blob: 45c75649b9e083e489e1a935a2617a8d0461749f [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/timer.h>
21#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/kernel.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070023#include <linux/freezer.h>
Christoph Hellwigc5478de2005-09-06 14:04:26 +020024#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/interrupt.h>
26#include <linux/blkdev.h>
27#include <linux/delay.h>
28
29#include <scsi/scsi.h>
Christoph Hellwigbeb40482006-06-10 18:01:03 +020030#include <scsi/scsi_cmnd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <scsi/scsi_dbg.h>
32#include <scsi/scsi_device.h>
33#include <scsi/scsi_eh.h>
James Smartc829c392006-03-13 08:28:57 -050034#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <scsi/scsi_host.h>
36#include <scsi/scsi_ioctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include "scsi_priv.h"
39#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070040#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Kei Tokunagabf816232010-04-01 20:41:40 +090042#include <trace/events/scsi.h>
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#define SENSE_TIMEOUT (10*HZ)
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46/*
47 * These should *probably* be handled by the host itself.
48 * Since it is allowed to sleep, it probably should.
49 */
50#define BUS_RESET_SETTLE_TIME (10)
51#define HOST_RESET_SETTLE_TIME (10)
52
53/* called with shost->host_lock held */
54void scsi_eh_wakeup(struct Scsi_Host *shost)
55{
56 if (shost->host_busy == shost->host_failed) {
Kei Tokunagabf816232010-04-01 20:41:40 +090057 trace_scsi_eh_wakeup(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -050058 wake_up_process(shost->ehandler);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 SCSI_LOG_ERROR_RECOVERY(5,
60 printk("Waking error handler thread\n"));
61 }
62}
Tejun Heof8bbfc22006-05-19 21:07:05 +090063
64/**
65 * scsi_schedule_eh - schedule EH for SCSI host
66 * @shost: SCSI host to invoke error handling on.
67 *
68 * Schedule SCSI EH without scmd.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080069 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090070void scsi_schedule_eh(struct Scsi_Host *shost)
71{
72 unsigned long flags;
73
74 spin_lock_irqsave(shost->host_lock, flags);
75
76 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
77 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
78 shost->host_eh_scheduled++;
79 scsi_eh_wakeup(shost);
80 }
81
82 spin_unlock_irqrestore(shost->host_lock, flags);
83}
84EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86/**
87 * scsi_eh_scmd_add - add scsi cmd to error handling.
88 * @scmd: scmd to run eh on.
89 * @eh_flag: optional SCSI_EH flag.
90 *
91 * Return value:
92 * 0 on failure.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080093 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070094int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
95{
96 struct Scsi_Host *shost = scmd->device->host;
97 unsigned long flags;
James Bottomley939647e2005-09-18 15:05:20 -050098 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
James Bottomley3ed7a472005-09-19 09:50:04 -0500100 if (!shost->ehandler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 return 0;
102
103 spin_lock_irqsave(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500104 if (scsi_host_set_state(shost, SHOST_RECOVERY))
105 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
106 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
James Bottomley939647e2005-09-18 15:05:20 -0500108 ret = 1;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200109 scmd->eh_eflags |= eh_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 shost->host_failed++;
112 scsi_eh_wakeup(shost);
James Bottomley939647e2005-09-18 15:05:20 -0500113 out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 spin_unlock_irqrestore(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500115 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116}
117
118/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 * scsi_times_out - Timeout function for normal scsi commands.
Jens Axboe242f9dc2008-09-14 05:55:09 -0700120 * @req: request that is timing out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 *
122 * Notes:
123 * We do not need to lock this. There is the potential for a race
124 * only in that the normal completion handling might run, but if the
125 * normal completion function determines that the timer has already
126 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800127 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700128enum blk_eh_timer_return scsi_times_out(struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700130 struct scsi_cmnd *scmd = req->special;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700131 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500132
Kei Tokunagabf816232010-04-01 20:41:40 +0900133 trace_scsi_dispatch_cmd_timeout(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 scsi_log_completion(scmd, TIMEOUT_ERROR);
135
James Smartc829c392006-03-13 08:28:57 -0500136 if (scmd->device->host->transportt->eh_timed_out)
Christoph Hellwigfa990782008-11-05 15:03:15 +0100137 rtn = scmd->device->host->transportt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500138 else if (scmd->device->host->hostt->eh_timed_out)
Christoph Hellwigfa990782008-11-05 15:03:15 +0100139 rtn = scmd->device->host->hostt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500140
Christoph Hellwigfa990782008-11-05 15:03:15 +0100141 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
142 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
James Bottomley939647e2005-09-18 15:05:20 -0500143 scmd->result |= DID_TIME_OUT << 16;
Christoph Hellwigfa990782008-11-05 15:03:15 +0100144 rtn = BLK_EH_HANDLED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 }
Jens Axboe242f9dc2008-09-14 05:55:09 -0700146
Christoph Hellwigfa990782008-11-05 15:03:15 +0100147 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
150/**
151 * scsi_block_when_processing_errors - Prevent cmds from being queued.
152 * @sdev: Device on which we are performing recovery.
153 *
154 * Description:
155 * We block until the host is out of error recovery, and then check to
156 * see whether the host or the device is offline.
157 *
158 * Return value:
159 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800160 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161int scsi_block_when_processing_errors(struct scsi_device *sdev)
162{
163 int online;
164
James Bottomley939647e2005-09-18 15:05:20 -0500165 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167 online = scsi_device_online(sdev);
168
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700169 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 online));
171
172 return online;
173}
174EXPORT_SYMBOL(scsi_block_when_processing_errors);
175
176#ifdef CONFIG_SCSI_LOGGING
177/**
178 * scsi_eh_prt_fail_stats - Log info on failures.
179 * @shost: scsi host being recovered.
180 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800181 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
183 struct list_head *work_q)
184{
185 struct scsi_cmnd *scmd;
186 struct scsi_device *sdev;
187 int total_failures = 0;
188 int cmd_failed = 0;
189 int cmd_cancel = 0;
190 int devices_failed = 0;
191
192 shost_for_each_device(sdev, shost) {
193 list_for_each_entry(scmd, work_q, eh_entry) {
194 if (scmd->device == sdev) {
195 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200196 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 ++cmd_cancel;
198 else
199 ++cmd_failed;
200 }
201 }
202
203 if (cmd_cancel || cmd_failed) {
204 SCSI_LOG_ERROR_RECOVERY(3,
James Bottomley9ccfc752005-10-02 11:45:08 -0500205 sdev_printk(KERN_INFO, sdev,
206 "%s: cmds failed: %d, cancel: %d\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700207 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500208 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 cmd_cancel = 0;
210 cmd_failed = 0;
211 ++devices_failed;
212 }
213 }
214
215 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
216 " devices require eh work\n",
217 total_failures, devices_failed));
218}
219#endif
220
221/**
222 * scsi_check_sense - Examine scsi cmd sense
223 * @scmd: Cmd to have sense checked.
224 *
225 * Return value:
226 * SUCCESS or FAILED or NEEDS_RETRY
227 *
228 * Notes:
229 * When a deferred error is detected the current command has
230 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800231 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232static int scsi_check_sense(struct scsi_cmnd *scmd)
233{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700234 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 struct scsi_sense_hdr sshdr;
236
237 if (! scsi_command_normalize_sense(scmd, &sshdr))
238 return FAILED; /* no valid sense data */
239
240 if (scsi_sense_is_deferred(&sshdr))
241 return NEEDS_RETRY;
242
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700243 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
244 sdev->scsi_dh_data->scsi_dh->check_sense) {
245 int rc;
246
247 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
248 if (rc != SCSI_RETURN_NOT_HANDLED)
249 return rc;
250 /* handler does not care. Drop down to default handling */
251 }
252
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 /*
254 * Previous logic looked for FILEMARK, EOM or ILI which are
255 * mainly associated with tapes and returned SUCCESS.
