blob: 1de30eb83bb097664056209ec6d06a470385b76d [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
Linus Torvalds2f9e8252010-08-10 15:22:42 -0700323 if (scmd->request->cmd_flags & REQ_HARDBARRIER)
James Bottomley77a42292010-05-04 16:51:40 -0400324 /*
325 * barrier requests should always retry on UA
326 * otherwise block will get a spurious error
327 */
328 return NEEDS_RETRY;
329 else
330 /*
331 * for normal (non barrier) commands, pass the
332 * UA upwards for a determination in the
333 * completion functions
334 */
335 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
337 /* these three are not supported */
338 case COPY_ABORTED:
339 case VOLUME_OVERFLOW:
340 case MISCOMPARE:
341 return SUCCESS;
342
343 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800344 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
345 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
346 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
347 return SUCCESS;
348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 return NEEDS_RETRY;
350
351 case HARDWARE_ERROR:
352 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700353 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 else
355 return SUCCESS;
356
357 case ILLEGAL_REQUEST:
358 case BLANK_CHECK:
359 case DATA_PROTECT:
360 default:
361 return SUCCESS;
362 }
363}
364
Vasu Dev4a840672009-10-22 15:46:33 -0700365static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
366{
367 struct scsi_host_template *sht = sdev->host->hostt;
368 struct scsi_device *tmp_sdev;
369
370 if (!sht->change_queue_depth ||
371 sdev->queue_depth >= sdev->max_queue_depth)
372 return;
373
374 if (time_before(jiffies,
375 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
376 return;
377
378 if (time_before(jiffies,
379 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
380 return;
381
382 /*
383 * Walk all devices of a target and do
384 * ramp up on them.
385 */
386 shost_for_each_device(tmp_sdev, sdev->host) {
387 if (tmp_sdev->channel != sdev->channel ||
388 tmp_sdev->id != sdev->id ||
389 tmp_sdev->queue_depth == sdev->max_queue_depth)
390 continue;
391 /*
392 * call back into LLD to increase queue_depth by one
393 * with ramp up reason code.
394 */
395 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
396 SCSI_QDEPTH_RAMP_UP);
397 sdev->last_queue_ramp_up = jiffies;
398 }
399}
400
Mike Christie42a6a912009-10-15 17:46:44 -0700401static void scsi_handle_queue_full(struct scsi_device *sdev)
402{
403 struct scsi_host_template *sht = sdev->host->hostt;
404 struct scsi_device *tmp_sdev;
405
406 if (!sht->change_queue_depth)
407 return;
408
409 shost_for_each_device(tmp_sdev, sdev->host) {
410 if (tmp_sdev->channel != sdev->channel ||
411 tmp_sdev->id != sdev->id)
412 continue;
413 /*
414 * We do not know the number of commands that were at
415 * the device when we got the queue full so we start
416 * from the highest possible value and work our way down.
417 */
418 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
419 SCSI_QDEPTH_QFULL);
420 }
421}
422
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423/**
424 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
425 * @scmd: SCSI cmd to examine.
426 *
427 * Notes:
428 * This is *only* called when we are examining the status of commands
429 * queued during error recovery. the main difference here is that we
430 * don't allow for the possibility of retries here, and we are a lot
431 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800432 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
434{
435 /*
436 * first check the host byte, to see if there is anything in there
437 * that would indicate what we need to do.
438 */
439 if (host_byte(scmd->result) == DID_RESET) {
440 /*
441 * rats. we are already in the error handler, so we now
442 * get to try and figure out what to do next. if the sense
443 * is valid, we have a pretty good idea of what to do.
444 * if not, we mark it as FAILED.
445 */
446 return scsi_check_sense(scmd);
447 }
448 if (host_byte(scmd->result) != DID_OK)
449 return FAILED;
450
451 /*
452 * next, check the message byte.
453 */
454 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
455 return FAILED;
456
457 /*
458 * now, check the status byte to see if this indicates
459 * anything special.
460 */
461 switch (status_byte(scmd->result)) {
462 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700463 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 case COMMAND_TERMINATED:
465 return SUCCESS;
466 case CHECK_CONDITION:
467 return scsi_check_sense(scmd);
468 case CONDITION_GOOD:
469 case INTERMEDIATE_GOOD:
470 case INTERMEDIATE_C_GOOD:
471 /*
472 * who knows? FIXME(eric)
473 */
474 return SUCCESS;
Michael Reed5f91bb02009-08-10 11:59:28 -0500475 case RESERVATION_CONFLICT:
James Bottomley67110df2010-08-06 15:17:24 -0500476 if (scmd->cmnd[0] == TEST_UNIT_READY)
477 /* it is a success, we probed the device and
478 * found it */
479 return SUCCESS;
480 /* otherwise, we failed to send the command */
481 return FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700483 scsi_handle_queue_full(scmd->device);
484 /* fall through */
485 case BUSY:
Hannes Reinecke3eb3a922010-07-29 10:10:16 +0200486 return NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 default:
488 return FAILED;
489 }
490 return FAILED;
491}
492
493/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 * scsi_eh_done - Completion function for error handling.
495 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800496 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497static void scsi_eh_done(struct scsi_cmnd *scmd)
498{
Michael Reed85631672005-12-07 21:46:27 -0600499 struct completion *eh_action;
500
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100501 SCSI_LOG_ERROR_RECOVERY(3,
502 printk("%s scmd: %p result: %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700503 __func__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600504
505 eh_action = scmd->device->host->eh_action;
506 if (eh_action)
507 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508}
509
510/**
Brian King292148f2007-01-30 17:51:17 -0600511 * scsi_try_host_reset - ask host adapter to reset itself
512 * @scmd: SCSI cmd to send hsot reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800513 */
Brian King292148f2007-01-30 17:51:17 -0600514static int scsi_try_host_reset(struct scsi_cmnd *scmd)
515{
516 unsigned long flags;
517 int rtn;
518
519 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700520 __func__));
Brian King292148f2007-01-30 17:51:17 -0600521
522 if (!scmd->device->host->hostt->eh_host_reset_handler)
523 return FAILED;
524
525 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
526
527 if (rtn == SUCCESS) {
528 if (!scmd->device->host->hostt->skip_settle_delay)
529 ssleep(HOST_RESET_SETTLE_TIME);
530 spin_lock_irqsave(scmd->device->host->host_lock, flags);
531 scsi_report_bus_reset(scmd->device->host,
532 scmd_channel(scmd));
533 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
534 }
535
536 return rtn;
537}
538
539/**
540 * scsi_try_bus_reset - ask host to perform a bus reset
541 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800542 */
Brian King292148f2007-01-30 17:51:17 -0600543static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
544{
545 unsigned long flags;
546 int rtn;
547
548 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700549 __func__));
Brian King292148f2007-01-30 17:51:17 -0600550
551 if (!scmd->device->host->hostt->eh_bus_reset_handler)
552 return FAILED;
553
554 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
555
556 if (rtn == SUCCESS) {
557 if (!scmd->device->host->hostt->skip_settle_delay)
558 ssleep(BUS_RESET_SETTLE_TIME);
559 spin_lock_irqsave(scmd->device->host->host_lock, flags);
560 scsi_report_bus_reset(scmd->device->host,
561 scmd_channel(scmd));
562 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
563 }
564
565 return rtn;
566}
567
Mike Christie30bd7df2008-02-29 18:25:19 -0600568static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
569{
570 sdev->was_reset = 1;
571 sdev->expecting_cc_ua = 1;
572}
573
574/**
575 * scsi_try_target_reset - Ask host to perform a target reset
576 * @scmd: SCSI cmd used to send a target reset
577 *
578 * Notes:
579 * There is no timeout for this operation. if this operation is
580 * unreliable for a given host, then the host itself needs to put a
581 * timer on it, and set the host back to a consistent state prior to
582 * returning.
