu - send START_UNIT if needed
937 * @shost: scsi host being recovered.
938 * @eh_done_q: list_head for processed commands.
939 *
940 * Notes:
941 * If commands are failing due to not ready, initializing command required,
942 * try revalidating the device, which will end up sending a start unit.
943 **/
944 static int scsi_eh_stu(struct Scsi_Host *shost,
945 struct list_head *work_q,
946 struct list_head *done_q)
947 {
948 struct list_head *lh, *lh_sf;
949 struct scsi_cmnd *scmd, *stu_scmd;
950 struct scsi_device *sdev;
951
952 shost_for_each_device(sdev, shost) {
953 stu_scmd = NULL;
954 list_for_each_entry(scmd, work_q, eh_entry)
955 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
956 scsi_check_sense(scmd) == FAILED ) {
957 stu_scmd = scmd;
958 break;
959 }
960
961 if (!stu_scmd)
962 continue;
963
964 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
965 " 0x%p\n", current->comm, sdev));
966
967 if (!scsi_eh_try_stu(stu_scmd)) {
968 if (!scsi_device_online(sdev) ||
969 !scsi_eh_tur(stu_scmd)) {
970 list_for_each_safe(lh, lh_sf, work_q) {
971 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
972 if (scmd->device == sdev)
973 scsi_eh_finish_cmd(scmd, done_q);
974 }
975 }
976 } else {
977 SCSI_LOG_ERROR_RECOVERY(3,
978 printk("%s: START_UNIT failed to sdev:"
979 " 0x%p\n", current->comm, sdev));
980 }
981 }
982
983 return list_empty(work_q);
984 }
985
986
987 /**
988 * scsi_eh_bus_device_reset - send bdr if needed
989 * @shost: scsi host being recovered.
990 * @eh_done_q: list_head for processed commands.
991 *
992 * Notes:
993 * Try a bus device reset. still, look to see whether we have multiple
994 * devices that are jammed or not - if we have multiple devices, it
995 * makes no sense to try bus_device_reset - we really would need to try
996 * a bus_reset instead.
997 **/
998 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
999 struct list_head *work_q,
1000 struct list_head *done_q)
1001 {
1002 struct list_head *lh, *lh_sf;
1003 struct scsi_cmnd *scmd, *bdr_scmd;
1004 struct scsi_device *sdev;
1005 int rtn;
1006
1007 shost_for_each_device(sdev, shost) {
1008 bdr_scmd = NULL;
1009 list_for_each_entry(scmd, work_q, eh_entry)
1010 if (scmd->device == sdev) {
1011 bdr_scmd = scmd;
1012 break;
1013 }
1014
1015 if (!bdr_scmd)
1016 continue;
1017
1018 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1019 " 0x%p\n", current->comm,
1020 sdev));
1021 rtn = scsi_try_bus_device_reset(bdr_scmd);
1022 if (rtn == SUCCESS) {
1023 if (!scsi_device_online(sdev) ||
1024 !scsi_eh_tur(bdr_scmd)) {
1025 list_for_each_safe(lh, lh_sf,
1026 work_q) {
1027 scmd = list_entry(lh, struct
1028 scsi_cmnd,
1029 eh_entry);
1030 if (scmd->device == sdev)
1031 scsi_eh_finish_cmd(scmd,
1032 done_q);
1033 }
1034 }
1035 } else {
1036 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1037 " failed sdev:"
1038 "0x%p\n",
1039 current->comm,
1040 sdev));
1041 }
1042 }
1043
1044 return list_empty(work_q);
1045 }
1046
1047 /**
1048 * scsi_try_bus_reset - ask host to perform a bus reset
1049 * @scmd: SCSI cmd to send bus reset.
1050 **/
1051 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
1052 {
1053 unsigned long flags;
1054 int rtn;
1055
1056 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
1057 __FUNCTION__));
1058 scmd->owner = SCSI_OWNER_LOWLEVEL;
1059 scmd->serial_number_at_timeout = scmd->serial_number;
1060
1061 if (!scmd->device->host->hostt->eh_bus_reset_handler)
1062 return FAILED;
1063
1064 spin_lock_irqsave(scmd->device->host->host_lock, flags);
1065 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
1066 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
1067
1068 if (rtn == SUCCESS) {
1069 if (!scmd->device->host->hostt->skip_settle_delay)
1070 ssleep(BUS_RESET_SETTLE_TIME);
1071 spin_lock_irqsave(scmd->device->host->host_lock, flags);
1072 scsi_report_bus_reset(scmd->device->host, scmd->device->channel);
1073 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
1074 }
1075
1076 return rtn;
1077 }
1078
1079 /**
1080 * scsi_try_host_reset - ask host adapter to reset itself
1081 * @scmd: SCSI cmd to send hsot reset.
