blob: 0c2c73be197469289e17defb2bebb28febc4814e [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.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080065 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090066void 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.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080089 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070090int 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/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 * scsi_times_out - Timeout function for normal scsi commands.
Jens Axboe242f9dc2008-09-14 05:55:09 -0700116 * @req: request that is timing out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 *
118 * Notes:
119 * We do not need to lock this. There is the potential for a race
120 * only in that the normal completion handling might run, but if the
121 * normal completion function determines that the timer has already
122 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800123 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700124enum blk_eh_timer_return scsi_times_out(struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700126 struct scsi_cmnd *scmd = req->special;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700127 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500128
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 scsi_log_completion(scmd, TIMEOUT_ERROR);
130
James Smartc829c392006-03-13 08:28:57 -0500131 if (scmd->device->host->transportt->eh_timed_out)
Christoph Hellwigfa990782008-11-05 15:03:15 +0100132 rtn = scmd->device->host->transportt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500133 else if (scmd->device->host->hostt->eh_timed_out)
Christoph Hellwigfa990782008-11-05 15:03:15 +0100134 rtn = scmd->device->host->hostt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500135
Christoph Hellwigfa990782008-11-05 15:03:15 +0100136 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
137 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
James Bottomley939647e2005-09-18 15:05:20 -0500138 scmd->result |= DID_TIME_OUT << 16;
Christoph Hellwigfa990782008-11-05 15:03:15 +0100139 rtn = BLK_EH_HANDLED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 }
Jens Axboe242f9dc2008-09-14 05:55:09 -0700141
Christoph Hellwigfa990782008-11-05 15:03:15 +0100142 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
145/**
146 * scsi_block_when_processing_errors - Prevent cmds from being queued.
147 * @sdev: Device on which we are performing recovery.
148 *
149 * Description:
150 * We block until the host is out of error recovery, and then check to
151 * see whether the host or the device is offline.
152 *
153 * Return value:
154 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800155 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156int scsi_block_when_processing_errors(struct scsi_device *sdev)
157{
158 int online;
159
James Bottomley939647e2005-09-18 15:05:20 -0500160 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 online = scsi_device_online(sdev);
163
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700164 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 online));
166
167 return online;
168}
169EXPORT_SYMBOL(scsi_block_when_processing_errors);
170
171#ifdef CONFIG_SCSI_LOGGING
172/**
173 * scsi_eh_prt_fail_stats - Log info on failures.
174 * @shost: scsi host being recovered.
175 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800176 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
178 struct list_head *work_q)
179{
180 struct scsi_cmnd *scmd;
181 struct scsi_device *sdev;
182 int total_failures = 0;
183 int cmd_failed = 0;
184 int cmd_cancel = 0;
185 int devices_failed = 0;
186
187 shost_for_each_device(sdev, shost) {
188 list_for_each_entry(scmd, work_q, eh_entry) {
189 if (scmd->device == sdev) {
190 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200191 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 ++cmd_cancel;
193 else
194 ++cmd_failed;
195 }
196 }
197
198 if (cmd_cancel || cmd_failed) {
199 SCSI_LOG_ERROR_RECOVERY(3,
James Bottomley9ccfc752005-10-02 11:45:08 -0500200 sdev_printk(KERN_INFO, sdev,
201 "%s: cmds failed: %d, cancel: %d\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700202 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500203 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 cmd_cancel = 0;
205 cmd_failed = 0;
206 ++devices_failed;
207 }
208 }
209
210 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
211 " devices require eh work\n",
212 total_failures, devices_failed));
213}
214#endif
215
216/**
217 * scsi_check_sense - Examine scsi cmd sense
218 * @scmd: Cmd to have sense checked.
219 *
220 * Return value:
221 * SUCCESS or FAILED or NEEDS_RETRY
222 *
223 * Notes:
224 * When a deferred error is detected the current command has
225 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800226 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227static int scsi_check_sense(struct scsi_cmnd *scmd)
228{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700229 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 struct scsi_sense_hdr sshdr;
231
232 if (! scsi_command_normalize_sense(scmd, &sshdr))
233 return FAILED; /* no valid sense data */
234
235 if (scsi_sense_is_deferred(&sshdr))
236 return NEEDS_RETRY;
237
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700238 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
239 sdev->scsi_dh_data->scsi_dh->check_sense) {
240 int rc;
241
242 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
243 if (rc != SCSI_RETURN_NOT_HANDLED)
244 return rc;
245 /* handler does not care. Drop down to default handling */
246 }
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 /*
249 * Previous logic looked for FILEMARK, EOM or ILI which are
250 * mainly associated with tapes and returned SUCCESS.
251 */
252 if (sshdr.response_code == 0x70) {
253 /* fixed format */
254 if (scmd->sense_buffer[2] & 0xe0)
255 return SUCCESS;
256 } else {
257 /*
258 * descriptor format: look for "stream commands sense data
259 * descriptor" (see SSC-3). Assume single sense data
260 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
261 */
262 if ((sshdr.additional_length > 3) &&
263 (scmd->sense_buffer[8] == 0x4) &&
264 (scmd->sense_buffer[11] & 0xe0))
265 return SUCCESS;
266 }
267
268 switch (sshdr.sense_key) {
269 case NO_SENSE:
270 return SUCCESS;
271 case RECOVERED_ERROR:
272 return /* soft_error */ SUCCESS;
273
274 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400275 if (sshdr.asc == 0x10) /* DIF */
276 return SUCCESS;
277
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 return NEEDS_RETRY;
279 case NOT_READY:
280 case UNIT_ATTENTION:
281 /*
282 * if we are expecting a cc/ua because of a bus reset that we
283 * performed, treat this just as a retry. otherwise this is
284 * information that we should pass up to the upper-level driver
285 * so that we can deal with it there.
286 */
287 if (scmd->device->expecting_cc_ua) {
288 scmd->device->expecting_cc_ua = 0;
289 return NEEDS_RETRY;
290 }
291 /*
292 * if the device is in the process of becoming ready, we
293 * should retry.
294 */
295 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
296 return NEEDS_RETRY;
297 /*
298 * if the device is not started, we need to wake
299 * the error handler to start the motor
300 */
301 if (scmd->device->allow_restart &&
302 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
303 return FAILED;
304 return SUCCESS;
305
306 /* these three are not supported */
307 case COPY_ABORTED:
308 case VOLUME_OVERFLOW:
309 case MISCOMPARE:
310 return SUCCESS;
311
312 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800313 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
314 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
315 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
316 return SUCCESS;
317 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 return NEEDS_RETRY;
319
320 case HARDWARE_ERROR:
321 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700322 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 else
324 return SUCCESS;
325
326 case ILLEGAL_REQUEST:
327 case BLANK_CHECK:
328 case DATA_PROTECT:
329 default:
330 return SUCCESS;
331 }
332}
333
334/**
335 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
336 * @scmd: SCSI cmd to examine.
