blob: ebaca4ca4a135787a6ff3fa477a722dbc4b2ec45 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/kernel.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070022#include <linux/freezer.h>
Christoph Hellwigc5478de2005-09-06 14:04:26 +020023#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/interrupt.h>
25#include <linux/blkdev.h>
26#include <linux/delay.h>
27
28#include <scsi/scsi.h>
Christoph Hellwigbeb40482006-06-10 18:01:03 +020029#include <scsi/scsi_cmnd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <scsi/scsi_dbg.h>
31#include <scsi/scsi_device.h>
32#include <scsi/scsi_eh.h>
James Smartc829c392006-03-13 08:28:57 -050033#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <scsi/scsi_host.h>
35#include <scsi/scsi_ioctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37#include "scsi_priv.h"
38#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070039#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#define SENSE_TIMEOUT (10*HZ)
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
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 */
51void scsi_eh_wakeup(struct Scsi_Host *shost)
52{
53 if (shost->host_busy == shost->host_failed) {
James Bottomley3ed7a472005-09-19 09:50:04 -050054 wake_up_process(shost->ehandler);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 SCSI_LOG_ERROR_RECOVERY(5,
56 printk("Waking error handler thread\n"));
57 }
58}
Tejun Heof8bbfc22006-05-19 21:07:05 +090059
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 **/
66void 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}
80EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
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 **/
90int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
91{
92 struct Scsi_Host *shost = scmd->device->host;
93 unsigned long flags;
James Bottomley939647e2005-09-18 15:05:20 -050094 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
James Bottomley3ed7a472005-09-19 09:50:04 -050096 if (!shost->ehandler)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 return 0;
98
99 spin_lock_irqsave(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500100 if (scsi_host_set_state(shost, SHOST_RECOVERY))
101 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
102 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
James Bottomley939647e2005-09-18 15:05:20 -0500104 ret = 1;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200105 scmd->eh_eflags |= eh_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 shost->host_failed++;
108 scsi_eh_wakeup(shost);
James Bottomley939647e2005-09-18 15:05:20 -0500109 out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 spin_unlock_irqrestore(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500111 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112}
113
114/**
115 * scsi_add_timer - Start timeout timer for a single scsi command.
116 * @scmd: scsi command that is about to start running.
117 * @timeout: amount of time to allow this command to run.
118 * @complete: timeout function to call if timer isn't canceled.
119 *
120 * Notes:
121 * This should be turned into an inline function. Each scsi command
122 * has its own timer, and as it is added to the queue, we set up the
123 * timer. When the command completes, we cancel the timer.
124 **/
125void scsi_add_timer(struct scsi_cmnd *scmd, int timeout,
126 void (*complete)(struct scsi_cmnd *))
127{
128
129 /*
130 * If the clock was already running for this command, then
131 * first delete the timer. The timer handling code gets rather
132 * confused if we don't do this.
133 */
134 if (scmd->eh_timeout.function)
135 del_timer(&scmd->eh_timeout);
136
137 scmd->eh_timeout.data = (unsigned long)scmd;
138 scmd->eh_timeout.expires = jiffies + timeout;
139 scmd->eh_timeout.function = (void (*)(unsigned long)) complete;
140
141 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
142 " %d, (%p)\n", __FUNCTION__,
143 scmd, timeout, complete));
144
145 add_timer(&scmd->eh_timeout);
146}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148/**
149 * scsi_delete_timer - Delete/cancel timer for a given function.
150 * @scmd: Cmd that we are canceling timer for
151 *
152 * Notes:
153 * This should be turned into an inline function.
154 *
155 * Return value:
156 * 1 if we were able to detach the timer. 0 if we blew it, and the
157 * timer function has already started to run.
158 **/
159int scsi_delete_timer(struct scsi_cmnd *scmd)
160{
161 int rtn;
162
163 rtn = del_timer(&scmd->eh_timeout);
164
165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
166 " rtn: %d\n", __FUNCTION__,
167 scmd, rtn));
168
169 scmd->eh_timeout.data = (unsigned long)NULL;
170 scmd->eh_timeout.function = NULL;
171
172 return rtn;
173}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175/**
176 * scsi_times_out - Timeout function for normal scsi commands.
177 * @scmd: Cmd that is timing out.
178 *
179 * Notes:
180 * We do not need to lock this. There is the potential for a race
181 * only in that the normal completion handling might run, but if the
182 * normal completion function determines that the timer has already
183 * fired, then it mustn't do anything.
184 **/
185void scsi_times_out(struct scsi_cmnd *scmd)
186{
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500187 enum scsi_eh_timer_return (* eh_timed_out)(struct scsi_cmnd *);
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 scsi_log_completion(scmd, TIMEOUT_ERROR);
190
James Smartc829c392006-03-13 08:28:57 -0500191 if (scmd->device->host->transportt->eh_timed_out)
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500192 eh_timed_out = scmd->device->host->transportt->eh_timed_out;
193 else if (scmd->device->host->hostt->eh_timed_out)
194 eh_timed_out = scmd->device->host->hostt->eh_timed_out;
195 else
196 eh_timed_out = NULL;
197
198 if (eh_timed_out)
199 switch (eh_timed_out(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 case EH_HANDLED:
201 __scsi_done(scmd);
202 return;
203 case EH_RESET_TIMER:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 scsi_add_timer(scmd, scmd->timeout_per_command,
205 scsi_times_out);
206 return;
207 case EH_NOT_HANDLED:
208 break;
209 }
210
211 if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
James Bottomley939647e2005-09-18 15:05:20 -0500212 scmd->result |= DID_TIME_OUT << 16;
213 __scsi_done(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 }
215}
216
217/**
218 * scsi_block_when_processing_errors - Prevent cmds from being queued.
219 * @sdev: Device on which we are performing recovery.
220 *
221 * Description:
222 * We block until the host is out of error recovery, and then check to
223 * see whether the host or the device is offline.
224 *
225 * Return value:
226 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
227 **/
228int scsi_block_when_processing_errors(struct scsi_device *sdev)
229{
230 int online;
231
James Bottomley939647e2005-09-18 15:05:20 -0500232 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
234 online = scsi_device_online(sdev);
235
236 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__,
237 online));
238
239 return online;
240}
241EXPORT_SYMBOL(scsi_block_when_processing_errors);
242
243#ifdef CONFIG_SCSI_LOGGING
244/**
245 * scsi_eh_prt_fail_stats - Log info on failures.
246 * @shost: scsi host being recovered.
247 * @work_q: Queue of scsi cmds to process.
248 **/
249static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
250 struct list_head *work_q)
251{
252 struct scsi_cmnd *scmd;
253 struct scsi_device *sdev;
254 int total_failures = 0;
255 int cmd_failed = 0;
256 int cmd_cancel = 0;
257 int devices_failed = 0;
258
259 shost_for_each_device(sdev, shost) {
260 list_for_each_entry(scmd, work_q, eh_entry) {
261 if (scmd->device == sdev) {
262 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200263 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 ++cmd_cancel;
265 else
266 ++cmd_failed;
267 }
268 }
269
270 if (cmd_cancel || cmd_failed) {
271 SCSI_LOG_ERROR_RECOVERY(3,
James Bottomley9ccfc752005-10-02 11:45:08 -0500272 sdev_printk(KERN_INFO, sdev,
273 "%s: cmds failed: %d, cancel: %d\n",
274 __FUNCTION__, cmd_failed,
275 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 cmd_cancel = 0;
277 cmd_failed = 0;
278 ++devices_failed;
279 }
280 }
281
282 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
283 " devices require eh work\n",
284 total_failures, devices_failed));
285}
286#endif
287
288/**
289 * scsi_check_sense - Examine scsi cmd sense
290 * @scmd: Cmd to have sense checked.