583 */
584static int scsi_try_target_reset(struct scsi_cmnd *scmd)
585{
586 unsigned long flags;
587 int rtn;
588
589 if (!scmd->device->host->hostt->eh_target_reset_handler)
590 return FAILED;
591
592 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
593 if (rtn == SUCCESS) {
594 spin_lock_irqsave(scmd->device->host->host_lock, flags);
595 __starget_for_each_device(scsi_target(scmd->device), NULL,
596 __scsi_report_device_reset);
597 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
598 }
599
600 return rtn;
601}
602
Brian King292148f2007-01-30 17:51:17 -0600603/**
604 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
605 * @scmd: SCSI cmd used to send BDR
606 *
607 * Notes:
608 * There is no timeout for this operation. if this operation is
609 * unreliable for a given host, then the host itself needs to put a
610 * timer on it, and set the host back to a consistent state prior to
611 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800612 */
Brian King292148f2007-01-30 17:51:17 -0600613static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
614{
615 int rtn;
616
617 if (!scmd->device->host->hostt->eh_device_reset_handler)
618 return FAILED;
619
620 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600621 if (rtn == SUCCESS)
622 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600623 return rtn;
624}
625
626static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
627{
628 if (!scmd->device->host->hostt->eh_abort_handler)
629 return FAILED;
630
631 return scmd->device->host->hostt->eh_abort_handler(scmd);
632}
633
634/**
635 * scsi_try_to_abort_cmd - Ask host to abort a running command.
636 * @scmd: SCSI cmd to abort from Lower Level.
637 *
638 * Notes:
639 * This function will not return until the user's completion function
640 * has been called. there is no timeout on this operation. if the
641 * author of the low-level driver wishes this operation to be timed,
642 * they can provide this facility themselves. helper functions in
643 * scsi_error.c can be supplied to make this easier to do.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800644 */
Brian King292148f2007-01-30 17:51:17 -0600645static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
646{
647 /*
648 * scsi_done was called just after the command timed out and before
649 * we had a chance to process it. (db)
650 */
651 if (scmd->serial_number == 0)
652 return SUCCESS;
653 return __scsi_try_to_abort_cmd(scmd);
654}
655
656static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
657{
658 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
659 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600660 if (scsi_try_target_reset(scmd) != SUCCESS)
661 if (scsi_try_bus_reset(scmd) != SUCCESS)
662 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600663}
664
665/**
Boaz Harroshe1c23462007-10-08 16:36:45 +0200666 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100667 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200668 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200669 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300670 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200671 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100672 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200673 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200674 * as part of the error recovery process. If @sense_bytes is 0 the command
675 * sent must be one that does not transfer any data. If @sense_bytes != 0
676 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
677 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800678 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200679void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
680 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
f59114b2005-04-17 15:00:23 -0500682 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
Christoph Hellwig631c2282006-07-08 20:42:15 +0200684 /*
685 * We need saved copies of a number of fields - this is because
686 * error handling may need to overwrite these with different values
687 * to run different commands, and once error handling is complete,
688 * we will need to restore these values prior to running the actual
689 * command.
690 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200691 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300692 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200693 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200694 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200695 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200696 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400697 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400698 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200699
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400700 scmd->prot_op = SCSI_PROT_NORMAL;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300701 scmd->cmnd = ses->eh_cmnd;
702 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200703 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200704 scmd->request->next_rq = NULL;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200705
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200706 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200707 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
708 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200709 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200710 scmd->sdb.length);
711 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200712 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200713 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200714 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200715 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200716 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200717 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200718 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200719 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300720 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200721 memcpy(scmd->cmnd, cmnd, cmnd_size);
722 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
723 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200724 }
725
726 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200727
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200728 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500730 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
Christoph Hellwig631c2282006-07-08 20:42:15 +0200732 /*
733 * Zero the sense buffer. The scsi spec mandates that any
734 * untransferred sense data should be interpreted as being zero.
735 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900736 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200737}
738EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200739
Boaz Harroshe1c23462007-10-08 16:36:45 +0200740/**
741 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
742 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200743 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +0200744 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200745 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800746 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200747void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
748{
749 /*
750 * Restore original data
751 */
752 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300753 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200754 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200755 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200756 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200757 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400758 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400759 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200760}
761EXPORT_SYMBOL(scsi_eh_restore_cmnd);
762
763/**
764 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
765 * @scmd: SCSI command structure to hijack
766 * @cmnd: CDB to send
767 * @cmnd_size: size in bytes of @cmnd
768 * @timeout: timeout for this request
769 * @sense_bytes: size of sense data to copy or 0
770 *
771 * This function is used to send a scsi command down to a target device
772 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
773 *
774 * Return value:
775 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800776 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200777static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
778 int cmnd_size, int timeout, unsigned sense_bytes)
779{
780 struct scsi_device *sdev = scmd->device;
781 struct Scsi_Host *shost = sdev->host;
782 DECLARE_COMPLETION_ONSTACK(done);
783 unsigned long timeleft;
784 unsigned long flags;
785 struct scsi_eh_save ses;
786 int rtn;
787
788 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100789 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
f59114b2005-04-17 15:00:23 -0500791 spin_lock_irqsave(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 scsi_log_send(scmd);
f59114b2005-04-17 15:00:23 -0500793 shost->hostt->queuecommand(scmd, scsi_eh_done);
794 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100796 timeleft = wait_for_completion_timeout(&done, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
f59114b2005-04-17 15:00:23 -0500798 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100800 scsi_log_completion(scmd, SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100802 SCSI_LOG_ERROR_RECOVERY(3,
803 printk("%s: scmd: %p, timeleft: %ld\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700804 __func__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
806 /*
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100807 * If there is time left scsi_eh_done got called, and we will
808 * examine the actual status codes to see whether the command
809 * actually did complete normally, else tell the host to forget
810 * about this command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100812 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 rtn = scsi_eh_completed_normally(scmd);
814 SCSI_LOG_ERROR_RECOVERY(3,
815 printk("%s: scsi_eh_completed_normally %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700816 __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100817
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 switch (rtn) {
819 case SUCCESS:
820 case NEEDS_RETRY:
821 case FAILED:
822 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +0200823 case ADD_TO_MLQUEUE:
824 rtn = NEEDS_RETRY;
825 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 default:
827 rtn = FAILED;
828 break;
829 }
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100830 } else {
Brian King292148f2007-01-30 17:51:17 -0600831 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100832 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 }
834
Boaz Harroshe1c23462007-10-08 16:36:45 +0200835 scsi_eh_restore_cmnd(scmd, &ses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 return rtn;
837}
838
839/**
840 * scsi_request_sense - Request sense data from a particular target.
