blob: b6c86cce57bfa2ecf9ce2ee52b5d39fa2f67d790 [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 *
5 * SCSI error/timeout handling
6 * Initial versions: Eric Youngdale. Based upon conversations with
Jesper Juhl0bf8c862011-03-21 20:47:31 +01007 * Leonard Zubkoff and David Miller at Linux Expo,
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * ideas originating from all over the place.
9 *
10 * Restructured scsi_unjam_host and associated functions.
11 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 *
13 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
Jesper Juhl0bf8c862011-03-21 20:47:31 +010014 * minor cleanups.
Linus Torvalds1da177e2005-04-16 15:20:36 -070015 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
16 */
17
18#include <linux/module.h>
19#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/timer.h>
22#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/kernel.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070024#include <linux/freezer.h>
Christoph Hellwigc5478de2005-09-06 14:04:26 +020025#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/interrupt.h>
27#include <linux/blkdev.h>
28#include <linux/delay.h>
Hannes Reineckefc736482013-04-25 08:10:00 +020029#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31#include <scsi/scsi.h>
Christoph Hellwigbeb40482006-06-10 18:01:03 +020032#include <scsi/scsi_cmnd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <scsi/scsi_dbg.h>
34#include <scsi/scsi_device.h>
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -050035#include <scsi/scsi_driver.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <scsi/scsi_eh.h>
Sagi Grimberg7708c162015-07-08 17:58:52 +030037#include <scsi/scsi_common.h>
James Smartc829c392006-03-13 08:28:57 -050038#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <scsi/scsi_host.h>
40#include <scsi/scsi_ioctl.h>
Christoph Hellwigee14c672015-08-27 14:16:59 +020041#include <scsi/scsi_dh.h>
Martin Wilck29cfc2a2018-04-18 01:35:10 +020042#include <scsi/scsi_devinfo.h>
Christoph Hellwig176aa9d2014-10-11 12:06:47 +020043#include <scsi/sg.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#include "scsi_priv.h"
46#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070047#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Kei Tokunagabf816232010-04-01 20:41:40 +090049#include <trace/events/scsi.h>
50
Damien Le Moal29087692017-04-24 16:51:12 +090051#include <asm/unaligned.h>
52
James Bottomley14216562012-07-25 23:55:55 +040053static void scsi_eh_done(struct scsi_cmnd *scmd);
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055/*
56 * These should *probably* be handled by the host itself.
57 * Since it is allowed to sleep, it probably should.
58 */
59#define BUS_RESET_SETTLE_TIME (10)
60#define HOST_RESET_SETTLE_TIME (10)
61
David Jeffery3eef6252011-05-19 14:41:12 -040062static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
Bart Van Asscheb8e162f2021-04-15 15:08:11 -070063static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *,
64 struct scsi_cmnd *);
David Jeffery3eef6252011-05-19 14:41:12 -040065
Linus Torvalds1da177e2005-04-16 15:20:36 -070066void scsi_eh_wakeup(struct Scsi_Host *shost)
67{
Bart Van Asschef0317e82017-12-04 10:06:24 -080068 lockdep_assert_held(shost->host_lock);
69
Ming Leic84b0232018-06-24 22:03:26 +080070 if (scsi_host_busy(shost) == shost->host_failed) {
Kei Tokunagabf816232010-04-01 20:41:40 +090071 trace_scsi_eh_wakeup(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -050072 wake_up_process(shost->ehandler);
Hannes Reinecke91921e02014-06-25 16:39:59 +020073 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
74 "Waking error handler thread\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 }
76}
Tejun Heof8bbfc22006-05-19 21:07:05 +090077
78/**
79 * scsi_schedule_eh - schedule EH for SCSI host
80 * @shost: SCSI host to invoke error handling on.
81 *
82 * Schedule SCSI EH without scmd.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080083 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090084void scsi_schedule_eh(struct Scsi_Host *shost)
85{
86 unsigned long flags;
87
88 spin_lock_irqsave(shost->host_lock, flags);
89
90 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
91 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
92 shost->host_eh_scheduled++;
93 scsi_eh_wakeup(shost);
94 }
95
96 spin_unlock_irqrestore(shost->host_lock, flags);
97}
98EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Hannes Reineckeb4562022013-10-23 10:51:21 +0200100static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
101{
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100102 if (!shost->last_reset || shost->eh_deadline == -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200103 return 0;
104
Hannes Reinecke76ad3e52013-11-11 13:44:55 +0100105 /*
106 * 32bit accesses are guaranteed to be atomic
107 * (on all supported architectures), so instead
108 * of using a spinlock we can as well double check
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100109 * if eh_deadline has been set to 'off' during the
Hannes Reinecke76ad3e52013-11-11 13:44:55 +0100110 * time_before call.
111 */
112 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100113 shost->eh_deadline > -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200114 return 0;
115
116 return 1;
117}
118
Mike Christie2a242d52020-10-01 10:35:53 -0500119static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd)
120{
121 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
122 return true;
123
124 return ++cmd->retries <= cmd->allowed;
125}
126
Muneendra Kumar60bee272021-01-07 03:19:05 +0530127static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd)
128{
129 struct scsi_device *sdev = cmd->device;
130 struct Scsi_Host *host = sdev->host;
131
132 if (host->hostt->eh_should_retry_cmd)
133 return host->hostt->eh_should_retry_cmd(cmd);
134
135 return true;
136}
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138/**
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100139 * scmd_eh_abort_handler - Handle command aborts
140 * @work: command to be aborted.
Bart Van Assche923f46f2018-02-12 10:38:05 -0800141 *
142 * Note: this function must be called only for a command that has timed out.
143 * Because the block layer marks a request as complete before it calls
144 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
145 * timed out do not have any effect. Hence it is safe to call
146 * scsi_finish_command() from this function.
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100147 */
148void
149scmd_eh_abort_handler(struct work_struct *work)
150{
151 struct scsi_cmnd *scmd =
152 container_of(work, struct scsi_cmnd, abort_work.work);
153 struct scsi_device *sdev = scmd->device;
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700154 enum scsi_disposition rtn;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100155
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100156 if (scsi_host_eh_past_deadline(sdev->host)) {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100157 SCSI_LOG_ERROR_RECOVERY(3,
158 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100159 "eh timeout, not aborting\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100160 } else {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100161 SCSI_LOG_ERROR_RECOVERY(3,
162 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100163 "aborting command\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100164 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
165 if (rtn == SUCCESS) {
Ulrich Obergfell8922a902014-06-04 13:34:57 +0200166 set_host_byte(scmd, DID_TIME_OUT);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100167 if (scsi_host_eh_past_deadline(sdev->host)) {
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100170 "eh timeout, not retrying "
171 "aborted command\n"));
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100172 } else if (!scsi_noretry_cmd(scmd) &&
Muneendra Kumar60bee272021-01-07 03:19:05 +0530173 scsi_cmd_retry_allowed(scmd) &&
174 scsi_eh_should_retry_cmd(scmd)) {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100175 SCSI_LOG_ERROR_RECOVERY(3,
176 scmd_printk(KERN_WARNING, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100177 "retry aborted command\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100178 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100179 return;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100180 } else {
181 SCSI_LOG_ERROR_RECOVERY(3,
182 scmd_printk(KERN_WARNING, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100183 "finish aborted command\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100184 scsi_finish_command(scmd);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100185 return;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100186 }
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100187 } else {
188 SCSI_LOG_ERROR_RECOVERY(3,
189 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100190 "cmd abort %s\n",
Hannes Reinecke883a0302014-10-24 14:27:04 +0200191 (rtn == FAST_IO_FAIL) ?
192 "not send" : "failed"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100193 }
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100194 }
195
Hannes Reineckea0658632017-04-06 15:36:35 +0200196 scsi_eh_scmd_add(scmd);
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100197}
198
199/**
200 * scsi_abort_command - schedule a command abort
201 * @scmd: scmd to abort.
202 *
203 * We only need to abort commands after a command timeout
204 */
205static int
206scsi_abort_command(struct scsi_cmnd *scmd)
207{
208 struct scsi_device *sdev = scmd->device;
209 struct Scsi_Host *shost = sdev->host;
210 unsigned long flags;
211
212 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
213 /*
214 * Retry after abort failed, escalate to next level.
215 */
216 SCSI_LOG_ERROR_RECOVERY(3,
217 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100218 "previous abort failed\n"));
Bart Van Asschefcc95a72014-06-02 11:50:52 +0200219 BUG_ON(delayed_work_pending(&scmd->abort_work));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100220 return FAILED;
221 }
222
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100223 spin_lock_irqsave(shost->host_lock, flags);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100224 if (shost->eh_deadline != -1 && !shost->last_reset)
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100225 shost->last_reset = jiffies;
226 spin_unlock_irqrestore(shost->host_lock, flags);
227
228 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
229 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100230 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100231 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
232 return SUCCESS;
233}
234
235/**
Hannes Reinecke7a38dc02017-04-06 15:36:29 +0200236 * scsi_eh_reset - call into ->eh_action to reset internal counters
237 * @scmd: scmd to run eh on.
238 *
239 * The scsi driver might be carrying internal state about the
240 * devices, so we need to call into the driver to reset the
241 * internal state once the error handler is started.