256 */
257 if (sshdr.response_code == 0x70) {
258 /* fixed format */
259 if (scmd->sense_buffer[2] & 0xe0)
260 return SUCCESS;
261 } else {
262 /*
263 * descriptor format: look for "stream commands sense data
264 * descriptor" (see SSC-3). Assume single sense data
265 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
266 */
267 if ((sshdr.additional_length > 3) &&
268 (scmd->sense_buffer[8] == 0x4) &&
269 (scmd->sense_buffer[11] & 0xe0))
270 return SUCCESS;
271 }
272
273 switch (sshdr.sense_key) {
274 case NO_SENSE:
275 return SUCCESS;
276 case RECOVERED_ERROR:
277 return /* soft_error */ SUCCESS;
278
279 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400280 if (sshdr.asc == 0x10) /* DIF */
281 return SUCCESS;
282
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 return NEEDS_RETRY;
284 case NOT_READY:
285 case UNIT_ATTENTION:
286 /*
287 * if we are expecting a cc/ua because of a bus reset that we
288 * performed, treat this just as a retry. otherwise this is
289 * information that we should pass up to the upper-level driver
290 * so that we can deal with it there.
291 */
292 if (scmd->device->expecting_cc_ua) {
293 scmd->device->expecting_cc_ua = 0;
294 return NEEDS_RETRY;
295 }
296 /*
297 * if the device is in the process of becoming ready, we
298 * should retry.
299 */
300 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
301 return NEEDS_RETRY;
302 /*
303 * if the device is not started, we need to wake
304 * the error handler to start the motor
305 */
306 if (scmd->device->allow_restart &&
307 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
308 return FAILED;
James Bottomley77a42292010-05-04 16:51:40 -0400309
Mike Christie6e499492010-04-25 07:03:57 -0500310 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
311 scmd_printk(KERN_WARNING, scmd,
312 "Warning! Received an indication that the "
313 "LUN assignments on this target have "
314 "changed. The Linux SCSI layer does not "
315 "automatically remap LUN assignments.\n");
316 else if (sshdr.asc == 0x3f)
317 scmd_printk(KERN_WARNING, scmd,
318 "Warning! Received an indication that the "
319 "operating parameters on this target have "
320 "changed. The Linux SCSI layer does not "
321 "automatically adjust these parameters.\n");
322
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100323 /*
324 * Pass the UA upwards for a determination in the completion
325 * functions.
326 */
327 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
329 /* these three are not supported */
330 case COPY_ABORTED:
331 case VOLUME_OVERFLOW:
332 case MISCOMPARE:
333 return SUCCESS;
334
335 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800336 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
337 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
338 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
339 return SUCCESS;
340 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 return NEEDS_RETRY;
342
343 case HARDWARE_ERROR:
344 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700345 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 else
347 return SUCCESS;
348
349 case ILLEGAL_REQUEST:
350 case BLANK_CHECK:
351 case DATA_PROTECT:
352 default:
353 return SUCCESS;
354 }
355}
356
Vasu Dev4a840672009-10-22 15:46:33 -0700357static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
358{
359 struct scsi_host_template *sht = sdev->host->hostt;
360 struct scsi_device *tmp_sdev;
361
362 if (!sht->change_queue_depth ||
363 sdev->queue_depth >= sdev->max_queue_depth)
364 return;
365
366 if (time_before(jiffies,
367 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
368 return;
369
370 if (time_before(jiffies,
371 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
372 return;
373
374 /*
375 * Walk all devices of a target and do
376 * ramp up on them.
377 */
378 shost_for_each_device(tmp_sdev, sdev->host) {
379 if (tmp_sdev->channel != sdev->channel ||
380 tmp_sdev->id != sdev->id ||
381 tmp_sdev->queue_depth == sdev->max_queue_depth)
382 continue;
383 /*
384 * call back into LLD to increase queue_depth by one
385 * with ramp up reason code.
386 */
387 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
388 SCSI_QDEPTH_RAMP_UP);
389 sdev->last_queue_ramp_up = jiffies;
390 }
391}
392
Mike Christie42a6a912009-10-15 17:46:44 -0700393static void scsi_handle_queue_full(struct scsi_device *sdev)
394{
395 struct scsi_host_template *sht = sdev->host->hostt;
396 struct scsi_device *tmp_sdev;
397
398 if (!sht->change_queue_depth)
399 return;
400
401 shost_for_each_device(tmp_sdev, sdev->host) {
402 if (tmp_sdev->channel != sdev->channel ||
403 tmp_sdev->id != sdev->id)
404 continue;
405 /*
406 * We do not know the number of commands that were at
407 * the device when we got the queue full so we start
408 * from the highest possible value and work our way down.
409 */
410 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
411 SCSI_QDEPTH_QFULL);
412 }
413}
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415/**
416 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
417 * @scmd: SCSI cmd to examine.
418 *
419 * Notes:
420 * This is *only* called when we are examining the status of commands
421 * queued during error recovery. the main difference here is that we
422 * don't allow for the possibility of retries here, and we are a lot
423 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800424 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
426{
427 /*
428 * first check the host byte, to see if there is anything in there
429 * that would indicate what we need to do.
430 */
431 if (host_byte(scmd->result) == DID_RESET) {
432 /*
433 * rats. we are already in the error handler, so we now
434 * get to try and figure out what to do next. if the sense
435 * is valid, we have a pretty good idea of what to do.
436 * if not, we mark it as FAILED.
437 */
438 return scsi_check_sense(scmd);
439 }
440 if (host_byte(scmd->result) != DID_OK)
441 return FAILED;
442
443 /*
444 * next, check the message byte.
445 */
446 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
447 return FAILED;
448
449 /*
450 * now, check the status byte to see if this indicates
451 * anything special.
452 */
453 switch (status_byte(scmd->result)) {
454 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700455 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 case COMMAND_TERMINATED:
457 return SUCCESS;
458 case CHECK_CONDITION:
459 return scsi_check_sense(scmd);
460 case CONDITION_GOOD:
461 case INTERMEDIATE_GOOD:
462 case INTERMEDIATE_C_GOOD:
463 /*
464 * who knows? FIXME(eric)
465 */
466 return SUCCESS;
Michael Reed5f91bb02009-08-10 11:59:28 -0500467 case RESERVATION_CONFLICT:
James Bottomley67110df2010-08-06 15:17:24 -0500468 if (scmd->cmnd[0] == TEST_UNIT_READY)
469 /* it is a success, we probed the device and
470 * found it */
471 return SUCCESS;
472 /* otherwise, we failed to send the command */
473 return FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700475 scsi_handle_queue_full(scmd->device);
476 /* fall through */
477 case BUSY:
Hannes Reinecke3eb3a922010-07-29 10:10:16 +0200478 return NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 default:
480 return FAILED;
481 }
482 return FAILED;
483}
484
485/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 * scsi_eh_done - Completion function for error handling.
