aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/scsi/scsi_error.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/scsi/scsi_error.c')
-rw-r--r--drivers/scsi/scsi_error.c2050
1 files changed, 2050 insertions, 0 deletions
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
new file mode 100644
index 000000000000..9bc597bd13ba
--- /dev/null
+++ b/drivers/scsi/scsi_error.c
@@ -0,0 +1,2050 @@
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 */
49void 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 **/
66int 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 **/
106void 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}
128EXPORT_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 **/
141int 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}
156EXPORT_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 **/
168void 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 **/
206int 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}
219EXPORT_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 **/
227static 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 **/
279static 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 HARDWARE_ERROR:
354 if (scmd->device->retry_hwerror)
355 return NEEDS_RETRY;
356 else
357 return SUCCESS;
358
359 case ILLEGAL_REQUEST:
360 case BLANK_CHECK:
361 case DATA_PROTECT:
362 default:
363 return SUCCESS;
364 }
365}
366
367/**
368 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
369 * @scmd: SCSI cmd to examine.
370 *
371 * Notes:
372 * This is *only* called when we are examining the status of commands
373 * queued during error recovery. the main difference here is that we
374 * don't allow for the possibility of retries here, and we are a lot
375 * more restrictive about what we consider acceptable.
376 **/
377static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
378{
379 /*
380 * first check the host byte, to see if there is anything in there
381 * that would indicate what we need to do.
382 */
383 if (host_byte(scmd->result) == DID_RESET) {
384 /*
385 * rats. we are already in the error handler, so we now
386 * get to try and figure out what to do next. if the sense
387 * is valid, we have a pretty good idea of what to do.
388 * if not, we mark it as FAILED.
389 */
390 return scsi_check_sense(scmd);
391 }
392 if (host_byte(scmd->result) != DID_OK)
393 return FAILED;
394
395 /*
396 * next, check the message byte.
397 */
398 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
399 return FAILED;
400
401 /*
402 * now, check the status byte to see if this indicates
403 * anything special.
404 */
405 switch (status_byte(scmd->result)) {
406 case GOOD:
407 case COMMAND_TERMINATED:
408 return SUCCESS;
409 case CHECK_CONDITION:
410 return scsi_check_sense(scmd);
411 case CONDITION_GOOD:
412 case INTERMEDIATE_GOOD:
413 case INTERMEDIATE_C_GOOD:
414 /*
415 * who knows? FIXME(eric)
416 */
417 return SUCCESS;
418 case BUSY:
419 case QUEUE_FULL:
420 case RESERVATION_CONFLICT:
421 default:
422 return FAILED;
423 }
424 return FAILED;
425}
426
427/**
428 * scsi_eh_times_out - timeout function for error handling.
429 * @scmd: Cmd that is timing out.
430 *
431 * Notes:
432 * During error handling, the kernel thread will be sleeping waiting
433 * for some action to complete on the device. our only job is to
434 * record that it timed out, and to wake up the thread.
435 **/
436static void scsi_eh_times_out(struct scsi_cmnd *scmd)
437{
438 scsi_eh_eflags_set(scmd, SCSI_EH_REC_TIMEOUT);
439 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd:%p\n", __FUNCTION__,
440 scmd));
441
442 if (scmd->device->host->eh_action)
443 up(scmd->device->host->eh_action);
444}
445
446/**
447 * scsi_eh_done - Completion function for error handling.
448 * @scmd: Cmd that is done.
449 **/
450static void scsi_eh_done(struct scsi_cmnd *scmd)
451{
452 /*
453 * if the timeout handler is already running, then just set the
454 * flag which says we finished late, and return. we have no
455 * way of stopping the timeout handler from running, so we must
456 * always defer to it.
457 */
458 if (del_timer(&scmd->eh_timeout)) {
459 scmd->request->rq_status = RQ_SCSI_DONE;
460 scmd->owner = SCSI_OWNER_ERROR_HANDLER;
461
462 SCSI_LOG_ERROR_RECOVERY(3, printk("%s scmd: %p result: %x\n",
463 __FUNCTION__, scmd, scmd->result));
464
465 if (scmd->device->host->eh_action)
466 up(scmd->device->host->eh_action);
467 }
468}
469
470/**
471 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery.
472 * @scmd: SCSI Cmd to send.
473 * @timeout: Timeout for cmd.
474 *
475 * Notes:
476 * The initialization of the structures is quite a bit different in
477 * this case, and furthermore, there is a different completion handler
478 * vs scsi_dispatch_cmd.
479 * Return value:
480 * SUCCESS or FAILED or NEEDS_RETRY
481 **/
482static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout)
483{
484 struct Scsi_Host *host = scmd->device->host;
485 DECLARE_MUTEX_LOCKED(sem);
486 unsigned long flags;
487 int rtn = SUCCESS;
488
489 /*
490 * we will use a queued command if possible, otherwise we will
491 * emulate the queuing and calling of completion function ourselves.
492 */
493 scmd->owner = SCSI_OWNER_LOWLEVEL;
494
495 if (scmd->device->scsi_level <= SCSI_2)
496 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
497 (scmd->device->lun << 5 & 0xe0);
498
499 scsi_add_timer(scmd, timeout, scsi_eh_times_out);
500
501 /*
502 * set up the semaphore so we wait for the command to complete.
503 */
504 scmd->device->host->eh_action = &sem;
505 scmd->request->rq_status = RQ_SCSI_BUSY;
506
507 spin_lock_irqsave(scmd->device->host->host_lock, flags);
508 scsi_log_send(scmd);
509 host->hostt->queuecommand(scmd, scsi_eh_done);
510 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
511
512 down(&sem);
513 scsi_log_completion(scmd, SUCCESS);
514
515 scmd->device->host->eh_action = NULL;
516
517 /*
518 * see if timeout. if so, tell the host to forget about it.
519 * in other words, we don't want a callback any more.
520 */
521 if (scsi_eh_eflags_chk(scmd, SCSI_EH_REC_TIMEOUT)) {
522 scsi_eh_eflags_clr(scmd, SCSI_EH_REC_TIMEOUT);
523 scmd->owner = SCSI_OWNER_LOWLEVEL;
524
525 /*
526 * as far as the low level driver is
527 * concerned, this command is still active, so
528 * we must give the low level driver a chance
529 * to abort it. (db)
530 *
531 * FIXME(eric) - we are not tracking whether we could
532 * abort a timed out command or not. not sure how
533 * we should treat them differently anyways.
534 */
535 spin_lock_irqsave(scmd->device->host->host_lock, flags);
536 if (scmd->device->host->hostt->eh_abort_handler)
537 scmd->device->host->hostt->eh_abort_handler(scmd);
538 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
539
540 scmd->request->rq_status = RQ_SCSI_DONE;
541 scmd->owner = SCSI_OWNER_ERROR_HANDLER;
542
543 rtn = FAILED;
544 }
545
546 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd: %p, rtn:%x\n",
547 __FUNCTION__, scmd, rtn));
548
549 /*
550 * now examine the actual status codes to see whether the command
551 * actually did complete normally.