242 */
243static void scsi_eh_reset(struct scsi_cmnd *scmd)
244{
Bart Van Asscheaa8e25e2021-08-09 16:03:05 -0700245 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
Hannes Reinecke7a38dc02017-04-06 15:36:29 +0200246 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
247 if (sdrv->eh_reset)
248 sdrv->eh_reset(scmd);
249 }
250}
251
Bart Van Assche3bd6f432017-12-04 10:06:23 -0800252static void scsi_eh_inc_host_failed(struct rcu_head *head)
253{
Bart Van Assche3be88282018-02-22 11:30:20 -0800254 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
255 struct Scsi_Host *shost = scmd->device->host;
Bart Van Assche3bd6f432017-12-04 10:06:23 -0800256 unsigned long flags;
257
258 spin_lock_irqsave(shost->host_lock, flags);
259 shost->host_failed++;
260 scsi_eh_wakeup(shost);
261 spin_unlock_irqrestore(shost->host_lock, flags);
262}
263
Hannes Reinecke7a38dc02017-04-06 15:36:29 +0200264/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 * scsi_eh_scmd_add - add scsi cmd to error handling.
266 * @scmd: scmd to run eh on.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800267 */
Hannes Reineckea0658632017-04-06 15:36:35 +0200268void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
270 struct Scsi_Host *shost = scmd->device->host;
271 unsigned long flags;
Hannes Reinecke2171b6d2017-04-06 15:36:34 +0200272 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
Hannes Reinecke2171b6d2017-04-06 15:36:34 +0200274 WARN_ON_ONCE(!shost->ehandler);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
276 spin_lock_irqsave(shost->host_lock, flags);
Hannes Reinecke2171b6d2017-04-06 15:36:34 +0200277 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
278 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
279 WARN_ON_ONCE(ret);
280 }
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100281 if (shost->eh_deadline != -1 && !shost->last_reset)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200282 shost->last_reset = jiffies;
283
Hannes Reinecke7a38dc02017-04-06 15:36:29 +0200284 scsi_eh_reset(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 spin_unlock_irqrestore(shost->host_lock, flags);
Bart Van Assche3bd6f432017-12-04 10:06:23 -0800287 /*
288 * Ensure that all tasks observe the host state change before the
289 * host_failed change.
290 */
Bart Van Assche3be88282018-02-22 11:30:20 -0800291 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292}
293
294/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 * scsi_times_out - Timeout function for normal scsi commands.
Jens Axboe242f9dc2008-09-14 05:55:09 -0700296 * @req: request that is timing out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 *
298 * Notes:
299 * We do not need to lock this. There is the potential for a race
300 * only in that the normal completion handling might run, but if the
301 * normal completion function determines that the timer has already
302 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800303 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700304enum blk_eh_timer_return scsi_times_out(struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
Bart Van Asschebed22132017-08-25 13:46:32 -0700306 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
Christoph Hellwig66005932018-05-29 15:52:29 +0200307 enum blk_eh_timer_return rtn = BLK_EH_DONE;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100308 struct Scsi_Host *host = scmd->device->host;
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500309
Kei Tokunagabf816232010-04-01 20:41:40 +0900310 trace_scsi_dispatch_cmd_timeout(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 scsi_log_completion(scmd, TIMEOUT_ERROR);
312
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100313 if (host->eh_deadline != -1 && !host->last_reset)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200314 host->last_reset = jiffies;
315
Christoph Hellwigb6a05c82017-01-30 13:18:58 +0100316 if (host->hostt->eh_timed_out)
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100317 rtn = host->hostt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500318
Christoph Hellwig66005932018-05-29 15:52:29 +0200319 if (rtn == BLK_EH_DONE) {
Keith Busch065990b2018-07-23 08:37:51 -0600320 /*
Keith Buschf1342702018-11-26 09:54:29 -0700321 * Set the command to complete first in order to prevent a real
322 * completion from releasing the command while error handling
323 * is using it. If the command was already completed, then the
324 * lower level driver beat the timeout handler, and it is safe
325 * to return without escalating error recovery.
Keith Busch065990b2018-07-23 08:37:51 -0600326 *
Keith Buschf1342702018-11-26 09:54:29 -0700327 * If timeout handling lost the race to a real completion, the
328 * block layer may ignore that due to a fake timeout injection,
329 * so return RESET_TIMER to allow error handling another shot
330 * at this command.
Keith Busch065990b2018-07-23 08:37:51 -0600331 */
Keith Buschf1342702018-11-26 09:54:29 -0700332 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
333 return BLK_EH_RESET_TIMER;
Hannes Reineckea0658632017-04-06 15:36:35 +0200334 if (scsi_abort_command(scmd) != SUCCESS) {
Hannes Reinecke2171b6d2017-04-06 15:36:34 +0200335 set_host_byte(scmd, DID_TIME_OUT);
Hannes Reineckea0658632017-04-06 15:36:35 +0200336 scsi_eh_scmd_add(scmd);
Hannes Reinecke2171b6d2017-04-06 15:36:34 +0200337 }
Hannes Reineckea33c0702014-06-13 14:01:45 +0200338 }
Jens Axboe242f9dc2008-09-14 05:55:09 -0700339
Christoph Hellwigfa990782008-11-05 15:03:15 +0100340 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341}
342
343/**
344 * scsi_block_when_processing_errors - Prevent cmds from being queued.
345 * @sdev: Device on which we are performing recovery.
346 *
347 * Description:
348 * We block until the host is out of error recovery, and then check to
349 * see whether the host or the device is offline.
350 *
351 * Return value:
352 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800353 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354int scsi_block_when_processing_errors(struct scsi_device *sdev)
355{
356 int online;
357
James Bottomley939647e2005-09-18 15:05:20 -0500358 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
360 online = scsi_device_online(sdev);
361
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 return online;
363}
364EXPORT_SYMBOL(scsi_block_when_processing_errors);
365
366#ifdef CONFIG_SCSI_LOGGING
367/**
368 * scsi_eh_prt_fail_stats - Log info on failures.
369 * @shost: scsi host being recovered.
370 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800371 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
373 struct list_head *work_q)
374{
375 struct scsi_cmnd *scmd;
376 struct scsi_device *sdev;
377 int total_failures = 0;
378 int cmd_failed = 0;
379 int cmd_cancel = 0;
380 int devices_failed = 0;
381
382 shost_for_each_device(sdev, shost) {
383 list_for_each_entry(scmd, work_q, eh_entry) {
384 if (scmd->device == sdev) {
385 ++total_failures;
Hannes Reineckea0658632017-04-06 15:36:35 +0200386 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 ++cmd_cancel;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100388 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 ++cmd_failed;
390 }
391 }
392
393 if (cmd_cancel || cmd_failed) {
394 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea3a790d2014-10-24 14:27:03 +0200395 shost_printk(KERN_INFO, shost,
James Bottomley9ccfc752005-10-02 11:45:08 -0500396 "%s: cmds failed: %d, cancel: %d\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700397 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500398 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 cmd_cancel = 0;
400 cmd_failed = 0;
401 ++devices_failed;
402 }
403 }
404
Hannes Reinecke91921e02014-06-25 16:39:59 +0200405 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
406 "Total of %d commands on %d"
407 " devices require eh work\n",
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100408 total_failures, devices_failed));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409}
410#endif
411
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400412 /**
413 * scsi_report_lun_change - Set flag on all *other* devices on the same target
414 * to indicate that a UNIT ATTENTION is expected.
415 * @sdev: Device reporting the UNIT ATTENTION
416 */
417static void scsi_report_lun_change(struct scsi_device *sdev)
418{
419 sdev->sdev_target->expecting_lun_change = 1;
420}
421
422/**
423 * scsi_report_sense - Examine scsi sense information and log messages for
424 * certain conditions, also issue uevents for some of them.
425 * @sdev: Device reporting the sense code
426 * @sshdr: sshdr to be examined
427 */
428static void scsi_report_sense(struct scsi_device *sdev,
429 struct scsi_sense_hdr *sshdr)
430{
431 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
432
433 if (sshdr->sense_key == UNIT_ATTENTION) {
434 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
435 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
436 sdev_printk(KERN_WARNING, sdev,
437 "Inquiry data has changed");
438 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
439 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
440 scsi_report_lun_change(sdev);
441 sdev_printk(KERN_WARNING, sdev,
442 "Warning! Received an indication that the "
443 "LUN assignments on this target have "
444 "changed. The Linux SCSI layer does not "
445 "automatically remap LUN assignments.\n");
446 } else if (sshdr->asc == 0x3f)
447 sdev_printk(KERN_WARNING, sdev,
448 "Warning! Received an indication that the "
449 "operating parameters on this target have "
450 "changed. The Linux SCSI layer does not "
451 "automatically adjust these parameters.\n");
452
453 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
454 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
455 sdev_printk(KERN_WARNING, sdev,
456 "Warning! Received an indication that the "
457 "LUN reached a thin provisioning soft "
458 "threshold.\n");
459 }
460
Hannes Reineckecf3431b2017-10-17 09:11:24 +0200461 if (sshdr->asc == 0x29) {
462 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
463 sdev_printk(KERN_WARNING, sdev,
464 "Power-on or device reset occurred\n");
465 }
466
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400467 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
468 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
469 sdev_printk(KERN_WARNING, sdev,
470 "Mode parameters changed");
Hannes Reinecke14c3e672015-07-06 13:41:53 +0200471 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
472 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
473 sdev_printk(KERN_WARNING, sdev,
474 "Asymmetric access state changed");
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400475 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
476 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
477 sdev_printk(KERN_WARNING, sdev,
478 "Capacity data has changed");
479 } else if (sshdr->asc == 0x2a)
480 sdev_printk(KERN_WARNING, sdev,
481 "Parameters changed");
482 }
483
484 if (evt_type != SDEV_EVT_MAXBITS) {
485 set_bit(evt_type, sdev->pending_events);
486 schedule_work(&sdev->event_work);
487 }
488}
489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490/**
491 * scsi_check_sense - Examine scsi cmd sense
492 * @scmd: Cmd to have sense checked.