337 *
338 * Notes:
339 * This is *only* called when we are examining the status of commands
340 * queued during error recovery. the main difference here is that we
341 * don't allow for the possibility of retries here, and we are a lot
342 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800343 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
345{
346 /*
347 * first check the host byte, to see if there is anything in there
348 * that would indicate what we need to do.
349 */
350 if (host_byte(scmd->result) == DID_RESET) {
351 /*
352 * rats. we are already in the error handler, so we now
353 * get to try and figure out what to do next. if the sense
354 * is valid, we have a pretty good idea of what to do.
355 * if not, we mark it as FAILED.
356 */
357 return scsi_check_sense(scmd);
358 }
359 if (host_byte(scmd->result) != DID_OK)
360 return FAILED;
361
362 /*
363 * next, check the message byte.
364 */
365 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
366 return FAILED;
367
368 /*
369 * now, check the status byte to see if this indicates
370 * anything special.
371 */
372 switch (status_byte(scmd->result)) {
373 case GOOD:
374 case COMMAND_TERMINATED:
375 return SUCCESS;
376 case CHECK_CONDITION:
377 return scsi_check_sense(scmd);
378 case CONDITION_GOOD:
379 case INTERMEDIATE_GOOD:
380 case INTERMEDIATE_C_GOOD:
381 /*
382 * who knows? FIXME(eric)
383 */
384 return SUCCESS;
385 case BUSY:
386 case QUEUE_FULL:
387 case RESERVATION_CONFLICT:
388 default:
389 return FAILED;
390 }
391 return FAILED;
392}
393
394/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 * scsi_eh_done - Completion function for error handling.
396 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800397 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398static void scsi_eh_done(struct scsi_cmnd *scmd)
399{
Michael Reed85631672005-12-07 21:46:27 -0600400 struct completion *eh_action;
401
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100402 SCSI_LOG_ERROR_RECOVERY(3,
403 printk("%s scmd: %p result: %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700404 __func__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600405
406 eh_action = scmd->device->host->eh_action;
407 if (eh_action)
408 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409}
410
411/**
Brian King292148f2007-01-30 17:51:17 -0600412 * scsi_try_host_reset - ask host adapter to reset itself
413 * @scmd: SCSI cmd to send hsot reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800414 */
Brian King292148f2007-01-30 17:51:17 -0600415static int scsi_try_host_reset(struct scsi_cmnd *scmd)
416{
417 unsigned long flags;
418 int rtn;
419
420 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700421 __func__));
Brian King292148f2007-01-30 17:51:17 -0600422
423 if (!scmd->device->host->hostt->eh_host_reset_handler)
424 return FAILED;
425
426 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
427
428 if (rtn == SUCCESS) {
429 if (!scmd->device->host->hostt->skip_settle_delay)
430 ssleep(HOST_RESET_SETTLE_TIME);
431 spin_lock_irqsave(scmd->device->host->host_lock, flags);
432 scsi_report_bus_reset(scmd->device->host,
433 scmd_channel(scmd));
434 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
435 }
436
437 return rtn;
438}
439
440/**
441 * scsi_try_bus_reset - ask host to perform a bus reset
442 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800443 */
Brian King292148f2007-01-30 17:51:17 -0600444static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
445{
446 unsigned long flags;
447 int rtn;
448
449 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700450 __func__));
Brian King292148f2007-01-30 17:51:17 -0600451
452 if (!scmd->device->host->hostt->eh_bus_reset_handler)
453 return FAILED;
454
455 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
456
457 if (rtn == SUCCESS) {
458 if (!scmd->device->host->hostt->skip_settle_delay)
459 ssleep(BUS_RESET_SETTLE_TIME);
460 spin_lock_irqsave(scmd->device->host->host_lock, flags);
461 scsi_report_bus_reset(scmd->device->host,
462 scmd_channel(scmd));
463 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
464 }
465
466 return rtn;
467}
468
Mike Christie30bd7df2008-02-29 18:25:19 -0600469static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
470{
471 sdev->was_reset = 1;
472 sdev->expecting_cc_ua = 1;
473}
474
475/**
476 * scsi_try_target_reset - Ask host to perform a target reset
477 * @scmd: SCSI cmd used to send a target reset
478 *
479 * Notes:
480 * There is no timeout for this operation. if this operation is
481 * unreliable for a given host, then the host itself needs to put a
482 * timer on it, and set the host back to a consistent state prior to
483 * returning.
484 */
485static int scsi_try_target_reset(struct scsi_cmnd *scmd)
486{
487 unsigned long flags;
488 int rtn;
489
490 if (!scmd->device->host->hostt->eh_target_reset_handler)
491 return FAILED;
492
493 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
494 if (rtn == SUCCESS) {
495 spin_lock_irqsave(scmd->device->host->host_lock, flags);
496 __starget_for_each_device(scsi_target(scmd->device), NULL,
497 __scsi_report_device_reset);
498 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
499 }
500
501 return rtn;
502}
503
Brian King292148f2007-01-30 17:51:17 -0600504/**
505 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
506 * @scmd: SCSI cmd used to send BDR
507 *
508 * Notes:
509 * There is no timeout for this operation. if this operation is
510 * unreliable for a given host, then the host itself needs to put a
511 * timer on it, and set the host back to a consistent state prior to
512 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800513 */
Brian King292148f2007-01-30 17:51:17 -0600514static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
515{
516 int rtn;
517
518 if (!scmd->device->host->hostt->eh_device_reset_handler)
519 return FAILED;
520
521 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600522 if (rtn == SUCCESS)
523 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600524 return rtn;
525}
526
527static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
528{
529 if (!scmd->device->host->hostt->eh_abort_handler)
530 return FAILED;
531
532 return scmd->device->host->hostt->eh_abort_handler(scmd);
533}
534
535/**
536 * scsi_try_to_abort_cmd - Ask host to abort a running command.
537 * @scmd: SCSI cmd to abort from Lower Level.
538 *
539 * Notes:
540 * This function will not return until the user's completion function
541 * has been called. there is no timeout on this operation. if the
542 * author of the low-level driver wishes this operation to be timed,
543 * they can provide this facility themselves. helper functions in
544 * scsi_error.c can be supplied to make this easier to do.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800545 */
Brian King292148f2007-01-30 17:51:17 -0600546static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
547{
548 /*
549 * scsi_done was called just after the command timed out and before
550 * we had a chance to process it. (db)
551 */
552 if (scmd->serial_number == 0)
553 return SUCCESS;
554 return __scsi_try_to_abort_cmd(scmd);
555}
556
557static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
558{
559 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
560 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600561 if (scsi_try_target_reset(scmd) != SUCCESS)
562 if (scsi_try_bus_reset(scmd) != SUCCESS)
563 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600564}
565
566/**
Boaz Harroshe1c23462007-10-08 16:36:45 +0200567 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100568 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200569 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200570 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300571 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200572 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100573 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200574 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200575 * as part of the error recovery process. If @sense_bytes is 0 the command
576 * sent must be one that does not transfer any data. If @sense_bytes != 0
577 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
578 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800579 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200580void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
581 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
f59114b2005-04-17 15:00:23 -0500583 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Christoph Hellwig631c2282006-07-08 20:42:15 +0200585 /*
586 * We need saved copies of a number of fields - this is because
587 * error handling may need to overwrite these with different values
588 * to run different commands, and once error handling is complete,
589 * we will need to restore these values prior to running the actual
590 * command.