552 */
553 if (rtn == SUCCESS) {
554 rtn = scsi_eh_completed_normally(scmd);
555 SCSI_LOG_ERROR_RECOVERY(3,
556 printk("%s: scsi_eh_completed_normally %x\n",
557 __FUNCTION__, rtn));
558 switch (rtn) {
559 case SUCCESS:
560 case NEEDS_RETRY:
561 case FAILED:
562 break;
563 default:
564 rtn = FAILED;
565 break;
566 }
567 }
568
569 return rtn;
570}
571
572/**
573 * scsi_request_sense - Request sense data from a particular target.
574 * @scmd: SCSI cmd for request sense.
575 *
576 * Notes:
577 * Some hosts automatically obtain this information, others require
578 * that we obtain it on our own. This function will *not* return until
579 * the command either times out, or it completes.
580 **/
581static int scsi_request_sense(struct scsi_cmnd *scmd)
582{
583 static unsigned char generic_sense[6] =
584 {REQUEST_SENSE, 0, 0, 0, 252, 0};
585 unsigned char *scsi_result;
586 int saved_result;
587 int rtn;
588
589 memcpy(scmd->cmnd, generic_sense, sizeof(generic_sense));
590
591 scsi_result = kmalloc(252, GFP_ATOMIC | (scmd->device->host->hostt->unchecked_isa_dma) ? __GFP_DMA : 0);
592
593
594 if (unlikely(!scsi_result)) {
595 printk(KERN_ERR "%s: cannot allocate scsi_result.\n",
596 __FUNCTION__);
597 return FAILED;
598 }
599
600 /*
601 * zero the sense buffer. some host adapters automatically always
602 * request sense, so it is not a good idea that
603 * scmd->request_buffer and scmd->sense_buffer point to the same
604 * address (db). 0 is not a valid sense code.
605 */
606 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
607 memset(scsi_result, 0, 252);
608
609 saved_result = scmd->result;
610 scmd->request_buffer = scsi_result;
611 scmd->request_bufflen = 252;
612 scmd->use_sg = 0;
613 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
614 scmd->sc_data_direction = DMA_FROM_DEVICE;
615 scmd->underflow = 0;
616
617 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT);
618
619 /* last chance to have valid sense data */
620 if(!SCSI_SENSE_VALID(scmd)) {
621 memcpy(scmd->sense_buffer, scmd->request_buffer,
622 sizeof(scmd->sense_buffer));
623 }
624
625 kfree(scsi_result);
626
627 /*
628 * when we eventually call scsi_finish, we really wish to complete
629 * the original request, so let's restore the original data. (db)
630 */
631 scsi_setup_cmd_retry(scmd);
632 scmd->result = saved_result;
633 return rtn;
634}
635
636/**
637 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
638 * @scmd: Original SCSI cmd that eh has finished.
639 * @done_q: Queue for processed commands.
640 *
641 * Notes:
642 * We don't want to use the normal command completion while we are are
643 * still handling errors - it may cause other commands to be queued,
644 * and that would disturb what we are doing. thus we really want to
645 * keep a list of pending commands for final completion, and once we
646 * are ready to leave error handling we handle completion for real.
647 **/
648static void scsi_eh_finish_cmd(struct scsi_cmnd *scmd,
649 struct list_head *done_q)
650{
651 scmd->device->host->host_failed--;
652 scmd->state = SCSI_STATE_BHQUEUE;
653
654 scsi_eh_eflags_clr_all(scmd);
655
656 /*
657 * set this back so that the upper level can correctly free up
658 * things.
659 */
660 scsi_setup_cmd_retry(scmd);
661 list_move_tail(&scmd->eh_entry, done_q);
662}
663
664/**
665 * scsi_eh_get_sense - Get device sense data.
666 * @work_q: Queue of commands to process.
667 * @done_q: Queue of proccessed commands..
668 *
669 * Description:
670 * See if we need to request sense information. if so, then get it
671 * now, so we have a better idea of what to do.
672 *
673 * Notes:
674 * This has the unfortunate side effect that if a shost adapter does
675 * not automatically request sense information, that we end up shutting
676 * it down before we request it.
677 *
678 * All drivers should request sense information internally these days,
679 * so for now all I have to say is tough noogies if you end up in here.
680 *
681 * XXX: Long term this code should go away, but that needs an audit of
682 * all LLDDs first.
683 **/
684static int scsi_eh_get_sense(struct list_head *work_q,
685 struct list_head *done_q)
686{
687 struct list_head *lh, *lh_sf;
688 struct scsi_cmnd *scmd;
689 int rtn;
690
691 list_for_each_safe(lh, lh_sf, work_q) {
692 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
693 if (scsi_eh_eflags_chk(scmd, SCSI_EH_CANCEL_CMD) ||
694 SCSI_SENSE_VALID(scmd))
695 continue;
696
697 SCSI_LOG_ERROR_RECOVERY(2, printk("%s: requesting sense"
698 " for id: %d\n",
699 current->comm,
700 scmd->device->id));
701 rtn = scsi_request_sense(scmd);
702 if (rtn != SUCCESS)
703 continue;
704
705 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
706 " result %x\n", scmd,
707 scmd->result));
708 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
709
710 rtn = scsi_decide_disposition(scmd);
711
712 /*
713 * if the result was normal, then just pass it along to the
714 * upper level.
715 */
716 if (rtn == SUCCESS)
717 /* we don't want this command reissued, just
718 * finished with the sense data, so set
719 * retries to the max allowed to ensure it
720 * won't get reissued */
721 scmd->retries = scmd->allowed;
722 else if (rtn != NEEDS_RETRY)
723 continue;
724
725 scsi_eh_finish_cmd(scmd, done_q);
726 }
727
728 return list_empty(work_q);
729}
730
731/**
732 * scsi_try_to_abort_cmd - Ask host to abort a running command.
733 * @scmd: SCSI cmd to abort from Lower Level.
734 *
735 * Notes:
736 * This function will not return until the user's completion function
737 * has been called. there is no timeout on this operation. if the
738 * author of the low-level driver wishes this operation to be timed,
739 * they can provide this facility themselves. helper functions in
740 * scsi_error.c can be supplied to make this easier to do.
741 **/
742static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
743{
744 unsigned long flags;
745 int rtn = FAILED;
746
747 if (!scmd->device->host->hostt->eh_abort_handler)
748 return rtn;
749
750 /*
751 * scsi_done was called just after the command timed out and before
752 * we had a chance to process it. (db)
753 */
754 if (scmd->serial_number == 0)
755 return SUCCESS;
756
757 scmd->owner = SCSI_OWNER_LOWLEVEL;
758
759 spin_lock_irqsave(scmd->device->host->host_lock, flags);
760 rtn = scmd->device->host->hostt->eh_abort_handler(scmd);
761 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
762
763 return rtn;
764}
765
766/**
767 * scsi_eh_tur - Send TUR to device.