487 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800488 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489static void scsi_eh_done(struct scsi_cmnd *scmd)
490{
Michael Reed85631672005-12-07 21:46:27 -0600491 struct completion *eh_action;
492
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100493 SCSI_LOG_ERROR_RECOVERY(3,
494 printk("%s scmd: %p result: %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700495 __func__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600496
497 eh_action = scmd->device->host->eh_action;
498 if (eh_action)
499 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500}
501
502/**
Brian King292148f2007-01-30 17:51:17 -0600503 * scsi_try_host_reset - ask host adapter to reset itself
504 * @scmd: SCSI cmd to send hsot reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800505 */
Brian King292148f2007-01-30 17:51:17 -0600506static int scsi_try_host_reset(struct scsi_cmnd *scmd)
507{
508 unsigned long flags;
509 int rtn;
510
511 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700512 __func__));
Brian King292148f2007-01-30 17:51:17 -0600513
514 if (!scmd->device->host->hostt->eh_host_reset_handler)
515 return FAILED;
516
517 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
518
519 if (rtn == SUCCESS) {
520 if (!scmd->device->host->hostt->skip_settle_delay)
521 ssleep(HOST_RESET_SETTLE_TIME);
522 spin_lock_irqsave(scmd->device->host->host_lock, flags);
523 scsi_report_bus_reset(scmd->device->host,
524 scmd_channel(scmd));
525 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
526 }
527
528 return rtn;
529}
530
531/**
532 * scsi_try_bus_reset - ask host to perform a bus reset
533 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800534 */
Brian King292148f2007-01-30 17:51:17 -0600535static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
536{
537 unsigned long flags;
538 int rtn;
539
540 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700541 __func__));
Brian King292148f2007-01-30 17:51:17 -0600542
543 if (!scmd->device->host->hostt->eh_bus_reset_handler)
544 return FAILED;
545
546 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
547
548 if (rtn == SUCCESS) {
549 if (!scmd->device->host->hostt->skip_settle_delay)
550 ssleep(BUS_RESET_SETTLE_TIME);
551 spin_lock_irqsave(scmd->device->host->host_lock, flags);
552 scsi_report_bus_reset(scmd->device->host,
553 scmd_channel(scmd));
554 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
555 }
556
557 return rtn;
558}
559
Mike Christie30bd7df2008-02-29 18:25:19 -0600560static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
561{
562 sdev->was_reset = 1;
563 sdev->expecting_cc_ua = 1;
564}
565
566/**
567 * scsi_try_target_reset - Ask host to perform a target reset
568 * @scmd: SCSI cmd used to send a target reset
569 *
570 * Notes:
571 * There is no timeout for this operation. if this operation is
572 * unreliable for a given host, then the host itself needs to put a
573 * timer on it, and set the host back to a consistent state prior to
574 * returning.
575 */
576static int scsi_try_target_reset(struct scsi_cmnd *scmd)
577{
578 unsigned long flags;
579 int rtn;
580
581 if (!scmd->device->host->hostt->eh_target_reset_handler)
582 return FAILED;
583
584 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
585 if (rtn == SUCCESS) {
586 spin_lock_irqsave(scmd->device->host->host_lock, flags);
587 __starget_for_each_device(scsi_target(scmd->device), NULL,
588 __scsi_report_device_reset);
589 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
590 }
591
592 return rtn;
593}
594
Brian King292148f2007-01-30 17:51:17 -0600595/**
596 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
597 * @scmd: SCSI cmd used to send BDR
598 *
599 * Notes:
600 * There is no timeout for this operation. if this operation is
601 * unreliable for a given host, then the host itself needs to put a
602 * timer on it, and set the host back to a consistent state prior to
603 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800604 */
Brian King292148f2007-01-30 17:51:17 -0600605static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
606{
607 int rtn;
608
609 if (!scmd->device->host->hostt->eh_device_reset_handler)
610 return FAILED;
611
612 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600613 if (rtn == SUCCESS)
614 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600615 return rtn;
616}
617
James Bottomley459dbf72010-11-17 10:10:57 -0600618static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600619{
620 if (!scmd->device->host->hostt->eh_abort_handler)
621 return FAILED;
622
623 return scmd->device->host->hostt->eh_abort_handler(scmd);
624}
625
Brian King292148f2007-01-30 17:51:17 -0600626static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
627{
James Bottomley459dbf72010-11-17 10:10:57 -0600628 if (scsi_try_to_abort_cmd(scmd) != SUCCESS)
Brian King292148f2007-01-30 17:51:17 -0600629 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600630 if (scsi_try_target_reset(scmd) != SUCCESS)
631 if (scsi_try_bus_reset(scmd) != SUCCESS)
632 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600633}
634
635/**
Boaz Harroshe1c23462007-10-08 16:36:45 +0200636 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100637 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200638 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200639 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300640 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200641 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100642 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200643 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200644 * as part of the error recovery process. If @sense_bytes is 0 the command
645 * sent must be one that does not transfer any data. If @sense_bytes != 0
646 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
647 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800648 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200649void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
650 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651{
f59114b2005-04-17 15:00:23 -0500652 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Christoph Hellwig631c2282006-07-08 20:42:15 +0200654 /*
655 * We need saved copies of a number of fields - this is because
656 * error handling may need to overwrite these with different values
657 * to run different commands, and once error handling is complete,
658 * we will need to restore these values prior to running the actual
659 * command.
660 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200661 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300662 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200663 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200664 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200665 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200666 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400667 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400668 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200669
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400670 scmd->prot_op = SCSI_PROT_NORMAL;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300671 scmd->cmnd = ses->eh_cmnd;
672 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200673 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200674 scmd->request->next_rq = NULL;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200675
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200676 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200677 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
678 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200679 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200680 scmd->sdb.length);
681 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200682 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200683 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200684 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200685 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200686 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200687 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200688 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200689 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300690 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200691 memcpy(scmd->cmnd, cmnd, cmnd_size);
692 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
693 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200694 }
695
696 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200697
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200698 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500700 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
Christoph Hellwig631c2282006-07-08 20:42:15 +0200702 /*
703 * Zero the sense buffer. The scsi spec mandates that any
704 * untransferred sense data should be interpreted as being zero.
705 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900706 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200707}
708EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200709
Boaz Harroshe1c23462007-10-08 16:36:45 +0200710/**
711 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
712 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200713 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +0200714 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200715 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800716 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200717void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
718{
719 /*
720 * Restore original data
721 */
722 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300723 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200724 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200725 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200726 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200727 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400728 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400729 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200730}
731EXPORT_SYMBOL(scsi_eh_restore_cmnd);
732
733/**
734 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
735 * @scmd: SCSI command structure to hijack
736 * @cmnd: CDB to send
737 * @cmnd_size: size in bytes of @cmnd
738 * @timeout: timeout for this request
739 * @sense_bytes: size of sense data to copy or 0
740 *
741 * This function is used to send a scsi command down to a target device
742 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
743 *
744 * Return value:
745 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800746 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200747static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
748 int cmnd_size, int timeout, unsigned sense_bytes)
749{
750 struct scsi_device *sdev = scmd->device;
751 struct Scsi_Host *shost = sdev->host;
752 DECLARE_COMPLETION_ONSTACK(done);
753 unsigned long timeleft;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200754 struct scsi_eh_save ses;
755 int rtn;
756
757 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100758 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 scsi_log_send(scmd);
Jeff Garzikf2812332010-11-16 02:10:29 -0500761 scmd->scsi_done = scsi_eh_done;
762 shost->hostt->queuecommand(shost, scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100764 timeleft = wait_for_completion_timeout(&done, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
f59114b2005-04-17 15:00:23 -0500766 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100768 scsi_log_completion(scmd, SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100770 SCSI_LOG_ERROR_RECOVERY(3,
771 printk("%s: scmd: %p, timeleft: %ld\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700772 __func__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
774 /*
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100775 * If there is time left scsi_eh_done got called, and we will
776 * examine the actual status codes to see whether the command
777 * actually did complete normally, else tell the host to forget
778 * about this command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100780 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 rtn = scsi_eh_completed_normally(scmd);
782 SCSI_LOG_ERROR_RECOVERY(3,
783 printk("%s: scsi_eh_completed_normally %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700784 __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 switch (rtn) {
787 case SUCCESS:
788 case NEEDS_RETRY:
789 case FAILED:
790 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +0200791 case ADD_TO_MLQUEUE:
792 rtn = NEEDS_RETRY;
793 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 default:
795 rtn = FAILED;
796 break;
797 }
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100798 } else {
Brian King292148f2007-01-30 17:51:17 -0600799 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100800 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 }
802
Boaz Harroshe1c23462007-10-08 16:36:45 +0200803 scsi_eh_restore_cmnd(scmd, &ses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 return rtn;
805}
806
807/**
808 * scsi_request_sense - Request sense data from a particular target.
