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/kernel.h>
22 #include <linux/freezer.h>
23 #include <linux/kthread.h>
24 #include <linux/interrupt.h>
25 #include <linux/blkdev.h>
26 #include <linux/delay.h>
27
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_transport.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_ioctl.h>
36
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
39 #include "scsi_transport_api.h"
40
41 #define SENSE_TIMEOUT (10*HZ)
42
43 /*
44 * These should *probably* be handled by the host itself.
45 * Since it is allowed to sleep, it probably should.
46 */
47 #define BUS_RESET_SETTLE_TIME (10)
48 #define HOST_RESET_SETTLE_TIME (10)
49
50 /* called with shost->host_lock held */
51 void scsi_eh_wakeup(struct Scsi_Host *shost)
52 {
53 if (shost->host_busy == shost->host_failed) {
54 wake_up_process(shost->ehandler);
55 SCSI_LOG_ERROR_RECOVERY(5,
56 printk("Waking error handler thread\n"));
57 }
58 }
59
60 /**
61 * scsi_schedule_eh - schedule EH for SCSI host
62 * @shost: SCSI host to invoke error handling on.
63 *
64 * Schedule SCSI EH without scmd.
65 */
66 void scsi_schedule_eh(struct Scsi_Host *shost)
67 {
68 unsigned long flags;
69
70 spin_lock_irqsave(shost->host_lock, flags);
71
72 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
73 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
74 shost->host_eh_scheduled++;
75 scsi_eh_wakeup(shost);
76 }
77
78 spin_unlock_irqrestore(shost->host_lock, flags);
79 }
80 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
81
82 /**
83 * scsi_eh_scmd_add - add scsi cmd to error handling.
84 * @scmd: scmd to run eh on.
85 * @eh_flag: optional SCSI_EH flag.
86 *
87 * Return value:
88 * 0 on failure.
89 */
90 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
91 {
92 struct Scsi_Host *shost = scmd->device->host;
93 unsigned long flags;
94 int ret = 0;
95
96 if (!shost->ehandler)
97 return 0;
98
99 spin_lock_irqsave(shost->host_lock, flags);
100 if (scsi_host_set_state(shost, SHOST_RECOVERY))
101 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
102 goto out_unlock;
103
104 ret = 1;
105 scmd->eh_eflags |= eh_flag;
106 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
107 shost->host_failed++;
108 scsi_eh_wakeup(shost);
109 out_unlock:
110 spin_unlock_irqrestore(shost->host_lock, flags);
111 return ret;
112 }
113
114 /**
115 * scsi_times_out - Timeout function for normal scsi commands.
116 * @req: request that is timing out.
117 *
118 * Notes:
119 * We do not need to lock this. There is the potential for a race
120 * only in that the normal completion handling might run, but if the
121 * normal completion function determines that the timer has already
122 * fired, then it mustn't do anything.
123 */
124 enum blk_eh_timer_return scsi_times_out(struct request *req)
125 {
126 struct scsi_cmnd *scmd = req->special;
127 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
128
129 scsi_log_completion(scmd, TIMEOUT_ERROR);
130
131 if (scmd->device->host->transportt->eh_timed_out)
132 rtn = scmd->device->host->transportt->eh_timed_out(scmd);
133 else if (scmd->device->host->hostt->eh_timed_out)
134 rtn = scmd->device->host->hostt->eh_timed_out(scmd);
135
136 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
137 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
138 scmd->result |= DID_TIME_OUT << 16;
139 rtn = BLK_EH_HANDLED;
140 }
141
142 return rtn;
143 }
144
145 /**
146 * scsi_block_when_processing_errors - Prevent cmds from being queued.
147 * @sdev: Device on which we are performing recovery.
148 *
149 * Description:
150 * We block until the host is out of error recovery, and then check to
151 * see whether the host or the device is offline.
152 *
153 * Return value:
154 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
155 */
156 int scsi_block_when_processing_errors(struct scsi_device *sdev)
157 {
158 int online;
159
160 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
161
162 online = scsi_device_online(sdev);
163
164 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
165 online));
166
167 return online;
168 }
169 EXPORT_SYMBOL(scsi_block_when_processing_errors);
170
171 #ifdef CONFIG_SCSI_LOGGING
172 /**
173 * scsi_eh_prt_fail_stats - Log info on failures.
174 * @shost: scsi host being recovered.
175 * @work_q: Queue of scsi cmds to process.
176 */
177 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
178 struct list_head *work_q)
179 {
180 struct scsi_cmnd *scmd;
181 struct scsi_device *sdev;
182 int total_failures = 0;
183 int cmd_failed = 0;
184 int cmd_cancel = 0;
185 int devices_failed = 0;
186
187 shost_for_each_device(sdev, shost) {
188 list_for_each_entry(scmd, work_q, eh_entry) {
189 if (scmd->device == sdev) {
190 ++total_failures;
191 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
192 ++cmd_cancel;
193 else
194 ++cmd_failed;
195 }
196 }
197
198 if (cmd_cancel || cmd_failed) {
199 SCSI_LOG_ERROR_RECOVERY(3,
200 sdev_printk(KERN_INFO, sdev,
201 "%s: cmds failed: %d, cancel: %d\n",
202 __func__, cmd_failed,
203 cmd_cancel));
204 cmd_cancel = 0;
205 cmd_failed = 0;
206 ++devices_failed;
207 }
208 }
209
210 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
211 " devices require eh work\n",
212 total_failures, devices_failed));
213 }
214 #endif
215
216 /**
217 * scsi_check_sense - Examine scsi cmd sense
218 * @scmd: Cmd to have sense checked.
219 *
220 * Return value:
221 * SUCCESS or FAILED or NEEDS_RETRY
222 *
223 * Notes:
224 * When a deferred error is detected the current command has
225 * not been executed and needs retrying.
226 */
227 static int scsi_check_sense(struct scsi_cmnd *scmd)
228 {
229 struct scsi_device *sdev = scmd->device;
230 struct scsi_sense_hdr sshdr;
231
232 if (! scsi_command_normalize_sense(scmd, &sshdr))
233 return FAILED; /* no valid sense data */
234
235 if (scsi_sense_is_deferred(&sshdr))
236 return NEEDS_RETRY;
237
238 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
239 sdev->scsi_dh_data->scsi_dh->check_sense) {
240 int rc;
241
242 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
243 if (rc != SCSI_RETURN_NOT_HANDLED)
244 return rc;
245 /* handler does not care. Drop down to default handling */
246 }
247
248 /*
249 * Previous logic looked for FILEMARK, EOM or ILI which are
250 * mainly associated with tapes and returned SUCCESS.
251 */
252 if (sshdr.response_code == 0x70) {
253 /* fixed format */
254 if (scmd->sense_buffer[2] & 0xe0)
255 return SUCCESS;
256 } else {
257 /*
258 * descriptor format: look for "stream commands sense data
259 * descriptor" (see SSC-3). Assume single sense data
260 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
261 */
262 if ((sshdr.additional_length > 3) &&
263 (scmd->sense_buffer[8] == 0x4) &&
264 (scmd->sense_buffer[11] & 0xe0))
265 return SUCCESS;
266 }
267
268 switch (sshdr.sense_key) {
269 case NO_SENSE:
270 return SUCCESS;
271 case RECOVERED_ERROR:
272 return /* soft_error */ SUCCESS;
273
274 case ABORTED_COMMAND:
275 if (sshdr.asc == 0x10) /* DIF */
276 return SUCCESS;
277
278 return NEEDS_RETRY;
279 case NOT_READY:
280 case UNIT_ATTENTION:
281 /*
282 * if we are expecting a cc/ua because of a bus reset that we
283 * performed, treat this just as a retry. otherwise this is
284 * information that we should pass up to the upper-level driver
285 * so that we can deal with it there.