841 * @scmd: SCSI cmd for request sense.
842 *
843 * Notes:
844 * Some hosts automatically obtain this information, others require
845 * that we obtain it on our own. This function will *not* return until
846 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800847 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848static int scsi_request_sense(struct scsi_cmnd *scmd)
849{
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200850 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851}
852
853/**
854 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
855 * @scmd: Original SCSI cmd that eh has finished.
856 * @done_q: Queue for processed commands.
857 *
858 * Notes:
859 * We don't want to use the normal command completion while we are are
860 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -0500861 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 * keep a list of pending commands for final completion, and once we
863 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800864 */
Tejun Heo041c5fc2006-01-23 13:09:36 +0900865void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
867 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200868 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 list_move_tail(&scmd->eh_entry, done_q);
870}
Tejun Heo041c5fc2006-01-23 13:09:36 +0900871EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
873/**
874 * scsi_eh_get_sense - Get device sense data.
875 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -0500876 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 *
878 * Description:
879 * See if we need to request sense information. if so, then get it
880 * now, so we have a better idea of what to do.
881 *
882 * Notes:
883 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -0500884 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 * it down before we request it.
886 *
887 * All drivers should request sense information internally these days,
888 * so for now all I have to say is tough noogies if you end up in here.
889 *
890 * XXX: Long term this code should go away, but that needs an audit of
891 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800892 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800893int scsi_eh_get_sense(struct list_head *work_q,
894 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895{
Christoph Hellwig937abea2005-06-19 13:43:56 +0200896 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 int rtn;
898
Christoph Hellwig937abea2005-06-19 13:43:56 +0200899 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200900 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 SCSI_SENSE_VALID(scmd))
902 continue;
903
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400904 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
905 "%s: requesting sense\n",
906 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 rtn = scsi_request_sense(scmd);
908 if (rtn != SUCCESS)
909 continue;
910
911 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
912 " result %x\n", scmd,
913 scmd->result));
914 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
915
916 rtn = scsi_decide_disposition(scmd);
917
918 /*
919 * if the result was normal, then just pass it along to the
920 * upper level.
921 */
922 if (rtn == SUCCESS)
923 /* we don't want this command reissued, just
924 * finished with the sense data, so set
925 * retries to the max allowed to ensure it
926 * won't get reissued */
927 scmd->retries = scmd->allowed;
928 else if (rtn != NEEDS_RETRY)
929 continue;
930
931 scsi_eh_finish_cmd(scmd, done_q);
932 }
933
934 return list_empty(work_q);
935}
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800936EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937
938/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500940 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 *
942 * Return value:
943 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800944 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945static int scsi_eh_tur(struct scsi_cmnd *scmd)
946{
947 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
948 int retry_cnt = 1, rtn;
949
950retry_tur:
Mike Christie0db99e32006-08-26 03:00:22 -0400951 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700954 __func__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +0200955
956 switch (rtn) {
957 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 if (retry_cnt--)
959 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200960 /*FALLTHRU*/
961 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -0500962 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200963 default:
964 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -0500965 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966}
967
968/**
Rob Landleyeb448202007-11-03 13:30:39 -0500969 * scsi_eh_abort_cmds - abort pending commands.
970 * @work_q: &list_head for pending commands.
971 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 *
973 * Decription:
974 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -0500975 * running command. This only works out to be the case if we have one
976 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 * no sense to try and abort the command, since as far as the shost
978 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800979 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980static int scsi_eh_abort_cmds(struct list_head *work_q,
981 struct list_head *done_q)
982{
Christoph Hellwig937abea2005-06-19 13:43:56 +0200983 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 int rtn;
985
Christoph Hellwig937abea2005-06-19 13:43:56 +0200986 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200987 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 continue;
989 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
990 "0x%p\n", current->comm,
991 scmd));
992 rtn = scsi_try_to_abort_cmd(scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +0100993 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200994 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 if (!scsi_device_online(scmd->device) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +0100996 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 !scsi_eh_tur(scmd)) {
998 scsi_eh_finish_cmd(scmd, done_q);
999 }
1000
1001 } else
1002 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1003 " cmd failed:"
1004 "0x%p\n",
1005 current->comm,
1006 scmd));
1007 }
1008
1009 return list_empty(work_q);
1010}
1011
1012/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001014 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 *
1016 * Return value:
1017 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001018 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1020{
1021 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
Christoph Hellwig631c2282006-07-08 20:42:15 +02001023 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -05001024 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Brian Kinged773e62007-03-29 15:25:52 -05001026 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -06001027 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -05001028
Christoph Hellwig631c2282006-07-08 20:42:15 +02001029 if (rtn == SUCCESS)
1030 return 0;
1031 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 return 1;
1034}
1035
1036 /**
1037 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001038 * @shost: &scsi host being recovered.
1039 * @work_q: &list_head for pending commands.