809 * @scmd: SCSI cmd for request sense.
810 *
811 * Notes:
812 * Some hosts automatically obtain this information, others require
813 * that we obtain it on our own. This function will *not* return until
814 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800815 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816static int scsi_request_sense(struct scsi_cmnd *scmd)
817{
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200818 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819}
820
821/**
822 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
823 * @scmd: Original SCSI cmd that eh has finished.
824 * @done_q: Queue for processed commands.
825 *
826 * Notes:
827 * We don't want to use the normal command completion while we are are
828 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -0500829 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 * keep a list of pending commands for final completion, and once we
831 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800832 */
Tejun Heo041c5fc2006-01-23 13:09:36 +0900833void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834{
835 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200836 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 list_move_tail(&scmd->eh_entry, done_q);
838}
Tejun Heo041c5fc2006-01-23 13:09:36 +0900839EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841/**
842 * scsi_eh_get_sense - Get device sense data.
843 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -0500844 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 *
846 * Description:
847 * See if we need to request sense information. if so, then get it
848 * now, so we have a better idea of what to do.
849 *
850 * Notes:
851 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -0500852 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 * it down before we request it.
854 *
855 * All drivers should request sense information internally these days,
856 * so for now all I have to say is tough noogies if you end up in here.
857 *
858 * XXX: Long term this code should go away, but that needs an audit of
859 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800860 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800861int scsi_eh_get_sense(struct list_head *work_q,
862 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
Christoph Hellwig937abea2005-06-19 13:43:56 +0200864 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 int rtn;
866
Christoph Hellwig937abea2005-06-19 13:43:56 +0200867 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200868 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 SCSI_SENSE_VALID(scmd))
870 continue;
871
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400872 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
873 "%s: requesting sense\n",
874 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 rtn = scsi_request_sense(scmd);
876 if (rtn != SUCCESS)
877 continue;
878
879 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
880 " result %x\n", scmd,
881 scmd->result));
882 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
883
884 rtn = scsi_decide_disposition(scmd);
885
886 /*
887 * if the result was normal, then just pass it along to the
888 * upper level.
889 */
890 if (rtn == SUCCESS)
891 /* we don't want this command reissued, just
892 * finished with the sense data, so set
893 * retries to the max allowed to ensure it
894 * won't get reissued */
895 scmd->retries = scmd->allowed;
896 else if (rtn != NEEDS_RETRY)
897 continue;
898
899 scsi_eh_finish_cmd(scmd, done_q);
900 }
901
902 return list_empty(work_q);
903}
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800904EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500908 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 *
910 * Return value:
911 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800912 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913static int scsi_eh_tur(struct scsi_cmnd *scmd)
914{
915 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
916 int retry_cnt = 1, rtn;
917
918retry_tur:
Mike Christie0db99e32006-08-26 03:00:22 -0400919 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700922 __func__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +0200923
924 switch (rtn) {
925 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 if (retry_cnt--)
927 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200928 /*FALLTHRU*/
929 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -0500930 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200931 default:
932 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -0500933 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934}
935
936/**
Rob Landleyeb448202007-11-03 13:30:39 -0500937 * scsi_eh_abort_cmds - abort pending commands.
938 * @work_q: &list_head for pending commands.
939 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 *
941 * Decription:
942 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -0500943 * running command. This only works out to be the case if we have one
944 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 * no sense to try and abort the command, since as far as the shost
946 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800947 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948static int scsi_eh_abort_cmds(struct list_head *work_q,
949 struct list_head *done_q)
950{
Christoph Hellwig937abea2005-06-19 13:43:56 +0200951 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 int rtn;
953
Christoph Hellwig937abea2005-06-19 13:43:56 +0200954 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200955 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 continue;
957 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
958 "0x%p\n", current->comm,
959 scmd));
960 rtn = scsi_try_to_abort_cmd(scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +0100961 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200962 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 if (!scsi_device_online(scmd->device) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +0100964 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 !scsi_eh_tur(scmd)) {
966 scsi_eh_finish_cmd(scmd, done_q);
967 }
968
969 } else
970 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
971 " cmd failed:"
972 "0x%p\n",
973 current->comm,
974 scmd));
975 }
976
977 return list_empty(work_q);
978}
979
980/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500982 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 *
984 * Return value:
985 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800986 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
988{
989 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
Christoph Hellwig631c2282006-07-08 20:42:15 +0200991 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -0500992 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Brian Kinged773e62007-03-29 15:25:52 -0500994 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -0600995 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -0500996
Christoph Hellwig631c2282006-07-08 20:42:15 +0200997 if (rtn == SUCCESS)
998 return 0;
999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 return 1;
1002}
1003
1004 /**
1005 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001006 * @shost: &scsi host being recovered.
1007 * @work_q: &list_head for pending commands.
1008 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 *
1010 * Notes:
1011 * If commands are failing due to not ready, initializing command required,
1012 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001013 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014static int scsi_eh_stu(struct Scsi_Host *shost,
1015 struct list_head *work_q,
1016 struct list_head *done_q)
1017{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001018 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 struct scsi_device *sdev;
1020
1021 shost_for_each_device(sdev, shost) {
1022 stu_scmd = NULL;
1023 list_for_each_entry(scmd, work_q, eh_entry)
1024 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1025 scsi_check_sense(scmd) == FAILED ) {
1026 stu_scmd = scmd;
1027 break;
1028 }
1029
1030 if (!stu_scmd)
1031 continue;
1032
1033 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1034 " 0x%p\n", current->comm, sdev));
1035
1036 if (!scsi_eh_try_stu(stu_scmd)) {
1037 if (!scsi_device_online(sdev) ||
1038 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001039 list_for_each_entry_safe(scmd, next,
1040 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 if (scmd->device == sdev)
1042 scsi_eh_finish_cmd(scmd, done_q);
1043 }
1044 }
1045 } else {
1046 SCSI_LOG_ERROR_RECOVERY(3,
1047 printk("%s: START_UNIT failed to sdev:"
1048 " 0x%p\n", current->comm, sdev));
1049 }
1050 }
1051
1052 return list_empty(work_q);
1053}
1054
1055
1056/**
1057 * scsi_eh_bus_device_reset - send bdr if needed
1058 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001059 * @work_q: &list_head for pending commands.
1060 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 *
1062 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001063 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 * devices that are jammed or not - if we have multiple devices, it
1065 * makes no sense to try bus_device_reset - we really would need to try
1066 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001067 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1069 struct list_head *work_q,
1070 struct list_head *done_q)
1071{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001072 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 struct scsi_device *sdev;
1074 int rtn;
1075
1076 shost_for_each_device(sdev, shost) {
1077 bdr_scmd = NULL;
1078 list_for_each_entry(scmd, work_q, eh_entry)
1079 if (scmd->device == sdev) {
1080 bdr_scmd = scmd;
1081 break;
1082 }
1083
1084 if (!bdr_scmd)
1085 continue;
1086
1087 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1088 " 0x%p\n", current->comm,
1089 sdev));
1090 rtn = scsi_try_bus_device_reset(bdr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001091 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 if (!scsi_device_online(sdev) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001093 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001095 list_for_each_entry_safe(scmd, next,
1096 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 if (scmd->device == sdev)
1098 scsi_eh_finish_cmd(scmd,
1099 done_q);
1100 }
1101 }
1102 } else {
1103 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1104 " failed sdev:"
1105 "0x%p\n",
1106 current->comm,
1107 sdev));
1108 }
1109 }
1110
1111 return list_empty(work_q);
1112}
1113
1114/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001115 * scsi_eh_target_reset - send target reset if needed
1116 * @shost: scsi host being recovered.