1082 **/
1083 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
1084 {
1085 unsigned long flags;
1086 int rtn;
1087
1088 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
1089 __FUNCTION__));
1090 scmd->owner = SCSI_OWNER_LOWLEVEL;
1091 scmd->serial_number_at_timeout = scmd->serial_number;
1092
1093 if (!scmd->device->host->hostt->eh_host_reset_handler)
1094 return FAILED;
1095
1096 spin_lock_irqsave(scmd->device->host->host_lock, flags);
1097 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
1098 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
1099
1100 if (rtn == SUCCESS) {
1101 if (!scmd->device->host->hostt->skip_settle_delay)
1102 ssleep(HOST_RESET_SETTLE_TIME);
1103 spin_lock_irqsave(scmd->device->host->host_lock, flags);
1104 scsi_report_bus_reset(scmd->device->host, scmd->device->channel);
1105 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
1106 }
1107
1108 return rtn;
1109 }
1110
1111 /**
1112 * scsi_eh_bus_reset - send a bus reset
1113 * @shost: scsi host being recovered.
1114 * @eh_done_q: list_head for processed commands.
1115 **/
1116 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1117 struct list_head *work_q,
1118 struct list_head *done_q)
1119 {
1120 struct list_head *lh, *lh_sf;
1121 struct scsi_cmnd *scmd;
1122 struct scsi_cmnd *chan_scmd;
1123 unsigned int channel;
1124 int rtn;
1125
1126 /*
1127 * we really want to loop over the various channels, and do this on
1128 * a channel by channel basis. we should also check to see if any
1129 * of the failed commands are on soft_reset devices, and if so, skip
1130 * the reset.
1131 */
1132
1133 for (channel = 0; channel <= shost->max_channel; channel++) {
1134 chan_scmd = NULL;
1135 list_for_each_entry(scmd, work_q, eh_entry) {
1136 if (channel == scmd->device->channel) {
1137 chan_scmd = scmd;
1138 break;
1139 /*
1140 * FIXME add back in some support for
1141 * soft_reset devices.
1142 */
1143 }
1144 }
1145
1146 if (!chan_scmd)
1147 continue;
1148 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1149 " %d\n", current->comm,
1150 channel));
1151 rtn = scsi_try_bus_reset(chan_scmd);
1152 if (rtn == SUCCESS) {
1153 list_for_each_safe(lh, lh_sf, work_q) {
1154 scmd = list_entry(lh, struct scsi_cmnd,
1155 eh_entry);
1156 if (channel == scmd->device->channel)
1157 if (!scsi_device_online(scmd->device) ||
1158 !scsi_eh_tur(scmd))
1159 scsi_eh_finish_cmd(scmd,
1160 done_q);
1161 }
1162 } else {
1163 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1164 " failed chan: %d\n",
1165 current->comm,
1166 channel));
1167 }
1168 }
1169 return list_empty(work_q);
1170 }
1171
1172 /**
1173 * scsi_eh_host_reset - send a host reset
1174 * @work_q: list_head for processed commands.
1175 * @done_q: list_head for processed commands.
1176 **/
1177 static int scsi_eh_host_reset(struct list_head *work_q,
1178 struct list_head *done_q)
1179 {
1180 int rtn;
1181 struct list_head *lh, *lh_sf;
1182 struct scsi_cmnd *scmd;
1183
1184 if (!list_empty(work_q)) {
1185 scmd = list_entry(work_q->next,
1186 struct scsi_cmnd, eh_entry);
1187
1188 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1189 , current->comm));
1190
1191 rtn = scsi_try_host_reset(scmd);
1192 if (rtn == SUCCESS) {
1193 list_for_each_safe(lh, lh_sf, work_q) {
1194 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
1195 if (!scsi_device_online(scmd->device) ||
1196 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1197 !scsi_eh_tur(scmd))
1198 scsi_eh_finish_cmd(scmd, done_q);
1199 }
1200 } else {
1201 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1202 " failed\n",
1203 current->comm));
1204 }
1205 }
1206 return list_empty(work_q);
1207 }
1208
1209 /**
1210 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1211 * @work_q: list_head for processed commands.
1212 * @done_q: list_head for processed commands.