591 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200592 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300593 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200594 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200595 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200596 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200597 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400598 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400599 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200600
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400601 scmd->prot_op = SCSI_PROT_NORMAL;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300602 scmd->cmnd = ses->eh_cmnd;
603 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200604 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200605 scmd->request->next_rq = NULL;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200606
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200607 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200608 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
609 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200610 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200611 scmd->sdb.length);
612 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200613 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200614 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200615 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200616 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200617 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200618 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200619 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200620 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300621 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200622 memcpy(scmd->cmnd, cmnd, cmnd_size);
623 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
624 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200625 }
626
627 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200628
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200629 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500631 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
Christoph Hellwig631c2282006-07-08 20:42:15 +0200633 /*
634 * Zero the sense buffer. The scsi spec mandates that any
635 * untransferred sense data should be interpreted as being zero.
636 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900637 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200638}
639EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200640
Boaz Harroshe1c23462007-10-08 16:36:45 +0200641/**
642 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
643 * @scmd: SCSI command structure to restore
644 * @ses: saved information from a coresponding call to scsi_prep_eh_cmnd
645 *
646 * Undo any damage done by above scsi_prep_eh_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800647 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200648void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
649{
650 /*
651 * Restore original data
652 */
653 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300654 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200655 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200656 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200657 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200658 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400659 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400660 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200661}
662EXPORT_SYMBOL(scsi_eh_restore_cmnd);
663
664/**
665 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
666 * @scmd: SCSI command structure to hijack
667 * @cmnd: CDB to send
668 * @cmnd_size: size in bytes of @cmnd
669 * @timeout: timeout for this request
670 * @sense_bytes: size of sense data to copy or 0
671 *
672 * This function is used to send a scsi command down to a target device
673 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
674 *
675 * Return value:
676 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800677 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200678static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
679 int cmnd_size, int timeout, unsigned sense_bytes)
680{
681 struct scsi_device *sdev = scmd->device;
682 struct Scsi_Host *shost = sdev->host;
683 DECLARE_COMPLETION_ONSTACK(done);
684 unsigned long timeleft;
685 unsigned long flags;
686 struct scsi_eh_save ses;
687 int rtn;
688
689 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100690 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
f59114b2005-04-17 15:00:23 -0500692 spin_lock_irqsave(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 scsi_log_send(scmd);
f59114b2005-04-17 15:00:23 -0500694 shost->hostt->queuecommand(scmd, scsi_eh_done);
695 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100697 timeleft = wait_for_completion_timeout(&done, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
f59114b2005-04-17 15:00:23 -0500699 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100701 scsi_log_completion(scmd, SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100703 SCSI_LOG_ERROR_RECOVERY(3,
704 printk("%s: scmd: %p, timeleft: %ld\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700705 __func__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
707 /*
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100708 * If there is time left scsi_eh_done got called, and we will
709 * examine the actual status codes to see whether the command
710 * actually did complete normally, else tell the host to forget
711 * about this command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100713 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 rtn = scsi_eh_completed_normally(scmd);
715 SCSI_LOG_ERROR_RECOVERY(3,
716 printk("%s: scsi_eh_completed_normally %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700717 __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 switch (rtn) {
720 case SUCCESS:
721 case NEEDS_RETRY:
722 case FAILED:
723 break;
724 default:
725 rtn = FAILED;
726 break;
727 }
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100728 } else {
Brian King292148f2007-01-30 17:51:17 -0600729 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100730 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
732
Boaz Harroshe1c23462007-10-08 16:36:45 +0200733 scsi_eh_restore_cmnd(scmd, &ses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 return rtn;
735}
736
737/**
738 * scsi_request_sense - Request sense data from a particular target.
739 * @scmd: SCSI cmd for request sense.
740 *
741 * Notes:
742 * Some hosts automatically obtain this information, others require
743 * that we obtain it on our own. This function will *not* return until
744 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800745 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746static int scsi_request_sense(struct scsi_cmnd *scmd)
747{
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200748 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749}
750
751/**
752 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
753 * @scmd: Original SCSI cmd that eh has finished.
754 * @done_q: Queue for processed commands.
755 *
756 * Notes:
757 * We don't want to use the normal command completion while we are are
758 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -0500759 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 * keep a list of pending commands for final completion, and once we
761 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800762 */
Tejun Heo041c5fc2006-01-23 13:09:36 +0900763void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764{
765 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200766 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 list_move_tail(&scmd->eh_entry, done_q);
768}
Tejun Heo041c5fc2006-01-23 13:09:36 +0900769EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
771/**
772 * scsi_eh_get_sense - Get device sense data.
773 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -0500774 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 *
776 * Description:
777 * See if we need to request sense information. if so, then get it
778 * now, so we have a better idea of what to do.
779 *
780 * Notes:
781 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -0500782 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 * it down before we request it.
784 *
785 * All drivers should request sense information internally these days,
786 * so for now all I have to say is tough noogies if you end up in here.
787 *
788 * XXX: Long term this code should go away, but that needs an audit of
789 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800790 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800791int scsi_eh_get_sense(struct list_head *work_q,
792 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200794 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 int rtn;
796
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200797 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200798 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 SCSI_SENSE_VALID(scmd))
800 continue;
801
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400802 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
803 "%s: requesting sense\n",
804 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 rtn = scsi_request_sense(scmd);
806 if (rtn != SUCCESS)
807 continue;
808
809 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
810 " result %x\n", scmd,
811 scmd->result));
812 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
813
814 rtn = scsi_decide_disposition(scmd);
815
816 /*
817 * if the result was normal, then just pass it along to the
818 * upper level.
819 */
820 if (rtn == SUCCESS)
821 /* we don't want this command reissued, just
822 * finished with the sense data, so set
823 * retries to the max allowed to ensure it
824 * won't get reissued */
825 scmd->retries = scmd->allowed;
826 else if (rtn != NEEDS_RETRY)
827 continue;
828
829 scsi_eh_finish_cmd(scmd, done_q);
830 }
831
832 return list_empty(work_q);
833}
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800834EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
836/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500838 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 *
840 * Return value:
841 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800842 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843static int scsi_eh_tur(struct scsi_cmnd *scmd)
844{
845 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
846 int retry_cnt = 1, rtn;
847
848retry_tur:
Mike Christie0db99e32006-08-26 03:00:22 -0400849 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700852 __func__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +0200853
854 switch (rtn) {
855 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 if (retry_cnt--)
857 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200858 /*FALLTHRU*/
859 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -0500860 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200861 default:
862 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -0500863 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864}
865
866/**
Rob Landleyeb448202007-11-03 13:30:39 -0500867 * scsi_eh_abort_cmds - abort pending commands.