286 */
287 if (scmd->device->expecting_cc_ua) {
288 scmd->device->expecting_cc_ua = 0;
289 return NEEDS_RETRY;
290 }
291 /*
292 * if the device is in the process of becoming ready, we
293 * should retry.
294 */
295 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
296 return NEEDS_RETRY;
297 /*
298 * if the device is not started, we need to wake
299 * the error handler to start the motor
300 */
301 if (scmd->device->allow_restart &&
302 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
303 return FAILED;
304 return SUCCESS;
305
306 /* these three are not supported */
307 case COPY_ABORTED:
308 case VOLUME_OVERFLOW:
309 case MISCOMPARE:
310 return SUCCESS;
311
312 case MEDIUM_ERROR:
313 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
314 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
315 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
316 return SUCCESS;
317 }
318 return NEEDS_RETRY;
319
320 case HARDWARE_ERROR:
321 if (scmd->device->retry_hwerror)
322 return ADD_TO_MLQUEUE;
323 else
324 return SUCCESS;
325
326 case ILLEGAL_REQUEST:
327 case BLANK_CHECK:
328 case DATA_PROTECT:
329 default:
330 return SUCCESS;
331 }
332 }
333
334 /**
335 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
336 * @scmd: SCSI cmd to examine.
337 *
338 * Notes:
339 * This is *only* called when we are examining the status of commands
340 * queued during error recovery. the main difference here is that we
341 * don't allow for the possibility of retries here, and we are a lot
342 * more restrictive about what we consider acceptable.
343 */
344 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
345 {
346 /*
347 * first check the host byte, to see if there is anything in there
348 * that would indicate what we need to do.
349 */
350 if (host_byte(scmd->result) == DID_RESET) {
351 /*
352 * rats. we are already in the error handler, so we now
353 * get to try and figure out what to do next. if the sense
354 * is valid, we have a pretty good idea of what to do.
355 * if not, we mark it as FAILED.
356 */
357 return scsi_check_sense(scmd);
358 }
359 if (host_byte(scmd->result) != DID_OK)
360 return FAILED;
361
362 /*
363 * next, check the message byte.
364 */
365 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
366 return FAILED;
367
368 /*
369 * now, check the status byte to see if this indicates
370 * anything special.
371 */
372 switch (status_byte(scmd->result)) {
373 case GOOD:
374 case COMMAND_TERMINATED:
375 return SUCCESS;
376 case CHECK_CONDITION:
377 return scsi_check_sense(scmd);
378 case CONDITION_GOOD:
379 case INTERMEDIATE_GOOD:
380 case INTERMEDIATE_C_GOOD:
381 /*
382 * who knows? FIXME(eric)
383 */
384 return SUCCESS;
385 case BUSY:
386 case QUEUE_FULL:
387 case RESERVATION_CONFLICT:
388 default:
389 return FAILED;
390 }
391 return FAILED;
392 }
393
394 /**
395 * scsi_eh_done - Completion function for error handling.
396 * @scmd: Cmd that is done.
397 */
398 static void scsi_eh_done(struct scsi_cmnd *scmd)
399 {
400 struct completion *eh_action;
401
402 SCSI_LOG_ERROR_RECOVERY(3,
403 printk("%s scmd: %p result: %x\n",
404 __func__, scmd, scmd->result));
405
406 eh_action = scmd->device->host->eh_action;
407 if (eh_action)
408 complete(eh_action);
409 }
410
411 /**
412 * scsi_try_host_reset - ask host adapter to reset itself
413 * @scmd: SCSI cmd to send hsot reset.
414 */
415 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
416 {
417 unsigned long flags;
418 int rtn;
419
420 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
421 __func__));
422
423 if (!scmd->device->host->hostt->eh_host_reset_handler)
424 return FAILED;
425
426 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
427
428 if (rtn == SUCCESS) {
429 if (!scmd->device->host->hostt->skip_settle_delay)
430 ssleep(HOST_RESET_SETTLE_TIME);
431 spin_lock_irqsave(scmd->device->host->host_lock, flags);
432 scsi_report_bus_reset(scmd->device->host,
433 scmd_channel(scmd));
434 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
435 }
436
437 return rtn;
438 }
439
440 /**
441 * scsi_try_bus_reset - ask host to perform a bus reset
442 * @scmd: SCSI cmd to send bus reset.
443 */
444 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
445 {
446 unsigned long flags;
447 int rtn;
448
449 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
450 __func__));
451
452 if (!scmd->device->host->hostt->eh_bus_reset_handler)
453 return FAILED;
454
455 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
456
457 if (rtn == SUCCESS) {
458 if (!scmd->device->host->hostt->skip_settle_delay)
459 ssleep(BUS_RESET_SETTLE_TIME);
460 spin_lock_irqsave(scmd->device->host->host_lock, flags);
461 scsi_report_bus_reset(scmd->device->host,
462 scmd_channel(scmd));
463 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
464 }
465
466 return rtn;
467 }
468
469 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
470 {
471 sdev->was_reset = 1;
472 sdev->expecting_cc_ua = 1;
473 }
474
475 /**
476 * scsi_try_target_reset - Ask host to perform a target reset
477 * @scmd: SCSI cmd used to send a target reset
478 *
479 * Notes:
480 * There is no timeout for this operation. if this operation is
481 * unreliable for a given host, then the host itself needs to put a
482 * timer on it, and set the host back to a consistent state prior to
483 * returning.
484 */
485 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
486 {
487 unsigned long flags;
488 int rtn;
489
490 if (!scmd->device->host->hostt->eh_target_reset_handler)
491 return FAILED;
492
493 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
494 if (rtn == SUCCESS) {
495 spin_lock_irqsave(scmd->device->host->host_lock, flags);
496 __starget_for_each_device(scsi_target(scmd->device), NULL,
497 __scsi_report_device_reset);
498 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
499 }
500
501 return rtn;
502 }
503
504 /**
505 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
506 * @scmd: SCSI cmd used to send BDR
507 *
508 * Notes:
509 * There is no timeout for this operation. if this operation is
510 * unreliable for a given host, then the host itself needs to put a
511 * timer on it, and set the host back to a consistent state prior to
512 * returning.
513 */
514 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
515 {
516 int rtn;
517
518 if (!scmd->device->host->hostt->eh_device_reset_handler)
519 return FAILED;
520
521 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
522 if (rtn == SUCCESS)
523 __scsi_report_device_reset(scmd->device, NULL);
524 return rtn;
525 }
526
527 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
528 {
529 if (!scmd->device->host->hostt->eh_abort_handler)
530 return FAILED;
531
532 return scmd->device->host->hostt->eh_abort_handler(scmd);
533 }
534
535 /**
536 * scsi_try_to_abort_cmd - Ask host to abort a running command.
537 * @scmd: SCSI cmd to abort from Lower Level.
538 *
539 * Notes:
540 * This function will not return until the user's completion function
541 * has been called. there is no timeout on this operation. if the
542 * author of the low-level driver wishes this operation to be timed,
543 * they can provide this facility themselves. helper functions in
544 * scsi_error.c can be supplied to make this easier to do.