291 *
292 * Return value:
293 * SUCCESS or FAILED or NEEDS_RETRY
294 *
295 * Notes:
296 * When a deferred error is detected the current command has
297 * not been executed and needs retrying.
298 **/
299static int scsi_check_sense(struct scsi_cmnd *scmd)
300{
301 struct scsi_sense_hdr sshdr;
302
303 if (! scsi_command_normalize_sense(scmd, &sshdr))
304 return FAILED; /* no valid sense data */
305
306 if (scsi_sense_is_deferred(&sshdr))
307 return NEEDS_RETRY;
308
309 /*
310 * Previous logic looked for FILEMARK, EOM or ILI which are
311 * mainly associated with tapes and returned SUCCESS.
312 */
313 if (sshdr.response_code == 0x70) {
314 /* fixed format */
315 if (scmd->sense_buffer[2] & 0xe0)
316 return SUCCESS;
317 } else {
318 /*
319 * descriptor format: look for "stream commands sense data
320 * descriptor" (see SSC-3). Assume single sense data
321 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
322 */
323 if ((sshdr.additional_length > 3) &&
324 (scmd->sense_buffer[8] == 0x4) &&
325 (scmd->sense_buffer[11] & 0xe0))
326 return SUCCESS;
327 }
328
329 switch (sshdr.sense_key) {
330 case NO_SENSE:
331 return SUCCESS;
332 case RECOVERED_ERROR:
333 return /* soft_error */ SUCCESS;
334
335 case ABORTED_COMMAND:
336 return NEEDS_RETRY;
337 case NOT_READY:
338 case UNIT_ATTENTION:
339 /*
340 * if we are expecting a cc/ua because of a bus reset that we
341 * performed, treat this just as a retry. otherwise this is
342 * information that we should pass up to the upper-level driver
343 * so that we can deal with it there.
344 */
345 if (scmd->device->expecting_cc_ua) {
346 scmd->device->expecting_cc_ua = 0;
347 return NEEDS_RETRY;
348 }
349 /*
350 * if the device is in the process of becoming ready, we
351 * should retry.
352 */
353 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
354 return NEEDS_RETRY;
355 /*
356 * if the device is not started, we need to wake
357 * the error handler to start the motor
358 */
359 if (scmd->device->allow_restart &&
360 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
361 return FAILED;
362 return SUCCESS;
363
364 /* these three are not supported */
365 case COPY_ABORTED:
366 case VOLUME_OVERFLOW:
367 case MISCOMPARE:
368 return SUCCESS;
369
370 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800371 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
372 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
373 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
374 return SUCCESS;
375 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 return NEEDS_RETRY;
377
378 case HARDWARE_ERROR:
379 if (scmd->device->retry_hwerror)
380 return NEEDS_RETRY;
381 else
382 return SUCCESS;
383
384 case ILLEGAL_REQUEST:
385 case BLANK_CHECK:
386 case DATA_PROTECT:
387 default:
388 return SUCCESS;
389 }
390}
391
392/**
393 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
394 * @scmd: SCSI cmd to examine.
395 *
396 * Notes:
397 * This is *only* called when we are examining the status of commands
398 * queued during error recovery. the main difference here is that we
399 * don't allow for the possibility of retries here, and we are a lot
400 * more restrictive about what we consider acceptable.
401 **/
402static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
403{
404 /*
405 * first check the host byte, to see if there is anything in there
406 * that would indicate what we need to do.
407 */
408 if (host_byte(scmd->result) == DID_RESET) {
409 /*
410 * rats. we are already in the error handler, so we now
411 * get to try and figure out what to do next. if the sense
412 * is valid, we have a pretty good idea of what to do.
413 * if not, we mark it as FAILED.
414 */
415 return scsi_check_sense(scmd);
416 }
417 if (host_byte(scmd->result) != DID_OK)
418 return FAILED;
419
420 /*
421 * next, check the message byte.
422 */
423 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
424 return FAILED;
425
426 /*
427 * now, check the status byte to see if this indicates
428 * anything special.
429 */
430 switch (status_byte(scmd->result)) {
431 case GOOD:
432 case COMMAND_TERMINATED:
433 return SUCCESS;
434 case CHECK_CONDITION:
435 return scsi_check_sense(scmd);
436 case CONDITION_GOOD:
437 case INTERMEDIATE_GOOD:
438 case INTERMEDIATE_C_GOOD:
439 /*
440 * who knows? FIXME(eric)
441 */
442 return SUCCESS;
443 case BUSY:
444 case QUEUE_FULL:
445 case RESERVATION_CONFLICT:
446 default:
447 return FAILED;
448 }
449 return FAILED;
450}
451
452/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 * scsi_eh_done - Completion function for error handling.
454 * @scmd: Cmd that is done.
455 **/
456static void scsi_eh_done(struct scsi_cmnd *scmd)
457{
Michael Reed85631672005-12-07 21:46:27 -0600458 struct completion *eh_action;
459
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100460 SCSI_LOG_ERROR_RECOVERY(3,
461 printk("%s scmd: %p result: %x\n",
462 __FUNCTION__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600463
464 eh_action = scmd->device->host->eh_action;
465 if (eh_action)
466 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
469/**
Brian King292148f2007-01-30 17:51:17 -0600470 * scsi_try_host_reset - ask host adapter to reset itself
471 * @scmd: SCSI cmd to send hsot reset.
472 **/
473static int scsi_try_host_reset(struct scsi_cmnd *scmd)
474{
475 unsigned long flags;
476 int rtn;
477
478 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
479 __FUNCTION__));
480
481 if (!scmd->device->host->hostt->eh_host_reset_handler)
482 return FAILED;
483
484 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
485
486 if (rtn == SUCCESS) {
487 if (!scmd->device->host->hostt->skip_settle_delay)
488 ssleep(HOST_RESET_SETTLE_TIME);
489 spin_lock_irqsave(scmd->device->host->host_lock, flags);
490 scsi_report_bus_reset(scmd->device->host,
491 scmd_channel(scmd));
492 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
493 }
494
495 return rtn;
496}
497
498/**
499 * scsi_try_bus_reset - ask host to perform a bus reset
500 * @scmd: SCSI cmd to send bus reset.