1213 *
1214 **/
1215 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1216 struct list_head *done_q)
1217 {
1218 struct list_head *lh, *lh_sf;
1219 struct scsi_cmnd *scmd;
1220
1221 list_for_each_safe(lh, lh_sf, work_q) {
1222 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
1223 printk(KERN_INFO "scsi: Device offlined - not"
1224 " ready after error recovery: host"
1225 " %d channel %d id %d lun %d\n",
1226 scmd->device->host->host_no,
1227 scmd->device->channel,
1228 scmd->device->id,
1229 scmd->device->lun);
1230 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1231 if (scsi_eh_eflags_chk(scmd, SCSI_EH_CANCEL_CMD)) {
1232 /*
1233 * FIXME: Handle lost cmds.
1234 */
1235 }
1236 scsi_eh_finish_cmd(scmd, done_q);
1237 }
1238 return;
1239 }
1240
1241 /**
1242 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1243 * @scmd: SCSI cmd to examine.
1244 *
1245 * Notes:
1246 * This is *only* called when we are examining the status after sending
1247 * out the actual data command. any commands that are queued for error
1248 * recovery (e.g. test_unit_ready) do *not* come through here.
1249 *
1250 * When this routine returns failed, it means the error handler thread
1251 * is woken. In cases where the error code indicates an error that
1252 * doesn't require the error handler read (i.e. we don't need to
1253 * abort/reset), this function should return SUCCESS.
1254 **/
1255 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1256 {
1257 int rtn;
1258
1259 /*
1260 * if the device is offline, then we clearly just pass the result back
1261 * up to the top level.
1262 */
1263 if (!scsi_device_online(scmd->device)) {
1264 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1265 " as SUCCESS\n",
1266 __FUNCTION__));
1267 return SUCCESS;
1268 }
1269
1270 /*
1271 * first check the host byte, to see if there is anything in there
1272 * that would indicate what we need to do.
1273 */
1274 switch (host_byte(scmd->result)) {
1275 case DID_PASSTHROUGH:
1276 /*
1277 * no matter what, pass this through to the upper layer.
1278 * nuke this special code so that it looks like we are saying
1279 * did_ok.
1280 */
1281 scmd->result &= 0xff00ffff;
1282 return SUCCESS;
1283 case DID_OK:
1284 /*
1285 * looks good. drop through, and check the next byte.
1286 */
1287 break;
1288 case DID_NO_CONNECT:
1289 case DID_BAD_TARGET:
1290 case DID_ABORT:
1291 /*
1292 * note - this means that we just report the status back
1293 * to the top level driver, not that we actually think
1294 * that it indicates SUCCESS.
1295 */
1296 return SUCCESS;
1297 /*
1298 * when the low level driver returns did_soft_error,
1299 * it is responsible for keeping an internal retry counter
1300 * in order to avoid endless loops (db)
1301 *
1302 * actually this is a bug in this function here. we should
1303 * be mindful of the maximum number of retries specified
1304 * and not get stuck in a loop.
1305 */
1306 case DID_SOFT_ERROR:
1307 goto maybe_retry;
1308 case DID_IMM_RETRY:
1309 return NEEDS_RETRY;
1310
1311 case DID_ERROR:
1312 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1313 status_byte(scmd->result) == RESERVATION_CONFLICT)
1314 /*
1315 * execute reservation conflict processing code
1316 * lower down
1317 */
1318 break;
1319 /* fallthrough */
1320
1321 case DID_BUS_BUSY:
1322 case DID_PARITY:
1323 goto maybe_retry;
1324 case DID_TIME_OUT:
1325 /*
1326 * when we scan the bus, we get timeout messages for
1327 * these commands if there is no device available.
1328 * other hosts report did_no_connect for the same thing.
1329 */
1330 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1331 scmd->cmnd[0] == INQUIRY)) {
1332 return SUCCESS;
1333 } else {
1334 return FAILED;
1335 }
1336 case DID_RESET:
1337 return SUCCESS;
1338 default:
1339 return FAILED;
1340 }
1341
1342 /*
1343 * next, check the message byte.
1344 */
1345 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1346 return FAILED;
1347
1348 /*
1349 * check the status byte to see if this indicates anything special.
1350 */
1351 switch (status_byte(scmd->result)) {
1352 case QUEUE_FULL:
1353 /*
1354 * the case of trying to send too many commands to a
1355 * tagged queueing device.
1356 */
1357 case BUSY:
1358 /*
1359 * device can't talk to us at the moment. Should only
1360 * occur (SAM-3) when the task queue is empty, so will cause
1361 * the empty queue handling to trigger a stall in the
1362 * device.