545 */
546 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
547 {
548 /*
549 * scsi_done was called just after the command timed out and before
550 * we had a chance to process it. (db)
551 */
552 if (scmd->serial_number == 0)
553 return SUCCESS;
554 return __scsi_try_to_abort_cmd(scmd);
555 }
556
557 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
558 {
559 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
560 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
561 if (scsi_try_target_reset(scmd) != SUCCESS)
562 if (scsi_try_bus_reset(scmd) != SUCCESS)
563 scsi_try_host_reset(scmd);
564 }
565
566 /**
567 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
568 * @scmd: SCSI command structure to hijack
569 * @ses: structure to save restore information
570 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
571 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
572 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
573 *
574 * This function is used to save a scsi command information before re-execution
575 * as part of the error recovery process. If @sense_bytes is 0 the command
576 * sent must be one that does not transfer any data. If @sense_bytes != 0
577 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
578 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
579 */
580 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
581 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
582 {
583 struct scsi_device *sdev = scmd->device;
584
585 /*
586 * We need saved copies of a number of fields - this is because
587 * error handling may need to overwrite these with different values
588 * to run different commands, and once error handling is complete,
589 * we will need to restore these values prior to running the actual
590 * command.
591 */
592 ses->cmd_len = scmd->cmd_len;
593 ses->cmnd = scmd->cmnd;
594 ses->data_direction = scmd->sc_data_direction;
595 ses->sdb = scmd->sdb;
596 ses->next_rq = scmd->request->next_rq;
597 ses->result = scmd->result;
598 ses->underflow = scmd->underflow;
599 ses->prot_op = scmd->prot_op;
600
601 scmd->prot_op = SCSI_PROT_NORMAL;
602 scmd->cmnd = ses->eh_cmnd;
603 memset(scmd->cmnd, 0, BLK_MAX_CDB);
604 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
605 scmd->request->next_rq = NULL;
606
607 if (sense_bytes) {
608 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
609 sense_bytes);
610 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
611 scmd->sdb.length);
612 scmd->sdb.table.sgl = &ses->sense_sgl;
613 scmd->sc_data_direction = DMA_FROM_DEVICE;
614 scmd->sdb.table.nents = 1;
615 scmd->cmnd[0] = REQUEST_SENSE;
616 scmd->cmnd[4] = scmd->sdb.length;
617 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
618 } else {
619 scmd->sc_data_direction = DMA_NONE;
620 if (cmnd) {
621 BUG_ON(cmnd_size > BLK_MAX_CDB);
622 memcpy(scmd->cmnd, cmnd, cmnd_size);
623 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
624 }
625 }
626
627 scmd->underflow = 0;
628
629 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
630 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
631 (sdev->lun << 5 & 0xe0);
632
633 /*
634 * Zero the sense buffer. The scsi spec mandates that any
635 * untransferred sense data should be interpreted as being zero.
636 */
637 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
638 }
639 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
640
641 /**
642 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
643 * @scmd: SCSI command structure to restore
644 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
645 *
646 * Undo any damage done by above scsi_eh_prep_cmnd().
647 */
648 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
649 {
650 /*
651 * Restore original data
652 */
653 scmd->cmd_len = ses->cmd_len;
654 scmd->cmnd = ses->cmnd;
655 scmd->sc_data_direction = ses->data_direction;
656 scmd->sdb = ses->sdb;
657 scmd->request->next_rq = ses->next_rq;
658 scmd->result = ses->result;
659 scmd->underflow = ses->underflow;
660 scmd->prot_op = ses->prot_op;
661 }
662 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
663
664 /**
665 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
666 * @scmd: SCSI command structure to hijack
667 * @cmnd: CDB to send
668 * @cmnd_size: size in bytes of @cmnd
669 * @timeout: timeout for this request
670 * @sense_bytes: size of sense data to copy or 0
671 *
672 * This function is used to send a scsi command down to a target device
673 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
674 *
675 * Return value:
676 * SUCCESS or FAILED or NEEDS_RETRY
677 */
678 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
679 int cmnd_size, int timeout, unsigned sense_bytes)
680 {
681 struct scsi_device *sdev = scmd->device;
682 struct Scsi_Host *shost = sdev->host;
683 DECLARE_COMPLETION_ONSTACK(done);
684 unsigned long timeleft;
685 unsigned long flags;
686 struct scsi_eh_save ses;
687 int rtn;
688
689 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
690 shost->eh_action = &done;
691
692 spin_lock_irqsave(shost->host_lock, flags);
693 scsi_log_send(scmd);
694 shost->hostt->queuecommand(scmd, scsi_eh_done);
695 spin_unlock_irqrestore(shost->host_lock, flags);
696
697 timeleft = wait_for_completion_timeout(&done, timeout);
698
699 shost->eh_action = NULL;
700
701 scsi_log_completion(scmd, SUCCESS);
702
703 SCSI_LOG_ERROR_RECOVERY(3,
704 printk("%s: scmd: %p, timeleft: %ld\n",
705 __func__, scmd, timeleft));
706
707 /*
708 * If there is time left scsi_eh_done got called, and we will
709 * examine the actual status codes to see whether the command
710 * actually did complete normally, else tell the host to forget
711 * about this command.
712 */
713 if (timeleft) {
714 rtn = scsi_eh_completed_normally(scmd);
715 SCSI_LOG_ERROR_RECOVERY(3,
716 printk("%s: scsi_eh_completed_normally %x\n",
717 __func__, rtn));
718
719 switch (rtn) {
720 case SUCCESS:
721 case NEEDS_RETRY:
722 case FAILED:
723 break;
724 case ADD_TO_MLQUEUE:
725 rtn = NEEDS_RETRY;
726 break;
727 default:
728 rtn = FAILED;
729 break;
730 }
731 } else {
732 scsi_abort_eh_cmnd(scmd);
733 rtn = FAILED;
734 }
735
736 scsi_eh_restore_cmnd(scmd, &ses);
737 return rtn;
738 }
739
740 /**
741 * scsi_request_sense - Request sense data from a particular target.
742 * @scmd: SCSI cmd for request sense.
743 *
744 * Notes:
745 * Some hosts automatically obtain this information, others require
746 * that we obtain it on our own. This function will *not* return until
747 * the command either times out, or it completes.
748 */
749 static int scsi_request_sense(struct scsi_cmnd *scmd)
750 {
751 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
752 }
753
754 /**
755 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
756 * @scmd: Original SCSI cmd that eh has finished.
757 * @done_q: Queue for processed commands.
758 *
759 * Notes:
760 * We don't want to use the normal command completion while we are are
761 * still handling errors - it may cause other commands to be queued,
762 * and that would disturb what we are doing. Thus we really want to
763 * keep a list of pending commands for final completion, and once we
764 * are ready to leave error handling we handle completion for real.
765 */
766 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
767 {
768 scmd->device->host->host_failed--;
769 scmd->eh_eflags = 0;
770 list_move_tail(&scmd->eh_entry, done_q);
771 }
772 EXPORT_SYMBOL(scsi_eh_finish_cmd);
773
774 /**
775 * scsi_eh_get_sense - Get device sense data.
776 * @work_q: Queue of commands to process.
777 * @done_q: Queue of processed commands.
778 *
779 * Description:
780 * See if we need to request sense information. if so, then get it
781 * now, so we have a better idea of what to do.
782 *
783 * Notes:
784 * This has the unfortunate side effect that if a shost adapter does
785 * not automatically request sense information, we end up shutting
786 * it down before we request it.
787 *
788 * All drivers should request sense information internally these days,
789 * so for now all I have to say is tough noogies if you end up in here.
790 *
791 * XXX: Long term this code should go away, but that needs an audit of
792 * all LLDDs first.
793 */
794 int scsi_eh_get_sense(struct list_head *work_q,
795 struct list_head *done_q)
796 {
797 struct scsi_cmnd *scmd, *next;
798 int rtn;
799
800 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
801 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
802 SCSI_SENSE_VALID(scmd))
803 continue;
804
805 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
806 "%s: requesting sense\n",
807 current->comm));
808 rtn = scsi_request_sense(scmd);
809 if (rtn != SUCCESS)
810 continue;
811
812 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
813 " result %x\n", scmd,
814 scmd->result));
815 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
816
817 rtn = scsi_decide_disposition(scmd);
818
819 /*
820 * if the result was normal, then just pass it along to the
821 * upper level.