493 *
494 * Return value:
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200495 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 *
497 * Notes:
498 * When a deferred error is detected the current command has
499 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800500 */
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700501enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700503 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 struct scsi_sense_hdr sshdr;
505
506 if (! scsi_command_normalize_sense(scmd, &sshdr))
507 return FAILED; /* no valid sense data */
508
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400509 scsi_report_sense(sdev, &sshdr);
510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 if (scsi_sense_is_deferred(&sshdr))
512 return NEEDS_RETRY;
513
Christoph Hellwigee14c672015-08-27 14:16:59 +0200514 if (sdev->handler && sdev->handler->check_sense) {
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700515 enum scsi_disposition rc;
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700516
Christoph Hellwigee14c672015-08-27 14:16:59 +0200517 rc = sdev->handler->check_sense(sdev, &sshdr);
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700518 if (rc != SCSI_RETURN_NOT_HANDLED)
519 return rc;
520 /* handler does not care. Drop down to default handling */
521 }
522
Christoph Hellwige925cc42014-11-06 15:11:22 -0600523 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
524 /*
525 * nasty: for mid-layer issued TURs, we need to return the
526 * actual sense data without any recovery attempt. For eh
527 * issued ones, we need to try to recover and interpret
528 */
529 return SUCCESS;
530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 /*
532 * Previous logic looked for FILEMARK, EOM or ILI which are
533 * mainly associated with tapes and returned SUCCESS.
534 */
535 if (sshdr.response_code == 0x70) {
536 /* fixed format */
537 if (scmd->sense_buffer[2] & 0xe0)
538 return SUCCESS;
539 } else {
540 /*
541 * descriptor format: look for "stream commands sense data
542 * descriptor" (see SSC-3). Assume single sense data
543 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
544 */
545 if ((sshdr.additional_length > 3) &&
546 (scmd->sense_buffer[8] == 0x4) &&
547 (scmd->sense_buffer[11] & 0xe0))
548 return SUCCESS;
549 }
550
551 switch (sshdr.sense_key) {
552 case NO_SENSE:
553 return SUCCESS;
554 case RECOVERED_ERROR:
555 return /* soft_error */ SUCCESS;
556
557 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400558 if (sshdr.asc == 0x10) /* DIF */
559 return SUCCESS;
560
Martin Wilck29cfc2a2018-04-18 01:35:10 +0200561 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
562 return ADD_TO_MLQUEUE;
Martin Wilckc3606522018-04-18 01:35:11 +0200563 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
564 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
565 return ADD_TO_MLQUEUE;
Martin Wilck29cfc2a2018-04-18 01:35:10 +0200566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 return NEEDS_RETRY;
568 case NOT_READY:
569 case UNIT_ATTENTION:
570 /*
571 * if we are expecting a cc/ua because of a bus reset that we
572 * performed, treat this just as a retry. otherwise this is
573 * information that we should pass up to the upper-level driver
574 * so that we can deal with it there.
575 */
576 if (scmd->device->expecting_cc_ua) {
TARUISI Hiroakidfcf7772011-08-11 20:25:20 +0900577 /*
578 * Because some device does not queue unit
579 * attentions correctly, we carefully check
580 * additional sense code and qualifier so as
581 * not to squash media change unit attention.
582 */
583 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
584 scmd->device->expecting_cc_ua = 0;
585 return NEEDS_RETRY;
586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 }
588 /*
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400589 * we might also expect a cc/ua if another LUN on the target
590 * reported a UA with an ASC/ASCQ of 3F 0E -
591 * REPORTED LUNS DATA HAS CHANGED.
592 */
593 if (scmd->device->sdev_target->expecting_lun_change &&
594 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
595 return NEEDS_RETRY;
596 /*
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100597 * if the device is in the process of becoming ready, we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 * should retry.
599 */
600 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
601 return NEEDS_RETRY;
602 /*
603 * if the device is not started, we need to wake
604 * the error handler to start the motor
605 */
606 if (scmd->device->allow_restart &&
607 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
608 return FAILED;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100609 /*
610 * Pass the UA upwards for a determination in the completion
611 * functions.
612 */
613 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100615 /* these are not supported */
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +0200616 case DATA_PROTECT:
617 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
618 /* Thin provisioning hard threshold reached */
619 set_host_byte(scmd, DID_ALLOC_FAILURE);
620 return SUCCESS;
621 }
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -0500622 fallthrough;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 case COPY_ABORTED:
624 case VOLUME_OVERFLOW:
625 case MISCOMPARE:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100626 case BLANK_CHECK:
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200627 set_host_byte(scmd, DID_TARGET_FAILURE);
628 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
630 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800631 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
632 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
633 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
Hannes Reinecke7e782af2013-07-01 15:16:26 +0200634 set_host_byte(scmd, DID_MEDIUM_ERROR);
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200635 return SUCCESS;
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 return NEEDS_RETRY;
638
639 case HARDWARE_ERROR:
640 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700641 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 else
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200643 set_host_byte(scmd, DID_TARGET_FAILURE);
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -0500644 fallthrough;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
646 case ILLEGAL_REQUEST:
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500647 if (sshdr.asc == 0x20 || /* Invalid command operation code */
648 sshdr.asc == 0x21 || /* Logical block address out of range */
Hannes Reineckea8bbb2a2017-10-17 09:10:56 +0200649 sshdr.asc == 0x22 || /* Invalid function */
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500650 sshdr.asc == 0x24 || /* Invalid field in cdb */
Martin Wilckd0b7a902017-09-27 14:44:19 +0200651 sshdr.asc == 0x26 || /* Parameter value invalid */
652 sshdr.asc == 0x27) { /* Write protected */
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200653 set_host_byte(scmd, DID_TARGET_FAILURE);
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500654 }
655 return SUCCESS;
656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 default:
658 return SUCCESS;
659 }
660}
Hannes Reinecke3852e372016-04-04 11:44:01 +0200661EXPORT_SYMBOL_GPL(scsi_check_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Vasu Dev4a840672009-10-22 15:46:33 -0700663static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
664{
665 struct scsi_host_template *sht = sdev->host->hostt;
666 struct scsi_device *tmp_sdev;
667
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100668 if (!sht->track_queue_depth ||
Vasu Dev4a840672009-10-22 15:46:33 -0700669 sdev->queue_depth >= sdev->max_queue_depth)
670 return;
671
672 if (time_before(jiffies,
673 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
674 return;
675
676 if (time_before(jiffies,
677 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
678 return;
679
680 /*
681 * Walk all devices of a target and do
682 * ramp up on them.
683 */
684 shost_for_each_device(tmp_sdev, sdev->host) {
685 if (tmp_sdev->channel != sdev->channel ||
686 tmp_sdev->id != sdev->id ||
687 tmp_sdev->queue_depth == sdev->max_queue_depth)
688 continue;
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100689
Christoph Hellwigdb5ed4d2014-11-13 15:08:42 +0100690 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
Vasu Dev4a840672009-10-22 15:46:33 -0700691 sdev->last_queue_ramp_up = jiffies;
692 }
693}
694
Mike Christie42a6a912009-10-15 17:46:44 -0700695static void scsi_handle_queue_full(struct scsi_device *sdev)
696{
697 struct scsi_host_template *sht = sdev->host->hostt;
698 struct scsi_device *tmp_sdev;
699
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100700 if (!sht->track_queue_depth)
Mike Christie42a6a912009-10-15 17:46:44 -0700701 return;
702
703 shost_for_each_device(tmp_sdev, sdev->host) {
704 if (tmp_sdev->channel != sdev->channel ||
705 tmp_sdev->id != sdev->id)
706 continue;
707 /*
708 * We do not know the number of commands that were at
709 * the device when we got the queue full so we start
710 * from the highest possible value and work our way down.
711 */
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100712 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
Mike Christie42a6a912009-10-15 17:46:44 -0700713 }
714}
715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716/**
717 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
718 * @scmd: SCSI cmd to examine.
719 *
720 * Notes:
721 * This is *only* called when we are examining the status of commands
722 * queued during error recovery. the main difference here is that we
723 * don't allow for the possibility of retries here, and we are a lot
724 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800725 */
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700726static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
728 /*
729 * first check the host byte, to see if there is anything in there
730 * that would indicate what we need to do.
731 */
732 if (host_byte(scmd->result) == DID_RESET) {
733 /*
734 * rats. we are already in the error handler, so we now
735 * get to try and figure out what to do next. if the sense
736 * is valid, we have a pretty good idea of what to do.
737 * if not, we mark it as FAILED.
738 */
739 return scsi_check_sense(scmd);
740 }
741 if (host_byte(scmd->result) != DID_OK)
742 return FAILED;
743
744 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 * now, check the status byte to see if this indicates
746 * anything special.
747 */
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +0200748 switch (get_status_byte(scmd)) {
749 case SAM_STAT_GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700750 scsi_handle_queue_ramp_up(scmd->device);
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -0500751 fallthrough;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +0200752 case SAM_STAT_COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 return SUCCESS;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +0200754 case SAM_STAT_CHECK_CONDITION:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 return scsi_check_sense(scmd);
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +0200756 case SAM_STAT_CONDITION_MET:
757 case SAM_STAT_INTERMEDIATE:
758 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 /*
760 * who knows? FIXME(eric)
761 */
762 return SUCCESS;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +0200763 case SAM_STAT_RESERVATION_CONFLICT:
James Bottomley67110df2010-08-06 15:17:24 -0500764 if (scmd->cmnd[0] == TEST_UNIT_READY)
765 /* it is a success, we probed the device and
766 * found it */
767 return SUCCESS;
768 /* otherwise, we failed to send the command */
769 return FAILED;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +0200770 case SAM_STAT_TASK_SET_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700771 scsi_handle_queue_full(scmd->device);
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -0500772 fallthrough;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +0200773 case SAM_STAT_BUSY:
Hannes Reinecke3eb3a922010-07-29 10:10:16 +0200774 return NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 default:
776 return FAILED;
777 }
778 return FAILED;
779}
780
781/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 * scsi_eh_done - Completion function for error handling.