1363 */
1364 return ADD_TO_MLQUEUE;
1365 case GOOD:
1366 case COMMAND_TERMINATED:
1367 return SUCCESS;
1368 case CHECK_CONDITION:
1369 rtn = scsi_check_sense(scmd);
1370 if (rtn == NEEDS_RETRY)
1371 goto maybe_retry;
1372 /* if rtn == FAILED, we have no sense information;
1373 * returning FAILED will wake the error handler thread
1374 * to collect the sense and redo the decide
1375 * disposition */
1376 return rtn;
1377 case CONDITION_GOOD:
1378 case INTERMEDIATE_GOOD:
1379 case INTERMEDIATE_C_GOOD:
1380 /*
1381 * who knows? FIXME(eric)
1382 */
1383 return SUCCESS;
1384
1385 case RESERVATION_CONFLICT:
1386 printk(KERN_INFO "scsi: reservation conflict: host"
1387 " %d channel %d id %d lun %d\n",
1388 scmd->device->host->host_no, scmd->device->channel,
1389 scmd->device->id, scmd->device->lun);
1390 return SUCCESS; /* causes immediate i/o error */
1391 default:
1392 return FAILED;
1393 }
1394 return FAILED;
1395
1396 maybe_retry:
1397
1398 /* we requeue for retry because the error was retryable, and
1399 * the request was not marked fast fail. Note that above,
1400 * even if the request is marked fast fail, we still requeue
1401 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1402 if ((++scmd->retries) < scmd->allowed
1403 && !blk_noretry_request(scmd->request)) {
1404 return NEEDS_RETRY;
1405 } else {
1406 /*
1407 * no more retries - report this one back to upper level.
1408 */
1409 return SUCCESS;
1410 }
1411 }
1412
1413 /**
1414 * scsi_eh_lock_done - done function for eh door lock request
1415 * @scmd: SCSI command block for the door lock request
1416 *
1417 * Notes:
1418 * We completed the asynchronous door lock request, and it has either
1419 * locked the door or failed. We must free the command structures
1420 * associated with this request.
1421 **/
1422 static void scsi_eh_lock_done(struct scsi_cmnd *scmd)
1423 {
1424 struct scsi_request *sreq = scmd->sc_request;
1425
1426 scsi_release_request(sreq);
1427 }
1428
1429
1430 /**
1431 * scsi_eh_lock_door - Prevent medium removal for the specified device
1432 * @sdev: SCSI device to prevent medium removal
1433 *
1434 * Locking:
1435 * We must be called from process context; scsi_allocate_request()
1436 * may sleep.
1437 *
1438 * Notes:
1439 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1440 * head of the devices request queue, and continue.
1441 *
1442 * Bugs:
1443 * scsi_allocate_request() may sleep waiting for existing requests to
1444 * be processed. However, since we haven't kicked off any request
1445 * processing for this host, this may deadlock.
1446 *
1447 * If scsi_allocate_request() fails for what ever reason, we
1448 * completely forget to lock the door.
1449 **/
1450 static void scsi_eh_lock_door(struct scsi_device *sdev)
1451 {
1452 struct scsi_request *sreq = scsi_allocate_request(sdev, GFP_KERNEL);
1453
1454 if (unlikely(!sreq)) {
1455 printk(KERN_ERR "%s: request allocate failed,"
1456 "prevent media removal cmd not sent\n", __FUNCTION__);
1457 return;
1458 }
1459
1460 sreq->sr_cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1461 sreq->sr_cmnd[1] = 0;
1462 sreq->sr_cmnd[2] = 0;
1463 sreq->sr_cmnd[3] = 0;
1464 sreq->sr_cmnd[4] = SCSI_REMOVAL_PREVENT;
1465 sreq->sr_cmnd[5] = 0;
1466 sreq->sr_data_direction = DMA_NONE;
1467 sreq->sr_bufflen = 0;
1468 sreq->sr_buffer = NULL;
1469 sreq->sr_allowed = 5;
1470 sreq->sr_done = scsi_eh_lock_done;
1471 sreq->sr_timeout_per_command = 10 * HZ;
1472 sreq->sr_cmd_len = COMMAND_SIZE(sreq->sr_cmnd[0]);
1473
1474 scsi_insert_special_req(sreq, 1);
1475 }
1476
1477
1478 /**
1479 * scsi_restart_operations - restart io operations to the specified host.
1480 * @shost: Host we are restarting.
1481 *
1482 * Notes:
1483 * When we entered the error handler, we blocked all further i/o to
1484 * this device. we need to 'reverse' this process.
1485 **/
1486 static void scsi_restart_operations(struct Scsi_Host *shost)
1487 {
1488 struct scsi_device *sdev;
1489
1490 /*
1491 * If the door was locked, we need to insert a door lock request
1492 * onto the head of the SCSI request queue for the device. There
1493 * is no point trying to lock the door of an off-line device.