768 * @scmd: Scsi cmd to send TUR
769 *
770 * Return value:
771 * 0 - Device is ready. 1 - Device NOT ready.
772 **/
773static int scsi_eh_tur(struct scsi_cmnd *scmd)
774{
775 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
776 int retry_cnt = 1, rtn;
777
778retry_tur:
779 memcpy(scmd->cmnd, tur_command, sizeof(tur_command));
780
781 /*
782 * zero the sense buffer. the scsi spec mandates that any
783 * untransferred sense data should be interpreted as being zero.
784 */
785 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
786
787 scmd->request_buffer = NULL;
788 scmd->request_bufflen = 0;
789 scmd->use_sg = 0;
790 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
791 scmd->underflow = 0;
792 scmd->sc_data_direction = DMA_NONE;
793
794 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT);
795
796 /*
797 * when we eventually call scsi_finish, we really wish to complete
798 * the original request, so let's restore the original data. (db)
799 */
800 scsi_setup_cmd_retry(scmd);
801
802 /*
803 * hey, we are done. let's look to see what happened.
804 */
805 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
806 __FUNCTION__, scmd, rtn));
807 if (rtn == SUCCESS)
808 return 0;
809 else if (rtn == NEEDS_RETRY)
810 if (retry_cnt--)
811 goto retry_tur;
812 return 1;
813}
814
815/**
816 * scsi_eh_abort_cmds - abort canceled commands.
817 * @shost: scsi host being recovered.
818 * @eh_done_q: list_head for processed commands.
819 *
820 * Decription:
821 * Try and see whether or not it makes sense to try and abort the
822 * running command. this only works out to be the case if we have one
823 * command that has timed out. if the command simply failed, it makes
824 * no sense to try and abort the command, since as far as the shost
825 * adapter is concerned, it isn't running.
826 **/
827static int scsi_eh_abort_cmds(struct list_head *work_q,
828 struct list_head *done_q)
829{
830 struct list_head *lh, *lh_sf;
831 struct scsi_cmnd *scmd;
832 int rtn;
833
834 list_for_each_safe(lh, lh_sf, work_q) {
835 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
836 if (!scsi_eh_eflags_chk(scmd, SCSI_EH_CANCEL_CMD))
837 continue;
838 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
839 "0x%p\n", current->comm,
840 scmd));
841 rtn = scsi_try_to_abort_cmd(scmd);
842 if (rtn == SUCCESS) {
843 scsi_eh_eflags_clr(scmd, SCSI_EH_CANCEL_CMD);
844 if (!scsi_device_online(scmd->device) ||
845 !scsi_eh_tur(scmd)) {
846 scsi_eh_finish_cmd(scmd, done_q);
847 }
848
849 } else
850 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
851 " cmd failed:"
852 "0x%p\n",
853 current->comm,
854 scmd));
855 }
856
857 return list_empty(work_q);
858}
859
860/**
861 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
862 * @scmd: SCSI cmd used to send BDR
863 *
864 * Notes:
865 * There is no timeout for this operation. if this operation is
866 * unreliable for a given host, then the host itself needs to put a
867 * timer on it, and set the host back to a consistent state prior to
868 * returning.
869 **/
870static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
871{
872 unsigned long flags;
873 int rtn = FAILED;
874
875 if (!scmd->device->host->hostt->eh_device_reset_handler)
876 return rtn;
877
878 scmd->owner = SCSI_OWNER_LOWLEVEL;
879
880 spin_lock_irqsave(scmd->device->host->host_lock, flags);
881 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
882 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
883
884 if (rtn == SUCCESS) {
885 scmd->device->was_reset = 1;
886 scmd->device->expecting_cc_ua = 1;
887 }
888
889 return rtn;
890}
891
892/**
893 * scsi_eh_try_stu - Send START_UNIT to device.
894 * @scmd: Scsi cmd to send START_UNIT
895 *
896 * Return value:
897 * 0 - Device is ready. 1 - Device NOT ready.
898 **/
899static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
900{
901 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
902 int rtn;
903
904 if (!scmd->device->allow_restart)
905 return 1;
906
907 memcpy(scmd->cmnd, stu_command, sizeof(stu_command));
908
909 /*
910 * zero the sense buffer. the scsi spec mandates that any
911 * untransferred sense data should be interpreted as being zero.
912 */
913 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
914
915 scmd->request_buffer = NULL;
916 scmd->request_bufflen = 0;
917 scmd->use_sg = 0;
918 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
919 scmd->underflow = 0;
920 scmd->sc_data_direction = DMA_NONE;
921
922 rtn = scsi_send_eh_cmnd(scmd, START_UNIT_TIMEOUT);
923
924 /*
925 * when we eventually call scsi_finish, we really wish to complete
926 * the original request, so let's restore the original data. (db)
927 */
928 scsi_setup_cmd_retry(scmd);
929
930 /*
931 * hey, we are done. let's look to see what happened.
932 */
933 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
934 __FUNCTION__, scmd, rtn));
935 if (rtn == SUCCESS)
936 return 0;
937 return 1;
938}
939
940 /**
941 * scsi_eh_stu - send START_UNIT if needed
942 * @shost: scsi host being recovered.
943 * @eh_done_q: list_head for processed commands.
944 *
945 * Notes:
946 * If commands are failing due to not ready, initializing command required,
947 * try revalidating the device, which will end up sending a start unit.
948 **/
949static int scsi_eh_stu(struct Scsi_Host *shost,
950 struct list_head *work_q,
951 struct list_head *done_q)
952{
953 struct list_head *lh, *lh_sf;
954 struct scsi_cmnd *scmd, *stu_scmd;
955 struct scsi_device *sdev;
956
957 shost_for_each_device(sdev, shost) {
958 stu_scmd = NULL;
959 list_for_each_entry(scmd, work_q, eh_entry)
960 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
961 scsi_check_sense(scmd) == FAILED ) {
962 stu_scmd = scmd;
963 break;
964 }
965
966 if (!stu_scmd)
967 continue;
968
969 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
970 " 0x%p\n", current->comm, sdev));
971
972 if (!scsi_eh_try_stu(stu_scmd)) {
973 if (!scsi_device_online(sdev) ||
974 !scsi_eh_tur(stu_scmd)) {
975 list_for_each_safe(lh, lh_sf, work_q) {
976 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
977 if (scmd->device == sdev)
978 scsi_eh_finish_cmd(scmd, done_q);
979 }
980 }
981 } else {
982 SCSI_LOG_ERROR_RECOVERY(3,
983 printk("%s: START_UNIT failed to sdev:"
984 " 0x%p\n", current->comm, sdev));
985 }
986 }
987
988 return list_empty(work_q);
989}
990
991
992/**
993 * scsi_eh_bus_device_reset - send bdr if needed
994 * @shost: scsi host being recovered.
995 * @eh_done_q: list_head for processed commands.