783 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800784 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785static void scsi_eh_done(struct scsi_cmnd *scmd)
786{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100787 struct completion *eh_action;
Michael Reed85631672005-12-07 21:46:27 -0600788
Hannes Reinecke91921e02014-06-25 16:39:59 +0200789 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100790 "%s result: %x\n", __func__, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600791
792 eh_action = scmd->device->host->eh_action;
793 if (eh_action)
794 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795}
796
797/**
Brian King292148f2007-01-30 17:51:17 -0600798 * scsi_try_host_reset - ask host adapter to reset itself
Geert Uytterhoevenc2b3ebd2013-05-17 13:22:29 +0200799 * @scmd: SCSI cmd to send host reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800800 */
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700801static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600802{
803 unsigned long flags;
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700804 enum scsi_disposition rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100805 struct Scsi_Host *host = scmd->device->host;
806 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600807
Hannes Reinecke91921e02014-06-25 16:39:59 +0200808 SCSI_LOG_ERROR_RECOVERY(3,
809 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
Brian King292148f2007-01-30 17:51:17 -0600810
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100811 if (!hostt->eh_host_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600812 return FAILED;
813
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100814 rtn = hostt->eh_host_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600815
816 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100817 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600818 ssleep(HOST_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100819 spin_lock_irqsave(host->host_lock, flags);
820 scsi_report_bus_reset(host, scmd_channel(scmd));
821 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600822 }
823
824 return rtn;
825}
826
827/**
828 * scsi_try_bus_reset - ask host to perform a bus reset
829 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800830 */
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700831static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600832{
833 unsigned long flags;
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700834 enum scsi_disposition rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100835 struct Scsi_Host *host = scmd->device->host;
836 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600837
Hannes Reinecke91921e02014-06-25 16:39:59 +0200838 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
839 "%s: Snd Bus RST\n", __func__));
Brian King292148f2007-01-30 17:51:17 -0600840
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100841 if (!hostt->eh_bus_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600842 return FAILED;
843
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100844 rtn = hostt->eh_bus_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600845
846 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100847 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600848 ssleep(BUS_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100849 spin_lock_irqsave(host->host_lock, flags);
850 scsi_report_bus_reset(host, scmd_channel(scmd));
851 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600852 }
853
854 return rtn;
855}
856
Mike Christie30bd7df2008-02-29 18:25:19 -0600857static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
858{
859 sdev->was_reset = 1;
860 sdev->expecting_cc_ua = 1;
861}
862
863/**
864 * scsi_try_target_reset - Ask host to perform a target reset
865 * @scmd: SCSI cmd used to send a target reset
866 *
867 * Notes:
868 * There is no timeout for this operation. if this operation is
869 * unreliable for a given host, then the host itself needs to put a
870 * timer on it, and set the host back to a consistent state prior to
871 * returning.
872 */
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700873static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
Mike Christie30bd7df2008-02-29 18:25:19 -0600874{
875 unsigned long flags;
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700876 enum scsi_disposition rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100877 struct Scsi_Host *host = scmd->device->host;
878 struct scsi_host_template *hostt = host->hostt;
Mike Christie30bd7df2008-02-29 18:25:19 -0600879
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100880 if (!hostt->eh_target_reset_handler)
Mike Christie30bd7df2008-02-29 18:25:19 -0600881 return FAILED;
882
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100883 rtn = hostt->eh_target_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600884 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100885 spin_lock_irqsave(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600886 __starget_for_each_device(scsi_target(scmd->device), NULL,
887 __scsi_report_device_reset);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100888 spin_unlock_irqrestore(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600889 }
890
891 return rtn;
892}
893
Brian King292148f2007-01-30 17:51:17 -0600894/**
895 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
896 * @scmd: SCSI cmd used to send BDR
897 *
898 * Notes:
899 * There is no timeout for this operation. if this operation is
900 * unreliable for a given host, then the host itself needs to put a
901 * timer on it, and set the host back to a consistent state prior to
902 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800903 */
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700904static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600905{
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700906 enum scsi_disposition rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100907 struct scsi_host_template *hostt = scmd->device->host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600908
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100909 if (!hostt->eh_device_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600910 return FAILED;
911
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100912 rtn = hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600913 if (rtn == SUCCESS)
914 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600915 return rtn;
916}
917
Hannes Reinecke883a0302014-10-24 14:27:04 +0200918/**
919 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
Randy Dunlap6583f6f2014-12-29 09:40:56 -0800920 * @hostt: SCSI driver host template
Hannes Reinecke883a0302014-10-24 14:27:04 +0200921 * @scmd: SCSI cmd used to send a target reset
922 *
923 * Return value:
924 * SUCCESS, FAILED, or FAST_IO_FAIL
925 *
926 * Notes:
927 * SUCCESS does not necessarily indicate that the command
928 * has been aborted; it only indicates that the LLDDs
929 * has cleared all references to that command.
930 * LLDDs should return FAILED only if an abort was required
931 * but could not be executed. LLDDs should return FAST_IO_FAIL
932 * if the device is temporarily unavailable (eg due to a
933 * link down on FibreChannel)
934 */
Bart Van Asscheb8e162f2021-04-15 15:08:11 -0700935static enum scsi_disposition
936scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600937{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100938 if (!hostt->eh_abort_handler)
Brian King292148f2007-01-30 17:51:17 -0600939 return FAILED;
940
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100941 return hostt->eh_abort_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600942}
943
Brian King292148f2007-01-30 17:51:17 -0600944static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
945{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100946 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
Brian King292148f2007-01-30 17:51:17 -0600947 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600948 if (scsi_try_target_reset(scmd) != SUCCESS)
949 if (scsi_try_bus_reset(scmd) != SUCCESS)
950 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600951}
952
953/**
Santosh Y3b729f72012-04-08 18:47:09 +0530954 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100955 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200956 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200957 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300958 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200959 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100960 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200961 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200962 * as part of the error recovery process. If @sense_bytes is 0 the command
963 * sent must be one that does not transfer any data. If @sense_bytes != 0
964 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
965 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800966 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200967void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
968 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969{
f59114b2005-04-17 15:00:23 -0500970 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Christoph Hellwig631c2282006-07-08 20:42:15 +0200972 /*
973 * We need saved copies of a number of fields - this is because
974 * error handling may need to overwrite these with different values
975 * to run different commands, and once error handling is complete,
976 * we will need to restore these values prior to running the actual
977 * command.
978 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200979 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300980 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200981 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200982 ses->sdb = scmd->sdb;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200983 ses->result = scmd->result;
Damien Le Moal8f8fed02019-10-01 16:48:39 +0900984 ses->resid_len = scmd->req.resid_len;
Alan Stern12265702008-07-21 10:25:52 -0400985 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400986 ses->prot_op = scmd->prot_op;
Hannes Reinecke8e8c9d02017-04-06 15:36:33 +0200987 ses->eh_eflags = scmd->eh_eflags;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200988
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400989 scmd->prot_op = SCSI_PROT_NORMAL;
James Bottomleyc69e6f82014-04-10 13:36:11 -0700990 scmd->eh_eflags = 0;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300991 scmd->cmnd = ses->eh_cmnd;
992 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200993 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Alan Stern644373a2014-03-28 10:51:15 -0700994 scmd->result = 0;
Damien Le Moal8f8fed02019-10-01 16:48:39 +0900995 scmd->req.resid_len = 0;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200996
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200997 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200998 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
999 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +02001000 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001001 scmd->sdb.length);
1002 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +02001003 scmd->sc_data_direction = DMA_FROM_DEVICE;
Tony Battersby0c958ec2015-07-16 11:40:41 -04001004 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +02001005 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001006 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +02001007 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +02001008 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +02001009 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +02001010 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001011 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +02001012 memcpy(scmd->cmnd, cmnd, cmnd_size);
1013 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1014 }
Christoph Hellwig631c2282006-07-08 20:42:15 +02001015 }
1016
1017 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001018
Boaz Harrosh55db6c12007-10-08 16:35:19 +02001019 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -05001021 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
Christoph Hellwig631c2282006-07-08 20:42:15 +02001023 /*
1024 * Zero the sense buffer. The scsi spec mandates that any
1025 * untransferred sense data should be interpreted as being zero.