1494 */
1495 shost_for_each_device(sdev, shost) {
1496 if (scsi_device_online(sdev) && sdev->locked)
1497 scsi_eh_lock_door(sdev);
1498 }
1499
1500 /*
1501 * next free up anything directly waiting upon the host. this
1502 * will be requests for character device operations, and also for
1503 * ioctls to queued block devices.
1504 */
1505 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1506 __FUNCTION__));
1507
1508 clear_bit(SHOST_RECOVERY, &shost->shost_state);
1509
1510 wake_up(&shost->host_wait);
1511
1512 /*
1513 * finally we need to re-initiate requests that may be pending. we will
1514 * have had everything blocked while error handling is taking place, and
1515 * now that error recovery is done, we will need to ensure that these
1516 * requests are started.
1517 */
1518 scsi_run_host_queues(shost);
1519 }
1520
1521 /**
1522 * scsi_eh_ready_devs - check device ready state and recover if not.
1523 * @shost: host to be recovered.
1524 * @eh_done_q: list_head for processed commands.
1525 *
1526 **/
1527 static void scsi_eh_ready_devs(struct Scsi_Host *shost,
1528 struct list_head *work_q,
1529 struct list_head *done_q)
1530 {
1531 if (!scsi_eh_stu(shost, work_q, done_q))
1532 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1533 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1534 if (!scsi_eh_host_reset(work_q, done_q))
1535 scsi_eh_offline_sdevs(work_q, done_q);
1536 }
1537
1538 /**
1539 * scsi_eh_flush_done_q - finish processed commands or retry them.
1540 * @done_q: list_head of processed commands.
1541 *
1542 **/
1543 static void scsi_eh_flush_done_q(struct list_head *done_q)
1544 {
1545 struct list_head *lh, *lh_sf;
1546 struct scsi_cmnd *scmd;
1547
1548 list_for_each_safe(lh, lh_sf, done_q) {
1549 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
1550 list_del_init(lh);
1551 if (scsi_device_online(scmd->device) &&
1552 !blk_noretry_request(scmd->request) &&
1553 (++scmd->retries < scmd->allowed)) {
1554 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1555 " retry cmd: %p\n",
1556 current->comm,
1557 scmd));
1558 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1559 } else {
1560 if (!scmd->result)
1561 scmd->result |= (DRIVER_TIMEOUT << 24);
1562 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1563 " cmd: %p\n",
1564 current->comm, scmd));
1565 scsi_finish_command(scmd);
1566 }
1567 }
1568 }
1569
1570 /**
1571 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1572 * @shost: Host to unjam.
1573 *
1574 * Notes:
1575 * When we come in here, we *know* that all commands on the bus have
1576 * either completed, failed or timed out. we also know that no further
1577 * commands are being sent to the host, so things are relatively quiet
1578 * and we have freedom to fiddle with things as we wish.
1579 *
1580 * This is only the *default* implementation. it is possible for
1581 * individual drivers to supply their own version of this function, and
1582 * if the maintainer wishes to do this, it is strongly suggested that
1583 * this function be taken as a template and modified. this function
1584 * was designed to correctly handle problems for about 95% of the
1585 * different cases out there, and it should always provide at least a
1586 * reasonable amount of error recovery.
1587 *
1588 * Any command marked 'failed' or 'timeout' must eventually have
1589 * scsi_finish_cmd() called for it. we do all of the retry stuff
1590 * here, so when we restart the host after we return it should have an
1591 * empty queue.
1592 **/
1593 static void scsi_unjam_host(struct Scsi_Host *shost)
1594 {
1595 unsigned long flags;
1596 LIST_HEAD(eh_work_q);
1597 LIST_HEAD(eh_done_q);
1598
1599 spin_lock_irqsave(shost->host_lock, flags);
1600 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1601 spin_unlock_irqrestore(shost->host_lock, flags);
1602
1603 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1604
1605 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1606 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1607 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1608
1609 scsi_eh_flush_done_q(&eh_done_q);
1610 }
1611
1612 /**
1613 * scsi_error_handler - Handle errors/timeouts of SCSI cmds.
1614 * @data: Host for which we are running.
1615 *
1616 * Notes:
1617 * This is always run in the context of a kernel thread. The idea is
1618 * that we start this thing up when the kernel starts up (one per host
1619 * that we detect), and it immediately goes to sleep and waits for some
1620 * event (i.e. failure). When this takes place, we have the job of
1621 * trying to unjam the bus and restarting things.