996 *
997 * Notes:
998 * Try a bus device reset. still, look to see whether we have multiple
999 * devices that are jammed or not - if we have multiple devices, it
1000 * makes no sense to try bus_device_reset - we really would need to try
1001 * a bus_reset instead.
1002 **/
1003static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1004 struct list_head *work_q,
1005 struct list_head *done_q)
1006{
1007 struct list_head *lh, *lh_sf;
1008 struct scsi_cmnd *scmd, *bdr_scmd;
1009 struct scsi_device *sdev;
1010 int rtn;
1011
1012 shost_for_each_device(sdev, shost) {
1013 bdr_scmd = NULL;
1014 list_for_each_entry(scmd, work_q, eh_entry)
1015 if (scmd->device == sdev) {
1016 bdr_scmd = scmd;
1017 break;
1018 }
1019
1020 if (!bdr_scmd)
1021 continue;
1022
1023 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1024 " 0x%p\n", current->comm,
1025 sdev));
1026 rtn = scsi_try_bus_device_reset(bdr_scmd);
1027 if (rtn == SUCCESS) {
1028 if (!scsi_device_online(sdev) ||
1029 !scsi_eh_tur(bdr_scmd)) {
1030 list_for_each_safe(lh, lh_sf,
1031 work_q) {
1032 scmd = list_entry(lh, struct
1033 scsi_cmnd,
1034 eh_entry);
1035 if (scmd->device == sdev)
1036 scsi_eh_finish_cmd(scmd,
1037 done_q);
1038 }
1039 }
1040 } else {
1041 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1042 " failed sdev:"
1043 "0x%p\n",
1044 current->comm,
1045 sdev));
1046 }
1047 }
1048
1049 return list_empty(work_q);
1050}
1051
1052/**
1053 * scsi_try_bus_reset - ask host to perform a bus reset
1054 * @scmd: SCSI cmd to send bus reset.
1055 **/
1056static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
1057{
1058 unsigned long flags;
1059 int rtn;
1060
1061 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
1062 __FUNCTION__));
1063 scmd->owner = SCSI_OWNER_LOWLEVEL;
1064 scmd->serial_number_at_timeout = scmd->serial_number;
1065
1066 if (!scmd->device->host->hostt->eh_bus_reset_handler)
1067 return FAILED;
1068
1069 spin_lock_irqsave(scmd->device->host->host_lock, flags);
1070 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
1071 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
1072
1073 if (rtn == SUCCESS) {
1074 if (!scmd->device->host->hostt->skip_settle_delay)
1075 ssleep(BUS_RESET_SETTLE_TIME);
1076 spin_lock_irqsave(scmd->device->host->host_lock, flags);
1077 scsi_report_bus_reset(scmd->device->host, scmd->device->channel);
1078 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
1079 }
1080
1081 return rtn;
1082}
1083
1084/**
1085 * scsi_try_host_reset - ask host adapter to reset itself
1086 * @scmd: SCSI cmd to send hsot reset.
1087 **/
1088static int scsi_try_host_reset(struct scsi_cmnd *scmd)
1089{
1090 unsigned long flags;
1091 int rtn;
1092
1093 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
1094 __FUNCTION__));
1095 scmd->owner = SCSI_OWNER_LOWLEVEL;
1096 scmd->serial_number_at_timeout = scmd->serial_number;
1097
1098 if (!scmd->device->host->hostt->eh_host_reset_handler)
1099 return FAILED;
1100
1101 spin_lock_irqsave(scmd->device->host->host_lock, flags);
1102 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
1103 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
1104
1105 if (rtn == SUCCESS) {
1106 if (!scmd->device->host->hostt->skip_settle_delay)
1107 ssleep(HOST_RESET_SETTLE_TIME);
1108 spin_lock_irqsave(scmd->device->host->host_lock, flags);
1109 scsi_report_bus_reset(scmd->device->host, scmd->device->channel);
1110 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
1111 }
1112
1113 return rtn;
1114}
1115
1116/**
1117 * scsi_eh_bus_reset - send a bus reset
1118 * @shost: scsi host being recovered.
1119 * @eh_done_q: list_head for processed commands.
1120 **/
1121static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1122 struct list_head *work_q,
1123 struct list_head *done_q)
1124{
1125 struct list_head *lh, *lh_sf;
1126 struct scsi_cmnd *scmd;
1127 struct scsi_cmnd *chan_scmd;
1128 unsigned int channel;
1129 int rtn;
1130
1131 /*
1132 * we really want to loop over the various channels, and do this on
1133 * a channel by channel basis. we should also check to see if any
1134 * of the failed commands are on soft_reset devices, and if so, skip
1135 * the reset.
1136 */
1137
1138 for (channel = 0; channel <= shost->max_channel; channel++) {
1139 chan_scmd = NULL;
1140 list_for_each_entry(scmd, work_q, eh_entry) {
1141 if (channel == scmd->device->channel) {
1142 chan_scmd = scmd;
1143 break;
1144 /*
1145 * FIXME add back in some support for
1146 * soft_reset devices.
1147 */
1148 }
1149 }
1150
1151 if (!chan_scmd)
1152 continue;
1153 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1154 " %d\n", current->comm,
1155 channel));
1156 rtn = scsi_try_bus_reset(chan_scmd);
1157 if (rtn == SUCCESS) {
1158 list_for_each_safe(lh, lh_sf, work_q) {
1159 scmd = list_entry(lh, struct scsi_cmnd,
1160 eh_entry);
1161 if (channel == scmd->device->channel)
1162 if (!scsi_device_online(scmd->device) ||
1163 !scsi_eh_tur(scmd))
1164 scsi_eh_finish_cmd(scmd,
1165 done_q);
1166 }
1167 } else {
1168 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1169 " failed chan: %d\n",
1170 current->comm,
1171 channel));
1172 }
1173 }
1174 return list_empty(work_q);
1175}
1176
1177/**
1178 * scsi_eh_host_reset - send a host reset
1179 * @work_q: list_head for processed commands.
1180 * @done_q: list_head for processed commands.
1181 **/
1182static int scsi_eh_host_reset(struct list_head *work_q,
1183 struct list_head *done_q)
1184{
1185 int rtn;
1186 struct list_head *lh, *lh_sf;
1187 struct scsi_cmnd *scmd;
1188
1189 if (!list_empty(work_q)) {
1190 scmd = list_entry(work_q->next,
1191 struct scsi_cmnd, eh_entry);
1192
1193 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1194 , current->comm));
1195
1196 rtn = scsi_try_host_reset(scmd);
1197 if (rtn == SUCCESS) {
1198 list_for_each_safe(lh, lh_sf, work_q) {
1199 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
1200 if (!scsi_device_online(scmd->device) ||
1201 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1202 !scsi_eh_tur(scmd))
1203 scsi_eh_finish_cmd(scmd, done_q);
1204 }
1205 } else {
1206 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1207 " failed\n",
1208 current->comm));
1209 }
1210 }
1211 return list_empty(work_q);
1212}
1213
1214/**
1215 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1216 * @work_q: list_head for processed commands.