501 **/
502static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
503{
504 unsigned long flags;
505 int rtn;
506
507 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
508 __FUNCTION__));
509
510 if (!scmd->device->host->hostt->eh_bus_reset_handler)
511 return FAILED;
512
513 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
514
515 if (rtn == SUCCESS) {
516 if (!scmd->device->host->hostt->skip_settle_delay)
517 ssleep(BUS_RESET_SETTLE_TIME);
518 spin_lock_irqsave(scmd->device->host->host_lock, flags);
519 scsi_report_bus_reset(scmd->device->host,
520 scmd_channel(scmd));
521 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
522 }
523
524 return rtn;
525}
526
527/**
528 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
529 * @scmd: SCSI cmd used to send BDR
530 *
531 * Notes:
532 * There is no timeout for this operation. if this operation is
533 * unreliable for a given host, then the host itself needs to put a
534 * timer on it, and set the host back to a consistent state prior to
535 * returning.
536 **/
537static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
538{
539 int rtn;
540
541 if (!scmd->device->host->hostt->eh_device_reset_handler)
542 return FAILED;
543
544 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
545 if (rtn == SUCCESS) {
546 scmd->device->was_reset = 1;
547 scmd->device->expecting_cc_ua = 1;
548 }
549
550 return rtn;
551}
552
553static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
554{
555 if (!scmd->device->host->hostt->eh_abort_handler)
556 return FAILED;
557
558 return scmd->device->host->hostt->eh_abort_handler(scmd);
559}
560
561/**
562 * scsi_try_to_abort_cmd - Ask host to abort a running command.
563 * @scmd: SCSI cmd to abort from Lower Level.
564 *
565 * Notes:
566 * This function will not return until the user's completion function
567 * has been called. there is no timeout on this operation. if the
568 * author of the low-level driver wishes this operation to be timed,
569 * they can provide this facility themselves. helper functions in
570 * scsi_error.c can be supplied to make this easier to do.
571 **/
572static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
573{
574 /*
575 * scsi_done was called just after the command timed out and before
576 * we had a chance to process it. (db)
577 */
578 if (scmd->serial_number == 0)
579 return SUCCESS;
580 return __scsi_try_to_abort_cmd(scmd);
581}
582
583static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
584{
585 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
586 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
587 if (scsi_try_bus_reset(scmd) != SUCCESS)
588 scsi_try_host_reset(scmd);
589}
590
591/**
Boaz Harroshe1c23462007-10-08 16:36:45 +0200592 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100593 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200594 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200595 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100596 * @cmnd_size: size in bytes of @cmnd
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200597 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100598 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200599 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200600 * as part of the error recovery process. If @sense_bytes is 0 the command
601 * sent must be one that does not transfer any data. If @sense_bytes != 0
602 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
603 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 **/
Boaz Harroshe1c23462007-10-08 16:36:45 +0200605void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
606 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
f59114b2005-04-17 15:00:23 -0500608 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Christoph Hellwig631c2282006-07-08 20:42:15 +0200610 /*
611 * We need saved copies of a number of fields - this is because
612 * error handling may need to overwrite these with different values
613 * to run different commands, and once error handling is complete,
614 * we will need to restore these values prior to running the actual
615 * command.
616 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200617 ses->cmd_len = scmd->cmd_len;
618 memcpy(ses->cmnd, scmd->cmnd, sizeof(scmd->cmnd));
619 ses->data_direction = scmd->sc_data_direction;
620 ses->bufflen = scmd->request_bufflen;
621 ses->buffer = scmd->request_buffer;
622 ses->use_sg = scmd->use_sg;
623 ses->resid = scmd->resid;
624 ses->result = scmd->result;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200625
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200626 if (sense_bytes) {
627 scmd->request_bufflen = min_t(unsigned,
628 sizeof(scmd->sense_buffer), sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200629 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
630 scmd->request_bufflen);
631 scmd->request_buffer = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200632 scmd->sc_data_direction = DMA_FROM_DEVICE;
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100633 scmd->use_sg = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200634 memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
635 scmd->cmnd[0] = REQUEST_SENSE;
636 scmd->cmnd[4] = scmd->request_bufflen;
637 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200638 } else {
639 scmd->request_buffer = NULL;
640 scmd->request_bufflen = 0;
641 scmd->sc_data_direction = DMA_NONE;
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100642 scmd->use_sg = 0;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200643 if (cmnd) {
644 memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
645 memcpy(scmd->cmnd, cmnd, cmnd_size);
646 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
647 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200648 }
649
650 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200651
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200652 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500654 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Christoph Hellwig631c2282006-07-08 20:42:15 +0200656 /*
657 * Zero the sense buffer. The scsi spec mandates that any
658 * untransferred sense data should be interpreted as being zero.
659 */
660 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
Boaz Harroshe1c23462007-10-08 16:36:45 +0200661}
662EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200663
Boaz Harroshe1c23462007-10-08 16:36:45 +0200664/**
665 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
666 * @scmd: SCSI command structure to restore
667 * @ses: saved information from a coresponding call to scsi_prep_eh_cmnd
668 *
669 * Undo any damage done by above scsi_prep_eh_cmnd().
670 **/
671void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
672{
673 /*
674 * Restore original data
675 */
676 scmd->cmd_len = ses->cmd_len;
677 memcpy(scmd->cmnd, ses->cmnd, sizeof(scmd->cmnd));
678 scmd->sc_data_direction = ses->data_direction;
679 scmd->request_bufflen = ses->bufflen;
680 scmd->request_buffer = ses->buffer;
681 scmd->use_sg = ses->use_sg;
682 scmd->resid = ses->resid;
683 scmd->result = ses->result;
684}
685EXPORT_SYMBOL(scsi_eh_restore_cmnd);
686
687/**
688 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
689 * @scmd: SCSI command structure to hijack
690 * @cmnd: CDB to send
691 * @cmnd_size: size in bytes of @cmnd
692 * @timeout: timeout for this request
693 * @sense_bytes: size of sense data to copy or 0
694 *
695 * This function is used to send a scsi command down to a target device
696 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
697 *
698 * Return value:
699 * SUCCESS or FAILED or NEEDS_RETRY
700 **/
701static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
702 int cmnd_size, int timeout, unsigned sense_bytes)
703{
704 struct scsi_device *sdev = scmd->device;
705 struct Scsi_Host *shost = sdev->host;
706 DECLARE_COMPLETION_ONSTACK(done);
707 unsigned long timeleft;
708 unsigned long flags;
709 struct scsi_eh_save ses;
710 int rtn;
711
712 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100713 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
f59114b2005-04-17 15:00:23 -0500715 spin_lock_irqsave(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 scsi_log_send(scmd);
f59114b2005-04-17 15:00:23 -0500717 shost->hostt->queuecommand(scmd, scsi_eh_done);
718 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100720 timeleft = wait_for_completion_timeout(&done, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
f59114b2005-04-17 15:00:23 -0500722 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100724 scsi_log_completion(scmd, SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100726 SCSI_LOG_ERROR_RECOVERY(3,
727 printk("%s: scmd: %p, timeleft: %ld\n",
728 __FUNCTION__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
730 /*
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100731 * If there is time left scsi_eh_done got called, and we will
732 * examine the actual status codes to see whether the command
733 * actually did complete normally, else tell the host to forget
734 * about this command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100736 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 rtn = scsi_eh_completed_normally(scmd);
738 SCSI_LOG_ERROR_RECOVERY(3,
739 printk("%s: scsi_eh_completed_normally %x\n",
740 __FUNCTION__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100741
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 switch (rtn) {
743 case SUCCESS:
744 case NEEDS_RETRY:
745 case FAILED:
746 break;
747 default:
748 rtn = FAILED;
749 break;
750 }
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100751 } else {
Brian King292148f2007-01-30 17:51:17 -0600752 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100753 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 }
755
Boaz Harroshe1c23462007-10-08 16:36:45 +0200756 scsi_eh_restore_cmnd(scmd, &ses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 return rtn;
758}
759
760/**
761 * scsi_request_sense - Request sense data from a particular target.