1622 **/
1623 int scsi_error_handler(void *data)
1624 {
1625 struct Scsi_Host *shost = (struct Scsi_Host *) data;
1626 int rtn;
1627 DECLARE_MUTEX_LOCKED(sem);
1628
1629 /*
1630 * Flush resources
1631 */
1632
1633 daemonize("scsi_eh_%d", shost->host_no);
1634
1635 current->flags |= PF_NOFREEZE;
1636
1637 shost->eh_wait = &sem;
1638 shost->ehandler = current;
1639
1640 /*
1641 * Wake up the thread that created us.
1642 */
1643 SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent of"
1644 " scsi_eh_%d\n",shost->host_no));
1645
1646 complete(shost->eh_notify);
1647
1648 while (1) {
1649 /*
1650 * If we get a signal, it means we are supposed to go
1651 * away and die. This typically happens if the user is
1652 * trying to unload a module.
1653 */
1654 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1655 " scsi_eh_%d"
1656 " sleeping\n",shost->host_no));
1657
1658 /*
1659 * Note - we always use down_interruptible with the semaphore
1660 * even if the module was loaded as part of the kernel. The
1661 * reason is that down() will cause this thread to be counted
1662 * in the load average as a running process, and down
1663 * interruptible doesn't. Given that we need to allow this
1664 * thread to die if the driver was loaded as a module, using
1665 * semaphores isn't unreasonable.
1666 */
1667 down_interruptible(&sem);
1668 if (shost->eh_kill)
1669 break;
1670
1671 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1672 " scsi_eh_%d waking"
1673 " up\n",shost->host_no));
1674
1675 shost->eh_active = 1;
1676
1677 /*
1678 * We have a host that is failing for some reason. Figure out
1679 * what we need to do to get it up and online again (if we can).
1680 * If we fail, we end up taking the thing offline.
1681 */
1682 if (shost->hostt->eh_strategy_handler)
1683 rtn = shost->hostt->eh_strategy_handler(shost);
1684 else
1685 scsi_unjam_host(shost);
1686
1687 shost->eh_active = 0;
1688
1689 /*
1690 * Note - if the above fails completely, the action is to take
1691 * individual devices offline and flush the queue of any
1692 * outstanding requests that may have been pending. When we
1693 * restart, we restart any I/O to any other devices on the bus
1694 * which are still online.
1695 */
1696 scsi_restart_operations(shost);
1697
1698 }
1699
1700 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler scsi_eh_%d"
1701 " exiting\n",shost->host_no));
1702
1703 /*
1704 * Make sure that nobody tries to wake us up again.
1705 */
1706 shost->eh_wait = NULL;
1707
1708 /*
1709 * Knock this down too. From this point on, the host is flying
1710 * without a pilot. If this is because the module is being unloaded,
1711 * that's fine. If the user sent a signal to this thing, we are
1712 * potentially in real danger.
1713 */
1714 shost->eh_active = 0;
1715 shost->ehandler = NULL;
1716
1717 /*
1718 * If anyone is waiting for us to exit (i.e. someone trying to unload
1719 * a driver), then wake up that process to let them know we are on
1720 * the way out the door.
1721 */
1722 complete_and_exit(shost->eh_notify, 0);
1723 return 0;
1724 }
1725
1726 /*
1727 * Function: scsi_report_bus_reset()
1728 *
1729 * Purpose: Utility function used by low-level drivers to report that
1730 * they have observed a bus reset on the bus being handled.
1731 *
1732 * Arguments: shost - Host in question
1733 * channel - channel on which reset was observed.
1734 *
1735 * Returns: Nothing
1736 *
1737 * Lock status: Host lock must be held.
1738 *
1739 * Notes: This only needs to be called if the reset is one which
1740 * originates from an unknown location. Resets originated
1741 * by the mid-level itself don't need to call this, but there
1742 * should be no harm.
1743 *
1744 * The main purpose of this is to make sure that a CHECK_CONDITION
1745 * is properly treated.
1746 */
1747 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1748 {
1749 struct scsi_device *sdev;
1750
1751 __shost_for_each_device(sdev, shost) {
1752 if (channel == sdev->channel) {
1753 sdev->was_reset = 1;
1754 sdev->expecting_cc_ua = 1;
1755 }
1756 }
1757 }
1758 EXPORT_SYMBOL(scsi_report_bus_reset);
1759
1760 /*
1761 * Function: scsi_report_device_reset()
1762 *
1763 * Purpose: Utility function used by low-level drivers to report that
1764 * they have observed a device reset on the device being handled.