1217 * @done_q: list_head for processed commands.
1218 *
1219 **/
1220static void scsi_eh_offline_sdevs(struct list_head *work_q,
1221 struct list_head *done_q)
1222{
1223 struct list_head *lh, *lh_sf;
1224 struct scsi_cmnd *scmd;
1225
1226 list_for_each_safe(lh, lh_sf, work_q) {
1227 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
1228 printk(KERN_INFO "scsi: Device offlined - not"
1229 " ready after error recovery: host"
1230 " %d channel %d id %d lun %d\n",
1231 scmd->device->host->host_no,
1232 scmd->device->channel,
1233 scmd->device->id,
1234 scmd->device->lun);
1235 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1236 if (scsi_eh_eflags_chk(scmd, SCSI_EH_CANCEL_CMD)) {
1237 /*
1238 * FIXME: Handle lost cmds.
1239 */
1240 }
1241 scsi_eh_finish_cmd(scmd, done_q);
1242 }
1243 return;
1244}
1245
1246/**
1247 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1248 * @scmd: SCSI cmd to examine.
1249 *
1250 * Notes:
1251 * This is *only* called when we are examining the status after sending
1252 * out the actual data command. any commands that are queued for error
1253 * recovery (e.g. test_unit_ready) do *not* come through here.
1254 *
1255 * When this routine returns failed, it means the error handler thread
1256 * is woken. In cases where the error code indicates an error that
1257 * doesn't require the error handler read (i.e. we don't need to
1258 * abort/reset), this function should return SUCCESS.
1259 **/
1260int scsi_decide_disposition(struct scsi_cmnd *scmd)
1261{
1262 int rtn;
1263
1264 /*
1265 * if the device is offline, then we clearly just pass the result back
1266 * up to the top level.
1267 */
1268 if (!scsi_device_online(scmd->device)) {
1269 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1270 " as SUCCESS\n",
1271 __FUNCTION__));
1272 return SUCCESS;
1273 }
1274
1275 /*
1276 * first check the host byte, to see if there is anything in there
1277 * that would indicate what we need to do.
1278 */
1279 switch (host_byte(scmd->result)) {
1280 case DID_PASSTHROUGH:
1281 /*
1282 * no matter what, pass this through to the upper layer.
1283 * nuke this special code so that it looks like we are saying
1284 * did_ok.
1285 */
1286 scmd->result &= 0xff00ffff;
1287 return SUCCESS;
1288 case DID_OK:
1289 /*
1290 * looks good. drop through, and check the next byte.
1291 */
1292 break;
1293 case DID_NO_CONNECT:
1294 case DID_BAD_TARGET:
1295 case DID_ABORT:
1296 /*
1297 * note - this means that we just report the status back
1298 * to the top level driver, not that we actually think
1299 * that it indicates SUCCESS.
1300 */
1301 return SUCCESS;
1302 /*
1303 * when the low level driver returns did_soft_error,
1304 * it is responsible for keeping an internal retry counter
1305 * in order to avoid endless loops (db)
1306 *
1307 * actually this is a bug in this function here. we should
1308 * be mindful of the maximum number of retries specified
1309 * and not get stuck in a loop.
1310 */
1311 case DID_SOFT_ERROR:
1312 goto maybe_retry;
1313 case DID_IMM_RETRY:
1314 return NEEDS_RETRY;
1315
1316 case DID_ERROR:
1317 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1318 status_byte(scmd->result) == RESERVATION_CONFLICT)
1319 /*
1320 * execute reservation conflict processing code
1321 * lower down
1322 */
1323 break;
1324 /* fallthrough */
1325
1326 case DID_BUS_BUSY:
1327 case DID_PARITY:
1328 goto maybe_retry;
1329 case DID_TIME_OUT:
1330 /*
1331 * when we scan the bus, we get timeout messages for
1332 * these commands if there is no device available.
1333 * other hosts report did_no_connect for the same thing.
1334 */
1335 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1336 scmd->cmnd[0] == INQUIRY)) {
1337 return SUCCESS;
1338 } else {
1339 return FAILED;
1340 }
1341 case DID_RESET:
1342 return SUCCESS;
1343 default:
1344 return FAILED;
1345 }
1346
1347 /*
1348 * next, check the message byte.
1349 */
1350 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1351 return FAILED;
1352
1353 /*
1354 * check the status byte to see if this indicates anything special.
1355 */
1356 switch (status_byte(scmd->result)) {
1357 case QUEUE_FULL:
1358 /*
1359 * the case of trying to send too many commands to a
1360 * tagged queueing device.
1361 */
1362 case BUSY:
1363 /*
1364 * device can't talk to us at the moment. Should only
1365 * occur (SAM-3) when the task queue is empty, so will cause
1366 * the empty queue handling to trigger a stall in the
1367 * device.
1368 */
1369 return ADD_TO_MLQUEUE;
1370 case GOOD:
1371 case COMMAND_TERMINATED:
1372 case TASK_ABORTED:
1373 return SUCCESS;
1374 case CHECK_CONDITION:
1375 rtn = scsi_check_sense(scmd);
1376 if (rtn == NEEDS_RETRY)
1377 goto maybe_retry;
1378 /* if rtn == FAILED, we have no sense information;
1379 * returning FAILED will wake the error handler thread
1380 * to collect the sense and redo the decide
1381 * disposition */
1382 return rtn;
1383 case CONDITION_GOOD:
1384 case INTERMEDIATE_GOOD:
1385 case INTERMEDIATE_C_GOOD:
1386 case ACA_ACTIVE:
1387 /*
1388 * who knows? FIXME(eric)
1389 */
1390 return SUCCESS;
1391
1392 case RESERVATION_CONFLICT:
1393 printk(KERN_INFO "scsi: reservation conflict: host"
1394 " %d channel %d id %d lun %d\n",
1395 scmd->device->host->host_no, scmd->device->channel,
1396 scmd->device->id, scmd->device->lun);
1397 return SUCCESS; /* causes immediate i/o error */
1398 default:
1399 return FAILED;
1400 }
1401 return FAILED;
1402
1403 maybe_retry:
1404
1405 /* we requeue for retry because the error was retryable, and
1406 * the request was not marked fast fail. Note that above,
1407 * even if the request is marked fast fail, we still requeue
1408 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1409 if ((++scmd->retries) < scmd->allowed
1410 && !blk_noretry_request(scmd->request)) {
1411 return NEEDS_RETRY;
1412 } else {
1413 /*
1414 * no more retries - report this one back to upper level.
1415 */
1416 return SUCCESS;
1417 }
1418}
1419
1420/**
1421 * scsi_eh_lock_done - done function for eh door lock request
1422 * @scmd: SCSI command block for the door lock request
1423 *
1424 * Notes:
1425 * We completed the asynchronous door lock request, and it has either
1426 * locked the door or failed. We must free the command structures
1427 * associated with this request.