868 * @work_q: &list_head for pending commands.
869 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 *
871 * Decription:
872 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -0500873 * running command. This only works out to be the case if we have one
874 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 * no sense to try and abort the command, since as far as the shost
876 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800877 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878static int scsi_eh_abort_cmds(struct list_head *work_q,
879 struct list_head *done_q)
880{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200881 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 int rtn;
883
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200884 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200885 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 continue;
887 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
888 "0x%p\n", current->comm,
889 scmd));
890 rtn = scsi_try_to_abort_cmd(scmd);
891 if (rtn == SUCCESS) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200892 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (!scsi_device_online(scmd->device) ||
894 !scsi_eh_tur(scmd)) {
895 scsi_eh_finish_cmd(scmd, done_q);
896 }
897
898 } else
899 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
900 " cmd failed:"
901 "0x%p\n",
902 current->comm,
903 scmd));
904 }
905
906 return list_empty(work_q);
907}
908
909/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500911 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 *
913 * Return value:
914 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800915 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
917{
918 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Christoph Hellwig631c2282006-07-08 20:42:15 +0200920 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -0500921 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
Brian Kinged773e62007-03-29 15:25:52 -0500923 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -0600924 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -0500925
Christoph Hellwig631c2282006-07-08 20:42:15 +0200926 if (rtn == SUCCESS)
927 return 0;
928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 return 1;
931}
932
933 /**
934 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -0500935 * @shost: &scsi host being recovered.
936 * @work_q: &list_head for pending commands.
937 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 *
939 * Notes:
940 * If commands are failing due to not ready, initializing command required,
941 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800942 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943static int scsi_eh_stu(struct Scsi_Host *shost,
944 struct list_head *work_q,
945 struct list_head *done_q)
946{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200947 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 struct scsi_device *sdev;
949
950 shost_for_each_device(sdev, shost) {
951 stu_scmd = NULL;
952 list_for_each_entry(scmd, work_q, eh_entry)
953 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
954 scsi_check_sense(scmd) == FAILED ) {
955 stu_scmd = scmd;
956 break;
957 }
958
959 if (!stu_scmd)
960 continue;
961
962 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
963 " 0x%p\n", current->comm, sdev));
964
965 if (!scsi_eh_try_stu(stu_scmd)) {
966 if (!scsi_device_online(sdev) ||
967 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200968 list_for_each_entry_safe(scmd, next,
969 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 if (scmd->device == sdev)
971 scsi_eh_finish_cmd(scmd, done_q);
972 }
973 }
974 } else {
975 SCSI_LOG_ERROR_RECOVERY(3,
976 printk("%s: START_UNIT failed to sdev:"
977 " 0x%p\n", current->comm, sdev));
978 }
979 }
980
981 return list_empty(work_q);
982}
983
984
985/**
986 * scsi_eh_bus_device_reset - send bdr if needed
987 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -0500988 * @work_q: &list_head for pending commands.
989 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 *
991 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -0500992 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 * devices that are jammed or not - if we have multiple devices, it
994 * makes no sense to try bus_device_reset - we really would need to try
995 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800996 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
998 struct list_head *work_q,
999 struct list_head *done_q)
1000{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001001 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 struct scsi_device *sdev;
1003 int rtn;
1004
1005 shost_for_each_device(sdev, shost) {
1006 bdr_scmd = NULL;
1007 list_for_each_entry(scmd, work_q, eh_entry)
1008 if (scmd->device == sdev) {
1009 bdr_scmd = scmd;
1010 break;
1011 }
1012
1013 if (!bdr_scmd)
1014 continue;
1015
1016 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1017 " 0x%p\n", current->comm,
1018 sdev));
1019 rtn = scsi_try_bus_device_reset(bdr_scmd);
1020 if (rtn == SUCCESS) {
1021 if (!scsi_device_online(sdev) ||
1022 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001023 list_for_each_entry_safe(scmd, next,
1024 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 if (scmd->device == sdev)
1026 scsi_eh_finish_cmd(scmd,
1027 done_q);
1028 }
1029 }
1030 } else {
1031 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1032 " failed sdev:"
1033 "0x%p\n",
1034 current->comm,
1035 sdev));
1036 }
1037 }
1038
1039 return list_empty(work_q);
1040}
1041
1042/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001043 * scsi_eh_target_reset - send target reset if needed
1044 * @shost: scsi host being recovered.
1045 * @work_q: &list_head for pending commands.
1046 * @done_q: &list_head for processed commands.
1047 *
1048 * Notes:
1049 * Try a target reset.
1050 */
1051static int scsi_eh_target_reset(struct Scsi_Host *shost,
1052 struct list_head *work_q,
1053 struct list_head *done_q)
1054{
1055 struct scsi_cmnd *scmd, *tgtr_scmd, *next;
James Bottomleyc82dc882008-09-12 16:46:51 -05001056 unsigned int id = 0;
Mike Christie30bd7df2008-02-29 18:25:19 -06001057 int rtn;
1058
James Bottomleyc82dc882008-09-12 16:46:51 -05001059 do {
Mike Christie30bd7df2008-02-29 18:25:19 -06001060 tgtr_scmd = NULL;
1061 list_for_each_entry(scmd, work_q, eh_entry) {
1062 if (id == scmd_id(scmd)) {
1063 tgtr_scmd = scmd;
1064 break;
1065 }
1066 }
James Bottomleyc82dc882008-09-12 16:46:51 -05001067 if (!tgtr_scmd) {
1068 /* not one exactly equal; find the next highest */
1069 list_for_each_entry(scmd, work_q, eh_entry) {
1070 if (scmd_id(scmd) > id &&
1071 (!tgtr_scmd ||
1072 scmd_id(tgtr_scmd) > scmd_id(scmd)))
1073 tgtr_scmd = scmd;
1074 }
1075 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001076 if (!tgtr_scmd)
James Bottomleyc82dc882008-09-12 16:46:51 -05001077 /* no more commands, that's it */
1078 break;
Mike Christie30bd7df2008-02-29 18:25:19 -06001079
1080 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1081 "to target %d\n",
1082 current->comm, id));
1083 rtn = scsi_try_target_reset(tgtr_scmd);
1084 if (rtn == SUCCESS) {
1085 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1086 if (id == scmd_id(scmd))
1087 if (!scsi_device_online(scmd->device) ||
1088 !scsi_eh_tur(tgtr_scmd))
1089 scsi_eh_finish_cmd(scmd,
1090 done_q);
1091 }
1092 } else
1093 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1094 " failed target: "
1095 "%d\n",
1096 current->comm, id));
James Bottomleyc82dc882008-09-12 16:46:51 -05001097 id++;
1098 } while(id != 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001099
1100 return list_empty(work_q);
1101}
1102
1103/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001105 * @shost: &scsi host being recovered.
1106 * @work_q: &list_head for pending commands.