1765 *
1766 * Arguments: shost - Host in question
1767 * channel - channel on which reset was observed
1768 * target - target on which reset was observed
1769 *
1770 * Returns: Nothing
1771 *
1772 * Lock status: Host lock must be held
1773 *
1774 * Notes: This only needs to be called if the reset is one which
1775 * originates from an unknown location. Resets originated
1776 * by the mid-level itself don't need to call this, but there
1777 * should be no harm.
1778 *
1779 * The main purpose of this is to make sure that a CHECK_CONDITION
1780 * is properly treated.
1781 */
1782 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1783 {
1784 struct scsi_device *sdev;
1785
1786 __shost_for_each_device(sdev, shost) {
1787 if (channel == sdev->channel &&
1788 target == sdev->id) {
1789 sdev->was_reset = 1;
1790 sdev->expecting_cc_ua = 1;
1791 }
1792 }
1793 }
1794 EXPORT_SYMBOL(scsi_report_device_reset);
1795
1796 static void
1797 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1798 {
1799 }
1800
1801 /*
1802 * Function: scsi_reset_provider
1803 *
1804 * Purpose: Send requested reset to a bus or device at any phase.
1805 *
1806 * Arguments: device - device to send reset to
1807 * flag - reset type (see scsi.h)
1808 *
1809 * Returns: SUCCESS/FAILURE.
1810 *
1811 * Notes: This is used by the SCSI Generic driver to provide
1812 * Bus/Device reset capability.
1813 */
1814 int
1815 scsi_reset_provider(struct scsi_device *dev, int flag)
1816 {
1817 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1818 struct request req;
1819 int rtn;
1820
1821 scmd->request = &req;
1822 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
1823 scmd->request->rq_status = RQ_SCSI_BUSY;
1824 scmd->state = SCSI_STATE_INITIALIZING;
1825 scmd->owner = SCSI_OWNER_MIDLEVEL;
1826
1827 memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd));
1828
1829 scmd->scsi_done = scsi_reset_provider_done_command;
1830 scmd->done = NULL;
1831 scmd->buffer = NULL;
1832 scmd->bufflen = 0;
1833 scmd->request_buffer = NULL;
1834 scmd->request_bufflen = 0;
1835 scmd->internal_timeout = NORMAL_TIMEOUT;
1836 scmd->abort_reason = DID_ABORT;
1837
1838 scmd->cmd_len = 0;
1839
1840 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1841 scmd->sc_request = NULL;
1842 scmd->sc_magic = SCSI_CMND_MAGIC;
1843
1844 init_timer(&scmd->eh_timeout);
1845
1846 /*
1847 * Sometimes the command can get back into the timer chain,
1848 * so use the pid as an identifier.
1849 */
1850 scmd->pid = 0;
1851
1852 switch (flag) {
1853 case SCSI_TRY_RESET_DEVICE:
1854 rtn = scsi_try_bus_device_reset(scmd);
1855 if (rtn == SUCCESS)
1856 break;
1857 /* FALLTHROUGH */
1858 case SCSI_TRY_RESET_BUS:
1859 rtn = scsi_try_bus_reset(scmd);
1860 if (rtn == SUCCESS)
1861 break;
1862 /* FALLTHROUGH */
1863 case SCSI_TRY_RESET_HOST:
1864 rtn = scsi_try_host_reset(scmd);
1865 break;
1866 default:
1867 rtn = FAILED;
1868 }
1869
1870 scsi_delete_timer(scmd);
1871 scsi_next_command(scmd);
1872 return rtn;
1873 }
1874 EXPORT_SYMBOL(scsi_reset_provider);
1875
1876 /**
1877 * scsi_normalize_sense - normalize main elements from either fixed or
1878 * descriptor sense data format into a common format.
1879 *
1880 * @sense_buffer: byte array containing sense data returned by device
1881 * @sb_len: number of valid bytes in sense_buffer
1882 * @sshdr: pointer to instance of structure that common
1883 * elements are written to.
1884 *
1885 * Notes:
1886 * The "main elements" from sense data are: response_code, sense_key,
1887 * asc, ascq and additional_length (only for descriptor format).
1888 *
1889 * Typically this function can be called after a device has
1890 * responded to a SCSI command with the CHECK_CONDITION status.