762 * @scmd: SCSI cmd for request sense.
763 *
764 * Notes:
765 * Some hosts automatically obtain this information, others require
766 * that we obtain it on our own. This function will *not* return until
767 * the command either times out, or it completes.
768 **/
769static int scsi_request_sense(struct scsi_cmnd *scmd)
770{
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200771 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772}
773
774/**
775 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
776 * @scmd: Original SCSI cmd that eh has finished.
777 * @done_q: Queue for processed commands.
778 *
779 * Notes:
780 * We don't want to use the normal command completion while we are are
781 * still handling errors - it may cause other commands to be queued,
782 * and that would disturb what we are doing. thus we really want to
783 * keep a list of pending commands for final completion, and once we
784 * are ready to leave error handling we handle completion for real.
785 **/
Tejun Heo041c5fc2006-01-23 13:09:36 +0900786void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
788 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200789 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 list_move_tail(&scmd->eh_entry, done_q);
791}
Tejun Heo041c5fc2006-01-23 13:09:36 +0900792EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
794/**
795 * scsi_eh_get_sense - Get device sense data.
796 * @work_q: Queue of commands to process.
797 * @done_q: Queue of proccessed commands..
798 *
799 * Description:
800 * See if we need to request sense information. if so, then get it
801 * now, so we have a better idea of what to do.
802 *
803 * Notes:
804 * This has the unfortunate side effect that if a shost adapter does
805 * not automatically request sense information, that we end up shutting
806 * it down before we request it.
807 *
808 * All drivers should request sense information internally these days,
809 * so for now all I have to say is tough noogies if you end up in here.
810 *
811 * XXX: Long term this code should go away, but that needs an audit of
812 * all LLDDs first.
813 **/
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800814int scsi_eh_get_sense(struct list_head *work_q,
815 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200817 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 int rtn;
819
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200820 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200821 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 SCSI_SENSE_VALID(scmd))
823 continue;
824
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400825 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
826 "%s: requesting sense\n",
827 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 rtn = scsi_request_sense(scmd);
829 if (rtn != SUCCESS)
830 continue;
831
832 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
833 " result %x\n", scmd,
834 scmd->result));
835 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
836
837 rtn = scsi_decide_disposition(scmd);
838
839 /*
840 * if the result was normal, then just pass it along to the
841 * upper level.
842 */
843 if (rtn == SUCCESS)
844 /* we don't want this command reissued, just
845 * finished with the sense data, so set
846 * retries to the max allowed to ensure it
847 * won't get reissued */
848 scmd->retries = scmd->allowed;
849 else if (rtn != NEEDS_RETRY)
850 continue;
851
852 scsi_eh_finish_cmd(scmd, done_q);
853 }
854
855 return list_empty(work_q);
856}
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800857EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
859/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 * scsi_eh_tur - Send TUR to device.
861 * @scmd: Scsi cmd to send TUR
862 *
863 * Return value:
864 * 0 - Device is ready. 1 - Device NOT ready.
865 **/
866static int scsi_eh_tur(struct scsi_cmnd *scmd)
867{
868 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
869 int retry_cnt = 1, rtn;
870
871retry_tur:
Mike Christie0db99e32006-08-26 03:00:22 -0400872 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
875 __FUNCTION__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +0200876
877 switch (rtn) {
878 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 if (retry_cnt--)
880 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200881 /*FALLTHRU*/
882 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -0500883 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200884 default:
885 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -0500886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887}
888
889/**
890 * scsi_eh_abort_cmds - abort canceled commands.
891 * @shost: scsi host being recovered.
892 * @eh_done_q: list_head for processed commands.
893 *
894 * Decription:
895 * Try and see whether or not it makes sense to try and abort the
896 * running command. this only works out to be the case if we have one
897 * command that has timed out. if the command simply failed, it makes
898 * no sense to try and abort the command, since as far as the shost
899 * adapter is concerned, it isn't running.
900 **/
901static int scsi_eh_abort_cmds(struct list_head *work_q,
902 struct list_head *done_q)
903{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200904 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 int rtn;
906
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200907 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200908 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 continue;
910 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
911 "0x%p\n", current->comm,
912 scmd));
913 rtn = scsi_try_to_abort_cmd(scmd);
914 if (rtn == SUCCESS) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200915 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 if (!scsi_device_online(scmd->device) ||
917 !scsi_eh_tur(scmd)) {
918 scsi_eh_finish_cmd(scmd, done_q);
919 }
920
921 } else
922 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
923 " cmd failed:"
924 "0x%p\n",
925 current->comm,
926 scmd));
927 }
928
929 return list_empty(work_q);
930}
931
932/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 * scsi_eh_try_stu - Send START_UNIT to device.
934 * @scmd: Scsi cmd to send START_UNIT
935 *
936 * Return value:
937 * 0 - Device is ready. 1 - Device NOT ready.
938 **/
939static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
940{
941 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Christoph Hellwig631c2282006-07-08 20:42:15 +0200943 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -0500944 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
Brian Kinged773e62007-03-29 15:25:52 -0500946 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
947 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
Brian Kinge555db92007-04-19 13:59:59 -0500948 scmd->device->timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -0500949
Christoph Hellwig631c2282006-07-08 20:42:15 +0200950 if (rtn == SUCCESS)
951 return 0;
952 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 return 1;
955}
956
957 /**
958 * scsi_eh_stu - send START_UNIT if needed
959 * @shost: scsi host being recovered.
960 * @eh_done_q: list_head for processed commands.
961 *
962 * Notes:
963 * If commands are failing due to not ready, initializing command required,
964 * try revalidating the device, which will end up sending a start unit.
965 **/
966static int scsi_eh_stu(struct Scsi_Host *shost,
967 struct list_head *work_q,
968 struct list_head *done_q)
969{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200970 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 struct scsi_device *sdev;
972
973 shost_for_each_device(sdev, shost) {
974 stu_scmd = NULL;
975 list_for_each_entry(scmd, work_q, eh_entry)
976 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
977 scsi_check_sense(scmd) == FAILED ) {
978 stu_scmd = scmd;
979 break;
980 }
981
982 if (!stu_scmd)
983 continue;
984
985 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
986 " 0x%p\n", current->comm, sdev));
987
988 if (!scsi_eh_try_stu(stu_scmd)) {
989 if (!scsi_device_online(sdev) ||
990 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200991 list_for_each_entry_safe(scmd, next,
992 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 if (scmd->device == sdev)
994 scsi_eh_finish_cmd(scmd, done_q);
995 }
996 }
997 } else {
998 SCSI_LOG_ERROR_RECOVERY(3,
999 printk("%s: START_UNIT failed to sdev:"
1000 " 0x%p\n", current->comm, sdev));
1001 }
1002 }
1003
1004 return list_empty(work_q);
1005}
1006
1007
1008/**
1009 * scsi_eh_bus_device_reset - send bdr if needed
1010 * @shost: scsi host being recovered.