822 */
823 if (rtn == SUCCESS)
824 /* we don't want this command reissued, just
825 * finished with the sense data, so set
826 * retries to the max allowed to ensure it
827 * won't get reissued */
828 scmd->retries = scmd->allowed;
829 else if (rtn != NEEDS_RETRY)
830 continue;
831
832 scsi_eh_finish_cmd(scmd, done_q);
833 }
834
835 return list_empty(work_q);
836 }
837 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
838
839 /**
840 * scsi_eh_tur - Send TUR to device.
841 * @scmd: &scsi_cmnd to send TUR
842 *
843 * Return value:
844 * 0 - Device is ready. 1 - Device NOT ready.
845 */
846 static int scsi_eh_tur(struct scsi_cmnd *scmd)
847 {
848 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
849 int retry_cnt = 1, rtn;
850
851 retry_tur:
852 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
853
854 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
855 __func__, scmd, rtn));
856
857 switch (rtn) {
858 case NEEDS_RETRY:
859 if (retry_cnt--)
860 goto retry_tur;
861 /*FALLTHRU*/
862 case SUCCESS:
863 return 0;
864 default:
865 return 1;
866 }
867 }
868
869 /**
870 * scsi_eh_abort_cmds - abort pending commands.
871 * @work_q: &list_head for pending commands.
872 * @done_q: &list_head for processed commands.
873 *
874 * Decription:
875 * Try and see whether or not it makes sense to try and abort the
876 * running command. This only works out to be the case if we have one
877 * command that has timed out. If the command simply failed, it makes
878 * no sense to try and abort the command, since as far as the shost
879 * adapter is concerned, it isn't running.
880 */
881 static int scsi_eh_abort_cmds(struct list_head *work_q,
882 struct list_head *done_q)
883 {
884 struct scsi_cmnd *scmd, *next;
885 int rtn;
886
887 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
888 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
889 continue;
890 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
891 "0x%p\n", current->comm,
892 scmd));
893 rtn = scsi_try_to_abort_cmd(scmd);
894 if (rtn == SUCCESS) {
895 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
896 if (!scsi_device_online(scmd->device) ||
897 !scsi_eh_tur(scmd)) {
898 scsi_eh_finish_cmd(scmd, done_q);
899 }
900
901 } else
902 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
903 " cmd failed:"
904 "0x%p\n",
905 current->comm,
906 scmd));
907 }
908
909 return list_empty(work_q);
910 }
911
912 /**
913 * scsi_eh_try_stu - Send START_UNIT to device.
914 * @scmd: &scsi_cmnd to send START_UNIT
915 *
916 * Return value:
917 * 0 - Device is ready. 1 - Device NOT ready.
918 */
919 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
920 {
921 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
922
923 if (scmd->device->allow_restart) {
924 int i, rtn = NEEDS_RETRY;
925
926 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
927 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
928
929 if (rtn == SUCCESS)
930 return 0;
931 }
932
933 return 1;
934 }
935
936 /**
937 * scsi_eh_stu - send START_UNIT if needed
938 * @shost: &scsi host being recovered.
939 * @work_q: &list_head for pending commands.
940 * @done_q: &list_head for processed commands.
941 *
942 * Notes:
943 * If commands are failing due to not ready, initializing command required,
944 * try revalidating the device, which will end up sending a start unit.
945 */
946 static int scsi_eh_stu(struct Scsi_Host *shost,
947 struct list_head *work_q,
948 struct list_head *done_q)
949 {
950 struct scsi_cmnd *scmd, *stu_scmd, *next;
951 struct scsi_device *sdev;
952
953 shost_for_each_device(sdev, shost) {
954 stu_scmd = NULL;
955 list_for_each_entry(scmd, work_q, eh_entry)
956 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
957 scsi_check_sense(scmd) == FAILED ) {
958 stu_scmd = scmd;
959 break;
960 }
961
962 if (!stu_scmd)
963 continue;
964
965 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
966 " 0x%p\n", current->comm, sdev));
967
968 if (!scsi_eh_try_stu(stu_scmd)) {
969 if (!scsi_device_online(sdev) ||
970 !scsi_eh_tur(stu_scmd)) {
971 list_for_each_entry_safe(scmd, next,
972 work_q, eh_entry) {
973 if (scmd->device == sdev)
974 scsi_eh_finish_cmd(scmd, done_q);
975 }
976 }
977 } else {
978 SCSI_LOG_ERROR_RECOVERY(3,
979 printk("%s: START_UNIT failed to sdev:"
980 " 0x%p\n", current->comm, sdev));
981 }
982 }
983
984 return list_empty(work_q);
985 }
986
987
988 /**
989 * scsi_eh_bus_device_reset - send bdr if needed
990 * @shost: scsi host being recovered.
991 * @work_q: &list_head for pending commands.
992 * @done_q: &list_head for processed commands.
993 *
994 * Notes:
995 * Try a bus device reset. Still, look to see whether we have multiple
996 * devices that are jammed or not - if we have multiple devices, it
997 * makes no sense to try bus_device_reset - we really would need to try
998 * a bus_reset instead.
999 */
1000 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1001 struct list_head *work_q,
1002 struct list_head *done_q)
1003 {
1004 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1005 struct scsi_device *sdev;
1006 int rtn;
1007
1008 shost_for_each_device(sdev, shost) {
1009 bdr_scmd = NULL;
1010 list_for_each_entry(scmd, work_q, eh_entry)
1011 if (scmd->device == sdev) {
1012 bdr_scmd = scmd;
1013 break;
1014 }
1015
1016 if (!bdr_scmd)
1017 continue;
1018
1019 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1020 " 0x%p\n", current->comm,
1021 sdev));
1022 rtn = scsi_try_bus_device_reset(bdr_scmd);
1023 if (rtn == SUCCESS) {
1024 if (!scsi_device_online(sdev) ||
1025 !scsi_eh_tur(bdr_scmd)) {
1026 list_for_each_entry_safe(scmd, next,
1027 work_q, eh_entry) {
1028 if (scmd->device == sdev)
1029 scsi_eh_finish_cmd(scmd,
1030 done_q);
1031 }
1032 }
1033 } else {
1034 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1035 " failed sdev:"
1036 "0x%p\n",
1037 current->comm,
1038 sdev));
1039 }
1040 }
1041
1042 return list_empty(work_q);
1043 }
1044
1045 /**
1046 * scsi_eh_target_reset - send target reset if needed
1047 * @shost: scsi host being recovered.
1048 * @work_q: &list_head for pending commands.
1049 * @done_q: &list_head for processed commands.
1050 *
1051 * Notes:
1052 * Try a target reset.