1428 **/
1429static void scsi_eh_lock_done(struct scsi_cmnd *scmd)
1430{
1431 struct scsi_request *sreq = scmd->sc_request;
1432
1433 scsi_release_request(sreq);
1434}
1435
1436
1437/**
1438 * scsi_eh_lock_door - Prevent medium removal for the specified device
1439 * @sdev: SCSI device to prevent medium removal
1440 *
1441 * Locking:
1442 * We must be called from process context; scsi_allocate_request()
1443 * may sleep.
1444 *
1445 * Notes:
1446 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1447 * head of the devices request queue, and continue.
1448 *
1449 * Bugs:
1450 * scsi_allocate_request() may sleep waiting for existing requests to
1451 * be processed. However, since we haven't kicked off any request
1452 * processing for this host, this may deadlock.
1453 *
1454 * If scsi_allocate_request() fails for what ever reason, we
1455 * completely forget to lock the door.
1456 **/
1457static void scsi_eh_lock_door(struct scsi_device *sdev)
1458{
1459 struct scsi_request *sreq = scsi_allocate_request(sdev, GFP_KERNEL);
1460
1461 if (unlikely(!sreq)) {
1462 printk(KERN_ERR "%s: request allocate failed,"
1463 "prevent media removal cmd not sent\n", __FUNCTION__);
1464 return;
1465 }
1466
1467 sreq->sr_cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1468 sreq->sr_cmnd[1] = 0;
1469 sreq->sr_cmnd[2] = 0;
1470 sreq->sr_cmnd[3] = 0;
1471 sreq->sr_cmnd[4] = SCSI_REMOVAL_PREVENT;
1472 sreq->sr_cmnd[5] = 0;
1473 sreq->sr_data_direction = DMA_NONE;
1474 sreq->sr_bufflen = 0;
1475 sreq->sr_buffer = NULL;
1476 sreq->sr_allowed = 5;
1477 sreq->sr_done = scsi_eh_lock_done;
1478 sreq->sr_timeout_per_command = 10 * HZ;
1479 sreq->sr_cmd_len = COMMAND_SIZE(sreq->sr_cmnd[0]);
1480
1481 scsi_insert_special_req(sreq, 1);
1482}
1483
1484
1485/**
1486 * scsi_restart_operations - restart io operations to the specified host.
1487 * @shost: Host we are restarting.
1488 *
1489 * Notes:
1490 * When we entered the error handler, we blocked all further i/o to
1491 * this device. we need to 'reverse' this process.
1492 **/
1493static void scsi_restart_operations(struct Scsi_Host *shost)
1494{
1495 struct scsi_device *sdev;
1496
1497 /*
1498 * If the door was locked, we need to insert a door lock request
1499 * onto the head of the SCSI request queue for the device. There
1500 * is no point trying to lock the door of an off-line device.
1501 */
1502 shost_for_each_device(sdev, shost) {
1503 if (scsi_device_online(sdev) && sdev->locked)
1504 scsi_eh_lock_door(sdev);
1505 }
1506
1507 /*
1508 * next free up anything directly waiting upon the host. this
1509 * will be requests for character device operations, and also for
1510 * ioctls to queued block devices.
1511 */
1512 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1513 __FUNCTION__));
1514
1515 clear_bit(SHOST_RECOVERY, &shost->shost_state);
1516
1517 wake_up(&shost->host_wait);
1518
1519 /*
1520 * finally we need to re-initiate requests that may be pending. we will
1521 * have had everything blocked while error handling is taking place, and
1522 * now that error recovery is done, we will need to ensure that these
1523 * requests are started.
1524 */
1525 scsi_run_host_queues(shost);
1526}
1527
1528/**
1529 * scsi_eh_ready_devs - check device ready state and recover if not.
1530 * @shost: host to be recovered.
1531 * @eh_done_q: list_head for processed commands.
1532 *
1533 **/
1534static void scsi_eh_ready_devs(struct Scsi_Host *shost,
1535 struct list_head *work_q,
1536 struct list_head *done_q)
1537{
1538 if (!scsi_eh_stu(shost, work_q, done_q))
1539 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1540 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1541 if (!scsi_eh_host_reset(work_q, done_q))
1542 scsi_eh_offline_sdevs(work_q, done_q);
1543}
1544
1545/**
1546 * scsi_eh_flush_done_q - finish processed commands or retry them.
1547 * @done_q: list_head of processed commands.
1548 *
1549 **/
1550static void scsi_eh_flush_done_q(struct list_head *done_q)
1551{
1552 struct list_head *lh, *lh_sf;
1553 struct scsi_cmnd *scmd;
1554
1555 list_for_each_safe(lh, lh_sf, done_q) {
1556 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
1557 list_del_init(lh);
1558 if (scsi_device_online(scmd->device) &&
1559 !blk_noretry_request(scmd->request) &&
1560 (++scmd->retries < scmd->allowed)) {
1561 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1562 " retry cmd: %p\n",
1563 current->comm,
1564 scmd));
1565 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1566 } else {
1567 if (!scmd->result)
1568 scmd->result |= (DRIVER_TIMEOUT << 24);
1569 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1570 " cmd: %p\n",
1571 current->comm, scmd));
1572 scsi_finish_command(scmd);
1573 }
1574 }
1575}
1576
1577/**
1578 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1579 * @shost: Host to unjam.
1580 *
1581 * Notes:
1582 * When we come in here, we *know* that all commands on the bus have
1583 * either completed, failed or timed out. we also know that no further
1584 * commands are being sent to the host, so things are relatively quiet
1585 * and we have freedom to fiddle with things as we wish.
1586 *
1587 * This is only the *default* implementation. it is possible for
1588 * individual drivers to supply their own version of this function, and
1589 * if the maintainer wishes to do this, it is strongly suggested that
1590 * this function be taken as a template and modified. this function
1591 * was designed to correctly handle problems for about 95% of the
1592 * different cases out there, and it should always provide at least a
1593 * reasonable amount of error recovery.
1594 *
1595 * Any command marked 'failed' or 'timeout' must eventually have
1596 * scsi_finish_cmd() called for it. we do all of the retry stuff
1597 * here, so when we restart the host after we return it should have an
1598 * empty queue.
1599 **/
1600static void scsi_unjam_host(struct Scsi_Host *shost)
1601{
1602 unsigned long flags;
1603 LIST_HEAD(eh_work_q);
1604 LIST_HEAD(eh_done_q);
1605
1606 spin_lock_irqsave(shost->host_lock, flags);
1607 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1608 spin_unlock_irqrestore(shost->host_lock, flags);
1609
1610 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1611
1612 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1613 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1614 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1615
1616 scsi_eh_flush_done_q(&eh_done_q);
1617}
1618
1619/**
1620 * scsi_error_handler - Handle errors/timeouts of SCSI cmds.
1621 * @data: Host for which we are running.