1011 * @eh_done_q: list_head for processed commands.
1012 *
1013 * Notes:
1014 * Try a bus device reset. still, look to see whether we have multiple
1015 * devices that are jammed or not - if we have multiple devices, it
1016 * makes no sense to try bus_device_reset - we really would need to try
1017 * a bus_reset instead.
1018 **/
1019static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1020 struct list_head *work_q,
1021 struct list_head *done_q)
1022{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001023 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 struct scsi_device *sdev;
1025 int rtn;
1026
1027 shost_for_each_device(sdev, shost) {
1028 bdr_scmd = NULL;
1029 list_for_each_entry(scmd, work_q, eh_entry)
1030 if (scmd->device == sdev) {
1031 bdr_scmd = scmd;
1032 break;
1033 }
1034
1035 if (!bdr_scmd)
1036 continue;
1037
1038 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1039 " 0x%p\n", current->comm,
1040 sdev));
1041 rtn = scsi_try_bus_device_reset(bdr_scmd);
1042 if (rtn == SUCCESS) {
1043 if (!scsi_device_online(sdev) ||
1044 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001045 list_for_each_entry_safe(scmd, next,
1046 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 if (scmd->device == sdev)
1048 scsi_eh_finish_cmd(scmd,
1049 done_q);
1050 }
1051 }
1052 } else {
1053 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1054 " failed sdev:"
1055 "0x%p\n",
1056 current->comm,
1057 sdev));
1058 }
1059 }
1060
1061 return list_empty(work_q);
1062}
1063
1064/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 * scsi_eh_bus_reset - send a bus reset
1066 * @shost: scsi host being recovered.
1067 * @eh_done_q: list_head for processed commands.
1068 **/
1069static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1070 struct list_head *work_q,
1071 struct list_head *done_q)
1072{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001073 struct scsi_cmnd *scmd, *chan_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 unsigned int channel;
1075 int rtn;
1076
1077 /*
1078 * we really want to loop over the various channels, and do this on
1079 * a channel by channel basis. we should also check to see if any
1080 * of the failed commands are on soft_reset devices, and if so, skip
1081 * the reset.
1082 */
1083
1084 for (channel = 0; channel <= shost->max_channel; channel++) {
1085 chan_scmd = NULL;
1086 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001087 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 chan_scmd = scmd;
1089 break;
1090 /*
1091 * FIXME add back in some support for
1092 * soft_reset devices.
1093 */
1094 }
1095 }
1096
1097 if (!chan_scmd)
1098 continue;
1099 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1100 " %d\n", current->comm,
1101 channel));
1102 rtn = scsi_try_bus_reset(chan_scmd);
1103 if (rtn == SUCCESS) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001104 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001105 if (channel == scmd_channel(scmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 if (!scsi_device_online(scmd->device) ||
1107 !scsi_eh_tur(scmd))
1108 scsi_eh_finish_cmd(scmd,
1109 done_q);
1110 }
1111 } else {
1112 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1113 " failed chan: %d\n",
1114 current->comm,
1115 channel));
1116 }
1117 }
1118 return list_empty(work_q);
1119}
1120
1121/**
1122 * scsi_eh_host_reset - send a host reset
1123 * @work_q: list_head for processed commands.
1124 * @done_q: list_head for processed commands.
1125 **/
1126static int scsi_eh_host_reset(struct list_head *work_q,
1127 struct list_head *done_q)
1128{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001129 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
1132 if (!list_empty(work_q)) {
1133 scmd = list_entry(work_q->next,
1134 struct scsi_cmnd, eh_entry);
1135
1136 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1137 , current->comm));
1138
1139 rtn = scsi_try_host_reset(scmd);
1140 if (rtn == SUCCESS) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001141 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 if (!scsi_device_online(scmd->device) ||
1143 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1144 !scsi_eh_tur(scmd))
1145 scsi_eh_finish_cmd(scmd, done_q);
1146 }
1147 } else {
1148 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1149 " failed\n",
1150 current->comm));
1151 }
1152 }
1153 return list_empty(work_q);
1154}
1155
1156/**
1157 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1158 * @work_q: list_head for processed commands.
1159 * @done_q: list_head for processed commands.
1160 *
1161 **/
1162static void scsi_eh_offline_sdevs(struct list_head *work_q,
1163 struct list_head *done_q)
1164{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001165 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001167 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001168 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1169 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001171 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 /*
1173 * FIXME: Handle lost cmds.
1174 */
1175 }
1176 scsi_eh_finish_cmd(scmd, done_q);
1177 }
1178 return;
1179}
1180
1181/**
1182 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1183 * @scmd: SCSI cmd to examine.
1184 *
1185 * Notes:
1186 * This is *only* called when we are examining the status after sending
1187 * out the actual data command. any commands that are queued for error
1188 * recovery (e.g. test_unit_ready) do *not* come through here.
1189 *
1190 * When this routine returns failed, it means the error handler thread
1191 * is woken. In cases where the error code indicates an error that
1192 * doesn't require the error handler read (i.e. we don't need to
1193 * abort/reset), this function should return SUCCESS.
1194 **/
1195int scsi_decide_disposition(struct scsi_cmnd *scmd)
1196{
1197 int rtn;
1198
1199 /*
1200 * if the device is offline, then we clearly just pass the result back
1201 * up to the top level.
1202 */
1203 if (!scsi_device_online(scmd->device)) {
1204 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1205 " as SUCCESS\n",
1206 __FUNCTION__));
1207 return SUCCESS;
1208 }
1209
1210 /*
1211 * first check the host byte, to see if there is anything in there
1212 * that would indicate what we need to do.
1213 */
1214 switch (host_byte(scmd->result)) {
1215 case DID_PASSTHROUGH:
1216 /*
1217 * no matter what, pass this through to the upper layer.
1218 * nuke this special code so that it looks like we are saying
1219 * did_ok.
1220 */
1221 scmd->result &= 0xff00ffff;
1222 return SUCCESS;
1223 case DID_OK:
1224 /*
1225 * looks good. drop through, and check the next byte.
1226 */
1227 break;
1228 case DID_NO_CONNECT:
1229 case DID_BAD_TARGET:
1230 case DID_ABORT:
1231 /*
1232 * note - this means that we just report the status back
1233 * to the top level driver, not that we actually think
1234 * that it indicates SUCCESS.
1235 */
1236 return SUCCESS;
1237 /*
1238 * when the low level driver returns did_soft_error,
1239 * it is responsible for keeping an internal retry counter
1240 * in order to avoid endless loops (db)
1241 *
1242 * actually this is a bug in this function here. we should
1243 * be mindful of the maximum number of retries specified
1244 * and not get stuck in a loop.