1107 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001108 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1110 struct list_head *work_q,
1111 struct list_head *done_q)
1112{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001113 struct scsi_cmnd *scmd, *chan_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 unsigned int channel;
1115 int rtn;
1116
1117 /*
1118 * we really want to loop over the various channels, and do this on
1119 * a channel by channel basis. we should also check to see if any
1120 * of the failed commands are on soft_reset devices, and if so, skip
1121 * the reset.
1122 */
1123
1124 for (channel = 0; channel <= shost->max_channel; channel++) {
1125 chan_scmd = NULL;
1126 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001127 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 chan_scmd = scmd;
1129 break;
1130 /*
1131 * FIXME add back in some support for
1132 * soft_reset devices.
1133 */
1134 }
1135 }
1136
1137 if (!chan_scmd)
1138 continue;
1139 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1140 " %d\n", current->comm,
1141 channel));
1142 rtn = scsi_try_bus_reset(chan_scmd);
1143 if (rtn == SUCCESS) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001144 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001145 if (channel == scmd_channel(scmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 if (!scsi_device_online(scmd->device) ||
1147 !scsi_eh_tur(scmd))
1148 scsi_eh_finish_cmd(scmd,
1149 done_q);
1150 }
1151 } else {
1152 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1153 " failed chan: %d\n",
1154 current->comm,
1155 channel));
1156 }
1157 }
1158 return list_empty(work_q);
1159}
1160
1161/**
1162 * scsi_eh_host_reset - send a host reset
1163 * @work_q: list_head for processed commands.
1164 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001165 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166static int scsi_eh_host_reset(struct list_head *work_q,
1167 struct list_head *done_q)
1168{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001169 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
1172 if (!list_empty(work_q)) {
1173 scmd = list_entry(work_q->next,
1174 struct scsi_cmnd, eh_entry);
1175
1176 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1177 , current->comm));
1178
1179 rtn = scsi_try_host_reset(scmd);
1180 if (rtn == SUCCESS) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001181 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 if (!scsi_device_online(scmd->device) ||
1183 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1184 !scsi_eh_tur(scmd))
1185 scsi_eh_finish_cmd(scmd, done_q);
1186 }
1187 } else {
1188 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1189 " failed\n",
1190 current->comm));
1191 }
1192 }
1193 return list_empty(work_q);
1194}
1195
1196/**
1197 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1198 * @work_q: list_head for processed commands.
1199 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001200 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201static void scsi_eh_offline_sdevs(struct list_head *work_q,
1202 struct list_head *done_q)
1203{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001204 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001206 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001207 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1208 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001210 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 /*
1212 * FIXME: Handle lost cmds.
1213 */
1214 }
1215 scsi_eh_finish_cmd(scmd, done_q);
1216 }
1217 return;
1218}
1219
1220/**
Mike Christie4a274462008-08-19 18:45:31 -05001221 * scsi_noretry_cmd - determinte if command should be failed fast
1222 * @scmd: SCSI cmd to examine.
1223 */
1224int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1225{
1226 switch (host_byte(scmd->result)) {
1227 case DID_OK:
1228 break;
1229 case DID_BUS_BUSY:
1230 return blk_failfast_transport(scmd->request);
1231 case DID_PARITY:
1232 return blk_failfast_dev(scmd->request);
1233 case DID_ERROR:
1234 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1235 status_byte(scmd->result) == RESERVATION_CONFLICT)
1236 return 0;
1237 /* fall through */
1238 case DID_SOFT_ERROR:
1239 return blk_failfast_driver(scmd->request);
1240 }
1241
1242 switch (status_byte(scmd->result)) {
1243 case CHECK_CONDITION:
1244 /*
1245 * assume caller has checked sense and determinted
1246 * the check condition was retryable.
1247 */
1248 return blk_failfast_dev(scmd->request);
1249 }
1250
1251 return 0;
1252}
1253
1254/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1256 * @scmd: SCSI cmd to examine.
1257 *
1258 * Notes:
1259 * This is *only* called when we are examining the status after sending
1260 * out the actual data command. any commands that are queued for error
1261 * recovery (e.g. test_unit_ready) do *not* come through here.
1262 *
1263 * When this routine returns failed, it means the error handler thread
1264 * is woken. In cases where the error code indicates an error that
1265 * doesn't require the error handler read (i.e. we don't need to
1266 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001267 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268int scsi_decide_disposition(struct scsi_cmnd *scmd)
1269{
1270 int rtn;
1271
1272 /*
1273 * if the device is offline, then we clearly just pass the result back
1274 * up to the top level.
1275 */
1276 if (!scsi_device_online(scmd->device)) {
1277 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1278 " as SUCCESS\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001279 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 return SUCCESS;
1281 }
1282
1283 /*
1284 * first check the host byte, to see if there is anything in there
1285 * that would indicate what we need to do.
1286 */
1287 switch (host_byte(scmd->result)) {
1288 case DID_PASSTHROUGH:
1289 /*
1290 * no matter what, pass this through to the upper layer.
1291 * nuke this special code so that it looks like we are saying
1292 * did_ok.
1293 */
1294 scmd->result &= 0xff00ffff;
1295 return SUCCESS;
1296 case DID_OK:
1297 /*
1298 * looks good. drop through, and check the next byte.
1299 */
1300 break;
1301 case DID_NO_CONNECT:
1302 case DID_BAD_TARGET:
1303 case DID_ABORT:
1304 /*
1305 * note - this means that we just report the status back
1306 * to the top level driver, not that we actually think
1307 * that it indicates SUCCESS.
1308 */
1309 return SUCCESS;
1310 /*
1311 * when the low level driver returns did_soft_error,
1312 * it is responsible for keeping an internal retry counter
1313 * in order to avoid endless loops (db)
1314 *
1315 * actually this is a bug in this function here. we should
1316 * be mindful of the maximum number of retries specified
1317 * and not get stuck in a loop.
1318 */
1319 case DID_SOFT_ERROR:
1320 goto maybe_retry;
1321 case DID_IMM_RETRY:
1322 return NEEDS_RETRY;
1323
bf341912005-04-12 17:49:09 -05001324 case DID_REQUEUE:
1325 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001326 case DID_TRANSPORT_DISRUPTED:
1327 /*
1328 * LLD/transport was disrupted during processing of the IO.
1329 * The transport class is now blocked/blocking,
1330 * and the transport will decide what to do with the IO
Mike Christie939c2282008-11-04 19:47:19 -06001331 * based on its timers and recovery capablilities if
1332 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001333 */
Mike Christie939c2282008-11-04 19:47:19 -06001334 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001335 case DID_TRANSPORT_FAILFAST:
1336 /*
1337 * The transport decided to failfast the IO (most likely
1338 * the fast io fail tmo fired), so send IO directly upwards.