1053 */
1054 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1055 struct list_head *work_q,
1056 struct list_head *done_q)
1057 {
1058 struct scsi_cmnd *scmd, *tgtr_scmd, *next;
1059 unsigned int id = 0;
1060 int rtn;
1061
1062 do {
1063 tgtr_scmd = NULL;
1064 list_for_each_entry(scmd, work_q, eh_entry) {
1065 if (id == scmd_id(scmd)) {
1066 tgtr_scmd = scmd;
1067 break;
1068 }
1069 }
1070 if (!tgtr_scmd) {
1071 /* not one exactly equal; find the next highest */
1072 list_for_each_entry(scmd, work_q, eh_entry) {
1073 if (scmd_id(scmd) > id &&
1074 (!tgtr_scmd ||
1075 scmd_id(tgtr_scmd) > scmd_id(scmd)))
1076 tgtr_scmd = scmd;
1077 }
1078 }
1079 if (!tgtr_scmd)
1080 /* no more commands, that's it */
1081 break;
1082
1083 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1084 "to target %d\n",
1085 current->comm, id));
1086 rtn = scsi_try_target_reset(tgtr_scmd);
1087 if (rtn == SUCCESS) {
1088 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1089 if (id == scmd_id(scmd))
1090 if (!scsi_device_online(scmd->device) ||
1091 !scsi_eh_tur(tgtr_scmd))
1092 scsi_eh_finish_cmd(scmd,
1093 done_q);
1094 }
1095 } else
1096 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1097 " failed target: "
1098 "%d\n",
1099 current->comm, id));
1100 id++;
1101 } while(id != 0);
1102
1103 return list_empty(work_q);
1104 }
1105
1106 /**
1107 * scsi_eh_bus_reset - send a bus reset
1108 * @shost: &scsi host being recovered.
1109 * @work_q: &list_head for pending commands.
1110 * @done_q: &list_head for processed commands.
1111 */
1112 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1113 struct list_head *work_q,
1114 struct list_head *done_q)
1115 {
1116 struct scsi_cmnd *scmd, *chan_scmd, *next;
1117 unsigned int channel;
1118 int rtn;
1119
1120 /*
1121 * we really want to loop over the various channels, and do this on
1122 * a channel by channel basis. we should also check to see if any
1123 * of the failed commands are on soft_reset devices, and if so, skip
1124 * the reset.
1125 */
1126
1127 for (channel = 0; channel <= shost->max_channel; channel++) {
1128 chan_scmd = NULL;
1129 list_for_each_entry(scmd, work_q, eh_entry) {
1130 if (channel == scmd_channel(scmd)) {
1131 chan_scmd = scmd;
1132 break;
1133 /*
1134 * FIXME add back in some support for
1135 * soft_reset devices.
1136 */
1137 }
1138 }
1139
1140 if (!chan_scmd)
1141 continue;
1142 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1143 " %d\n", current->comm,
1144 channel));
1145 rtn = scsi_try_bus_reset(chan_scmd);
1146 if (rtn == SUCCESS) {
1147 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1148 if (channel == scmd_channel(scmd))
1149 if (!scsi_device_online(scmd->device) ||
1150 !scsi_eh_tur(scmd))
1151 scsi_eh_finish_cmd(scmd,
1152 done_q);
1153 }
1154 } else {
1155 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1156 " failed chan: %d\n",
1157 current->comm,
1158 channel));
1159 }
1160 }
1161 return list_empty(work_q);
1162 }
1163
1164 /**
1165 * scsi_eh_host_reset - send a host reset
1166 * @work_q: list_head for processed commands.
1167 * @done_q: list_head for processed commands.
1168 */
1169 static int scsi_eh_host_reset(struct list_head *work_q,
1170 struct list_head *done_q)
1171 {
1172 struct scsi_cmnd *scmd, *next;
1173 int rtn;
1174
1175 if (!list_empty(work_q)) {
1176 scmd = list_entry(work_q->next,
1177 struct scsi_cmnd, eh_entry);
1178
1179 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1180 , current->comm));
1181
1182 rtn = scsi_try_host_reset(scmd);
1183 if (rtn == SUCCESS) {
1184 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1185 if (!scsi_device_online(scmd->device) ||
1186 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1187 !scsi_eh_tur(scmd))
1188 scsi_eh_finish_cmd(scmd, done_q);
1189 }
1190 } else {
1191 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1192 " failed\n",
1193 current->comm));
1194 }
1195 }
1196 return list_empty(work_q);
1197 }
1198
1199 /**
1200 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1201 * @work_q: list_head for processed commands.
1202 * @done_q: list_head for processed commands.
1203 */
1204 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1205 struct list_head *done_q)
1206 {
1207 struct scsi_cmnd *scmd, *next;
1208
1209 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1210 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1211 "not ready after error recovery\n");
1212 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1213 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1214 /*
1215 * FIXME: Handle lost cmds.
1216 */
1217 }
1218 scsi_eh_finish_cmd(scmd, done_q);
1219 }
1220 return;
1221 }
1222
1223 /**
1224 * scsi_noretry_cmd - determinte if command should be failed fast
1225 * @scmd: SCSI cmd to examine.
1226 */
1227 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1228 {
1229 switch (host_byte(scmd->result)) {
1230 case DID_OK:
1231 break;
1232 case DID_BUS_BUSY:
1233 return blk_failfast_transport(scmd->request);
1234 case DID_PARITY:
1235 return blk_failfast_dev(scmd->request);
1236 case DID_ERROR:
1237 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1238 status_byte(scmd->result) == RESERVATION_CONFLICT)
1239 return 0;
1240 /* fall through */
1241 case DID_SOFT_ERROR:
1242 return blk_failfast_driver(scmd->request);
1243 }
1244
1245 switch (status_byte(scmd->result)) {
1246 case CHECK_CONDITION:
1247 /*
1248 * assume caller has checked sense and determinted
1249 * the check condition was retryable.
1250 */
1251 return blk_failfast_dev(scmd->request);
1252 }
1253
1254 return 0;
1255 }
1256
1257 /**
1258 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1259 * @scmd: SCSI cmd to examine.
1260 *
1261 * Notes:
1262 * This is *only* called when we are examining the status after sending
1263 * out the actual data command. any commands that are queued for error
1264 * recovery (e.g. test_unit_ready) do *not* come through here.
1265 *
1266 * When this routine returns failed, it means the error handler thread
1267 * is woken. In cases where the error code indicates an error that
1268 * doesn't require the error handler read (i.e. we don't need to
1269 * abort/reset), this function should return SUCCESS.
1270 */
1271 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1272 {
1273 int rtn;
1274
1275 /*
1276 * if the device is offline, then we clearly just pass the result back
1277 * up to the top level.
1278 */
1279 if (!scsi_device_online(scmd->device)) {
1280 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1281 " as SUCCESS\n",
1282 __func__));
1283 return SUCCESS;
1284 }
1285
1286 /*
1287 * first check the host byte, to see if there is anything in there
1288 * that would indicate what we need to do.
1289 */
1290 switch (host_byte(scmd->result)) {
1291 case DID_PASSTHROUGH:
1292 /*
1293 * no matter what, pass this through to the upper layer.
1294 * nuke this special code so that it looks like we are saying
1295 * did_ok.
1296 */
1297 scmd->result &= 0xff00ffff;
1298 return SUCCESS;
1299 case DID_OK:
1300 /*
1301 * looks good. drop through, and check the next byte.
1302 */
1303 break;
1304 case DID_NO_CONNECT:
1305 case DID_BAD_TARGET:
1306 case DID_ABORT:
1307 /*
1308 * note - this means that we just report the status back
1309 * to the top level driver, not that we actually think
1310 * that it indicates SUCCESS.
1311 */
1312 return SUCCESS;
1313 /*
1314 * when the low level driver returns did_soft_error,
1315 * it is responsible for keeping an internal retry counter
1316 * in order to avoid endless loops (db)
1317 *
1318 * actually this is a bug in this function here. we should
1319 * be mindful of the maximum number of retries specified
1320 * and not get stuck in a loop.
1321 */
1322 case DID_SOFT_ERROR:
1323 goto maybe_retry;
1324 case DID_IMM_RETRY:
1325 return NEEDS_RETRY;
1326
1327 case DID_REQUEUE:
1328 return ADD_TO_MLQUEUE;
1329 case DID_TRANSPORT_DISRUPTED:
1330 /*
1331 * LLD/transport was disrupted during processing of the IO.