1040 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 *
1042 * Notes:
1043 * If commands are failing due to not ready, initializing command required,
1044 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001045 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046static int scsi_eh_stu(struct Scsi_Host *shost,
1047 struct list_head *work_q,
1048 struct list_head *done_q)
1049{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001050 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 struct scsi_device *sdev;
1052
1053 shost_for_each_device(sdev, shost) {
1054 stu_scmd = NULL;
1055 list_for_each_entry(scmd, work_q, eh_entry)
1056 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1057 scsi_check_sense(scmd) == FAILED ) {
1058 stu_scmd = scmd;
1059 break;
1060 }
1061
1062 if (!stu_scmd)
1063 continue;
1064
1065 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1066 " 0x%p\n", current->comm, sdev));
1067
1068 if (!scsi_eh_try_stu(stu_scmd)) {
1069 if (!scsi_device_online(sdev) ||
1070 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001071 list_for_each_entry_safe(scmd, next,
1072 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (scmd->device == sdev)
1074 scsi_eh_finish_cmd(scmd, done_q);
1075 }
1076 }
1077 } else {
1078 SCSI_LOG_ERROR_RECOVERY(3,
1079 printk("%s: START_UNIT failed to sdev:"
1080 " 0x%p\n", current->comm, sdev));
1081 }
1082 }
1083
1084 return list_empty(work_q);
1085}
1086
1087
1088/**
1089 * scsi_eh_bus_device_reset - send bdr if needed
1090 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001091 * @work_q: &list_head for pending commands.
1092 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 *
1094 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001095 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 * devices that are jammed or not - if we have multiple devices, it
1097 * makes no sense to try bus_device_reset - we really would need to try
1098 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001099 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1101 struct list_head *work_q,
1102 struct list_head *done_q)
1103{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001104 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 struct scsi_device *sdev;
1106 int rtn;
1107
1108 shost_for_each_device(sdev, shost) {
1109 bdr_scmd = NULL;
1110 list_for_each_entry(scmd, work_q, eh_entry)
1111 if (scmd->device == sdev) {
1112 bdr_scmd = scmd;
1113 break;
1114 }
1115
1116 if (!bdr_scmd)
1117 continue;
1118
1119 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1120 " 0x%p\n", current->comm,
1121 sdev));
1122 rtn = scsi_try_bus_device_reset(bdr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001123 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 if (!scsi_device_online(sdev) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001125 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001127 list_for_each_entry_safe(scmd, next,
1128 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 if (scmd->device == sdev)
1130 scsi_eh_finish_cmd(scmd,
1131 done_q);
1132 }
1133 }
1134 } else {
1135 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1136 " failed sdev:"
1137 "0x%p\n",
1138 current->comm,
1139 sdev));
1140 }
1141 }
1142
1143 return list_empty(work_q);
1144}
1145
1146/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001147 * scsi_eh_target_reset - send target reset if needed
1148 * @shost: scsi host being recovered.
1149 * @work_q: &list_head for pending commands.
1150 * @done_q: &list_head for processed commands.
1151 *
1152 * Notes:
1153 * Try a target reset.
1154 */
1155static int scsi_eh_target_reset(struct Scsi_Host *shost,
1156 struct list_head *work_q,
1157 struct list_head *done_q)
1158{
1159 struct scsi_cmnd *scmd, *tgtr_scmd, *next;
James Bottomleyc82dc882008-09-12 16:46:51 -05001160 unsigned int id = 0;
Mike Christie30bd7df2008-02-29 18:25:19 -06001161 int rtn;
1162
James Bottomleyc82dc882008-09-12 16:46:51 -05001163 do {
Mike Christie30bd7df2008-02-29 18:25:19 -06001164 tgtr_scmd = NULL;
1165 list_for_each_entry(scmd, work_q, eh_entry) {
1166 if (id == scmd_id(scmd)) {
1167 tgtr_scmd = scmd;
1168 break;
1169 }
1170 }
James Bottomleyc82dc882008-09-12 16:46:51 -05001171 if (!tgtr_scmd) {
1172 /* not one exactly equal; find the next highest */
1173 list_for_each_entry(scmd, work_q, eh_entry) {
1174 if (scmd_id(scmd) > id &&
1175 (!tgtr_scmd ||
1176 scmd_id(tgtr_scmd) > scmd_id(scmd)))
1177 tgtr_scmd = scmd;
1178 }
1179 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001180 if (!tgtr_scmd)
James Bottomleyc82dc882008-09-12 16:46:51 -05001181 /* no more commands, that's it */
1182 break;
Mike Christie30bd7df2008-02-29 18:25:19 -06001183
1184 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1185 "to target %d\n",
1186 current->comm, id));
1187 rtn = scsi_try_target_reset(tgtr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001188 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Mike Christie30bd7df2008-02-29 18:25:19 -06001189 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1190 if (id == scmd_id(scmd))
1191 if (!scsi_device_online(scmd->device) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001192 rtn == FAST_IO_FAIL ||
Mike Christie30bd7df2008-02-29 18:25:19 -06001193 !scsi_eh_tur(tgtr_scmd))
1194 scsi_eh_finish_cmd(scmd,
1195 done_q);
1196 }
1197 } else
1198 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1199 " failed target: "
1200 "%d\n",
1201 current->comm, id));
James Bottomleyc82dc882008-09-12 16:46:51 -05001202 id++;
1203 } while(id != 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001204
1205 return list_empty(work_q);
1206}
1207
1208/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001210 * @shost: &scsi host being recovered.
1211 * @work_q: &list_head for pending commands.
1212 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001213 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1215 struct list_head *work_q,
1216 struct list_head *done_q)
1217{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001218 struct scsi_cmnd *scmd, *chan_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 unsigned int channel;
1220 int rtn;
1221
1222 /*
1223 * we really want to loop over the various channels, and do this on
1224 * a channel by channel basis. we should also check to see if any
1225 * of the failed commands are on soft_reset devices, and if so, skip
1226 * the reset.
1227 */
1228
1229 for (channel = 0; channel <= shost->max_channel; channel++) {
1230 chan_scmd = NULL;
1231 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001232 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 chan_scmd = scmd;
1234 break;
1235 /*
1236 * FIXME add back in some support for
1237 * soft_reset devices.
1238 */
1239 }
1240 }
1241
1242 if (!chan_scmd)
1243 continue;
1244 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1245 " %d\n", current->comm,
1246 channel));
1247 rtn = scsi_try_bus_reset(chan_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001248 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001249 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001250 if (channel == scmd_channel(scmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 if (!scsi_device_online(scmd->device) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001252 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 !scsi_eh_tur(scmd))
1254 scsi_eh_finish_cmd(scmd,
1255 done_q);
1256 }
1257 } else {
1258 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1259 " failed chan: %d\n",
1260 current->comm,
1261 channel));
1262 }
1263 }
1264 return list_empty(work_q);
1265}
1266
1267/**
1268 * scsi_eh_host_reset - send a host reset
1269 * @work_q: list_head for processed commands.