1026 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09001027 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +02001028}
1029EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +02001030
Boaz Harroshe1c23462007-10-08 16:36:45 +02001031/**
Santosh Y3b729f72012-04-08 18:47:09 +05301032 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
Boaz Harroshe1c23462007-10-08 16:36:45 +02001033 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +02001034 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +02001035 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +02001036 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001037 */
Boaz Harroshe1c23462007-10-08 16:36:45 +02001038void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1039{
1040 /*
1041 * Restore original data
1042 */
1043 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001044 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +02001045 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001046 scmd->sdb = ses->sdb;
Boaz Harroshe1c23462007-10-08 16:36:45 +02001047 scmd->result = ses->result;
Damien Le Moal8f8fed02019-10-01 16:48:39 +09001048 scmd->req.resid_len = ses->resid_len;
Alan Stern12265702008-07-21 10:25:52 -04001049 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -04001050 scmd->prot_op = ses->prot_op;
Hannes Reinecke8e8c9d02017-04-06 15:36:33 +02001051 scmd->eh_eflags = ses->eh_eflags;
Boaz Harroshe1c23462007-10-08 16:36:45 +02001052}
1053EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1054
1055/**
Santosh Y3b729f72012-04-08 18:47:09 +05301056 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
Boaz Harroshe1c23462007-10-08 16:36:45 +02001057 * @scmd: SCSI command structure to hijack
1058 * @cmnd: CDB to send
1059 * @cmnd_size: size in bytes of @cmnd
1060 * @timeout: timeout for this request
1061 * @sense_bytes: size of sense data to copy or 0
1062 *
1063 * This function is used to send a scsi command down to a target device
1064 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1065 *
1066 * Return value:
1067 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001068 */
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001069static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1070 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
Boaz Harroshe1c23462007-10-08 16:36:45 +02001071{
1072 struct scsi_device *sdev = scmd->device;
1073 struct Scsi_Host *shost = sdev->host;
1074 DECLARE_COMPLETION_ONSTACK(done);
Bart Van Asschebbe9fb02019-06-17 08:18:19 -07001075 unsigned long timeleft = timeout, delay;
Boaz Harroshe1c23462007-10-08 16:36:45 +02001076 struct scsi_eh_save ses;
Hannes Reineckefc736482013-04-25 08:10:00 +02001077 const unsigned long stall_for = msecs_to_jiffies(100);
Boaz Harroshe1c23462007-10-08 16:36:45 +02001078 int rtn;
1079
Hannes Reineckefc736482013-04-25 08:10:00 +02001080retry:
Boaz Harroshe1c23462007-10-08 16:36:45 +02001081 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001082 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 scsi_log_send(scmd);
Jeff Garzikf2812332010-11-16 02:10:29 -05001085 scmd->scsi_done = scsi_eh_done;
Bart Van Asschebbe9fb02019-06-17 08:18:19 -07001086
1087 /*
1088 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1089 * change the SCSI device state after we have examined it and before
1090 * .queuecommand() is called.
1091 */
1092 mutex_lock(&sdev->state_mutex);
1093 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1094 mutex_unlock(&sdev->state_mutex);
1095 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1096 "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1097 SDEV_BLOCK));
1098 delay = min(timeleft, stall_for);
1099 timeleft -= delay;
1100 msleep(jiffies_to_msecs(delay));
1101 mutex_lock(&sdev->state_mutex);
1102 }
1103 if (sdev->sdev_state != SDEV_BLOCK)
1104 rtn = shost->hostt->queuecommand(shost, scmd);
1105 else
Bart Van Assche280e91b2021-04-15 15:08:10 -07001106 rtn = FAILED;
Bart Van Asschebbe9fb02019-06-17 08:18:19 -07001107 mutex_unlock(&sdev->state_mutex);
1108
Hannes Reineckefc736482013-04-25 08:10:00 +02001109 if (rtn) {
1110 if (timeleft > stall_for) {
1111 scsi_eh_restore_cmnd(scmd, &ses);
1112 timeleft -= stall_for;
1113 msleep(jiffies_to_msecs(stall_for));
1114 goto retry;
1115 }
1116 /* signal not to enter either branch of the if () below */
1117 timeleft = 0;
Alan Stern511833a2014-11-21 10:44:49 -05001118 rtn = FAILED;
Hannes Reineckefc736482013-04-25 08:10:00 +02001119 } else {
1120 timeleft = wait_for_completion_timeout(&done, timeout);
Hannes Reineckeac61d192014-05-08 08:09:56 +02001121 rtn = SUCCESS;
Hannes Reineckefc736482013-04-25 08:10:00 +02001122 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
f59114b2005-04-17 15:00:23 -05001124 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
Hannes Reineckefc736482013-04-25 08:10:00 +02001126 scsi_log_completion(scmd, rtn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
Hannes Reinecke91921e02014-06-25 16:39:59 +02001128 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01001129 "%s timeleft: %ld\n",
1130 __func__, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
1132 /*
Hannes Reineckefc736482013-04-25 08:10:00 +02001133 * If there is time left scsi_eh_done got called, and we will examine
1134 * the actual status codes to see whether the command actually did
1135 * complete normally, else if we have a zero return and no time left,
1136 * the command must still be pending, so abort it and return FAILED.
1137 * If we never actually managed to issue the command, because
1138 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1139 * value above (so don't execute either branch of the if)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001141 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 rtn = scsi_eh_completed_normally(scmd);
Hannes Reinecke91921e02014-06-25 16:39:59 +02001143 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1144 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001145
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 switch (rtn) {
1147 case SUCCESS:
1148 case NEEDS_RETRY:
1149 case FAILED:
1150 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +02001151 case ADD_TO_MLQUEUE:
1152 rtn = NEEDS_RETRY;
1153 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 default:
1155 rtn = FAILED;
1156 break;
1157 }
Alan Stern511833a2014-11-21 10:44:49 -05001158 } else if (rtn != FAILED) {
Brian King292148f2007-01-30 17:51:17 -06001159 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001160 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 }
1162
Boaz Harroshe1c23462007-10-08 16:36:45 +02001163 scsi_eh_restore_cmnd(scmd, &ses);
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -05001164
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 return rtn;
1166}
1167
1168/**
1169 * scsi_request_sense - Request sense data from a particular target.
1170 * @scmd: SCSI cmd for request sense.
1171 *
1172 * Notes:
1173 * Some hosts automatically obtain this information, others require
1174 * that we obtain it on our own. This function will *not* return until
1175 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001176 */
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001177static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178{
Martin K. Petersen0816c922013-05-10 10:36:04 -04001179 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180}
1181
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001182static enum scsi_disposition
1183scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
James Bottomley24510792013-11-11 13:44:53 +01001184{
Bart Van Asscheaa8e25e2021-08-09 16:03:05 -07001185 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
James Bottomley24510792013-11-11 13:44:53 +01001186 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1187 if (sdrv->eh_action)
1188 rtn = sdrv->eh_action(scmd, rtn);
1189 }
1190 return rtn;
1191}
1192
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193/**
1194 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1195 * @scmd: Original SCSI cmd that eh has finished.
1196 * @done_q: Queue for processed commands.
1197 *
1198 * Notes:
1199 * We don't want to use the normal command completion while we are are
1200 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -05001201 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 * keep a list of pending commands for final completion, and once we
1203 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001204 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001205void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 list_move_tail(&scmd->eh_entry, done_q);
1208}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001209EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
1211/**
1212 * scsi_eh_get_sense - Get device sense data.
1213 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -05001214 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 *
1216 * Description:
1217 * See if we need to request sense information. if so, then get it
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001218 * now, so we have a better idea of what to do.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 *
1220 * Notes:
1221 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -05001222 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 * it down before we request it.
1224 *
1225 * All drivers should request sense information internally these days,
1226 * so for now all I have to say is tough noogies if you end up in here.
1227 *
1228 * XXX: Long term this code should go away, but that needs an audit of
1229 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001230 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001231int scsi_eh_get_sense(struct list_head *work_q,
1232 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001234 struct scsi_cmnd *scmd, *next;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001235 struct Scsi_Host *shost;
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001236 enum scsi_disposition rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
jiang.biao2@zte.com.cn709c75b2015-07-31 17:52:10 +08001238 /*
1239 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1240 * should not get sense.
1241 */
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001242 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Hannes Reineckea0658632017-04-06 15:36:35 +02001243 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 SCSI_SENSE_VALID(scmd))
1245 continue;
1246
Hannes Reineckeb4562022013-10-23 10:51:21 +02001247 shost = scmd->device->host;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001248 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001249 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001250 scmd_printk(KERN_INFO, scmd,
1251 "%s: skip request sense, past eh deadline\n",
1252 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001253 break;
1254 }
Hannes Reinecked0672a02021-04-27 10:30:14 +02001255 if (!scsi_status_is_check_condition(scmd->result))
James Bottomleyd555a2a2014-03-28 10:50:17 -07001256 /*
1257 * don't request sense if there's no check condition
1258 * status because the error we're processing isn't one
1259 * that has a sense code (and some devices get
1260 * confused by sense requests out of the blue)
1261 */
1262 continue;
1263
Jeff Garzik3bf743e2005-10-24 18:04:06 -04001264 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1265 "%s: requesting sense\n",
1266 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 rtn = scsi_request_sense(scmd);
1268 if (rtn != SUCCESS)
1269 continue;
1270
Hannes Reinecke91921e02014-06-25 16:39:59 +02001271 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01001272 "sense requested, result %x\n", scmd->result));
Hannes Reinecked811b842014-10-24 14:26:45 +02001273 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
1275 rtn = scsi_decide_disposition(scmd);
1276
1277 /*
1278 * if the result was normal, then just pass it along to the
1279 * upper level.
1280 */
1281 if (rtn == SUCCESS)
Mike Christie2a242d52020-10-01 10:35:53 -05001282 /*
1283 * We don't want this command reissued, just finished
1284 * with the sense data, so set retries to the max
1285 * allowed to ensure it won't get reissued. If the user
1286 * has requested infinite retries, we also want to
1287 * finish this command, so force completion by setting
1288 * retries and allowed to the same value.