1245 */
1246 case DID_SOFT_ERROR:
1247 goto maybe_retry;
1248 case DID_IMM_RETRY:
1249 return NEEDS_RETRY;
1250
bf341912005-04-12 17:49:09 -05001251 case DID_REQUEUE:
1252 return ADD_TO_MLQUEUE;
1253
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 case DID_ERROR:
1255 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1256 status_byte(scmd->result) == RESERVATION_CONFLICT)
1257 /*
1258 * execute reservation conflict processing code
1259 * lower down
1260 */
1261 break;
1262 /* fallthrough */
1263
1264 case DID_BUS_BUSY:
1265 case DID_PARITY:
1266 goto maybe_retry;
1267 case DID_TIME_OUT:
1268 /*
1269 * when we scan the bus, we get timeout messages for
1270 * these commands if there is no device available.
1271 * other hosts report did_no_connect for the same thing.
1272 */
1273 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1274 scmd->cmnd[0] == INQUIRY)) {
1275 return SUCCESS;
1276 } else {
1277 return FAILED;
1278 }
1279 case DID_RESET:
1280 return SUCCESS;
1281 default:
1282 return FAILED;
1283 }
1284
1285 /*
1286 * next, check the message byte.
1287 */
1288 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1289 return FAILED;
1290
1291 /*
1292 * check the status byte to see if this indicates anything special.
1293 */
1294 switch (status_byte(scmd->result)) {
1295 case QUEUE_FULL:
1296 /*
1297 * the case of trying to send too many commands to a
1298 * tagged queueing device.
1299 */
1300 case BUSY:
1301 /*
1302 * device can't talk to us at the moment. Should only
1303 * occur (SAM-3) when the task queue is empty, so will cause
1304 * the empty queue handling to trigger a stall in the
1305 * device.
1306 */
1307 return ADD_TO_MLQUEUE;
1308 case GOOD:
1309 case COMMAND_TERMINATED:
1310 case TASK_ABORTED:
1311 return SUCCESS;
1312 case CHECK_CONDITION:
1313 rtn = scsi_check_sense(scmd);
1314 if (rtn == NEEDS_RETRY)
1315 goto maybe_retry;
1316 /* if rtn == FAILED, we have no sense information;
1317 * returning FAILED will wake the error handler thread
1318 * to collect the sense and redo the decide
1319 * disposition */
1320 return rtn;
1321 case CONDITION_GOOD:
1322 case INTERMEDIATE_GOOD:
1323 case INTERMEDIATE_C_GOOD:
1324 case ACA_ACTIVE:
1325 /*
1326 * who knows? FIXME(eric)
1327 */
1328 return SUCCESS;
1329
1330 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001331 sdev_printk(KERN_INFO, scmd->device,
1332 "reservation conflict\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 return SUCCESS; /* causes immediate i/o error */
1334 default:
1335 return FAILED;
1336 }
1337 return FAILED;
1338
1339 maybe_retry:
1340
1341 /* we requeue for retry because the error was retryable, and
1342 * the request was not marked fast fail. Note that above,
1343 * even if the request is marked fast fail, we still requeue
1344 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001345 if ((++scmd->retries) <= scmd->allowed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 && !blk_noretry_request(scmd->request)) {
1347 return NEEDS_RETRY;
1348 } else {
1349 /*
1350 * no more retries - report this one back to upper level.
1351 */
1352 return SUCCESS;
1353 }
1354}
1355
1356/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 * scsi_eh_lock_door - Prevent medium removal for the specified device
1358 * @sdev: SCSI device to prevent medium removal
1359 *
1360 * Locking:
1361 * We must be called from process context; scsi_allocate_request()
1362 * may sleep.
1363 *
1364 * Notes:
1365 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1366 * head of the devices request queue, and continue.
1367 *
1368 * Bugs:
1369 * scsi_allocate_request() may sleep waiting for existing requests to
1370 * be processed. However, since we haven't kicked off any request
1371 * processing for this host, this may deadlock.
1372 *
1373 * If scsi_allocate_request() fails for what ever reason, we
1374 * completely forget to lock the door.
1375 **/
1376static void scsi_eh_lock_door(struct scsi_device *sdev)
1377{
Mike Christie6e68af62005-11-11 05:30:27 -06001378 unsigned char cmnd[MAX_COMMAND_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379
Mike Christie6e68af62005-11-11 05:30:27 -06001380 cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1381 cmnd[1] = 0;
1382 cmnd[2] = 0;
1383 cmnd[3] = 0;
1384 cmnd[4] = SCSI_REMOVAL_PREVENT;
1385 cmnd[5] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
brking@us.ibm.combb1d1072006-01-23 15:03:22 -06001387 scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ,
Mike Christie6e68af62005-11-11 05:30:27 -06001388 5, NULL, NULL, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389}
1390
1391
1392/**
1393 * scsi_restart_operations - restart io operations to the specified host.
1394 * @shost: Host we are restarting.
1395 *
1396 * Notes:
1397 * When we entered the error handler, we blocked all further i/o to
1398 * this device. we need to 'reverse' this process.
1399 **/
1400static void scsi_restart_operations(struct Scsi_Host *shost)
1401{
1402 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001403 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
1405 /*
1406 * If the door was locked, we need to insert a door lock request
1407 * onto the head of the SCSI request queue for the device. There
1408 * is no point trying to lock the door of an off-line device.
1409 */
1410 shost_for_each_device(sdev, shost) {
1411 if (scsi_device_online(sdev) && sdev->locked)
1412 scsi_eh_lock_door(sdev);
1413 }
1414
1415 /*
1416 * next free up anything directly waiting upon the host. this
1417 * will be requests for character device operations, and also for
1418 * ioctls to queued block devices.
1419 */
1420 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1421 __FUNCTION__));
1422
James Bottomley939647e2005-09-18 15:05:20 -05001423 spin_lock_irqsave(shost->host_lock, flags);
1424 if (scsi_host_set_state(shost, SHOST_RUNNING))
1425 if (scsi_host_set_state(shost, SHOST_CANCEL))
1426 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1427 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 wake_up(&shost->host_wait);
1430
1431 /*
1432 * finally we need to re-initiate requests that may be pending. we will
1433 * have had everything blocked while error handling is taking place, and
1434 * now that error recovery is done, we will need to ensure that these
1435 * requests are started.
1436 */
1437 scsi_run_host_queues(shost);
1438}
1439
1440/**
1441 * scsi_eh_ready_devs - check device ready state and recover if not.
1442 * @shost: host to be recovered.
1443 * @eh_done_q: list_head for processed commands.
1444 *
1445 **/
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001446void scsi_eh_ready_devs(struct Scsi_Host *shost,
1447 struct list_head *work_q,
1448 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449{
1450 if (!scsi_eh_stu(shost, work_q, done_q))
1451 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1452 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1453 if (!scsi_eh_host_reset(work_q, done_q))
1454 scsi_eh_offline_sdevs(work_q, done_q);
1455}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001456EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
1458/**
1459 * scsi_eh_flush_done_q - finish processed commands or retry them.
1460 * @done_q: list_head of processed commands.