1339 */
1340 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 case DID_ERROR:
1342 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1343 status_byte(scmd->result) == RESERVATION_CONFLICT)
1344 /*
1345 * execute reservation conflict processing code
1346 * lower down
1347 */
1348 break;
1349 /* fallthrough */
1350
1351 case DID_BUS_BUSY:
1352 case DID_PARITY:
1353 goto maybe_retry;
1354 case DID_TIME_OUT:
1355 /*
1356 * when we scan the bus, we get timeout messages for
1357 * these commands if there is no device available.
1358 * other hosts report did_no_connect for the same thing.
1359 */
1360 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1361 scmd->cmnd[0] == INQUIRY)) {
1362 return SUCCESS;
1363 } else {
1364 return FAILED;
1365 }
1366 case DID_RESET:
1367 return SUCCESS;
1368 default:
1369 return FAILED;
1370 }
1371
1372 /*
1373 * next, check the message byte.
1374 */
1375 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1376 return FAILED;
1377
1378 /*
1379 * check the status byte to see if this indicates anything special.
1380 */
1381 switch (status_byte(scmd->result)) {
1382 case QUEUE_FULL:
1383 /*
1384 * the case of trying to send too many commands to a
1385 * tagged queueing device.
1386 */
1387 case BUSY:
1388 /*
1389 * device can't talk to us at the moment. Should only
1390 * occur (SAM-3) when the task queue is empty, so will cause
1391 * the empty queue handling to trigger a stall in the
1392 * device.
1393 */
1394 return ADD_TO_MLQUEUE;
1395 case GOOD:
1396 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001398 case TASK_ABORTED:
1399 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 case CHECK_CONDITION:
1401 rtn = scsi_check_sense(scmd);
1402 if (rtn == NEEDS_RETRY)
1403 goto maybe_retry;
1404 /* if rtn == FAILED, we have no sense information;
1405 * returning FAILED will wake the error handler thread
1406 * to collect the sense and redo the decide
1407 * disposition */
1408 return rtn;
1409 case CONDITION_GOOD:
1410 case INTERMEDIATE_GOOD:
1411 case INTERMEDIATE_C_GOOD:
1412 case ACA_ACTIVE:
1413 /*
1414 * who knows? FIXME(eric)
1415 */
1416 return SUCCESS;
1417
1418 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001419 sdev_printk(KERN_INFO, scmd->device,
1420 "reservation conflict\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 return SUCCESS; /* causes immediate i/o error */
1422 default:
1423 return FAILED;
1424 }
1425 return FAILED;
1426
1427 maybe_retry:
1428
1429 /* we requeue for retry because the error was retryable, and
1430 * the request was not marked fast fail. Note that above,
1431 * even if the request is marked fast fail, we still requeue
1432 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001433 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001434 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 return NEEDS_RETRY;
1436 } else {
1437 /*
1438 * no more retries - report this one back to upper level.
1439 */
1440 return SUCCESS;
1441 }
1442}
1443
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001444static void eh_lock_door_done(struct request *req, int uptodate)
1445{
1446 __blk_put_request(req->q, req);
1447}
1448
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 * scsi_eh_lock_door - Prevent medium removal for the specified device
1451 * @sdev: SCSI device to prevent medium removal
1452 *
1453 * Locking:
1454 * We must be called from process context; scsi_allocate_request()
1455 * may sleep.
1456 *
1457 * Notes:
1458 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1459 * head of the devices request queue, and continue.
1460 *
1461 * Bugs:
1462 * scsi_allocate_request() may sleep waiting for existing requests to
1463 * be processed. However, since we haven't kicked off any request
1464 * processing for this host, this may deadlock.
1465 *
1466 * If scsi_allocate_request() fails for what ever reason, we
1467 * completely forget to lock the door.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001468 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469static void scsi_eh_lock_door(struct scsi_device *sdev)
1470{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001471 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001473 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1474 if (!req)
1475 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001477 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1478 req->cmd[1] = 0;
1479 req->cmd[2] = 0;
1480 req->cmd[3] = 0;
1481 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1482 req->cmd[5] = 0;
1483
1484 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1485
1486 req->cmd_type = REQ_TYPE_BLOCK_PC;
1487 req->cmd_flags |= REQ_QUIET;
1488 req->timeout = 10 * HZ;
1489 req->retries = 5;
1490
1491 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492}
1493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494/**
1495 * scsi_restart_operations - restart io operations to the specified host.
1496 * @shost: Host we are restarting.
1497 *
1498 * Notes:
1499 * When we entered the error handler, we blocked all further i/o to
1500 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001501 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502static void scsi_restart_operations(struct Scsi_Host *shost)
1503{
1504 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001505 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
1507 /*
1508 * If the door was locked, we need to insert a door lock request
1509 * onto the head of the SCSI request queue for the device. There
1510 * is no point trying to lock the door of an off-line device.
1511 */
1512 shost_for_each_device(sdev, shost) {
1513 if (scsi_device_online(sdev) && sdev->locked)
1514 scsi_eh_lock_door(sdev);
1515 }
1516
1517 /*
1518 * next free up anything directly waiting upon the host. this
1519 * will be requests for character device operations, and also for
1520 * ioctls to queued block devices.
1521 */
1522 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001523 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524
James Bottomley939647e2005-09-18 15:05:20 -05001525 spin_lock_irqsave(shost->host_lock, flags);
1526 if (scsi_host_set_state(shost, SHOST_RUNNING))
1527 if (scsi_host_set_state(shost, SHOST_CANCEL))
1528 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1529 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
1531 wake_up(&shost->host_wait);
1532
1533 /*
1534 * finally we need to re-initiate requests that may be pending. we will
1535 * have had everything blocked while error handling is taking place, and
1536 * now that error recovery is done, we will need to ensure that these
1537 * requests are started.
1538 */
1539 scsi_run_host_queues(shost);
1540}
1541
1542/**
1543 * scsi_eh_ready_devs - check device ready state and recover if not.
1544 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001545 * @work_q: &list_head for pending commands.
1546 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001547 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001548void scsi_eh_ready_devs(struct Scsi_Host *shost,
1549 struct list_head *work_q,
1550 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551{
1552 if (!scsi_eh_stu(shost, work_q, done_q))
1553 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06001554 if (!scsi_eh_target_reset(shost, work_q, done_q))
1555 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1556 if (!scsi_eh_host_reset(work_q, done_q))
1557 scsi_eh_offline_sdevs(work_q,
1558 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001560EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
1562/**
1563 * scsi_eh_flush_done_q - finish processed commands or retry them.
1564 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001565 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001566void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001568 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001570 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1571 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05001573 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06001574 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1576 " retry cmd: %p\n",
1577 current->comm,
1578 scmd));
1579 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1580 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07001581 /*
1582 * If just we got sense for the device (called
1583 * scsi_eh_get_sense), scmd->result is already
1584 * set, do not set DRIVER_TIMEOUT.
1585 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 if (!scmd->result)
1587 scmd->result |= (DRIVER_TIMEOUT << 24);
1588 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1589 " cmd: %p\n",
1590 current->comm, scmd));
1591 scsi_finish_command(scmd);
1592 }
1593 }
1594}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001595EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
1597/**
1598 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1599 * @shost: Host to unjam.