1117 * @work_q: &list_head for pending commands.
1118 * @done_q: &list_head for processed commands.
1119 *
1120 * Notes:
1121 * Try a target reset.
1122 */
1123static int scsi_eh_target_reset(struct Scsi_Host *shost,
1124 struct list_head *work_q,
1125 struct list_head *done_q)
1126{
James Bottomley98db5192010-10-25 15:53:41 -05001127 LIST_HEAD(tmp_list);
Mike Christie30bd7df2008-02-29 18:25:19 -06001128
James Bottomley98db5192010-10-25 15:53:41 -05001129 list_splice_init(work_q, &tmp_list);
1130
1131 while (!list_empty(&tmp_list)) {
1132 struct scsi_cmnd *next, *scmd;
1133 int rtn;
1134 unsigned int id;
1135
1136 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1137 id = scmd_id(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -06001138
1139 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1140 "to target %d\n",
1141 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001142 rtn = scsi_try_target_reset(scmd);
1143 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
Mike Christie30bd7df2008-02-29 18:25:19 -06001144 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1145 " failed target: "
1146 "%d\n",
1147 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001148 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1149 if (scmd_id(scmd) != id)
1150 continue;
1151
1152 if ((rtn == SUCCESS || rtn == FAST_IO_FAIL)
1153 && (!scsi_device_online(scmd->device) ||
1154 rtn == FAST_IO_FAIL || !scsi_eh_tur(scmd)))
1155 scsi_eh_finish_cmd(scmd, done_q);
1156 else
1157 /* push back on work queue for further processing */
1158 list_move(&scmd->eh_entry, work_q);
1159 }
1160 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001161
1162 return list_empty(work_q);
1163}
1164
1165/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001167 * @shost: &scsi host being recovered.
1168 * @work_q: &list_head for pending commands.
1169 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001170 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1172 struct list_head *work_q,
1173 struct list_head *done_q)
1174{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001175 struct scsi_cmnd *scmd, *chan_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 unsigned int channel;
1177 int rtn;
1178
1179 /*
1180 * we really want to loop over the various channels, and do this on
1181 * a channel by channel basis. we should also check to see if any
1182 * of the failed commands are on soft_reset devices, and if so, skip
1183 * the reset.
1184 */
1185
1186 for (channel = 0; channel <= shost->max_channel; channel++) {
1187 chan_scmd = NULL;
1188 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001189 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 chan_scmd = scmd;
1191 break;
1192 /*
1193 * FIXME add back in some support for
1194 * soft_reset devices.
1195 */
1196 }
1197 }
1198
1199 if (!chan_scmd)
1200 continue;
1201 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1202 " %d\n", current->comm,
1203 channel));
1204 rtn = scsi_try_bus_reset(chan_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001205 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001206 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001207 if (channel == scmd_channel(scmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 if (!scsi_device_online(scmd->device) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001209 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 !scsi_eh_tur(scmd))
1211 scsi_eh_finish_cmd(scmd,
1212 done_q);
1213 }
1214 } else {
1215 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1216 " failed chan: %d\n",
1217 current->comm,
1218 channel));
1219 }
1220 }
1221 return list_empty(work_q);
1222}
1223
1224/**
1225 * scsi_eh_host_reset - send a host reset
1226 * @work_q: list_head for processed commands.
1227 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001228 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229static int scsi_eh_host_reset(struct list_head *work_q,
1230 struct list_head *done_q)
1231{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001232 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234
1235 if (!list_empty(work_q)) {
1236 scmd = list_entry(work_q->next,
1237 struct scsi_cmnd, eh_entry);
1238
1239 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1240 , current->comm));
1241
1242 rtn = scsi_try_host_reset(scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001243 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001244 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 if (!scsi_device_online(scmd->device) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001246 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1248 !scsi_eh_tur(scmd))
1249 scsi_eh_finish_cmd(scmd, done_q);
1250 }
1251 } else {
1252 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1253 " failed\n",
1254 current->comm));
1255 }
1256 }
1257 return list_empty(work_q);
1258}
1259
1260/**
1261 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1262 * @work_q: list_head for processed commands.
1263 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001264 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265static void scsi_eh_offline_sdevs(struct list_head *work_q,
1266 struct list_head *done_q)
1267{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001268 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
Christoph Hellwig937abea2005-06-19 13:43:56 +02001270 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001271 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1272 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001274 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 /*
1276 * FIXME: Handle lost cmds.
1277 */
1278 }
1279 scsi_eh_finish_cmd(scmd, done_q);
1280 }
1281 return;
1282}
1283
1284/**
Mike Christie4a274462008-08-19 18:45:31 -05001285 * scsi_noretry_cmd - determinte if command should be failed fast
1286 * @scmd: SCSI cmd to examine.
1287 */
1288int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1289{
1290 switch (host_byte(scmd->result)) {
1291 case DID_OK:
1292 break;
1293 case DID_BUS_BUSY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001294 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
Mike Christie4a274462008-08-19 18:45:31 -05001295 case DID_PARITY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001296 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
Mike Christie4a274462008-08-19 18:45:31 -05001297 case DID_ERROR:
1298 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1299 status_byte(scmd->result) == RESERVATION_CONFLICT)
1300 return 0;
1301 /* fall through */
1302 case DID_SOFT_ERROR:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001303 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
Mike Christie4a274462008-08-19 18:45:31 -05001304 }
1305
1306 switch (status_byte(scmd->result)) {
1307 case CHECK_CONDITION:
1308 /*
1309 * assume caller has checked sense and determinted
1310 * the check condition was retryable.
1311 */
FUJITA Tomonorie96f6ab2010-07-09 09:38:26 +09001312 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1313 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1314 return 1;
Mike Christie4a274462008-08-19 18:45:31 -05001315 }
1316
1317 return 0;
1318}
1319
1320/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1322 * @scmd: SCSI cmd to examine.
1323 *
1324 * Notes:
1325 * This is *only* called when we are examining the status after sending
1326 * out the actual data command. any commands that are queued for error
1327 * recovery (e.g. test_unit_ready) do *not* come through here.
1328 *
1329 * When this routine returns failed, it means the error handler thread
1330 * is woken. In cases where the error code indicates an error that
1331 * doesn't require the error handler read (i.e. we don't need to
1332 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001333 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334int scsi_decide_disposition(struct scsi_cmnd *scmd)
1335{
1336 int rtn;
1337
1338 /*
1339 * if the device is offline, then we clearly just pass the result back
1340 * up to the top level.
1341 */
1342 if (!scsi_device_online(scmd->device)) {
1343 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1344 " as SUCCESS\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001345 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 return SUCCESS;
1347 }
1348
1349 /*
1350 * first check the host byte, to see if there is anything in there
1351 * that would indicate what we need to do.
1352 */
1353 switch (host_byte(scmd->result)) {
1354 case DID_PASSTHROUGH:
1355 /*
1356 * no matter what, pass this through to the upper layer.
1357 * nuke this special code so that it looks like we are saying
1358 * did_ok.
1359 */
1360 scmd->result &= 0xff00ffff;
1361 return SUCCESS;
1362 case DID_OK:
1363 /*
1364 * looks good. drop through, and check the next byte.