1461 *
1462 **/
Tejun Heo041c5fc2006-01-23 13:09:36 +09001463void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001465 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001467 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1468 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 if (scsi_device_online(scmd->device) &&
1470 !blk_noretry_request(scmd->request) &&
Brian King8884efa2006-02-24 17:10:04 -06001471 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1473 " retry cmd: %p\n",
1474 current->comm,
1475 scmd));
1476 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1477 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07001478 /*
1479 * If just we got sense for the device (called
1480 * scsi_eh_get_sense), scmd->result is already
1481 * set, do not set DRIVER_TIMEOUT.
1482 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 if (!scmd->result)
1484 scmd->result |= (DRIVER_TIMEOUT << 24);
1485 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1486 " cmd: %p\n",
1487 current->comm, scmd));
1488 scsi_finish_command(scmd);
1489 }
1490 }
1491}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001492EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
1494/**
1495 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1496 * @shost: Host to unjam.
1497 *
1498 * Notes:
1499 * When we come in here, we *know* that all commands on the bus have
1500 * either completed, failed or timed out. we also know that no further
1501 * commands are being sent to the host, so things are relatively quiet
1502 * and we have freedom to fiddle with things as we wish.
1503 *
1504 * This is only the *default* implementation. it is possible for
1505 * individual drivers to supply their own version of this function, and
1506 * if the maintainer wishes to do this, it is strongly suggested that
1507 * this function be taken as a template and modified. this function
1508 * was designed to correctly handle problems for about 95% of the
1509 * different cases out there, and it should always provide at least a
1510 * reasonable amount of error recovery.
1511 *
1512 * Any command marked 'failed' or 'timeout' must eventually have
1513 * scsi_finish_cmd() called for it. we do all of the retry stuff
1514 * here, so when we restart the host after we return it should have an
1515 * empty queue.
1516 **/
1517static void scsi_unjam_host(struct Scsi_Host *shost)
1518{
1519 unsigned long flags;
1520 LIST_HEAD(eh_work_q);
1521 LIST_HEAD(eh_done_q);
1522
1523 spin_lock_irqsave(shost->host_lock, flags);
1524 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1525 spin_unlock_irqrestore(shost->host_lock, flags);
1526
1527 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1528
1529 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1530 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1531 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1532
1533 scsi_eh_flush_done_q(&eh_done_q);
1534}
1535
1536/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001537 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 * @data: Host for which we are running.
1539 *
1540 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001541 * This is the main error handling loop. This is run as a kernel thread
1542 * for every SCSI host and handles all error handling activity.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 **/
1544int scsi_error_handler(void *data)
1545{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001546 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001549 * We use TASK_INTERRUPTIBLE so that the thread is not
1550 * counted against the load average as a running process.
1551 * We never actually get interrupted because kthread_run
1552 * disables singal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 */
James Bottomley3ed7a472005-09-19 09:50:04 -05001554 set_current_state(TASK_INTERRUPTIBLE);
1555 while (!kthread_should_stop()) {
Tejun Heoee7863b2006-05-15 20:57:20 +09001556 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05001557 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001558 SCSI_LOG_ERROR_RECOVERY(1,
1559 printk("Error handler scsi_eh_%d sleeping\n",
1560 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001561 schedule();
1562 set_current_state(TASK_INTERRUPTIBLE);
1563 continue;
1564 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
James Bottomley3ed7a472005-09-19 09:50:04 -05001566 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001567 SCSI_LOG_ERROR_RECOVERY(1,
1568 printk("Error handler scsi_eh_%d waking up\n",
1569 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 /*
1572 * We have a host that is failing for some reason. Figure out
1573 * what we need to do to get it up and online again (if we can).
1574 * If we fail, we end up taking the thing offline.
1575 */
Christoph Hellwig9227c332006-04-01 19:21:04 +02001576 if (shost->transportt->eh_strategy_handler)
1577 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 else
1579 scsi_unjam_host(shost);
1580
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 /*
1582 * Note - if the above fails completely, the action is to take
1583 * individual devices offline and flush the queue of any
1584 * outstanding requests that may have been pending. When we
1585 * restart, we restart any I/O to any other devices on the bus
1586 * which are still online.
1587 */
1588 scsi_restart_operations(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -05001589 set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04001591 __set_current_state(TASK_RUNNING);
1592
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001593 SCSI_LOG_ERROR_RECOVERY(1,
1594 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001595 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 return 0;
1597}
1598
1599/*
1600 * Function: scsi_report_bus_reset()
1601 *
1602 * Purpose: Utility function used by low-level drivers to report that
1603 * they have observed a bus reset on the bus being handled.
1604 *
1605 * Arguments: shost - Host in question
1606 * channel - channel on which reset was observed.
1607 *
1608 * Returns: Nothing
1609 *
1610 * Lock status: Host lock must be held.
1611 *
1612 * Notes: This only needs to be called if the reset is one which
1613 * originates from an unknown location. Resets originated
1614 * by the mid-level itself don't need to call this, but there
1615 * should be no harm.
1616 *
1617 * The main purpose of this is to make sure that a CHECK_CONDITION
1618 * is properly treated.
1619 */
1620void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1621{
1622 struct scsi_device *sdev;
1623
1624 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001625 if (channel == sdev_channel(sdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 sdev->was_reset = 1;
1627 sdev->expecting_cc_ua = 1;
1628 }
1629 }
1630}
1631EXPORT_SYMBOL(scsi_report_bus_reset);
1632
1633/*
1634 * Function: scsi_report_device_reset()
1635 *
1636 * Purpose: Utility function used by low-level drivers to report that
1637 * they have observed a device reset on the device being handled.
1638 *
1639 * Arguments: shost - Host in question
1640 * channel - channel on which reset was observed
1641 * target - target on which reset was observed
1642 *
1643 * Returns: Nothing
1644 *
1645 * Lock status: Host lock must be held
1646 *
1647 * Notes: This only needs to be called if the reset is one which
1648 * originates from an unknown location. Resets originated
1649 * by the mid-level itself don't need to call this, but there
1650 * should be no harm.
1651 *
1652 * The main purpose of this is to make sure that a CHECK_CONDITION
1653 * is properly treated.
1654 */
1655void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1656{
1657 struct scsi_device *sdev;
1658
1659 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001660 if (channel == sdev_channel(sdev) &&
1661 target == sdev_id(sdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 sdev->was_reset = 1;
1663 sdev->expecting_cc_ua = 1;
1664 }
1665 }
1666}
1667EXPORT_SYMBOL(scsi_report_device_reset);
1668
1669static void
1670scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1671{
1672}
1673
1674/*
1675 * Function: scsi_reset_provider
1676 *
1677 * Purpose: Send requested reset to a bus or device at any phase.
1678 *
1679 * Arguments: device - device to send reset to
1680 * flag - reset type (see scsi.h)
1681 *
1682 * Returns: SUCCESS/FAILURE.
1683 *
1684 * Notes: This is used by the SCSI Generic driver to provide
1685 * Bus/Device reset capability.