1289 */
1290 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1291 scmd->retries = scmd->allowed = 1;
1292 else
1293 scmd->retries = scmd->allowed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 else if (rtn != NEEDS_RETRY)
1295 continue;
1296
1297 scsi_eh_finish_cmd(scmd, done_q);
1298 }
1299
1300 return list_empty(work_q);
1301}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001302EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
1304/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001306 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 *
1308 * Return value:
1309 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001310 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311static int scsi_eh_tur(struct scsi_cmnd *scmd)
1312{
1313 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001314 int retry_cnt = 1;
1315 enum scsi_disposition rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
1317retry_tur:
Martin K. Petersen0816c922013-05-10 10:36:04 -04001318 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1319 scmd->device->eh_timeout, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
Hannes Reinecke91921e02014-06-25 16:39:59 +02001321 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01001322 "%s return: %x\n", __func__, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +02001323
1324 switch (rtn) {
1325 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 if (retry_cnt--)
1327 goto retry_tur;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001328 fallthrough;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001329 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -05001330 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001331 default:
1332 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -05001333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334}
1335
1336/**
David Jeffery3eef6252011-05-19 14:41:12 -04001337 * scsi_eh_test_devices - check if devices are responding from error recovery.
1338 * @cmd_list: scsi commands in error recovery.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001339 * @work_q: queue for commands which still need more error recovery
1340 * @done_q: queue for commands which are finished
1341 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
David Jeffery3eef6252011-05-19 14:41:12 -04001342 *
1343 * Decription:
1344 * Tests if devices are in a working state. Commands to devices now in
1345 * a working state are sent to the done_q while commands to devices which
1346 * are still failing to respond are returned to the work_q for more
1347 * processing.
1348 **/
1349static int scsi_eh_test_devices(struct list_head *cmd_list,
1350 struct list_head *work_q,
1351 struct list_head *done_q, int try_stu)
1352{
1353 struct scsi_cmnd *scmd, *next;
1354 struct scsi_device *sdev;
1355 int finish_cmds;
1356
1357 while (!list_empty(cmd_list)) {
1358 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1359 sdev = scmd->device;
1360
Hannes Reineckeb4562022013-10-23 10:51:21 +02001361 if (!try_stu) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001362 if (scsi_host_eh_past_deadline(sdev->host)) {
1363 /* Push items back onto work_q */
1364 list_splice_init(cmd_list, work_q);
Hannes Reineckeb4562022013-10-23 10:51:21 +02001365 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001366 sdev_printk(KERN_INFO, sdev,
1367 "%s: skip test device, past eh deadline",
1368 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001369 break;
1370 }
Hannes Reineckeb4562022013-10-23 10:51:21 +02001371 }
1372
David Jeffery3eef6252011-05-19 14:41:12 -04001373 finish_cmds = !scsi_device_online(scmd->device) ||
1374 (try_stu && !scsi_eh_try_stu(scmd) &&
1375 !scsi_eh_tur(scmd)) ||
1376 !scsi_eh_tur(scmd);
1377
1378 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1379 if (scmd->device == sdev) {
James Bottomley24510792013-11-11 13:44:53 +01001380 if (finish_cmds &&
1381 (try_stu ||
1382 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
David Jeffery3eef6252011-05-19 14:41:12 -04001383 scsi_eh_finish_cmd(scmd, done_q);
1384 else
1385 list_move_tail(&scmd->eh_entry, work_q);
1386 }
1387 }
1388 return list_empty(work_q);
1389}
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001393 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 *
1395 * Return value:
1396 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001397 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1399{
1400 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
Christoph Hellwig631c2282006-07-08 20:42:15 +02001402 if (scmd->device->allow_restart) {
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001403 int i;
1404 enum scsi_disposition rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Brian Kinged773e62007-03-29 15:25:52 -05001406 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -06001407 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -05001408
Christoph Hellwig631c2282006-07-08 20:42:15 +02001409 if (rtn == SUCCESS)
1410 return 0;
1411 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 return 1;
1414}
1415
1416 /**
1417 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001418 * @shost: &scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001419 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05001420 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 *
1422 * Notes:
1423 * If commands are failing due to not ready, initializing command required,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001424 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001425 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426static int scsi_eh_stu(struct Scsi_Host *shost,
1427 struct list_head *work_q,
1428 struct list_head *done_q)
1429{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001430 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 struct scsi_device *sdev;
1432
1433 shost_for_each_device(sdev, shost) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001434 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001435 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001436 sdev_printk(KERN_INFO, sdev,
1437 "%s: skip START_UNIT, past eh deadline\n",
1438 current->comm));
Ye Bin4dea1702020-05-18 15:44:20 +08001439 scsi_device_put(sdev);
Hannes Reineckeb4562022013-10-23 10:51:21 +02001440 break;
1441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 stu_scmd = NULL;
1443 list_for_each_entry(scmd, work_q, eh_entry)
1444 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1445 scsi_check_sense(scmd) == FAILED ) {
1446 stu_scmd = scmd;
1447 break;
1448 }
1449
1450 if (!stu_scmd)
1451 continue;
1452
Hannes Reinecke91921e02014-06-25 16:39:59 +02001453 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001454 sdev_printk(KERN_INFO, sdev,
1455 "%s: Sending START_UNIT\n",
1456 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
1458 if (!scsi_eh_try_stu(stu_scmd)) {
1459 if (!scsi_device_online(sdev) ||
1460 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001461 list_for_each_entry_safe(scmd, next,
1462 work_q, eh_entry) {
James Bottomley24510792013-11-11 13:44:53 +01001463 if (scmd->device == sdev &&
1464 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 scsi_eh_finish_cmd(scmd, done_q);
1466 }
1467 }
1468 } else {
1469 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001470 sdev_printk(KERN_INFO, sdev,
1471 "%s: START_UNIT failed\n",
1472 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 }
1474 }
1475
1476 return list_empty(work_q);
1477}
1478
1479
1480/**
1481 * scsi_eh_bus_device_reset - send bdr if needed
1482 * @shost: scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001483 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05001484 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 *
1486 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001487 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 * devices that are jammed or not - if we have multiple devices, it
1489 * makes no sense to try bus_device_reset - we really would need to try
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001490 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001491 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1493 struct list_head *work_q,
1494 struct list_head *done_q)
1495{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001496 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 struct scsi_device *sdev;
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001498 enum scsi_disposition rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
1500 shost_for_each_device(sdev, shost) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001501 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001502 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001503 sdev_printk(KERN_INFO, sdev,
1504 "%s: skip BDR, past eh deadline\n",
1505 current->comm));
Ye Bin4dea1702020-05-18 15:44:20 +08001506 scsi_device_put(sdev);
Hannes Reineckeb4562022013-10-23 10:51:21 +02001507 break;
1508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 bdr_scmd = NULL;
1510 list_for_each_entry(scmd, work_q, eh_entry)
1511 if (scmd->device == sdev) {
1512 bdr_scmd = scmd;
1513 break;
1514 }
1515
1516 if (!bdr_scmd)
1517 continue;
1518
Hannes Reinecke91921e02014-06-25 16:39:59 +02001519 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001520 sdev_printk(KERN_INFO, sdev,
1521 "%s: Sending BDR\n", current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 rtn = scsi_try_bus_device_reset(bdr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001523 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 if (!scsi_device_online(sdev) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001525 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001527 list_for_each_entry_safe(scmd, next,
1528 work_q, eh_entry) {
James Bottomley24510792013-11-11 13:44:53 +01001529 if (scmd->device == sdev &&
1530 scsi_eh_action(scmd, rtn) != FAILED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 scsi_eh_finish_cmd(scmd,
1532 done_q);
1533 }
1534 }
1535 } else {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001536 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001537 sdev_printk(KERN_INFO, sdev,
1538 "%s: BDR failed\n", current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 }
1540 }
1541
1542 return list_empty(work_q);
1543}
1544
1545/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001546 * scsi_eh_target_reset - send target reset if needed
1547 * @shost: scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001548 * @work_q: &list_head for pending commands.
Mike Christie30bd7df2008-02-29 18:25:19 -06001549 * @done_q: &list_head for processed commands.
1550 *
1551 * Notes:
1552 * Try a target reset.
1553 */
1554static int scsi_eh_target_reset(struct Scsi_Host *shost,
1555 struct list_head *work_q,
1556 struct list_head *done_q)
1557{
James Bottomley98db5192010-10-25 15:53:41 -05001558 LIST_HEAD(tmp_list);
David Jeffery3eef6252011-05-19 14:41:12 -04001559 LIST_HEAD(check_list);
Mike Christie30bd7df2008-02-29 18:25:19 -06001560
James Bottomley98db5192010-10-25 15:53:41 -05001561 list_splice_init(work_q, &tmp_list);
1562
1563 while (!list_empty(&tmp_list)) {
1564 struct scsi_cmnd *next, *scmd;
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001565 enum scsi_disposition rtn;
James Bottomley98db5192010-10-25 15:53:41 -05001566 unsigned int id;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001567
Hannes Reineckeb4562022013-10-23 10:51:21 +02001568 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001569 /* push back on work queue for further processing */
1570 list_splice_init(&check_list, work_q);
1571 list_splice_init(&tmp_list, work_q);
1572 SCSI_LOG_ERROR_RECOVERY(3,
1573 shost_printk(KERN_INFO, shost,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001574 "%s: Skip target reset, past eh deadline\n",
1575 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001576 return list_empty(work_q);
1577 }
James Bottomley98db5192010-10-25 15:53:41 -05001578
1579 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1580 id = scmd_id(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -06001581
Hannes Reinecke91921e02014-06-25 16:39:59 +02001582 SCSI_LOG_ERROR_RECOVERY(3,
1583 shost_printk(KERN_INFO, shost,
1584 "%s: Sending target reset to target %d\n",
1585 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001586 rtn = scsi_try_target_reset(scmd);
1587 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
Hannes Reinecke91921e02014-06-25 16:39:59 +02001588 SCSI_LOG_ERROR_RECOVERY(3,
1589 shost_printk(KERN_INFO, shost,
1590 "%s: Target reset failed"
1591 " target: %d\n",
1592 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001593 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1594 if (scmd_id(scmd) != id)
1595 continue;
1596
David Jeffery3eef6252011-05-19 14:41:12 -04001597 if (rtn == SUCCESS)
1598 list_move_tail(&scmd->eh_entry, &check_list);
1599 else if (rtn == FAST_IO_FAIL)
James Bottomley98db5192010-10-25 15:53:41 -05001600 scsi_eh_finish_cmd(scmd, done_q);
1601 else
1602 /* push back on work queue for further processing */
1603 list_move(&scmd->eh_entry, work_q);
1604 }
1605 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001606
David Jeffery3eef6252011-05-19 14:41:12 -04001607 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001608}
1609
1610/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001611 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001612 * @shost: &scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001613 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05001614 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001615 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1617 struct list_head *work_q,
1618 struct list_head *done_q)
1619{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001620 struct scsi_cmnd *scmd, *chan_scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001621 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 unsigned int channel;
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001623 enum scsi_disposition rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625 /*
1626 * we really want to loop over the various channels, and do this on
1627 * a channel by channel basis. we should also check to see if any
1628 * of the failed commands are on soft_reset devices, and if so, skip
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001629 * the reset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 */
1631
1632 for (channel = 0; channel <= shost->max_channel; channel++) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001633 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001634 list_splice_init(&check_list, work_q);
1635 SCSI_LOG_ERROR_RECOVERY(3,
1636 shost_printk(KERN_INFO, shost,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001637 "%s: skip BRST, past eh deadline\n",
1638 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001639 return list_empty(work_q);
1640 }
Hannes Reineckeb4562022013-10-23 10:51:21 +02001641
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 chan_scmd = NULL;
1643 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001644 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 chan_scmd = scmd;
1646 break;
1647 /*
1648 * FIXME add back in some support for
1649 * soft_reset devices.