1332 * The transport class is now blocked/blocking,
1333 * and the transport will decide what to do with the IO
1334 * based on its timers and recovery capablilities if
1335 * there are enough retries.
1336 */
1337 goto maybe_retry;
1338 case DID_TRANSPORT_FAILFAST:
1339 /*
1340 * The transport decided to failfast the IO (most likely
1341 * the fast io fail tmo fired), so send IO directly upwards.
1342 */
1343 return SUCCESS;
1344 case DID_ERROR:
1345 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1346 status_byte(scmd->result) == RESERVATION_CONFLICT)
1347 /*
1348 * execute reservation conflict processing code
1349 * lower down
1350 */
1351 break;
1352 /* fallthrough */
1353
1354 case DID_BUS_BUSY:
1355 case DID_PARITY:
1356 goto maybe_retry;
1357 case DID_TIME_OUT:
1358 /*
1359 * when we scan the bus, we get timeout messages for
1360 * these commands if there is no device available.
1361 * other hosts report did_no_connect for the same thing.
1362 */
1363 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1364 scmd->cmnd[0] == INQUIRY)) {
1365 return SUCCESS;
1366 } else {
1367 return FAILED;
1368 }
1369 case DID_RESET:
1370 return SUCCESS;
1371 default:
1372 return FAILED;
1373 }
1374
1375 /*
1376 * next, check the message byte.
1377 */
1378 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1379 return FAILED;
1380
1381 /*
1382 * check the status byte to see if this indicates anything special.
1383 */
1384 switch (status_byte(scmd->result)) {
1385 case QUEUE_FULL:
1386 /*
1387 * the case of trying to send too many commands to a
1388 * tagged queueing device.
1389 */
1390 case BUSY:
1391 /*
1392 * device can't talk to us at the moment. Should only
1393 * occur (SAM-3) when the task queue is empty, so will cause
1394 * the empty queue handling to trigger a stall in the
1395 * device.
1396 */
1397 return ADD_TO_MLQUEUE;
1398 case GOOD:
1399 case COMMAND_TERMINATED:
1400 return SUCCESS;
1401 case TASK_ABORTED:
1402 goto maybe_retry;
1403 case CHECK_CONDITION:
1404 rtn = scsi_check_sense(scmd);
1405 if (rtn == NEEDS_RETRY)
1406 goto maybe_retry;
1407 /* if rtn == FAILED, we have no sense information;
1408 * returning FAILED will wake the error handler thread
1409 * to collect the sense and redo the decide
1410 * disposition */
1411 return rtn;
1412 case CONDITION_GOOD:
1413 case INTERMEDIATE_GOOD:
1414 case INTERMEDIATE_C_GOOD:
1415 case ACA_ACTIVE:
1416 /*
1417 * who knows? FIXME(eric)
1418 */
1419 return SUCCESS;
1420
1421 case RESERVATION_CONFLICT:
1422 sdev_printk(KERN_INFO, scmd->device,
1423 "reservation conflict\n");
1424 return SUCCESS; /* causes immediate i/o error */
1425 default:
1426 return FAILED;
1427 }
1428 return FAILED;
1429
1430 maybe_retry:
1431
1432 /* we requeue for retry because the error was retryable, and
1433 * the request was not marked fast fail. Note that above,
1434 * even if the request is marked fast fail, we still requeue
1435 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1436 if ((++scmd->retries) <= scmd->allowed
1437 && !scsi_noretry_cmd(scmd)) {
1438 return NEEDS_RETRY;
1439 } else {
1440 /*
1441 * no more retries - report this one back to upper level.
1442 */
1443 return SUCCESS;
1444 }
1445 }
1446
1447 static void eh_lock_door_done(struct request *req, int uptodate)
1448 {
1449 __blk_put_request(req->q, req);
1450 }
1451
1452 /**
1453 * scsi_eh_lock_door - Prevent medium removal for the specified device
1454 * @sdev: SCSI device to prevent medium removal
1455 *
1456 * Locking:
1457 * We must be called from process context.
1458 *
1459 * Notes:
1460 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1461 * head of the devices request queue, and continue.
1462 */
1463 static void scsi_eh_lock_door(struct scsi_device *sdev)
1464 {
1465 struct request *req;
1466
1467 /*
1468 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1469 * request becomes available
1470 */
1471 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1472
1473 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1474 req->cmd[1] = 0;
1475 req->cmd[2] = 0;
1476 req->cmd[3] = 0;
1477 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1478 req->cmd[5] = 0;
1479
1480 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1481
1482 req->cmd_type = REQ_TYPE_BLOCK_PC;
1483 req->cmd_flags |= REQ_QUIET;
1484 req->timeout = 10 * HZ;
1485 req->retries = 5;
1486
1487 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1488 }
1489
1490 /**
1491 * scsi_restart_operations - restart io operations to the specified host.
1492 * @shost: Host we are restarting.
1493 *
1494 * Notes:
1495 * When we entered the error handler, we blocked all further i/o to
1496 * this device. we need to 'reverse' this process.
1497 */
1498 static void scsi_restart_operations(struct Scsi_Host *shost)
1499 {
1500 struct scsi_device *sdev;
1501 unsigned long flags;
1502
1503 /*
1504 * If the door was locked, we need to insert a door lock request
1505 * onto the head of the SCSI request queue for the device. There
1506 * is no point trying to lock the door of an off-line device.
1507 */
1508 shost_for_each_device(sdev, shost) {
1509 if (scsi_device_online(sdev) && sdev->locked)
1510 scsi_eh_lock_door(sdev);
1511 }
1512
1513 /*
1514 * next free up anything directly waiting upon the host. this
1515 * will be requests for character device operations, and also for
1516 * ioctls to queued block devices.
1517 */
1518 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1519 __func__));
1520
1521 spin_lock_irqsave(shost->host_lock, flags);
1522 if (scsi_host_set_state(shost, SHOST_RUNNING))
1523 if (scsi_host_set_state(shost, SHOST_CANCEL))
1524 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1525 spin_unlock_irqrestore(shost->host_lock, flags);
1526
1527 wake_up(&shost->host_wait);
1528
1529 /*
1530 * finally we need to re-initiate requests that may be pending. we will
1531 * have had everything blocked while error handling is taking place, and
1532 * now that error recovery is done, we will need to ensure that these
1533 * requests are started.
1534 */
1535 scsi_run_host_queues(shost);
1536 }
1537
1538 /**
1539 * scsi_eh_ready_devs - check device ready state and recover if not.
1540 * @shost: host to be recovered.
1541 * @work_q: &list_head for pending commands.
1542 * @done_q: &list_head for processed commands.
1543 */
1544 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1545 struct list_head *work_q,
1546 struct list_head *done_q)
1547 {
1548 if (!scsi_eh_stu(shost, work_q, done_q))
1549 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1550 if (!scsi_eh_target_reset(shost, work_q, done_q))
1551 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1552 if (!scsi_eh_host_reset(work_q, done_q))
1553 scsi_eh_offline_sdevs(work_q,
1554 done_q);
1555 }
1556 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1557
1558 /**
1559 * scsi_eh_flush_done_q - finish processed commands or retry them.
1560 * @done_q: list_head of processed commands.
1561 */
1562 void scsi_eh_flush_done_q(struct list_head *done_q)
1563 {
1564 struct scsi_cmnd *scmd, *next;
1565
1566 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1567 list_del_init(&scmd->eh_entry);
1568 if (scsi_device_online(scmd->device) &&
1569 !scsi_noretry_cmd(scmd) &&
1570 (++scmd->retries <= scmd->allowed)) {
1571 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1572 " retry cmd: %p\n",
1573 current->comm,
1574 scmd));
1575 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1576 } else {
1577 /*
1578 * If just we got sense for the device (called
1579 * scsi_eh_get_sense), scmd->result is already
1580 * set, do not set DRIVER_TIMEOUT.