1270 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001271 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272static int scsi_eh_host_reset(struct list_head *work_q,
1273 struct list_head *done_q)
1274{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001275 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
1278 if (!list_empty(work_q)) {
1279 scmd = list_entry(work_q->next,
1280 struct scsi_cmnd, eh_entry);
1281
1282 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1283 , current->comm));
1284
1285 rtn = scsi_try_host_reset(scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001286 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001287 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 if (!scsi_device_online(scmd->device) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001289 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1291 !scsi_eh_tur(scmd))
1292 scsi_eh_finish_cmd(scmd, done_q);
1293 }
1294 } else {
1295 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1296 " failed\n",
1297 current->comm));
1298 }
1299 }
1300 return list_empty(work_q);
1301}
1302
1303/**
1304 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1305 * @work_q: list_head for processed commands.
1306 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001307 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308static void scsi_eh_offline_sdevs(struct list_head *work_q,
1309 struct list_head *done_q)
1310{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001311 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
Christoph Hellwig937abea2005-06-19 13:43:56 +02001313 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001314 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1315 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001317 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 /*
1319 * FIXME: Handle lost cmds.
1320 */
1321 }
1322 scsi_eh_finish_cmd(scmd, done_q);
1323 }
1324 return;
1325}
1326
1327/**
Mike Christie4a274462008-08-19 18:45:31 -05001328 * scsi_noretry_cmd - determinte if command should be failed fast
1329 * @scmd: SCSI cmd to examine.
1330 */
1331int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1332{
1333 switch (host_byte(scmd->result)) {
1334 case DID_OK:
1335 break;
1336 case DID_BUS_BUSY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001337 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
Mike Christie4a274462008-08-19 18:45:31 -05001338 case DID_PARITY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001339 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
Mike Christie4a274462008-08-19 18:45:31 -05001340 case DID_ERROR:
1341 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1342 status_byte(scmd->result) == RESERVATION_CONFLICT)
1343 return 0;
1344 /* fall through */
1345 case DID_SOFT_ERROR:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001346 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
Mike Christie4a274462008-08-19 18:45:31 -05001347 }
1348
1349 switch (status_byte(scmd->result)) {
1350 case CHECK_CONDITION:
1351 /*
1352 * assume caller has checked sense and determinted
1353 * the check condition was retryable.
1354 */
FUJITA Tomonorie96f6ab2010-07-09 09:38:26 +09001355 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1356 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1357 return 1;
Mike Christie4a274462008-08-19 18:45:31 -05001358 }
1359
1360 return 0;
1361}
1362
1363/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1365 * @scmd: SCSI cmd to examine.
1366 *
1367 * Notes:
1368 * This is *only* called when we are examining the status after sending
1369 * out the actual data command. any commands that are queued for error
1370 * recovery (e.g. test_unit_ready) do *not* come through here.
1371 *
1372 * When this routine returns failed, it means the error handler thread
1373 * is woken. In cases where the error code indicates an error that
1374 * doesn't require the error handler read (i.e. we don't need to
1375 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001376 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377int scsi_decide_disposition(struct scsi_cmnd *scmd)
1378{
1379 int rtn;
1380
1381 /*
1382 * if the device is offline, then we clearly just pass the result back
1383 * up to the top level.
1384 */
1385 if (!scsi_device_online(scmd->device)) {
1386 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1387 " as SUCCESS\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001388 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 return SUCCESS;
1390 }
1391
1392 /*
1393 * first check the host byte, to see if there is anything in there
1394 * that would indicate what we need to do.
1395 */
1396 switch (host_byte(scmd->result)) {
1397 case DID_PASSTHROUGH:
1398 /*
1399 * no matter what, pass this through to the upper layer.
1400 * nuke this special code so that it looks like we are saying
1401 * did_ok.
1402 */
1403 scmd->result &= 0xff00ffff;
1404 return SUCCESS;
1405 case DID_OK:
1406 /*
1407 * looks good. drop through, and check the next byte.
1408 */
1409 break;
1410 case DID_NO_CONNECT:
1411 case DID_BAD_TARGET:
1412 case DID_ABORT:
1413 /*
1414 * note - this means that we just report the status back
1415 * to the top level driver, not that we actually think
1416 * that it indicates SUCCESS.
1417 */
1418 return SUCCESS;
1419 /*
1420 * when the low level driver returns did_soft_error,
1421 * it is responsible for keeping an internal retry counter
1422 * in order to avoid endless loops (db)
1423 *
1424 * actually this is a bug in this function here. we should
1425 * be mindful of the maximum number of retries specified
1426 * and not get stuck in a loop.
1427 */
1428 case DID_SOFT_ERROR:
1429 goto maybe_retry;
1430 case DID_IMM_RETRY:
1431 return NEEDS_RETRY;
1432
bf341912005-04-12 17:49:09 -05001433 case DID_REQUEUE:
1434 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001435 case DID_TRANSPORT_DISRUPTED:
1436 /*
1437 * LLD/transport was disrupted during processing of the IO.
1438 * The transport class is now blocked/blocking,
1439 * and the transport will decide what to do with the IO
Mike Christie939c2282008-11-04 19:47:19 -06001440 * based on its timers and recovery capablilities if
1441 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001442 */
Mike Christie939c2282008-11-04 19:47:19 -06001443 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001444 case DID_TRANSPORT_FAILFAST:
1445 /*
1446 * The transport decided to failfast the IO (most likely
1447 * the fast io fail tmo fired), so send IO directly upwards.
1448 */
1449 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 case DID_ERROR:
1451 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1452 status_byte(scmd->result) == RESERVATION_CONFLICT)
1453 /*
1454 * execute reservation conflict processing code
1455 * lower down
1456 */
1457 break;
1458 /* fallthrough */
1459
1460 case DID_BUS_BUSY:
1461 case DID_PARITY:
1462 goto maybe_retry;
1463 case DID_TIME_OUT:
1464 /*
1465 * when we scan the bus, we get timeout messages for
1466 * these commands if there is no device available.
1467 * other hosts report did_no_connect for the same thing.
1468 */
1469 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1470 scmd->cmnd[0] == INQUIRY)) {
1471 return SUCCESS;
1472 } else {
1473 return FAILED;
1474 }
1475 case DID_RESET:
1476 return SUCCESS;
1477 default:
1478 return FAILED;
1479 }
1480
1481 /*
1482 * next, check the message byte.
1483 */
1484 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1485 return FAILED;
1486
1487 /*
1488 * check the status byte to see if this indicates anything special.
1489 */
1490 switch (status_byte(scmd->result)) {
1491 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001492 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 /*
1494 * the case of trying to send too many commands to a
1495 * tagged queueing device.
1496 */
1497 case BUSY:
1498 /*
1499 * device can't talk to us at the moment. Should only
1500 * occur (SAM-3) when the task queue is empty, so will cause
1501 * the empty queue handling to trigger a stall in the
1502 * device.