1650 */
1651 }
1652 }
1653
1654 if (!chan_scmd)
1655 continue;
Hannes Reinecke91921e02014-06-25 16:39:59 +02001656 SCSI_LOG_ERROR_RECOVERY(3,
1657 shost_printk(KERN_INFO, shost,
1658 "%s: Sending BRST chan: %d\n",
1659 current->comm, channel));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 rtn = scsi_try_bus_reset(chan_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001661 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001662 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
David Jeffery3eef6252011-05-19 14:41:12 -04001663 if (channel == scmd_channel(scmd)) {
1664 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 scsi_eh_finish_cmd(scmd,
1666 done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001667 else
1668 list_move_tail(&scmd->eh_entry,
1669 &check_list);
1670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 }
1672 } else {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001673 SCSI_LOG_ERROR_RECOVERY(3,
1674 shost_printk(KERN_INFO, shost,
1675 "%s: BRST failed chan: %d\n",
1676 current->comm, channel));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 }
1678 }
David Jeffery3eef6252011-05-19 14:41:12 -04001679 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680}
1681
1682/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001683 * scsi_eh_host_reset - send a host reset
Randy Dunlap74cf2982014-08-16 14:15:11 -07001684 * @shost: host to be reset.
1685 * @work_q: &list_head for pending commands.
1686 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001687 */
Hannes Reinecke91921e02014-06-25 16:39:59 +02001688static int scsi_eh_host_reset(struct Scsi_Host *shost,
1689 struct list_head *work_q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 struct list_head *done_q)
1691{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001692 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001693 LIST_HEAD(check_list);
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001694 enum scsi_disposition rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
1696 if (!list_empty(work_q)) {
1697 scmd = list_entry(work_q->next,
1698 struct scsi_cmnd, eh_entry);
1699
Hannes Reinecke91921e02014-06-25 16:39:59 +02001700 SCSI_LOG_ERROR_RECOVERY(3,
1701 shost_printk(KERN_INFO, shost,
1702 "%s: Sending HRST\n",
1703 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
1705 rtn = scsi_try_host_reset(scmd);
David Jeffery3eef6252011-05-19 14:41:12 -04001706 if (rtn == SUCCESS) {
1707 list_splice_init(work_q, &check_list);
1708 } else if (rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001709 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 scsi_eh_finish_cmd(scmd, done_q);
1711 }
1712 } else {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001713 SCSI_LOG_ERROR_RECOVERY(3,
1714 shost_printk(KERN_INFO, shost,
1715 "%s: HRST failed\n",
1716 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 }
1718 }
David Jeffery3eef6252011-05-19 14:41:12 -04001719 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720}
1721
1722/**
1723 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
Randy Dunlap74cf2982014-08-16 14:15:11 -07001724 * @work_q: &list_head for pending commands.
1725 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001726 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727static void scsi_eh_offline_sdevs(struct list_head *work_q,
1728 struct list_head *done_q)
1729{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001730 struct scsi_cmnd *scmd, *next;
Bart Van Assche0db6ca82017-06-02 14:21:55 -07001731 struct scsi_device *sdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001733 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001734 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1735 "not ready after error recovery\n");
Bart Van Assche0db6ca82017-06-02 14:21:55 -07001736 sdev = scmd->device;
1737
1738 mutex_lock(&sdev->state_mutex);
1739 scsi_device_set_state(sdev, SDEV_OFFLINE);
1740 mutex_unlock(&sdev->state_mutex);
1741
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 scsi_eh_finish_cmd(scmd, done_q);
1743 }
1744 return;
1745}
1746
1747/**
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001748 * scsi_noretry_cmd - determine if command should be failed fast
Mike Christie4a274462008-08-19 18:45:31 -05001749 * @scmd: SCSI cmd to examine.
1750 */
1751int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1752{
Bart Van Asscheaa8e25e2021-08-09 16:03:05 -07001753 struct request *req = scsi_cmd_to_rq(scmd);
1754
Mike Christie4a274462008-08-19 18:45:31 -05001755 switch (host_byte(scmd->result)) {
1756 case DID_OK:
1757 break;
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001758 case DID_TIME_OUT:
1759 goto check_type;
Mike Christie4a274462008-08-19 18:45:31 -05001760 case DID_BUS_BUSY:
Bart Van Asscheaa8e25e2021-08-09 16:03:05 -07001761 return req->cmd_flags & REQ_FAILFAST_TRANSPORT;
Mike Christie4a274462008-08-19 18:45:31 -05001762 case DID_PARITY:
Bart Van Asscheaa8e25e2021-08-09 16:03:05 -07001763 return req->cmd_flags & REQ_FAILFAST_DEV;
Mike Christie4a274462008-08-19 18:45:31 -05001764 case DID_ERROR:
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +02001765 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
Mike Christie4a274462008-08-19 18:45:31 -05001766 return 0;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001767 fallthrough;
Mike Christie4a274462008-08-19 18:45:31 -05001768 case DID_SOFT_ERROR:
Bart Van Asscheaa8e25e2021-08-09 16:03:05 -07001769 return req->cmd_flags & REQ_FAILFAST_DRIVER;
Mike Christie4a274462008-08-19 18:45:31 -05001770 }
1771
Hannes Reinecked0672a02021-04-27 10:30:14 +02001772 if (!scsi_status_is_check_condition(scmd->result))
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001773 return 0;
Mike Christie4a274462008-08-19 18:45:31 -05001774
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001775check_type:
1776 /*
1777 * assume caller has checked sense and determined
1778 * the check condition was retryable.
1779 */
Bart Van Asscheaa8e25e2021-08-09 16:03:05 -07001780 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req))
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001781 return 1;
Damien Le Moal342c81e2020-09-10 16:48:41 +09001782
1783 return 0;
Mike Christie4a274462008-08-19 18:45:31 -05001784}
1785
1786/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1788 * @scmd: SCSI cmd to examine.
1789 *
1790 * Notes:
1791 * This is *only* called when we are examining the status after sending
1792 * out the actual data command. any commands that are queued for error
1793 * recovery (e.g. test_unit_ready) do *not* come through here.
1794 *
1795 * When this routine returns failed, it means the error handler thread
1796 * is woken. In cases where the error code indicates an error that
1797 * doesn't require the error handler read (i.e. we don't need to
1798 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001799 */
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001800enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801{
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07001802 enum scsi_disposition rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
1804 /*
1805 * if the device is offline, then we clearly just pass the result back
1806 * up to the top level.
1807 */
1808 if (!scsi_device_online(scmd->device)) {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001809 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1810 "%s: device offline - report as SUCCESS\n", __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 return SUCCESS;
1812 }
1813
1814 /*
1815 * first check the host byte, to see if there is anything in there
1816 * that would indicate what we need to do.
1817 */
1818 switch (host_byte(scmd->result)) {
1819 case DID_PASSTHROUGH:
1820 /*
1821 * no matter what, pass this through to the upper layer.
1822 * nuke this special code so that it looks like we are saying
1823 * did_ok.
1824 */
1825 scmd->result &= 0xff00ffff;
1826 return SUCCESS;
1827 case DID_OK:
1828 /*
1829 * looks good. drop through, and check the next byte.
1830 */
1831 break;
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001832 case DID_ABORT:
1833 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
Ulrich Obergfell8922a902014-06-04 13:34:57 +02001834 set_host_byte(scmd, DID_TIME_OUT);
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001835 return SUCCESS;
1836 }
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001837 fallthrough;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 case DID_NO_CONNECT:
1839 case DID_BAD_TARGET:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 /*
1841 * note - this means that we just report the status back
1842 * to the top level driver, not that we actually think
1843 * that it indicates SUCCESS.