1622 *
1623 * Notes:
1624 * This is always run in the context of a kernel thread. The idea is
1625 * that we start this thing up when the kernel starts up (one per host
1626 * that we detect), and it immediately goes to sleep and waits for some
1627 * event (i.e. failure). When this takes place, we have the job of
1628 * trying to unjam the bus and restarting things.
1629 **/
1630int scsi_error_handler(void *data)
1631{
1632 struct Scsi_Host *shost = (struct Scsi_Host *) data;
1633 int rtn;
1634 DECLARE_MUTEX_LOCKED(sem);
1635
1636 /*
1637 * Flush resources
1638 */
1639
1640 daemonize("scsi_eh_%d", shost->host_no);
1641
1642 current->flags |= PF_NOFREEZE;
1643
1644 shost->eh_wait = &sem;
1645 shost->ehandler = current;
1646
1647 /*
1648 * Wake up the thread that created us.
1649 */
1650 SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent of"
1651 " scsi_eh_%d\n",shost->host_no));
1652
1653 complete(shost->eh_notify);
1654
1655 while (1) {
1656 /*
1657 * If we get a signal, it means we are supposed to go
1658 * away and die. This typically happens if the user is
1659 * trying to unload a module.
1660 */
1661 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1662 " scsi_eh_%d"
1663 " sleeping\n",shost->host_no));
1664
1665 /*
1666 * Note - we always use down_interruptible with the semaphore
1667 * even if the module was loaded as part of the kernel. The
1668 * reason is that down() will cause this thread to be counted
1669 * in the load average as a running process, and down
1670 * interruptible doesn't. Given that we need to allow this
1671 * thread to die if the driver was loaded as a module, using
1672 * semaphores isn't unreasonable.
1673 */
1674 down_interruptible(&sem);
1675 if (shost->eh_kill)
1676 break;
1677
1678 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1679 " scsi_eh_%d waking"
1680 " up\n",shost->host_no));
1681
1682 shost->eh_active = 1;
1683
1684 /*
1685 * We have a host that is failing for some reason. Figure out
1686 * what we need to do to get it up and online again (if we can).
1687 * If we fail, we end up taking the thing offline.
1688 */
1689 if (shost->hostt->eh_strategy_handler)
1690 rtn = shost->hostt->eh_strategy_handler(shost);
1691 else
1692 scsi_unjam_host(shost);
1693
1694 shost->eh_active = 0;
1695
1696 /*
1697 * Note - if the above fails completely, the action is to take
1698 * individual devices offline and flush the queue of any
1699 * outstanding requests that may have been pending. When we
1700 * restart, we restart any I/O to any other devices on the bus
1701 * which are still online.
1702 */
1703 scsi_restart_operations(shost);
1704
1705 }
1706
1707 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler scsi_eh_%d"
1708 " exiting\n",shost->host_no));
1709
1710 /*
1711 * Make sure that nobody tries to wake us up again.
1712 */
1713 shost->eh_wait = NULL;
1714
1715 /*
1716 * Knock this down too. From this point on, the host is flying
1717 * without a pilot. If this is because the module is being unloaded,
1718 * that's fine. If the user sent a signal to this thing, we are
1719 * potentially in real danger.
1720 */
1721 shost->eh_active = 0;
1722 shost->ehandler = NULL;
1723
1724 /*
1725 * If anyone is waiting for us to exit (i.e. someone trying to unload
1726 * a driver), then wake up that process to let them know we are on
1727 * the way out the door.
1728 */
1729 complete_and_exit(shost->eh_notify, 0);
1730 return 0;
1731}
1732
1733/*
1734 * Function: scsi_report_bus_reset()
1735 *
1736 * Purpose: Utility function used by low-level drivers to report that
1737 * they have observed a bus reset on the bus being handled.
1738 *
1739 * Arguments: shost - Host in question
1740 * channel - channel on which reset was observed.
1741 *
1742 * Returns: Nothing
1743 *
1744 * Lock status: Host lock must be held.
1745 *
1746 * Notes: This only needs to be called if the reset is one which
1747 * originates from an unknown location. Resets originated
1748 * by the mid-level itself don't need to call this, but there
1749 * should be no harm.
1750 *
1751 * The main purpose of this is to make sure that a CHECK_CONDITION
1752 * is properly treated.
1753 */
1754void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1755{
1756 struct scsi_device *sdev;
1757
1758 __shost_for_each_device(sdev, shost) {
1759 if (channel == sdev->channel) {
1760 sdev->was_reset = 1;
1761 sdev->expecting_cc_ua = 1;
1762 }
1763 }
1764}
1765EXPORT_SYMBOL(scsi_report_bus_reset);
1766
1767/*
1768 * Function: scsi_report_device_reset()
1769 *
1770 * Purpose: Utility function used by low-level drivers to report that
1771 * they have observed a device reset on the device being handled.
1772 *
1773 * Arguments: shost - Host in question
1774 * channel - channel on which reset was observed
1775 * target - target on which reset was observed
1776 *
1777 * Returns: Nothing
1778 *
1779 * Lock status: Host lock must be held
1780 *
1781 * Notes: This only needs to be called if the reset is one which
1782 * originates from an unknown location. Resets originated
1783 * by the mid-level itself don't need to call this, but there
1784 * should be no harm.
1785 *
1786 * The main purpose of this is to make sure that a CHECK_CONDITION
1787 * is properly treated.
1788 */
1789void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1790{
1791 struct scsi_device *sdev;
1792
1793 __shost_for_each_device(sdev, shost) {
1794 if (channel == sdev->channel &&
1795 target == sdev->id) {
1796 sdev->was_reset = 1;
1797 sdev->expecting_cc_ua = 1;
1798 }
1799 }
1800}
1801EXPORT_SYMBOL(scsi_report_device_reset);
1802
1803static void
1804scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1805{
1806}
1807
1808/*
1809 * Function: scsi_reset_provider
1810 *
1811 * Purpose: Send requested reset to a bus or device at any phase.
1812 *
1813 * Arguments: device - device to send reset to
1814 * flag - reset type (see scsi.h)
1815 *
1816 * Returns: SUCCESS/FAILURE.
1817 *
1818 * Notes: This is used by the SCSI Generic driver to provide
1819 * Bus/Device reset capability.
1820 */
1821int
1822scsi_reset_provider(struct scsi_device *dev, int flag)
1823{
1824 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1825 struct request req;
1826 int rtn;
1827
1828 scmd->request = &req;
1829 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
1830 scmd->request->rq_status = RQ_SCSI_BUSY;
1831 scmd->state = SCSI_STATE_INITIALIZING;
1832 scmd->owner = SCSI_OWNER_MIDLEVEL;
1833
1834 memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd));
1835
1836 scmd->scsi_done = scsi_reset_provider_done_command;
1837 scmd->done = NULL;
1838 scmd->buffer = NULL;
1839 scmd->bufflen = 0;
1840 scmd->request_buffer = NULL;
1841 scmd->request_bufflen = 0;
1842 scmd->internal_timeout = NORMAL_TIMEOUT;
1843 scmd->abort_reason = DID_ABORT;
1844
1845 scmd->cmd_len = 0;
1846
1847 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1848 scmd->sc_request = NULL;
1849 scmd->sc_magic = SCSI_CMND_MAGIC;
1850
1851 init_timer(&scmd->eh_timeout);
1852
1853 /*
1854 * Sometimes the command can get back into the timer chain,
1855 * so use the pid as an identifier.