1600 *
1601 * Notes:
1602 * When we come in here, we *know* that all commands on the bus have
1603 * either completed, failed or timed out. we also know that no further
1604 * commands are being sent to the host, so things are relatively quiet
1605 * and we have freedom to fiddle with things as we wish.
1606 *
1607 * This is only the *default* implementation. it is possible for
1608 * individual drivers to supply their own version of this function, and
1609 * if the maintainer wishes to do this, it is strongly suggested that
1610 * this function be taken as a template and modified. this function
1611 * was designed to correctly handle problems for about 95% of the
1612 * different cases out there, and it should always provide at least a
1613 * reasonable amount of error recovery.
1614 *
1615 * Any command marked 'failed' or 'timeout' must eventually have
1616 * scsi_finish_cmd() called for it. we do all of the retry stuff
1617 * here, so when we restart the host after we return it should have an
1618 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001619 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620static void scsi_unjam_host(struct Scsi_Host *shost)
1621{
1622 unsigned long flags;
1623 LIST_HEAD(eh_work_q);
1624 LIST_HEAD(eh_done_q);
1625
1626 spin_lock_irqsave(shost->host_lock, flags);
1627 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1628 spin_unlock_irqrestore(shost->host_lock, flags);
1629
1630 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1631
1632 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1633 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1634 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1635
1636 scsi_eh_flush_done_q(&eh_done_q);
1637}
1638
1639/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001640 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 * @data: Host for which we are running.
1642 *
1643 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001644 * This is the main error handling loop. This is run as a kernel thread
1645 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001646 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647int scsi_error_handler(void *data)
1648{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001649 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001652 * We use TASK_INTERRUPTIBLE so that the thread is not
1653 * counted against the load average as a running process.
1654 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01001655 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 */
James Bottomley3ed7a472005-09-19 09:50:04 -05001657 set_current_state(TASK_INTERRUPTIBLE);
1658 while (!kthread_should_stop()) {
Tejun Heoee7863b2006-05-15 20:57:20 +09001659 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05001660 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001661 SCSI_LOG_ERROR_RECOVERY(1,
1662 printk("Error handler scsi_eh_%d sleeping\n",
1663 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001664 schedule();
1665 set_current_state(TASK_INTERRUPTIBLE);
1666 continue;
1667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
James Bottomley3ed7a472005-09-19 09:50:04 -05001669 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001670 SCSI_LOG_ERROR_RECOVERY(1,
1671 printk("Error handler scsi_eh_%d waking up\n",
1672 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 /*
1675 * We have a host that is failing for some reason. Figure out
1676 * what we need to do to get it up and online again (if we can).
1677 * If we fail, we end up taking the thing offline.
1678 */
Christoph Hellwig9227c332006-04-01 19:21:04 +02001679 if (shost->transportt->eh_strategy_handler)
1680 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 else
1682 scsi_unjam_host(shost);
1683
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 /*
1685 * Note - if the above fails completely, the action is to take
1686 * individual devices offline and flush the queue of any
1687 * outstanding requests that may have been pending. When we
1688 * restart, we restart any I/O to any other devices on the bus
1689 * which are still online.
1690 */
1691 scsi_restart_operations(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -05001692 set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04001694 __set_current_state(TASK_RUNNING);
1695
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001696 SCSI_LOG_ERROR_RECOVERY(1,
1697 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001698 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 return 0;
1700}
1701
1702/*
1703 * Function: scsi_report_bus_reset()
1704 *
1705 * Purpose: Utility function used by low-level drivers to report that
1706 * they have observed a bus reset on the bus being handled.
1707 *
1708 * Arguments: shost - Host in question
1709 * channel - channel on which reset was observed.
1710 *
1711 * Returns: Nothing
1712 *
1713 * Lock status: Host lock must be held.
1714 *
1715 * Notes: This only needs to be called if the reset is one which
1716 * originates from an unknown location. Resets originated
1717 * by the mid-level itself don't need to call this, but there
1718 * should be no harm.
1719 *
1720 * The main purpose of this is to make sure that a CHECK_CONDITION
1721 * is properly treated.
1722 */
1723void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1724{
1725 struct scsi_device *sdev;
1726
1727 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06001728 if (channel == sdev_channel(sdev))
1729 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 }
1731}
1732EXPORT_SYMBOL(scsi_report_bus_reset);
1733
1734/*
1735 * Function: scsi_report_device_reset()
1736 *
1737 * Purpose: Utility function used by low-level drivers to report that
1738 * they have observed a device reset on the device being handled.
1739 *
1740 * Arguments: shost - Host in question
1741 * channel - channel on which reset was observed
1742 * target - target on which reset was observed
1743 *
1744 * Returns: Nothing
1745 *
1746 * Lock status: Host lock must be held
1747 *
1748 * Notes: This only needs to be called if the reset is one which
1749 * originates from an unknown location. Resets originated
1750 * by the mid-level itself don't need to call this, but there
1751 * should be no harm.
1752 *
1753 * The main purpose of this is to make sure that a CHECK_CONDITION
1754 * is properly treated.
1755 */
1756void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1757{
1758 struct scsi_device *sdev;
1759
1760 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001761 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06001762 target == sdev_id(sdev))
1763 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 }
1765}
1766EXPORT_SYMBOL(scsi_report_device_reset);
1767
1768static void
1769scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1770{
1771}
1772
1773/*
1774 * Function: scsi_reset_provider
1775 *
1776 * Purpose: Send requested reset to a bus or device at any phase.
1777 *
1778 * Arguments: device - device to send reset to
1779 * flag - reset type (see scsi.h)
1780 *
1781 * Returns: SUCCESS/FAILURE.
1782 *
1783 * Notes: This is used by the SCSI Generic driver to provide
1784 * Bus/Device reset capability.
1785 */
1786int
1787scsi_reset_provider(struct scsi_device *dev, int flag)
1788{
1789 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
James Smartd7a1bb02006-03-08 14:50:12 -05001790 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05001792 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 int rtn;
1794
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02001795 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02001797
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001798 scmd->cmnd = req.cmd;
1799
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001801 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
1803 scmd->cmd_len = 0;
1804
1805 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
James Smartd7a1bb02006-03-08 14:50:12 -05001807 spin_lock_irqsave(shost->host_lock, flags);
1808 shost->tmf_in_progress = 1;
1809 spin_unlock_irqrestore(shost->host_lock, flags);
1810
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 switch (flag) {
1812 case SCSI_TRY_RESET_DEVICE:
1813 rtn = scsi_try_bus_device_reset(scmd);
1814 if (rtn == SUCCESS)
1815 break;
1816 /* FALLTHROUGH */
Mike Christie30bd7df2008-02-29 18:25:19 -06001817 case SCSI_TRY_RESET_TARGET:
1818 rtn = scsi_try_target_reset(scmd);
1819 if (rtn == SUCCESS)
1820 break;
1821 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 case SCSI_TRY_RESET_BUS:
1823 rtn = scsi_try_bus_reset(scmd);
1824 if (rtn == SUCCESS)
1825 break;
1826 /* FALLTHROUGH */
1827 case SCSI_TRY_RESET_HOST:
1828 rtn = scsi_try_host_reset(scmd);
1829 break;
1830 default:
1831 rtn = FAILED;
1832 }
1833
James Smartd7a1bb02006-03-08 14:50:12 -05001834 spin_lock_irqsave(shost->host_lock, flags);
1835 shost->tmf_in_progress = 0;
1836 spin_unlock_irqrestore(shost->host_lock, flags);
1837
1838 /*
1839 * be sure to wake up anyone who was sleeping or had their queue
1840 * suspended while we performed the TMF.