1686 */
1687int
1688scsi_reset_provider(struct scsi_device *dev, int flag)
1689{
1690 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
James Smartd7a1bb02006-03-08 14:50:12 -05001691 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05001693 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 int rtn;
1695
1696 scmd->request = &req;
1697 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02001698
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd));
1700
1701 scmd->scsi_done = scsi_reset_provider_done_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 scmd->request_buffer = NULL;
1703 scmd->request_bufflen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
1705 scmd->cmd_len = 0;
1706
1707 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
1709 init_timer(&scmd->eh_timeout);
1710
James Smartd7a1bb02006-03-08 14:50:12 -05001711 spin_lock_irqsave(shost->host_lock, flags);
1712 shost->tmf_in_progress = 1;
1713 spin_unlock_irqrestore(shost->host_lock, flags);
1714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 switch (flag) {
1716 case SCSI_TRY_RESET_DEVICE:
1717 rtn = scsi_try_bus_device_reset(scmd);
1718 if (rtn == SUCCESS)
1719 break;
1720 /* FALLTHROUGH */
1721 case SCSI_TRY_RESET_BUS:
1722 rtn = scsi_try_bus_reset(scmd);
1723 if (rtn == SUCCESS)
1724 break;
1725 /* FALLTHROUGH */
1726 case SCSI_TRY_RESET_HOST:
1727 rtn = scsi_try_host_reset(scmd);
1728 break;
1729 default:
1730 rtn = FAILED;
1731 }
1732
James Smartd7a1bb02006-03-08 14:50:12 -05001733 spin_lock_irqsave(shost->host_lock, flags);
1734 shost->tmf_in_progress = 0;
1735 spin_unlock_irqrestore(shost->host_lock, flags);
1736
1737 /*
1738 * be sure to wake up anyone who was sleeping or had their queue
1739 * suspended while we performed the TMF.
1740 */
1741 SCSI_LOG_ERROR_RECOVERY(3,
1742 printk("%s: waking up host to restart after TMF\n",
1743 __FUNCTION__));
1744
1745 wake_up(&shost->host_wait);
1746
1747 scsi_run_host_queues(shost);
1748
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 scsi_next_command(scmd);
1750 return rtn;
1751}
1752EXPORT_SYMBOL(scsi_reset_provider);
1753
1754/**
1755 * scsi_normalize_sense - normalize main elements from either fixed or
1756 * descriptor sense data format into a common format.
1757 *
1758 * @sense_buffer: byte array containing sense data returned by device
1759 * @sb_len: number of valid bytes in sense_buffer
1760 * @sshdr: pointer to instance of structure that common
1761 * elements are written to.
1762 *
1763 * Notes:
1764 * The "main elements" from sense data are: response_code, sense_key,
1765 * asc, ascq and additional_length (only for descriptor format).
1766 *
1767 * Typically this function can be called after a device has
1768 * responded to a SCSI command with the CHECK_CONDITION status.
1769 *
1770 * Return value:
1771 * 1 if valid sense data information found, else 0;
1772 **/
1773int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1774 struct scsi_sense_hdr *sshdr)
1775{
James Bottomley33aa6872005-08-28 11:31:14 -05001776 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 return 0;
1778
1779 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1780
1781 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05001782
1783 if (!scsi_sense_valid(sshdr))
1784 return 0;
1785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 if (sshdr->response_code >= 0x72) {
1787 /*
1788 * descriptor format
1789 */
1790 if (sb_len > 1)
1791 sshdr->sense_key = (sense_buffer[1] & 0xf);
1792 if (sb_len > 2)
1793 sshdr->asc = sense_buffer[2];
1794 if (sb_len > 3)
1795 sshdr->ascq = sense_buffer[3];
1796 if (sb_len > 7)
1797 sshdr->additional_length = sense_buffer[7];
1798 } else {
1799 /*
1800 * fixed format
1801 */
1802 if (sb_len > 2)
1803 sshdr->sense_key = (sense_buffer[2] & 0xf);
1804 if (sb_len > 7) {
1805 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1806 sb_len : (sense_buffer[7] + 8);
1807 if (sb_len > 12)
1808 sshdr->asc = sense_buffer[12];
1809 if (sb_len > 13)
1810 sshdr->ascq = sense_buffer[13];
1811 }
1812 }
1813
1814 return 1;
1815}
1816EXPORT_SYMBOL(scsi_normalize_sense);
1817
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1819 struct scsi_sense_hdr *sshdr)
1820{
1821 return scsi_normalize_sense(cmd->sense_buffer,
1822 sizeof(cmd->sense_buffer), sshdr);
1823}
1824EXPORT_SYMBOL(scsi_command_normalize_sense);
1825
1826/**
1827 * scsi_sense_desc_find - search for a given descriptor type in
1828 * descriptor sense data format.
1829 *
1830 * @sense_buffer: byte array of descriptor format sense data
1831 * @sb_len: number of valid bytes in sense_buffer
1832 * @desc_type: value of descriptor type to find
1833 * (e.g. 0 -> information)
1834 *
1835 * Notes:
1836 * only valid when sense data is in descriptor format
1837 *
1838 * Return value:
1839 * pointer to start of (first) descriptor if found else NULL
1840 **/
1841const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1842 int desc_type)
1843{
1844 int add_sen_len, add_len, desc_len, k;
1845 const u8 * descp;
1846
1847 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1848 return NULL;
1849 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1850 return NULL;
1851 add_sen_len = (add_sen_len < (sb_len - 8)) ?
1852 add_sen_len : (sb_len - 8);
1853 descp = &sense_buffer[8];
1854 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1855 descp += desc_len;
1856 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1857 desc_len = add_len + 2;
1858 if (descp[0] == desc_type)
1859 return descp;
1860 if (add_len < 0) // short descriptor ??
1861 break;
1862 }
1863 return NULL;
1864}
1865EXPORT_SYMBOL(scsi_sense_desc_find);
1866
1867/**
1868 * scsi_get_sense_info_fld - attempts to get information field from
1869 * sense data (either fixed or descriptor format)
1870 *
1871 * @sense_buffer: byte array of sense data
1872 * @sb_len: number of valid bytes in sense_buffer
1873 * @info_out: pointer to 64 integer where 8 or 4 byte information
1874 * field will be placed if found.
1875 *
1876 * Return value:
1877 * 1 if information field found, 0 if not found.
1878 **/
1879int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1880 u64 * info_out)
1881{
1882 int j;
1883 const u8 * ucp;
1884 u64 ull;
1885
1886 if (sb_len < 7)
1887 return 0;
1888 switch (sense_buffer[0] & 0x7f) {
1889 case 0x70:
1890 case 0x71:
1891 if (sense_buffer[0] & 0x80) {
1892 *info_out = (sense_buffer[3] << 24) +
1893 (sense_buffer[4] << 16) +
1894 (sense_buffer[5] << 8) + sense_buffer[6];
1895 return 1;
1896 } else
1897 return 0;
1898 case 0x72:
1899 case 0x73:
1900 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
1901 0 /* info desc */);
1902 if (ucp && (0xa == ucp[1])) {
1903 ull = 0;
1904 for (j = 0; j < 8; ++j) {
1905 if (j > 0)
1906 ull <<= 8;
1907 ull |= ucp[4 + j];
1908 }
1909 *info_out = ull;
1910 return 1;
1911 } else
1912 return 0;
1913 default:
1914 return 0;
1915 }
1916}
1917EXPORT_SYMBOL(scsi_get_sense_info_fld);