1856 */
1857 scmd->pid = 0;
1858
1859 switch (flag) {
1860 case SCSI_TRY_RESET_DEVICE:
1861 rtn = scsi_try_bus_device_reset(scmd);
1862 if (rtn == SUCCESS)
1863 break;
1864 /* FALLTHROUGH */
1865 case SCSI_TRY_RESET_BUS:
1866 rtn = scsi_try_bus_reset(scmd);
1867 if (rtn == SUCCESS)
1868 break;
1869 /* FALLTHROUGH */
1870 case SCSI_TRY_RESET_HOST:
1871 rtn = scsi_try_host_reset(scmd);
1872 break;
1873 default:
1874 rtn = FAILED;
1875 }
1876
1877 scsi_delete_timer(scmd);
1878 scsi_next_command(scmd);
1879 return rtn;
1880}
1881EXPORT_SYMBOL(scsi_reset_provider);
1882
1883/**
1884 * scsi_normalize_sense - normalize main elements from either fixed or
1885 * descriptor sense data format into a common format.
1886 *
1887 * @sense_buffer: byte array containing sense data returned by device
1888 * @sb_len: number of valid bytes in sense_buffer
1889 * @sshdr: pointer to instance of structure that common
1890 * elements are written to.
1891 *
1892 * Notes:
1893 * The "main elements" from sense data are: response_code, sense_key,
1894 * asc, ascq and additional_length (only for descriptor format).
1895 *
1896 * Typically this function can be called after a device has
1897 * responded to a SCSI command with the CHECK_CONDITION status.
1898 *
1899 * Return value:
1900 * 1 if valid sense data information found, else 0;
1901 **/
1902int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1903 struct scsi_sense_hdr *sshdr)
1904{
1905 if (!sense_buffer || !sb_len || (sense_buffer[0] & 0x70) != 0x70)
1906 return 0;
1907
1908 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1909
1910 sshdr->response_code = (sense_buffer[0] & 0x7f);
1911 if (sshdr->response_code >= 0x72) {
1912 /*
1913 * descriptor format
1914 */
1915 if (sb_len > 1)
1916 sshdr->sense_key = (sense_buffer[1] & 0xf);
1917 if (sb_len > 2)
1918 sshdr->asc = sense_buffer[2];
1919 if (sb_len > 3)
1920 sshdr->ascq = sense_buffer[3];
1921 if (sb_len > 7)
1922 sshdr->additional_length = sense_buffer[7];
1923 } else {
1924 /*
1925 * fixed format
1926 */
1927 if (sb_len > 2)
1928 sshdr->sense_key = (sense_buffer[2] & 0xf);
1929 if (sb_len > 7) {
1930 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1931 sb_len : (sense_buffer[7] + 8);
1932 if (sb_len > 12)
1933 sshdr->asc = sense_buffer[12];
1934 if (sb_len > 13)
1935 sshdr->ascq = sense_buffer[13];
1936 }
1937 }
1938
1939 return 1;
1940}
1941EXPORT_SYMBOL(scsi_normalize_sense);
1942
1943int scsi_request_normalize_sense(struct scsi_request *sreq,
1944 struct scsi_sense_hdr *sshdr)
1945{
1946 return scsi_normalize_sense(sreq->sr_sense_buffer,
1947 sizeof(sreq->sr_sense_buffer), sshdr);
1948}
1949EXPORT_SYMBOL(scsi_request_normalize_sense);
1950
1951int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1952 struct scsi_sense_hdr *sshdr)
1953{
1954 return scsi_normalize_sense(cmd->sense_buffer,
1955 sizeof(cmd->sense_buffer), sshdr);
1956}
1957EXPORT_SYMBOL(scsi_command_normalize_sense);
1958
1959/**
1960 * scsi_sense_desc_find - search for a given descriptor type in
1961 * descriptor sense data format.
1962 *
1963 * @sense_buffer: byte array of descriptor format sense data
1964 * @sb_len: number of valid bytes in sense_buffer
1965 * @desc_type: value of descriptor type to find
1966 * (e.g. 0 -> information)
1967 *
1968 * Notes:
1969 * only valid when sense data is in descriptor format
1970 *
1971 * Return value:
1972 * pointer to start of (first) descriptor if found else NULL
1973 **/
1974const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1975 int desc_type)
1976{
1977 int add_sen_len, add_len, desc_len, k;
1978 const u8 * descp;
1979
1980 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1981 return NULL;
1982 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1983 return NULL;
1984 add_sen_len = (add_sen_len < (sb_len - 8)) ?
1985 add_sen_len : (sb_len - 8);
1986 descp = &sense_buffer[8];
1987 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1988 descp += desc_len;
1989 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1990 desc_len = add_len + 2;
1991 if (descp[0] == desc_type)
1992 return descp;
1993 if (add_len < 0) // short descriptor ??
1994 break;
1995 }
1996 return NULL;
1997}
1998EXPORT_SYMBOL(scsi_sense_desc_find);
1999
2000/**
2001 * scsi_get_sense_info_fld - attempts to get information field from
2002 * sense data (either fixed or descriptor format)
2003 *
2004 * @sense_buffer: byte array of sense data
2005 * @sb_len: number of valid bytes in sense_buffer
2006 * @info_out: pointer to 64 integer where 8 or 4 byte information
2007 * field will be placed if found.
2008 *
2009 * Return value:
2010 * 1 if information field found, 0 if not found.
2011 **/
2012int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2013 u64 * info_out)
2014{
2015 int j;
2016 const u8 * ucp;
2017 u64 ull;
2018
2019 if (sb_len < 7)
2020 return 0;
2021 switch (sense_buffer[0] & 0x7f) {
2022 case 0x70:
2023 case 0x71:
2024 if (sense_buffer[0] & 0x80) {
2025 *info_out = (sense_buffer[3] << 24) +
2026 (sense_buffer[4] << 16) +
2027 (sense_buffer[5] << 8) + sense_buffer[6];
2028 return 1;
2029 } else
2030 return 0;
2031 case 0x72:
2032 case 0x73:
2033 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2034 0 /* info desc */);
2035 if (ucp && (0xa == ucp[1])) {
2036 ull = 0;
2037 for (j = 0; j < 8; ++j) {
2038 if (j > 0)
2039 ull <<= 8;
2040 ull |= ucp[4 + j];
2041 }
2042 *info_out = ull;
2043 return 1;
2044 } else
2045 return 0;
2046 default:
2047 return 0;
2048 }
2049}
2050EXPORT_SYMBOL(scsi_get_sense_info_fld);
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-18 10:45:20 -0500