1581 */
1582 if (!scmd->result)
1583 scmd->result |= (DRIVER_TIMEOUT << 24);
1584 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1585 " cmd: %p\n",
1586 current->comm, scmd));
1587 scsi_finish_command(scmd);
1588 }
1589 }
1590 }
1591 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1592
1593 /**
1594 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1595 * @shost: Host to unjam.
1596 *
1597 * Notes:
1598 * When we come in here, we *know* that all commands on the bus have
1599 * either completed, failed or timed out. we also know that no further
1600 * commands are being sent to the host, so things are relatively quiet
1601 * and we have freedom to fiddle with things as we wish.
1602 *
1603 * This is only the *default* implementation. it is possible for
1604 * individual drivers to supply their own version of this function, and
1605 * if the maintainer wishes to do this, it is strongly suggested that
1606 * this function be taken as a template and modified. this function
1607 * was designed to correctly handle problems for about 95% of the
1608 * different cases out there, and it should always provide at least a
1609 * reasonable amount of error recovery.
1610 *
1611 * Any command marked 'failed' or 'timeout' must eventually have
1612 * scsi_finish_cmd() called for it. we do all of the retry stuff
1613 * here, so when we restart the host after we return it should have an
1614 * empty queue.
1615 */
1616 static void scsi_unjam_host(struct Scsi_Host *shost)
1617 {
1618 unsigned long flags;
1619 LIST_HEAD(eh_work_q);
1620 LIST_HEAD(eh_done_q);
1621
1622 spin_lock_irqsave(shost->host_lock, flags);
1623 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1624 spin_unlock_irqrestore(shost->host_lock, flags);
1625
1626 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1627
1628 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1629 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1630 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1631
1632 scsi_eh_flush_done_q(&eh_done_q);
1633 }
1634
1635 /**
1636 * scsi_error_handler - SCSI error handler thread
1637 * @data: Host for which we are running.
1638 *
1639 * Notes:
1640 * This is the main error handling loop. This is run as a kernel thread
1641 * for every SCSI host and handles all error handling activity.
1642 */
1643 int scsi_error_handler(void *data)
1644 {
1645 struct Scsi_Host *shost = data;
1646
1647 /*
1648 * We use TASK_INTERRUPTIBLE so that the thread is not
1649 * counted against the load average as a running process.
1650 * We never actually get interrupted because kthread_run
1651 * disables signal delivery for the created thread.
1652 */
1653 set_current_state(TASK_INTERRUPTIBLE);
1654 while (!kthread_should_stop()) {
1655 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1656 shost->host_failed != shost->host_busy) {
1657 SCSI_LOG_ERROR_RECOVERY(1,
1658 printk("Error handler scsi_eh_%d sleeping\n",
1659 shost->host_no));
1660 schedule();
1661 set_current_state(TASK_INTERRUPTIBLE);
1662 continue;
1663 }
1664
1665 __set_current_state(TASK_RUNNING);
1666 SCSI_LOG_ERROR_RECOVERY(1,
1667 printk("Error handler scsi_eh_%d waking up\n",
1668 shost->host_no));
1669
1670 /*
1671 * We have a host that is failing for some reason. Figure out
1672 * what we need to do to get it up and online again (if we can).
1673 * If we fail, we end up taking the thing offline.
1674 */
1675 if (shost->transportt->eh_strategy_handler)
1676 shost->transportt->eh_strategy_handler(shost);
1677 else
1678 scsi_unjam_host(shost);
1679
1680 /*
1681 * Note - if the above fails completely, the action is to take
1682 * individual devices offline and flush the queue of any
1683 * outstanding requests that may have been pending. When we
1684 * restart, we restart any I/O to any other devices on the bus
1685 * which are still online.
1686 */
1687 scsi_restart_operations(shost);
1688 set_current_state(TASK_INTERRUPTIBLE);
1689 }
1690 __set_current_state(TASK_RUNNING);
1691
1692 SCSI_LOG_ERROR_RECOVERY(1,
1693 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1694 shost->ehandler = NULL;
1695 return 0;
1696 }
1697
1698 /*
1699 * Function: scsi_report_bus_reset()
1700 *
1701 * Purpose: Utility function used by low-level drivers to report that
1702 * they have observed a bus reset on the bus being handled.
1703 *
1704 * Arguments: shost - Host in question
1705 * channel - channel on which reset was observed.
1706 *
1707 * Returns: Nothing
1708 *
1709 * Lock status: Host lock must be held.
1710 *
1711 * Notes: This only needs to be called if the reset is one which
1712 * originates from an unknown location. Resets originated
1713 * by the mid-level itself don't need to call this, but there
1714 * should be no harm.
1715 *
1716 * The main purpose of this is to make sure that a CHECK_CONDITION
1717 * is properly treated.
1718 */
1719 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1720 {
1721 struct scsi_device *sdev;
1722
1723 __shost_for_each_device(sdev, shost) {
1724 if (channel == sdev_channel(sdev))
1725 __scsi_report_device_reset(sdev, NULL);
1726 }
1727 }
1728 EXPORT_SYMBOL(scsi_report_bus_reset);
1729
1730 /*
1731 * Function: scsi_report_device_reset()
1732 *
1733 * Purpose: Utility function used by low-level drivers to report that
1734 * they have observed a device reset on the device being handled.
1735 *
1736 * Arguments: shost - Host in question
1737 * channel - channel on which reset was observed
1738 * target - target on which reset was observed
1739 *
1740 * Returns: Nothing
1741 *
1742 * Lock status: Host lock must be held
1743 *
1744 * Notes: This only needs to be called if the reset is one which
1745 * originates from an unknown location. Resets originated
1746 * by the mid-level itself don't need to call this, but there
1747 * should be no harm.
1748 *
1749 * The main purpose of this is to make sure that a CHECK_CONDITION
1750 * is properly treated.
1751 */
1752 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1753 {
1754 struct scsi_device *sdev;
1755
1756 __shost_for_each_device(sdev, shost) {
1757 if (channel == sdev_channel(sdev) &&
1758 target == sdev_id(sdev))
1759 __scsi_report_device_reset(sdev, NULL);
1760 }
1761 }
1762 EXPORT_SYMBOL(scsi_report_device_reset);
1763
1764 static void
1765 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1766 {
1767 }
1768
1769 /*
1770 * Function: scsi_reset_provider
1771 *
1772 * Purpose: Send requested reset to a bus or device at any phase.
1773 *
1774 * Arguments: device - device to send reset to
1775 * flag - reset type (see scsi.h)
1776 *
1777 * Returns: SUCCESS/FAILURE.
1778 *
1779 * Notes: This is used by the SCSI Generic driver to provide
1780 * Bus/Device reset capability.