1503 */
1504 return ADD_TO_MLQUEUE;
1505 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -07001506 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001509 case TASK_ABORTED:
1510 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 case CHECK_CONDITION:
1512 rtn = scsi_check_sense(scmd);
1513 if (rtn == NEEDS_RETRY)
1514 goto maybe_retry;
1515 /* if rtn == FAILED, we have no sense information;
1516 * returning FAILED will wake the error handler thread
1517 * to collect the sense and redo the decide
1518 * disposition */
1519 return rtn;
1520 case CONDITION_GOOD:
1521 case INTERMEDIATE_GOOD:
1522 case INTERMEDIATE_C_GOOD:
1523 case ACA_ACTIVE:
1524 /*
1525 * who knows? FIXME(eric)
1526 */
1527 return SUCCESS;
1528
1529 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001530 sdev_printk(KERN_INFO, scmd->device,
1531 "reservation conflict\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 return SUCCESS; /* causes immediate i/o error */
1533 default:
1534 return FAILED;
1535 }
1536 return FAILED;
1537
1538 maybe_retry:
1539
1540 /* we requeue for retry because the error was retryable, and
1541 * the request was not marked fast fail. Note that above,
1542 * even if the request is marked fast fail, we still requeue
1543 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001544 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001545 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 return NEEDS_RETRY;
1547 } else {
1548 /*
1549 * no more retries - report this one back to upper level.
1550 */
1551 return SUCCESS;
1552 }
1553}
1554
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001555static void eh_lock_door_done(struct request *req, int uptodate)
1556{
1557 __blk_put_request(req->q, req);
1558}
1559
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 * scsi_eh_lock_door - Prevent medium removal for the specified device
1562 * @sdev: SCSI device to prevent medium removal
1563 *
1564 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001565 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 *
1567 * Notes:
1568 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1569 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001570 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571static void scsi_eh_lock_door(struct scsi_device *sdev)
1572{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001573 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
James Bottomley91bc31f2009-05-17 09:30:48 -05001575 /*
1576 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1577 * request becomes available
1578 */
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001579 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001581 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1582 req->cmd[1] = 0;
1583 req->cmd[2] = 0;
1584 req->cmd[3] = 0;
1585 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1586 req->cmd[5] = 0;
1587
1588 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1589
1590 req->cmd_type = REQ_TYPE_BLOCK_PC;
1591 req->cmd_flags |= REQ_QUIET;
1592 req->timeout = 10 * HZ;
1593 req->retries = 5;
1594
1595 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596}
1597
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598/**
1599 * scsi_restart_operations - restart io operations to the specified host.
1600 * @shost: Host we are restarting.
1601 *
1602 * Notes:
1603 * When we entered the error handler, we blocked all further i/o to
1604 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001605 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606static void scsi_restart_operations(struct Scsi_Host *shost)
1607{
1608 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001609 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
1611 /*
1612 * If the door was locked, we need to insert a door lock request
1613 * onto the head of the SCSI request queue for the device. There
1614 * is no point trying to lock the door of an off-line device.
1615 */
1616 shost_for_each_device(sdev, shost) {
1617 if (scsi_device_online(sdev) && sdev->locked)
1618 scsi_eh_lock_door(sdev);
1619 }
1620
1621 /*
1622 * next free up anything directly waiting upon the host. this
1623 * will be requests for character device operations, and also for
1624 * ioctls to queued block devices.
1625 */
1626 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001627 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
James Bottomley939647e2005-09-18 15:05:20 -05001629 spin_lock_irqsave(shost->host_lock, flags);
1630 if (scsi_host_set_state(shost, SHOST_RUNNING))
1631 if (scsi_host_set_state(shost, SHOST_CANCEL))
1632 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1633 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634
1635 wake_up(&shost->host_wait);
1636
1637 /*
1638 * finally we need to re-initiate requests that may be pending. we will
1639 * have had everything blocked while error handling is taking place, and
1640 * now that error recovery is done, we will need to ensure that these
1641 * requests are started.
1642 */
1643 scsi_run_host_queues(shost);
1644}
1645
1646/**
1647 * scsi_eh_ready_devs - check device ready state and recover if not.
1648 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001649 * @work_q: &list_head for pending commands.
1650 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001651 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001652void scsi_eh_ready_devs(struct Scsi_Host *shost,
1653 struct list_head *work_q,
1654 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655{
1656 if (!scsi_eh_stu(shost, work_q, done_q))
1657 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06001658 if (!scsi_eh_target_reset(shost, work_q, done_q))
1659 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1660 if (!scsi_eh_host_reset(work_q, done_q))
1661 scsi_eh_offline_sdevs(work_q,
1662 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001664EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
1666/**
1667 * scsi_eh_flush_done_q - finish processed commands or retry them.
1668 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001669 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001670void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001672 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
Christoph Hellwig937abea2005-06-19 13:43:56 +02001674 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1675 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05001677 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06001678 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1680 " retry cmd: %p\n",
1681 current->comm,
1682 scmd));
1683 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1684 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07001685 /*
1686 * If just we got sense for the device (called
1687 * scsi_eh_get_sense), scmd->result is already
1688 * set, do not set DRIVER_TIMEOUT.
1689 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 if (!scmd->result)
1691 scmd->result |= (DRIVER_TIMEOUT << 24);
1692 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1693 " cmd: %p\n",
1694 current->comm, scmd));
1695 scsi_finish_command(scmd);
1696 }
1697 }
1698}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001699EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
1701/**
1702 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1703 * @shost: Host to unjam.
1704 *
1705 * Notes:
1706 * When we come in here, we *know* that all commands on the bus have
1707 * either completed, failed or timed out. we also know that no further
1708 * commands are being sent to the host, so things are relatively quiet
1709 * and we have freedom to fiddle with things as we wish.
1710 *
1711 * This is only the *default* implementation. it is possible for
1712 * individual drivers to supply their own version of this function, and
1713 * if the maintainer wishes to do this, it is strongly suggested that
1714 * this function be taken as a template and modified. this function
1715 * was designed to correctly handle problems for about 95% of the
1716 * different cases out there, and it should always provide at least a
1717 * reasonable amount of error recovery.
1718 *
1719 * Any command marked 'failed' or 'timeout' must eventually have
1720 * scsi_finish_cmd() called for it. we do all of the retry stuff
1721 * here, so when we restart the host after we return it should have an
1722 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001723 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724static void scsi_unjam_host(struct Scsi_Host *shost)
1725{
1726 unsigned long flags;
1727 LIST_HEAD(eh_work_q);
1728 LIST_HEAD(eh_done_q);
1729
1730 spin_lock_irqsave(shost->host_lock, flags);
1731 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1732 spin_unlock_irqrestore(shost->host_lock, flags);
1733
1734 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1735
1736 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1737 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1738 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1739
1740 scsi_eh_flush_done_q(&eh_done_q);
1741}
1742
1743/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001744 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 * @data: Host for which we are running.