1891 *
1892 * Return value:
1893 * 1 if valid sense data information found, else 0;
1894 **/
1895 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1896 struct scsi_sense_hdr *sshdr)
1897 {
1898 if (!sense_buffer || !sb_len || (sense_buffer[0] & 0x70) != 0x70)
1899 return 0;
1900
1901 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1902
1903 sshdr->response_code = (sense_buffer[0] & 0x7f);
1904 if (sshdr->response_code >= 0x72) {
1905 /*
1906 * descriptor format
1907 */
1908 if (sb_len > 1)
1909 sshdr->sense_key = (sense_buffer[1] & 0xf);
1910 if (sb_len > 2)
1911 sshdr->asc = sense_buffer[2];
1912 if (sb_len > 3)
1913 sshdr->ascq = sense_buffer[3];
1914 if (sb_len > 7)
1915 sshdr->additional_length = sense_buffer[7];
1916 } else {
1917 /*
1918 * fixed format
1919 */
1920 if (sb_len > 2)
1921 sshdr->sense_key = (sense_buffer[2] & 0xf);
1922 if (sb_len > 7) {
1923 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1924 sb_len : (sense_buffer[7] + 8);
1925 if (sb_len > 12)
1926 sshdr->asc = sense_buffer[12];
1927 if (sb_len > 13)
1928 sshdr->ascq = sense_buffer[13];
1929 }
1930 }
1931
1932 return 1;
1933 }
1934 EXPORT_SYMBOL(scsi_normalize_sense);
1935
1936 int scsi_request_normalize_sense(struct scsi_request *sreq,
1937 struct scsi_sense_hdr *sshdr)
1938 {
1939 return scsi_normalize_sense(sreq->sr_sense_buffer,
1940 sizeof(sreq->sr_sense_buffer), sshdr);
1941 }
1942 EXPORT_SYMBOL(scsi_request_normalize_sense);
1943
1944 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1945 struct scsi_sense_hdr *sshdr)
1946 {
1947 return scsi_normalize_sense(cmd->sense_buffer,
1948 sizeof(cmd->sense_buffer), sshdr);
1949 }
1950 EXPORT_SYMBOL(scsi_command_normalize_sense);
1951
1952 /**
1953 * scsi_sense_desc_find - search for a given descriptor type in
1954 * descriptor sense data format.
1955 *
1956 * @sense_buffer: byte array of descriptor format sense data
1957 * @sb_len: number of valid bytes in sense_buffer
1958 * @desc_type: value of descriptor type to find
1959 * (e.g. 0 -> information)
1960 *
1961 * Notes:
1962 * only valid when sense data is in descriptor format
1963 *
1964 * Return value:
1965 * pointer to start of (first) descriptor if found else NULL
1966 **/
1967 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1968 int desc_type)
1969 {
1970 int add_sen_len, add_len, desc_len, k;
1971 const u8 * descp;
1972
1973 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1974 return NULL;
1975 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1976 return NULL;
1977 add_sen_len = (add_sen_len < (sb_len - 8)) ?
1978 add_sen_len : (sb_len - 8);
1979 descp = &sense_buffer[8];
1980 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1981 descp += desc_len;
1982 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1983 desc_len = add_len + 2;
1984 if (descp[0] == desc_type)
1985 return descp;
1986 if (add_len < 0) // short descriptor ??
1987 break;
1988 }
1989 return NULL;
1990 }
1991 EXPORT_SYMBOL(scsi_sense_desc_find);
1992
1993 /**
1994 * scsi_get_sense_info_fld - attempts to get information field from
1995 * sense data (either fixed or descriptor format)
1996 *
1997 * @sense_buffer: byte array of sense data
1998 * @sb_len: number of valid bytes in sense_buffer
1999 * @info_out: pointer to 64 integer where 8 or 4 byte information
2000 * field will be placed if found.
2001 *
2002 * Return value:
2003 * 1 if information field found, 0 if not found.
2004 **/
2005 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2006 u64 * info_out)
2007 {
2008 int j;
2009 const u8 * ucp;
2010 u64 ull;
2011
2012 if (sb_len < 7)
2013 return 0;
2014 switch (sense_buffer[0] & 0x7f) {
2015 case 0x70:
2016 case 0x71:
2017 if (sense_buffer[0] & 0x80) {
2018 *info_out = (sense_buffer[3] << 24) +
2019 (sense_buffer[4] << 16) +
2020 (sense_buffer[5] << 8) + sense_buffer[6];
2021 return 1;
2022 } else
2023 return 0;
2024 case 0x72:
2025 case 0x73:
2026 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2027 0 /* info desc */);
2028 if (ucp && (0xa == ucp[1])) {
2029 ull = 0;
2030 for (j = 0; j < 8; ++j) {
2031 if (j > 0)
2032 ull <<= 8;
2033 ull |= ucp[4 + j];
2034 }
2035 *info_out = ull;
2036 return 1;
2037 } else
2038 return 0;
2039 default:
2040 return 0;
2041 }
2042 }
2043 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2044
|
This page was automatically generated by the
LXR engine.
|