1841 */
1842 SCSI_LOG_ERROR_RECOVERY(3,
1843 printk("%s: waking up host to restart after TMF\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001844 __func__));
James Smartd7a1bb02006-03-08 14:50:12 -05001845
1846 wake_up(&shost->host_wait);
1847
1848 scsi_run_host_queues(shost);
1849
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 scsi_next_command(scmd);
1851 return rtn;
1852}
1853EXPORT_SYMBOL(scsi_reset_provider);
1854
1855/**
1856 * scsi_normalize_sense - normalize main elements from either fixed or
1857 * descriptor sense data format into a common format.
1858 *
1859 * @sense_buffer: byte array containing sense data returned by device
1860 * @sb_len: number of valid bytes in sense_buffer
1861 * @sshdr: pointer to instance of structure that common
1862 * elements are written to.
1863 *
1864 * Notes:
1865 * The "main elements" from sense data are: response_code, sense_key,
1866 * asc, ascq and additional_length (only for descriptor format).
1867 *
1868 * Typically this function can be called after a device has
1869 * responded to a SCSI command with the CHECK_CONDITION status.
1870 *
1871 * Return value:
1872 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001873 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1875 struct scsi_sense_hdr *sshdr)
1876{
James Bottomley33aa6872005-08-28 11:31:14 -05001877 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 return 0;
1879
1880 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1881
1882 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05001883
1884 if (!scsi_sense_valid(sshdr))
1885 return 0;
1886
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 if (sshdr->response_code >= 0x72) {
1888 /*
1889 * descriptor format
1890 */
1891 if (sb_len > 1)
1892 sshdr->sense_key = (sense_buffer[1] & 0xf);
1893 if (sb_len > 2)
1894 sshdr->asc = sense_buffer[2];
1895 if (sb_len > 3)
1896 sshdr->ascq = sense_buffer[3];
1897 if (sb_len > 7)
1898 sshdr->additional_length = sense_buffer[7];
1899 } else {
1900 /*
1901 * fixed format
1902 */
1903 if (sb_len > 2)
1904 sshdr->sense_key = (sense_buffer[2] & 0xf);
1905 if (sb_len > 7) {
1906 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1907 sb_len : (sense_buffer[7] + 8);
1908 if (sb_len > 12)
1909 sshdr->asc = sense_buffer[12];
1910 if (sb_len > 13)
1911 sshdr->ascq = sense_buffer[13];
1912 }
1913 }
1914
1915 return 1;
1916}
1917EXPORT_SYMBOL(scsi_normalize_sense);
1918
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1920 struct scsi_sense_hdr *sshdr)
1921{
1922 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09001923 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924}
1925EXPORT_SYMBOL(scsi_command_normalize_sense);
1926
1927/**
Rob Landleyeb448202007-11-03 13:30:39 -05001928 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 * @sense_buffer: byte array of descriptor format sense data
1930 * @sb_len: number of valid bytes in sense_buffer
1931 * @desc_type: value of descriptor type to find
1932 * (e.g. 0 -> information)
1933 *
1934 * Notes:
1935 * only valid when sense data is in descriptor format
1936 *
1937 * Return value:
1938 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001939 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1941 int desc_type)
1942{
1943 int add_sen_len, add_len, desc_len, k;
1944 const u8 * descp;
1945
1946 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1947 return NULL;
1948 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1949 return NULL;
1950 add_sen_len = (add_sen_len < (sb_len - 8)) ?
1951 add_sen_len : (sb_len - 8);
1952 descp = &sense_buffer[8];
1953 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1954 descp += desc_len;
1955 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1956 desc_len = add_len + 2;
1957 if (descp[0] == desc_type)
1958 return descp;
1959 if (add_len < 0) // short descriptor ??
1960 break;
1961 }
1962 return NULL;
1963}
1964EXPORT_SYMBOL(scsi_sense_desc_find);
1965
1966/**
Rob Landleyeb448202007-11-03 13:30:39 -05001967 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 * @sense_buffer: byte array of sense data
1969 * @sb_len: number of valid bytes in sense_buffer
1970 * @info_out: pointer to 64 integer where 8 or 4 byte information
1971 * field will be placed if found.
1972 *
1973 * Return value:
1974 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001975 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1977 u64 * info_out)
1978{
1979 int j;
1980 const u8 * ucp;
1981 u64 ull;
1982
1983 if (sb_len < 7)
1984 return 0;
1985 switch (sense_buffer[0] & 0x7f) {
1986 case 0x70:
1987 case 0x71:
1988 if (sense_buffer[0] & 0x80) {
1989 *info_out = (sense_buffer[3] << 24) +
1990 (sense_buffer[4] << 16) +
1991 (sense_buffer[5] << 8) + sense_buffer[6];
1992 return 1;
1993 } else
1994 return 0;
1995 case 0x72:
1996 case 0x73:
1997 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
1998 0 /* info desc */);
1999 if (ucp && (0xa == ucp[1])) {
2000 ull = 0;
2001 for (j = 0; j < 8; ++j) {
2002 if (j > 0)
2003 ull <<= 8;
2004 ull |= ucp[4 + j];
2005 }
2006 *info_out = ull;
2007 return 1;
2008 } else
2009 return 0;
2010 default:
2011 return 0;
2012 }
2013}
2014EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002015
2016/**
2017 * scsi_build_sense_buffer - build sense data in a buffer
2018 * @desc: Sense format (non zero == descriptor format,
2019 * 0 == fixed format)
2020 * @buf: Where to build sense data
2021 * @key: Sense key
2022 * @asc: Additional sense code
2023 * @ascq: Additional sense code qualifier
2024 *
2025 **/
2026void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2027{
2028 if (desc) {
2029 buf[0] = 0x72; /* descriptor, current */
2030 buf[1] = key;
2031 buf[2] = asc;
2032 buf[3] = ascq;
2033 buf[7] = 0;
2034 } else {
2035 buf[0] = 0x70; /* fixed, current */
2036 buf[2] = key;
2037 buf[7] = 0xa;
2038 buf[12] = asc;
2039 buf[13] = ascq;
2040 }
2041}
2042EXPORT_SYMBOL(scsi_build_sense_buffer);