1746 *
1747 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001748 * This is the main error handling loop. This is run as a kernel thread
1749 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001750 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751int scsi_error_handler(void *data)
1752{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001753 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001756 * We use TASK_INTERRUPTIBLE so that the thread is not
1757 * counted against the load average as a running process.
1758 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01001759 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 */
James Bottomley3ed7a472005-09-19 09:50:04 -05001761 set_current_state(TASK_INTERRUPTIBLE);
1762 while (!kthread_should_stop()) {
Tejun Heoee7863b2006-05-15 20:57:20 +09001763 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05001764 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001765 SCSI_LOG_ERROR_RECOVERY(1,
1766 printk("Error handler scsi_eh_%d sleeping\n",
1767 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001768 schedule();
1769 set_current_state(TASK_INTERRUPTIBLE);
1770 continue;
1771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
James Bottomley3ed7a472005-09-19 09:50:04 -05001773 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001774 SCSI_LOG_ERROR_RECOVERY(1,
1775 printk("Error handler scsi_eh_%d waking up\n",
1776 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 /*
1779 * We have a host that is failing for some reason. Figure out
1780 * what we need to do to get it up and online again (if we can).
1781 * If we fail, we end up taking the thing offline.
1782 */
Alan Sternbc4f2402010-06-17 10:41:42 -04001783 if (scsi_autopm_get_host(shost) != 0) {
1784 SCSI_LOG_ERROR_RECOVERY(1,
1785 printk(KERN_ERR "Error handler scsi_eh_%d "
1786 "unable to autoresume\n",
1787 shost->host_no));
1788 continue;
1789 }
1790
Christoph Hellwig9227c332006-04-01 19:21:04 +02001791 if (shost->transportt->eh_strategy_handler)
1792 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 else
1794 scsi_unjam_host(shost);
1795
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 /*
1797 * Note - if the above fails completely, the action is to take
1798 * individual devices offline and flush the queue of any
1799 * outstanding requests that may have been pending. When we
1800 * restart, we restart any I/O to any other devices on the bus
1801 * which are still online.
1802 */
1803 scsi_restart_operations(shost);
Alan Sternbc4f2402010-06-17 10:41:42 -04001804 scsi_autopm_put_host(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -05001805 set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04001807 __set_current_state(TASK_RUNNING);
1808
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001809 SCSI_LOG_ERROR_RECOVERY(1,
1810 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001811 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 return 0;
1813}
1814
1815/*
1816 * Function: scsi_report_bus_reset()
1817 *
1818 * Purpose: Utility function used by low-level drivers to report that
1819 * they have observed a bus reset on the bus being handled.
1820 *
1821 * Arguments: shost - Host in question
1822 * channel - channel on which reset was observed.
1823 *
1824 * Returns: Nothing
1825 *
1826 * Lock status: Host lock must be held.
1827 *
1828 * Notes: This only needs to be called if the reset is one which
1829 * originates from an unknown location. Resets originated
1830 * by the mid-level itself don't need to call this, but there
1831 * should be no harm.
1832 *
1833 * The main purpose of this is to make sure that a CHECK_CONDITION
1834 * is properly treated.
1835 */
1836void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1837{
1838 struct scsi_device *sdev;
1839
1840 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06001841 if (channel == sdev_channel(sdev))
1842 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 }
1844}
1845EXPORT_SYMBOL(scsi_report_bus_reset);
1846
1847/*
1848 * Function: scsi_report_device_reset()
1849 *
1850 * Purpose: Utility function used by low-level drivers to report that
1851 * they have observed a device reset on the device being handled.
1852 *
1853 * Arguments: shost - Host in question
1854 * channel - channel on which reset was observed
1855 * target - target on which reset was observed
1856 *
1857 * Returns: Nothing
1858 *
1859 * Lock status: Host lock must be held
1860 *
1861 * Notes: This only needs to be called if the reset is one which
1862 * originates from an unknown location. Resets originated
1863 * by the mid-level itself don't need to call this, but there
1864 * should be no harm.
1865 *
1866 * The main purpose of this is to make sure that a CHECK_CONDITION
1867 * is properly treated.
1868 */
1869void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1870{
1871 struct scsi_device *sdev;
1872
1873 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001874 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06001875 target == sdev_id(sdev))
1876 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 }
1878}
1879EXPORT_SYMBOL(scsi_report_device_reset);
1880
1881static void
1882scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1883{
1884}
1885
1886/*
1887 * Function: scsi_reset_provider
1888 *
1889 * Purpose: Send requested reset to a bus or device at any phase.
1890 *
1891 * Arguments: device - device to send reset to
1892 * flag - reset type (see scsi.h)
1893 *
1894 * Returns: SUCCESS/FAILURE.
1895 *
1896 * Notes: This is used by the SCSI Generic driver to provide
1897 * Bus/Device reset capability.
1898 */
1899int
1900scsi_reset_provider(struct scsi_device *dev, int flag)
1901{
Alan Sternbc4f2402010-06-17 10:41:42 -04001902 struct scsi_cmnd *scmd;
James Smartd7a1bb02006-03-08 14:50:12 -05001903 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05001905 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 int rtn;
1907
Alan Sternbc4f2402010-06-17 10:41:42 -04001908 if (scsi_autopm_get_host(shost) < 0)
1909 return FAILED;
1910
1911 scmd = scsi_get_command(dev, GFP_KERNEL);
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02001912 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02001914
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001915 scmd->cmnd = req.cmd;
1916
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001918 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
1920 scmd->cmd_len = 0;
1921
1922 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
James Smartd7a1bb02006-03-08 14:50:12 -05001924 spin_lock_irqsave(shost->host_lock, flags);
1925 shost->tmf_in_progress = 1;
1926 spin_unlock_irqrestore(shost->host_lock, flags);
1927
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 switch (flag) {
1929 case SCSI_TRY_RESET_DEVICE:
1930 rtn = scsi_try_bus_device_reset(scmd);
1931 if (rtn == SUCCESS)
1932 break;
1933 /* FALLTHROUGH */
Mike Christie30bd7df2008-02-29 18:25:19 -06001934 case SCSI_TRY_RESET_TARGET:
1935 rtn = scsi_try_target_reset(scmd);
1936 if (rtn == SUCCESS)
1937 break;
1938 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 case SCSI_TRY_RESET_BUS:
1940 rtn = scsi_try_bus_reset(scmd);
1941 if (rtn == SUCCESS)
1942 break;
1943 /* FALLTHROUGH */
1944 case SCSI_TRY_RESET_HOST:
1945 rtn = scsi_try_host_reset(scmd);
1946 break;
1947 default:
1948 rtn = FAILED;
1949 }
1950
James Smartd7a1bb02006-03-08 14:50:12 -05001951 spin_lock_irqsave(shost->host_lock, flags);
1952 shost->tmf_in_progress = 0;
1953 spin_unlock_irqrestore(shost->host_lock, flags);
1954
1955 /*
1956 * be sure to wake up anyone who was sleeping or had their queue
1957 * suspended while we performed the TMF.