1365 */
1366 break;
1367 case DID_NO_CONNECT:
1368 case DID_BAD_TARGET:
1369 case DID_ABORT:
1370 /*
1371 * note - this means that we just report the status back
1372 * to the top level driver, not that we actually think
1373 * that it indicates SUCCESS.
1374 */
1375 return SUCCESS;
1376 /*
1377 * when the low level driver returns did_soft_error,
1378 * it is responsible for keeping an internal retry counter
1379 * in order to avoid endless loops (db)
1380 *
1381 * actually this is a bug in this function here. we should
1382 * be mindful of the maximum number of retries specified
1383 * and not get stuck in a loop.
1384 */
1385 case DID_SOFT_ERROR:
1386 goto maybe_retry;
1387 case DID_IMM_RETRY:
1388 return NEEDS_RETRY;
1389
bf341912005-04-12 17:49:09 -05001390 case DID_REQUEUE:
1391 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001392 case DID_TRANSPORT_DISRUPTED:
1393 /*
1394 * LLD/transport was disrupted during processing of the IO.
1395 * The transport class is now blocked/blocking,
1396 * and the transport will decide what to do with the IO
Mike Christie939c2282008-11-04 19:47:19 -06001397 * based on its timers and recovery capablilities if
1398 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001399 */
Mike Christie939c2282008-11-04 19:47:19 -06001400 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001401 case DID_TRANSPORT_FAILFAST:
1402 /*
1403 * The transport decided to failfast the IO (most likely
1404 * the fast io fail tmo fired), so send IO directly upwards.
1405 */
1406 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 case DID_ERROR:
1408 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1409 status_byte(scmd->result) == RESERVATION_CONFLICT)
1410 /*
1411 * execute reservation conflict processing code
1412 * lower down
1413 */
1414 break;
1415 /* fallthrough */
1416
1417 case DID_BUS_BUSY:
1418 case DID_PARITY:
1419 goto maybe_retry;
1420 case DID_TIME_OUT:
1421 /*
1422 * when we scan the bus, we get timeout messages for
1423 * these commands if there is no device available.
1424 * other hosts report did_no_connect for the same thing.
1425 */
1426 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1427 scmd->cmnd[0] == INQUIRY)) {
1428 return SUCCESS;
1429 } else {
1430 return FAILED;
1431 }
1432 case DID_RESET:
1433 return SUCCESS;
1434 default:
1435 return FAILED;
1436 }
1437
1438 /*
1439 * next, check the message byte.
1440 */
1441 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1442 return FAILED;
1443
1444 /*
1445 * check the status byte to see if this indicates anything special.
1446 */
1447 switch (status_byte(scmd->result)) {
1448 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001449 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 /*
1451 * the case of trying to send too many commands to a
1452 * tagged queueing device.
1453 */
1454 case BUSY:
1455 /*
1456 * device can't talk to us at the moment. Should only
1457 * occur (SAM-3) when the task queue is empty, so will cause
1458 * the empty queue handling to trigger a stall in the
1459 * device.
1460 */
1461 return ADD_TO_MLQUEUE;
1462 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -07001463 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001466 case TASK_ABORTED:
1467 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 case CHECK_CONDITION:
1469 rtn = scsi_check_sense(scmd);
1470 if (rtn == NEEDS_RETRY)
1471 goto maybe_retry;
1472 /* if rtn == FAILED, we have no sense information;
1473 * returning FAILED will wake the error handler thread
1474 * to collect the sense and redo the decide
1475 * disposition */
1476 return rtn;
1477 case CONDITION_GOOD:
1478 case INTERMEDIATE_GOOD:
1479 case INTERMEDIATE_C_GOOD:
1480 case ACA_ACTIVE:
1481 /*
1482 * who knows? FIXME(eric)
1483 */
1484 return SUCCESS;
1485
1486 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001487 sdev_printk(KERN_INFO, scmd->device,
1488 "reservation conflict\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 return SUCCESS; /* causes immediate i/o error */
1490 default:
1491 return FAILED;
1492 }
1493 return FAILED;
1494
1495 maybe_retry:
1496
1497 /* we requeue for retry because the error was retryable, and
1498 * the request was not marked fast fail. Note that above,
1499 * even if the request is marked fast fail, we still requeue
1500 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001501 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001502 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 return NEEDS_RETRY;
1504 } else {
1505 /*
1506 * no more retries - report this one back to upper level.
1507 */
1508 return SUCCESS;
1509 }
1510}
1511
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001512static void eh_lock_door_done(struct request *req, int uptodate)
1513{
1514 __blk_put_request(req->q, req);
1515}
1516
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 * scsi_eh_lock_door - Prevent medium removal for the specified device
1519 * @sdev: SCSI device to prevent medium removal
1520 *
1521 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001522 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 *
1524 * Notes:
1525 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1526 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001527 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528static void scsi_eh_lock_door(struct scsi_device *sdev)
1529{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001530 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
James Bottomley91bc31f2009-05-17 09:30:48 -05001532 /*
1533 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1534 * request becomes available
1535 */
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001536 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001538 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1539 req->cmd[1] = 0;
1540 req->cmd[2] = 0;
1541 req->cmd[3] = 0;
1542 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1543 req->cmd[5] = 0;
1544
1545 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1546
1547 req->cmd_type = REQ_TYPE_BLOCK_PC;
1548 req->cmd_flags |= REQ_QUIET;
1549 req->timeout = 10 * HZ;
1550 req->retries = 5;
1551
1552 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553}
1554
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555/**
1556 * scsi_restart_operations - restart io operations to the specified host.
1557 * @shost: Host we are restarting.
1558 *
1559 * Notes:
1560 * When we entered the error handler, we blocked all further i/o to
1561 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001562 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563static void scsi_restart_operations(struct Scsi_Host *shost)
1564{
1565 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001566 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
1568 /*
1569 * If the door was locked, we need to insert a door lock request
1570 * onto the head of the SCSI request queue for the device. There
1571 * is no point trying to lock the door of an off-line device.
1572 */
1573 shost_for_each_device(sdev, shost) {
1574 if (scsi_device_online(sdev) && sdev->locked)
1575 scsi_eh_lock_door(sdev);
1576 }
1577
1578 /*
1579 * next free up anything directly waiting upon the host. this
1580 * will be requests for character device operations, and also for
1581 * ioctls to queued block devices.
1582 */
1583 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001584 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
James Bottomley939647e2005-09-18 15:05:20 -05001586 spin_lock_irqsave(shost->host_lock, flags);
1587 if (scsi_host_set_state(shost, SHOST_RUNNING))
1588 if (scsi_host_set_state(shost, SHOST_CANCEL))
1589 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1590 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
1592 wake_up(&shost->host_wait);
1593
1594 /*
1595 * finally we need to re-initiate requests that may be pending. we will
1596 * have had everything blocked while error handling is taking place, and
1597 * now that error recovery is done, we will need to ensure that these
1598 * requests are started.
1599 */
1600 scsi_run_host_queues(shost);
1601}
1602
1603/**
1604 * scsi_eh_ready_devs - check device ready state and recover if not.
1605 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001606 * @work_q: &list_head for pending commands.
1607 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001608 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001609void scsi_eh_ready_devs(struct Scsi_Host *shost,
1610 struct list_head *work_q,
1611 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612{
1613 if (!scsi_eh_stu(shost, work_q, done_q))
1614 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06001615 if (!scsi_eh_target_reset(shost, work_q, done_q))
1616 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1617 if (!scsi_eh_host_reset(work_q, done_q))
1618 scsi_eh_offline_sdevs(work_q,
1619 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001621EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
1623/**
1624 * scsi_eh_flush_done_q - finish processed commands or retry them.