1781 */
1782 int
1783 scsi_reset_provider(struct scsi_device *dev, int flag)
1784 {
1785 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1786 struct Scsi_Host *shost = dev->host;
1787 struct request req;
1788 unsigned long flags;
1789 int rtn;
1790
1791 blk_rq_init(NULL, &req);
1792 scmd->request = &req;
1793
1794 scmd->cmnd = req.cmd;
1795
1796 scmd->scsi_done = scsi_reset_provider_done_command;
1797 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1798
1799 scmd->cmd_len = 0;
1800
1801 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1802
1803 spin_lock_irqsave(shost->host_lock, flags);
1804 shost->tmf_in_progress = 1;
1805 spin_unlock_irqrestore(shost->host_lock, flags);
1806
1807 switch (flag) {
1808 case SCSI_TRY_RESET_DEVICE:
1809 rtn = scsi_try_bus_device_reset(scmd);
1810 if (rtn == SUCCESS)
1811 break;
1812 /* FALLTHROUGH */
1813 case SCSI_TRY_RESET_TARGET:
1814 rtn = scsi_try_target_reset(scmd);
1815 if (rtn == SUCCESS)
1816 break;
1817 /* FALLTHROUGH */
1818 case SCSI_TRY_RESET_BUS:
1819 rtn = scsi_try_bus_reset(scmd);
1820 if (rtn == SUCCESS)
1821 break;
1822 /* FALLTHROUGH */
1823 case SCSI_TRY_RESET_HOST:
1824 rtn = scsi_try_host_reset(scmd);
1825 break;
1826 default:
1827 rtn = FAILED;
1828 }
1829
1830 spin_lock_irqsave(shost->host_lock, flags);
1831 shost->tmf_in_progress = 0;
1832 spin_unlock_irqrestore(shost->host_lock, flags);
1833
1834 /*
1835 * be sure to wake up anyone who was sleeping or had their queue
1836 * suspended while we performed the TMF.
1837 */
1838 SCSI_LOG_ERROR_RECOVERY(3,
1839 printk("%s: waking up host to restart after TMF\n",
1840 __func__));
1841
1842 wake_up(&shost->host_wait);
1843
1844 scsi_run_host_queues(shost);
1845
1846 scsi_next_command(scmd);
1847 return rtn;
1848 }
1849 EXPORT_SYMBOL(scsi_reset_provider);
1850
1851 /**
1852 * scsi_normalize_sense - normalize main elements from either fixed or
1853 * descriptor sense data format into a common format.
1854 *
1855 * @sense_buffer: byte array containing sense data returned by device
1856 * @sb_len: number of valid bytes in sense_buffer
1857 * @sshdr: pointer to instance of structure that common
1858 * elements are written to.
1859 *
1860 * Notes:
1861 * The "main elements" from sense data are: response_code, sense_key,
1862 * asc, ascq and additional_length (only for descriptor format).
1863 *
1864 * Typically this function can be called after a device has
1865 * responded to a SCSI command with the CHECK_CONDITION status.
1866 *
1867 * Return value:
1868 * 1 if valid sense data information found, else 0;
1869 */
1870 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1871 struct scsi_sense_hdr *sshdr)
1872 {
1873 if (!sense_buffer || !sb_len)
1874 return 0;
1875
1876 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1877
1878 sshdr->response_code = (sense_buffer[0] & 0x7f);
1879
1880 if (!scsi_sense_valid(sshdr))
1881 return 0;
1882
1883 if (sshdr->response_code >= 0x72) {
1884 /*
1885 * descriptor format
1886 */
1887 if (sb_len > 1)
1888 sshdr->sense_key = (sense_buffer[1] & 0xf);
1889 if (sb_len > 2)
1890 sshdr->asc = sense_buffer[2];
1891 if (sb_len > 3)
1892 sshdr->ascq = sense_buffer[3];
1893 if (sb_len > 7)
1894 sshdr->additional_length = sense_buffer[7];
1895 } else {
1896 /*
1897 * fixed format
1898 */
1899 if (sb_len > 2)
1900 sshdr->sense_key = (sense_buffer[2] & 0xf);
1901 if (sb_len > 7) {
1902 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1903 sb_len : (sense_buffer[7] + 8);
1904 if (sb_len > 12)
1905 sshdr->asc = sense_buffer[12];
1906 if (sb_len > 13)
1907 sshdr->ascq = sense_buffer[13];
1908 }
1909 }
1910
1911 return 1;
1912 }
1913 EXPORT_SYMBOL(scsi_normalize_sense);
1914
1915 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1916 struct scsi_sense_hdr *sshdr)
1917 {
1918 return scsi_normalize_sense(cmd->sense_buffer,
1919 SCSI_SENSE_BUFFERSIZE, sshdr);
1920 }
1921 EXPORT_SYMBOL(scsi_command_normalize_sense);
1922
1923 /**
1924 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1925 * @sense_buffer: byte array of descriptor format sense data
1926 * @sb_len: number of valid bytes in sense_buffer
1927 * @desc_type: value of descriptor type to find
1928 * (e.g. 0 -> information)
1929 *
1930 * Notes:
1931 * only valid when sense data is in descriptor format
1932 *
1933 * Return value:
1934 * pointer to start of (first) descriptor if found else NULL
1935 */
1936 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1937 int desc_type)
1938 {
1939 int add_sen_len, add_len, desc_len, k;
1940 const u8 * descp;
1941
1942 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1943 return NULL;
1944 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1945 return NULL;
1946 add_sen_len = (add_sen_len < (sb_len - 8)) ?
1947 add_sen_len : (sb_len - 8);
1948 descp = &sense_buffer[8];
1949 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1950 descp += desc_len;
1951 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1952 desc_len = add_len + 2;
1953 if (descp[0] == desc_type)
1954 return descp;
1955 if (add_len < 0) // short descriptor ??
1956 break;
1957 }
1958 return NULL;
1959 }
1960 EXPORT_SYMBOL(scsi_sense_desc_find);
1961
1962 /**
1963 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1964 * @sense_buffer: byte array of sense data
1965 * @sb_len: number of valid bytes in sense_buffer
1966 * @info_out: pointer to 64 integer where 8 or 4 byte information
1967 * field will be placed if found.
1968 *
1969 * Return value:
1970 * 1 if information field found, 0 if not found.
1971 */
1972 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1973 u64 * info_out)
1974 {
1975 int j;
1976 const u8 * ucp;
1977 u64 ull;
1978
1979 if (sb_len < 7)
1980 return 0;
1981 switch (sense_buffer[0] & 0x7f) {
1982 case 0x70:
1983 case 0x71:
1984 if (sense_buffer[0] & 0x80) {
1985 *info_out = (sense_buffer[3] << 24) +
1986 (sense_buffer[4] << 16) +
1987 (sense_buffer[5] << 8) + sense_buffer[6];
1988 return 1;
1989 } else
1990 return 0;
1991 case 0x72:
1992 case 0x73:
1993 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
1994 0 /* info desc */);
1995 if (ucp && (0xa == ucp[1])) {
1996 ull = 0;
1997 for (j = 0; j < 8; ++j) {
1998 if (j > 0)
1999 ull <<= 8;
2000 ull |= ucp[4 + j];
2001 }
2002 *info_out = ull;
2003 return 1;
2004 } else
2005 return 0;
2006 default:
2007 return 0;
2008 }
2009 }
2010 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2011
2012 /**
2013 * scsi_build_sense_buffer - build sense data in a buffer
2014 * @desc: Sense format (non zero == descriptor format,
2015 * 0 == fixed format)
2016 * @buf: Where to build sense data
2017 * @key: Sense key
2018 * @asc: Additional sense code
2019 * @ascq: Additional sense code qualifier
2020 *
2021 **/
2022 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2023 {
2024 if (desc) {
2025 buf[0] = 0x72; /* descriptor, current */
2026 buf[1] = key;
2027 buf[2] = asc;
2028 buf[3] = ascq;
2029 buf[7] = 0;
2030 } else {
2031 buf[0] = 0x70; /* fixed, current */
2032 buf[2] = key;
2033 buf[7] = 0xa;
2034 buf[12] = asc;
2035 buf[13] = ascq;
2036 }
2037 }
2038 EXPORT_SYMBOL(scsi_build_sense_buffer);
2039
|
This page was automatically generated by the
LXR engine.
|