1958 */
1959 SCSI_LOG_ERROR_RECOVERY(3,
1960 printk("%s: waking up host to restart after TMF\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001961 __func__));
James Smartd7a1bb02006-03-08 14:50:12 -05001962
1963 wake_up(&shost->host_wait);
1964
1965 scsi_run_host_queues(shost);
1966
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 scsi_next_command(scmd);
Alan Sternbc4f2402010-06-17 10:41:42 -04001968 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 return rtn;
1970}
1971EXPORT_SYMBOL(scsi_reset_provider);
1972
1973/**
1974 * scsi_normalize_sense - normalize main elements from either fixed or
1975 * descriptor sense data format into a common format.
1976 *
1977 * @sense_buffer: byte array containing sense data returned by device
1978 * @sb_len: number of valid bytes in sense_buffer
1979 * @sshdr: pointer to instance of structure that common
1980 * elements are written to.
1981 *
1982 * Notes:
1983 * The "main elements" from sense data are: response_code, sense_key,
1984 * asc, ascq and additional_length (only for descriptor format).
1985 *
1986 * Typically this function can be called after a device has
1987 * responded to a SCSI command with the CHECK_CONDITION status.
1988 *
1989 * Return value:
1990 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001991 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1993 struct scsi_sense_hdr *sshdr)
1994{
James Bottomley33aa6872005-08-28 11:31:14 -05001995 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 return 0;
1997
1998 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1999
2000 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05002001
2002 if (!scsi_sense_valid(sshdr))
2003 return 0;
2004
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 if (sshdr->response_code >= 0x72) {
2006 /*
2007 * descriptor format
2008 */
2009 if (sb_len > 1)
2010 sshdr->sense_key = (sense_buffer[1] & 0xf);
2011 if (sb_len > 2)
2012 sshdr->asc = sense_buffer[2];
2013 if (sb_len > 3)
2014 sshdr->ascq = sense_buffer[3];
2015 if (sb_len > 7)
2016 sshdr->additional_length = sense_buffer[7];
2017 } else {
2018 /*
2019 * fixed format
2020 */
2021 if (sb_len > 2)
2022 sshdr->sense_key = (sense_buffer[2] & 0xf);
2023 if (sb_len > 7) {
2024 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2025 sb_len : (sense_buffer[7] + 8);
2026 if (sb_len > 12)
2027 sshdr->asc = sense_buffer[12];
2028 if (sb_len > 13)
2029 sshdr->ascq = sense_buffer[13];
2030 }
2031 }
2032
2033 return 1;
2034}
2035EXPORT_SYMBOL(scsi_normalize_sense);
2036
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2038 struct scsi_sense_hdr *sshdr)
2039{
2040 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09002041 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042}
2043EXPORT_SYMBOL(scsi_command_normalize_sense);
2044
2045/**
Rob Landleyeb448202007-11-03 13:30:39 -05002046 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 * @sense_buffer: byte array of descriptor format sense data
2048 * @sb_len: number of valid bytes in sense_buffer
2049 * @desc_type: value of descriptor type to find
2050 * (e.g. 0 -> information)
2051 *
2052 * Notes:
2053 * only valid when sense data is in descriptor format
2054 *
2055 * Return value:
2056 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002057 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2059 int desc_type)
2060{
2061 int add_sen_len, add_len, desc_len, k;
2062 const u8 * descp;
2063
2064 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2065 return NULL;
2066 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2067 return NULL;
2068 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2069 add_sen_len : (sb_len - 8);
2070 descp = &sense_buffer[8];
2071 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2072 descp += desc_len;
2073 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2074 desc_len = add_len + 2;
2075 if (descp[0] == desc_type)
2076 return descp;
2077 if (add_len < 0) // short descriptor ??
2078 break;
2079 }
2080 return NULL;
2081}
2082EXPORT_SYMBOL(scsi_sense_desc_find);
2083
2084/**
Rob Landleyeb448202007-11-03 13:30:39 -05002085 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 * @sense_buffer: byte array of sense data
2087 * @sb_len: number of valid bytes in sense_buffer
2088 * @info_out: pointer to 64 integer where 8 or 4 byte information
2089 * field will be placed if found.
2090 *
2091 * Return value:
2092 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002093 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2095 u64 * info_out)
2096{
2097 int j;
2098 const u8 * ucp;
2099 u64 ull;
2100
2101 if (sb_len < 7)
2102 return 0;
2103 switch (sense_buffer[0] & 0x7f) {
2104 case 0x70:
2105 case 0x71:
2106 if (sense_buffer[0] & 0x80) {
2107 *info_out = (sense_buffer[3] << 24) +
2108 (sense_buffer[4] << 16) +
2109 (sense_buffer[5] << 8) + sense_buffer[6];
2110 return 1;
2111 } else
2112 return 0;
2113 case 0x72:
2114 case 0x73:
2115 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2116 0 /* info desc */);
2117 if (ucp && (0xa == ucp[1])) {
2118 ull = 0;
2119 for (j = 0; j < 8; ++j) {
2120 if (j > 0)
2121 ull <<= 8;
2122 ull |= ucp[4 + j];
2123 }
2124 *info_out = ull;
2125 return 1;
2126 } else
2127 return 0;
2128 default:
2129 return 0;
2130 }
2131}
2132EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002133
2134/**
2135 * scsi_build_sense_buffer - build sense data in a buffer
2136 * @desc: Sense format (non zero == descriptor format,
2137 * 0 == fixed format)
2138 * @buf: Where to build sense data
2139 * @key: Sense key
2140 * @asc: Additional sense code
2141 * @ascq: Additional sense code qualifier
2142 *
2143 **/
2144void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2145{
2146 if (desc) {
2147 buf[0] = 0x72; /* descriptor, current */
2148 buf[1] = key;
2149 buf[2] = asc;
2150 buf[3] = ascq;
2151 buf[7] = 0;
2152 } else {
2153 buf[0] = 0x70; /* fixed, current */
2154 buf[2] = key;
2155 buf[7] = 0xa;
2156 buf[12] = asc;
2157 buf[13] = ascq;
2158 }
2159}
2160EXPORT_SYMBOL(scsi_build_sense_buffer);