1625 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001626 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001627void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001629 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
Christoph Hellwig937abea2005-06-19 13:43:56 +02001631 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1632 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05001634 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06001635 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1637 " retry cmd: %p\n",
1638 current->comm,
1639 scmd));
1640 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1641 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07001642 /*
1643 * If just we got sense for the device (called
1644 * scsi_eh_get_sense), scmd->result is already
1645 * set, do not set DRIVER_TIMEOUT.
1646 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 if (!scmd->result)
1648 scmd->result |= (DRIVER_TIMEOUT << 24);
1649 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1650 " cmd: %p\n",
1651 current->comm, scmd));
1652 scsi_finish_command(scmd);
1653 }
1654 }
1655}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001656EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657
1658/**
1659 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1660 * @shost: Host to unjam.
1661 *
1662 * Notes:
1663 * When we come in here, we *know* that all commands on the bus have
1664 * either completed, failed or timed out. we also know that no further
1665 * commands are being sent to the host, so things are relatively quiet
1666 * and we have freedom to fiddle with things as we wish.
1667 *
1668 * This is only the *default* implementation. it is possible for
1669 * individual drivers to supply their own version of this function, and
1670 * if the maintainer wishes to do this, it is strongly suggested that
1671 * this function be taken as a template and modified. this function
1672 * was designed to correctly handle problems for about 95% of the
1673 * different cases out there, and it should always provide at least a
1674 * reasonable amount of error recovery.
1675 *
1676 * Any command marked 'failed' or 'timeout' must eventually have
1677 * scsi_finish_cmd() called for it. we do all of the retry stuff
1678 * here, so when we restart the host after we return it should have an
1679 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001680 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681static void scsi_unjam_host(struct Scsi_Host *shost)
1682{
1683 unsigned long flags;
1684 LIST_HEAD(eh_work_q);
1685 LIST_HEAD(eh_done_q);
1686
1687 spin_lock_irqsave(shost->host_lock, flags);
1688 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1689 spin_unlock_irqrestore(shost->host_lock, flags);
1690
1691 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1692
1693 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1694 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1695 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1696
1697 scsi_eh_flush_done_q(&eh_done_q);
1698}
1699
1700/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001701 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 * @data: Host for which we are running.
1703 *
1704 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001705 * This is the main error handling loop. This is run as a kernel thread
1706 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001707 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708int scsi_error_handler(void *data)
1709{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001710 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001713 * We use TASK_INTERRUPTIBLE so that the thread is not
1714 * counted against the load average as a running process.
1715 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01001716 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 */
James Bottomley3ed7a472005-09-19 09:50:04 -05001718 set_current_state(TASK_INTERRUPTIBLE);
1719 while (!kthread_should_stop()) {
Tejun Heoee7863b2006-05-15 20:57:20 +09001720 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05001721 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001722 SCSI_LOG_ERROR_RECOVERY(1,
1723 printk("Error handler scsi_eh_%d sleeping\n",
1724 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001725 schedule();
1726 set_current_state(TASK_INTERRUPTIBLE);
1727 continue;
1728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
James Bottomley3ed7a472005-09-19 09:50:04 -05001730 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001731 SCSI_LOG_ERROR_RECOVERY(1,
1732 printk("Error handler scsi_eh_%d waking up\n",
1733 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 /*
1736 * We have a host that is failing for some reason. Figure out
1737 * what we need to do to get it up and online again (if we can).
1738 * If we fail, we end up taking the thing offline.
1739 */
Alan Sternbc4f2402010-06-17 10:41:42 -04001740 if (scsi_autopm_get_host(shost) != 0) {
1741 SCSI_LOG_ERROR_RECOVERY(1,
1742 printk(KERN_ERR "Error handler scsi_eh_%d "
1743 "unable to autoresume\n",
1744 shost->host_no));
1745 continue;
1746 }
1747
Christoph Hellwig9227c332006-04-01 19:21:04 +02001748 if (shost->transportt->eh_strategy_handler)
1749 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 else
1751 scsi_unjam_host(shost);
1752
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 /*
1754 * Note - if the above fails completely, the action is to take
1755 * individual devices offline and flush the queue of any
1756 * outstanding requests that may have been pending. When we
1757 * restart, we restart any I/O to any other devices on the bus
1758 * which are still online.
1759 */
1760 scsi_restart_operations(shost);
Alan Sternbc4f2402010-06-17 10:41:42 -04001761 scsi_autopm_put_host(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -05001762 set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04001764 __set_current_state(TASK_RUNNING);
1765
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001766 SCSI_LOG_ERROR_RECOVERY(1,
1767 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001768 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 return 0;
1770}
1771
1772/*
1773 * Function: scsi_report_bus_reset()
1774 *
1775 * Purpose: Utility function used by low-level drivers to report that
1776 * they have observed a bus reset on the bus being handled.
1777 *
1778 * Arguments: shost - Host in question
1779 * channel - channel on which reset was observed.
1780 *
1781 * Returns: Nothing
1782 *
1783 * Lock status: Host lock must be held.
1784 *
1785 * Notes: This only needs to be called if the reset is one which
1786 * originates from an unknown location. Resets originated
1787 * by the mid-level itself don't need to call this, but there
1788 * should be no harm.
1789 *
1790 * The main purpose of this is to make sure that a CHECK_CONDITION
1791 * is properly treated.
1792 */
1793void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1794{
1795 struct scsi_device *sdev;
1796
1797 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06001798 if (channel == sdev_channel(sdev))
1799 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 }
1801}
1802EXPORT_SYMBOL(scsi_report_bus_reset);
1803
1804/*
1805 * Function: scsi_report_device_reset()
1806 *
1807 * Purpose: Utility function used by low-level drivers to report that
1808 * they have observed a device reset on the device being handled.
1809 *
1810 * Arguments: shost - Host in question
1811 * channel - channel on which reset was observed
1812 * target - target on which reset was observed
1813 *
1814 * Returns: Nothing
1815 *
1816 * Lock status: Host lock must be held
1817 *
1818 * Notes: This only needs to be called if the reset is one which
1819 * originates from an unknown location. Resets originated
1820 * by the mid-level itself don't need to call this, but there
1821 * should be no harm.
1822 *
1823 * The main purpose of this is to make sure that a CHECK_CONDITION
1824 * is properly treated.
1825 */
1826void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1827{
1828 struct scsi_device *sdev;
1829
1830 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001831 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06001832 target == sdev_id(sdev))
1833 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 }
1835}
1836EXPORT_SYMBOL(scsi_report_device_reset);
1837
1838static void
1839scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1840{
1841}
1842
1843/*
1844 * Function: scsi_reset_provider
1845 *
1846 * Purpose: Send requested reset to a bus or device at any phase.
1847 *
1848 * Arguments: device - device to send reset to
1849 * flag - reset type (see scsi.h)
1850 *
1851 * Returns: SUCCESS/FAILURE.
1852 *
1853 * Notes: This is used by the SCSI Generic driver to provide
1854 * Bus/Device reset capability.
1855 */
1856int
1857scsi_reset_provider(struct scsi_device *dev, int flag)
1858{
Alan Sternbc4f2402010-06-17 10:41:42 -04001859 struct scsi_cmnd *scmd;
James Smartd7a1bb02006-03-08 14:50:12 -05001860 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05001862 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 int rtn;
1864
Alan Sternbc4f2402010-06-17 10:41:42 -04001865 if (scsi_autopm_get_host(shost) < 0)
1866 return FAILED;
1867
1868 scmd = scsi_get_command(dev, GFP_KERNEL);
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02001869 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02001871
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001872 scmd->cmnd = req.cmd;
1873
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001875 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
1877 scmd->cmd_len = 0;
1878
1879 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
James Smartd7a1bb02006-03-08 14:50:12 -05001881 spin_lock_irqsave(shost->host_lock, flags);
1882 shost->tmf_in_progress = 1;
1883 spin_unlock_irqrestore(shost->host_lock, flags);
1884
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 switch (flag) {
1886 case SCSI_TRY_RESET_DEVICE:
1887 rtn = scsi_try_bus_device_reset(scmd);
1888 if (rtn == SUCCESS)
1889 break;
1890 /* FALLTHROUGH */
Mike Christie30bd7df2008-02-29 18:25:19 -06001891 case SCSI_TRY_RESET_TARGET:
1892 rtn = scsi_try_target_reset(scmd);
1893 if (rtn == SUCCESS)
1894 break;
1895 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 case SCSI_TRY_RESET_BUS:
1897 rtn = scsi_try_bus_reset(scmd);
1898 if (rtn == SUCCESS)
1899 break;
1900 /* FALLTHROUGH */
1901 case SCSI_TRY_RESET_HOST:
1902 rtn = scsi_try_host_reset(scmd);
1903 break;
1904 default:
1905 rtn = FAILED;
1906 }
1907
James Smartd7a1bb02006-03-08 14:50:12 -05001908 spin_lock_irqsave(shost->host_lock, flags);
1909 shost->tmf_in_progress = 0;
1910 spin_unlock_irqrestore(shost->host_lock, flags);
1911
1912 /*
1913 * be sure to wake up anyone who was sleeping or had their queue
1914 * suspended while we performed the TMF.
1915 */
1916 SCSI_LOG_ERROR_RECOVERY(3,
1917 printk("%s: waking up host to restart after TMF\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001918 __func__));
James Smartd7a1bb02006-03-08 14:50:12 -05001919
1920 wake_up(&shost->host_wait);
1921
1922 scsi_run_host_queues(shost);
1923
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 scsi_next_command(scmd);
Alan Sternbc4f2402010-06-17 10:41:42 -04001925 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 return rtn;
1927}
1928EXPORT_SYMBOL(scsi_reset_provider);
1929
1930/**
1931 * scsi_normalize_sense - normalize main elements from either fixed or
1932 * descriptor sense data format into a common format.
1933 *
1934 * @sense_buffer: byte array containing sense data returned by device
1935 * @sb_len: number of valid bytes in sense_buffer
1936 * @sshdr: pointer to instance of structure that common
1937 * elements are written to.
1938 *
1939 * Notes:
1940 * The "main elements" from sense data are: response_code, sense_key,
1941 * asc, ascq and additional_length (only for descriptor format).
1942 *
1943 * Typically this function can be called after a device has
1944 * responded to a SCSI command with the CHECK_CONDITION status.
1945 *
1946 * Return value:
1947 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001948 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1950 struct scsi_sense_hdr *sshdr)
1951{
James Bottomley33aa6872005-08-28 11:31:14 -05001952 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 return 0;
1954
1955 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1956
1957 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05001958
1959 if (!scsi_sense_valid(sshdr))
1960 return 0;
1961
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 if (sshdr->response_code >= 0x72) {
1963 /*
1964 * descriptor format
1965 */
1966 if (sb_len > 1)
1967 sshdr->sense_key = (sense_buffer[1] & 0xf);
1968 if (sb_len > 2)
1969 sshdr->asc = sense_buffer[2];
1970 if (sb_len > 3)
1971 sshdr->ascq = sense_buffer[3];
1972 if (sb_len > 7)
1973 sshdr->additional_length = sense_buffer[7];
1974 } else {
1975 /*
1976 * fixed format
1977 */
1978 if (sb_len > 2)
1979 sshdr->sense_key = (sense_buffer[2] & 0xf);
1980 if (sb_len > 7) {
1981 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1982 sb_len : (sense_buffer[7] + 8);
1983 if (sb_len > 12)
1984 sshdr->asc = sense_buffer[12];
1985 if (sb_len > 13)
1986 sshdr->ascq = sense_buffer[13];
1987 }
1988 }
1989
1990 return 1;
1991}
1992EXPORT_SYMBOL(scsi_normalize_sense);
1993
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1995 struct scsi_sense_hdr *sshdr)
1996{
1997 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09001998 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999}
2000EXPORT_SYMBOL(scsi_command_normalize_sense);
2001
2002/**
Rob Landleyeb448202007-11-03 13:30:39 -05002003 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 * @sense_buffer: byte array of descriptor format sense data
2005 * @sb_len: number of valid bytes in sense_buffer
2006 * @desc_type: value of descriptor type to find
2007 * (e.g. 0 -> information)
2008 *
2009 * Notes:
2010 * only valid when sense data is in descriptor format
2011 *
2012 * Return value:
2013 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002014 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2016 int desc_type)
2017{
2018 int add_sen_len, add_len, desc_len, k;
2019 const u8 * descp;
2020
2021 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2022 return NULL;
2023 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2024 return NULL;
2025 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2026 add_sen_len : (sb_len - 8);
2027 descp = &sense_buffer[8];
2028 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2029 descp += desc_len;
2030 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2031 desc_len = add_len + 2;
2032 if (descp[0] == desc_type)
2033 return descp;
2034 if (add_len < 0) // short descriptor ??
2035 break;
2036 }
2037 return NULL;
2038}
2039EXPORT_SYMBOL(scsi_sense_desc_find);
2040
2041/**
Rob Landleyeb448202007-11-03 13:30:39 -05002042 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 * @sense_buffer: byte array of sense data
2044 * @sb_len: number of valid bytes in sense_buffer
2045 * @info_out: pointer to 64 integer where 8 or 4 byte information
2046 * field will be placed if found.
2047 *
2048 * Return value:
2049 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002050 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2052 u64 * info_out)
2053{
2054 int j;
2055 const u8 * ucp;
2056 u64 ull;
2057
2058 if (sb_len < 7)
2059 return 0;
2060 switch (sense_buffer[0] & 0x7f) {
2061 case 0x70:
2062 case 0x71:
2063 if (sense_buffer[0] & 0x80) {
2064 *info_out = (sense_buffer[3] << 24) +
2065 (sense_buffer[4] << 16) +
2066 (sense_buffer[5] << 8) + sense_buffer[6];
2067 return 1;
2068 } else
2069 return 0;
2070 case 0x72:
2071 case 0x73:
2072 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2073 0 /* info desc */);
2074 if (ucp && (0xa == ucp[1])) {
2075 ull = 0;
2076 for (j = 0; j < 8; ++j) {
2077 if (j > 0)
2078 ull <<= 8;
2079 ull |= ucp[4 + j];
2080 }
2081 *info_out = ull;
2082 return 1;
2083 } else
2084 return 0;
2085 default:
2086 return 0;
2087 }
2088}
2089EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002090
2091/**
2092 * scsi_build_sense_buffer - build sense data in a buffer
2093 * @desc: Sense format (non zero == descriptor format,
2094 * 0 == fixed format)
2095 * @buf: Where to build sense data
2096 * @key: Sense key
2097 * @asc: Additional sense code
2098 * @ascq: Additional sense code qualifier
2099 *
2100 **/
2101void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2102{
2103 if (desc) {
2104 buf[0] = 0x72; /* descriptor, current */
2105 buf[1] = key;
2106 buf[2] = asc;
2107 buf[3] = ascq;
2108 buf[7] = 0;
2109 } else {
2110 buf[0] = 0x70; /* fixed, current */
2111 buf[2] = key;
2112 buf[7] = 0xa;
2113 buf[12] = asc;
2114 buf[13] = ascq;
2115 }
2116}
2117EXPORT_SYMBOL(scsi_build_sense_buffer);