1844 */
1845 return SUCCESS;
Petros Koutoupisad950282017-10-30 16:38:10 -05001846 case DID_SOFT_ERROR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 /*
1848 * when the low level driver returns did_soft_error,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001849 * it is responsible for keeping an internal retry counter
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 * in order to avoid endless loops (db)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 goto maybe_retry;
1853 case DID_IMM_RETRY:
1854 return NEEDS_RETRY;
1855
bf341912005-04-12 17:49:09 -05001856 case DID_REQUEUE:
1857 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001858 case DID_TRANSPORT_DISRUPTED:
1859 /*
1860 * LLD/transport was disrupted during processing of the IO.
1861 * The transport class is now blocked/blocking,
1862 * and the transport will decide what to do with the IO
Mike Christie939c22882008-11-04 19:47:19 -06001863 * based on its timers and recovery capablilities if
1864 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001865 */
Mike Christie939c22882008-11-04 19:47:19 -06001866 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001867 case DID_TRANSPORT_FAILFAST:
1868 /*
1869 * The transport decided to failfast the IO (most likely
1870 * the fast io fail tmo fired), so send IO directly upwards.
1871 */
1872 return SUCCESS;
Muneendra Kumar962c8dc2021-01-07 03:19:04 +05301873 case DID_TRANSPORT_MARGINAL:
1874 /*
1875 * caller has decided not to do retries on
1876 * abort success, so send IO directly upwards
1877 */
1878 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 case DID_ERROR:
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +02001880 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 /*
1882 * execute reservation conflict processing code
1883 * lower down
1884 */
1885 break;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001886 fallthrough;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 case DID_BUS_BUSY:
1888 case DID_PARITY:
1889 goto maybe_retry;
1890 case DID_TIME_OUT:
1891 /*
1892 * when we scan the bus, we get timeout messages for
1893 * these commands if there is no device available.
1894 * other hosts report did_no_connect for the same thing.
1895 */
1896 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1897 scmd->cmnd[0] == INQUIRY)) {
1898 return SUCCESS;
1899 } else {
1900 return FAILED;
1901 }
1902 case DID_RESET:
1903 return SUCCESS;
1904 default:
1905 return FAILED;
1906 }
1907
1908 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 * check the status byte to see if this indicates anything special.
1910 */
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +02001911 switch (get_status_byte(scmd)) {
1912 case SAM_STAT_TASK_SET_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001913 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 /*
1915 * the case of trying to send too many commands to a
1916 * tagged queueing device.
1917 */
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001918 fallthrough;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +02001919 case SAM_STAT_BUSY:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 /*
1921 * device can't talk to us at the moment. Should only
1922 * occur (SAM-3) when the task queue is empty, so will cause
1923 * the empty queue handling to trigger a stall in the
1924 * device.
1925 */
1926 return ADD_TO_MLQUEUE;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +02001927 case SAM_STAT_GOOD:
Ewan D. Milne279afdf2013-08-08 15:07:48 -04001928 if (scmd->cmnd[0] == REPORT_LUNS)
1929 scmd->device->sdev_target->expecting_lun_change = 0;
Vasu Dev4a840672009-10-22 15:46:33 -07001930 scsi_handle_queue_ramp_up(scmd->device);
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001931 fallthrough;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +02001932 case SAM_STAT_COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 return SUCCESS;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +02001934 case SAM_STAT_TASK_ABORTED:
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001935 goto maybe_retry;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +02001936 case SAM_STAT_CHECK_CONDITION:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 rtn = scsi_check_sense(scmd);
1938 if (rtn == NEEDS_RETRY)
1939 goto maybe_retry;
1940 /* if rtn == FAILED, we have no sense information;
1941 * returning FAILED will wake the error handler thread
1942 * to collect the sense and redo the decide
1943 * disposition */
1944 return rtn;
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +02001945 case SAM_STAT_CONDITION_MET:
1946 case SAM_STAT_INTERMEDIATE:
1947 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
1948 case SAM_STAT_ACA_ACTIVE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 /*
1950 * who knows? FIXME(eric)
1951 */
1952 return SUCCESS;
1953
Hannes Reinecke3d45cefc2021-04-27 10:30:46 +02001954 case SAM_STAT_RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001955 sdev_printk(KERN_INFO, scmd->device,
1956 "reservation conflict\n");
Moger, Babu2082ebc2012-01-24 20:38:46 +00001957 set_host_byte(scmd, DID_NEXUS_FAILURE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 return SUCCESS; /* causes immediate i/o error */
1959 default:
1960 return FAILED;
1961 }
1962 return FAILED;
1963
Jianchao Wang8ef7fe42018-02-26 15:59:16 +08001964maybe_retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
1966 /* we requeue for retry because the error was retryable, and
1967 * the request was not marked fast fail. Note that above,
1968 * even if the request is marked fast fail, we still requeue
1969 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Mike Christie2a242d52020-10-01 10:35:53 -05001970 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 return NEEDS_RETRY;
1972 } else {
1973 /*
1974 * no more retries - report this one back to upper level.
1975 */
1976 return SUCCESS;
1977 }
1978}
1979
Christoph Hellwig2a842ac2017-06-03 09:38:04 +02001980static void eh_lock_door_done(struct request *req, blk_status_t status)
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001981{
Jens Axboe92bc5a22018-10-24 13:52:28 -06001982 blk_put_request(req);
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001983}
1984
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 * scsi_eh_lock_door - Prevent medium removal for the specified device
1987 * @sdev: SCSI device to prevent medium removal
1988 *
1989 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001990 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 *
1992 * Notes:
1993 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1994 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001995 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996static void scsi_eh_lock_door(struct scsi_device *sdev)
1997{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001998 struct request *req;
Christoph Hellwig82ed4db2017-01-27 09:46:29 +01001999 struct scsi_request *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
Christoph Hellwigda6269d2021-06-24 14:39:34 +02002001 req = blk_get_request(sdev->request_queue, REQ_OP_DRV_IN, 0);
Joe Lawrencea492f072014-08-28 08:15:21 -06002002 if (IS_ERR(req))
Joe Lawrenceeb571ee2014-07-02 15:35:16 -04002003 return;
Christoph Hellwig82ed4db2017-01-27 09:46:29 +01002004 rq = scsi_req(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
Christoph Hellwig82ed4db2017-01-27 09:46:29 +01002006 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
2007 rq->cmd[1] = 0;
2008 rq->cmd[2] = 0;
2009 rq->cmd[3] = 0;
2010 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
2011 rq->cmd[5] = 0;
2012 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
FUJITA Tomonorif0787272008-12-14 01:23:45 +09002013
Christoph Hellwige8064022016-10-20 15:12:13 +02002014 req->rq_flags |= RQF_QUIET;
FUJITA Tomonorif0787272008-12-14 01:23:45 +09002015 req->timeout = 10 * HZ;
Christoph Hellwig64c7f1d2017-04-05 19:18:12 +02002016 rq->retries = 5;
FUJITA Tomonorif0787272008-12-14 01:23:45 +09002017
Guoqing Jiang8eeed0b2021-01-25 05:49:57 +01002018 blk_execute_rq_nowait(NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019}
2020
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021/**
2022 * scsi_restart_operations - restart io operations to the specified host.
2023 * @shost: Host we are restarting.
2024 *
2025 * Notes:
2026 * When we entered the error handler, we blocked all further i/o to
2027 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002028 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029static void scsi_restart_operations(struct Scsi_Host *shost)
2030{
2031 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05002032 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
2034 /*
2035 * If the door was locked, we need to insert a door lock request
2036 * onto the head of the SCSI request queue for the device. There
2037 * is no point trying to lock the door of an off-line device.
2038 */
2039 shost_for_each_device(sdev, shost) {
Christoph Hellwig48379272014-11-03 19:36:40 +01002040 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 scsi_eh_lock_door(sdev);
Christoph Hellwig48379272014-11-03 19:36:40 +01002042 sdev->was_reset = 0;
2043 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 }
2045
2046 /*
2047 * next free up anything directly waiting upon the host. this
2048 * will be requests for character device operations, and also for
2049 * ioctls to queued block devices.
2050 */
Hannes Reineckeb4562022013-10-23 10:51:21 +02002051 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002052 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
James Bottomley939647e2005-09-18 15:05:20 -05002054 spin_lock_irqsave(shost->host_lock, flags);
2055 if (scsi_host_set_state(shost, SHOST_RUNNING))
2056 if (scsi_host_set_state(shost, SHOST_CANCEL))
2057 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2058 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
2060 wake_up(&shost->host_wait);
2061
2062 /*
2063 * finally we need to re-initiate requests that may be pending. we will
2064 * have had everything blocked while error handling is taking place, and
2065 * now that error recovery is done, we will need to ensure that these
2066 * requests are started.
2067 */
2068 scsi_run_host_queues(shost);
Dan Williams57fc2e32012-06-21 23:25:32 -07002069
2070 /*
2071 * if eh is active and host_eh_scheduled is pending we need to re-run
2072 * recovery. we do this check after scsi_run_host_queues() to allow
2073 * everything pent up since the last eh run a chance to make forward
2074 * progress before we sync again. Either we'll immediately re-run
2075 * recovery or scsi_device_unbusy() will wake us again when these
2076 * pending commands complete.
2077 */
2078 spin_lock_irqsave(shost->host_lock, flags);
2079 if (shost->host_eh_scheduled)
2080 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2081 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2082 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083}
2084
2085/**
2086 * scsi_eh_ready_devs - check device ready state and recover if not.
Randy Dunlap74cf2982014-08-16 14:15:11 -07002087 * @shost: host to be recovered.
2088 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05002089 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002090 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08002091void scsi_eh_ready_devs(struct Scsi_Host *shost,
2092 struct list_head *work_q,
2093 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094{
2095 if (!scsi_eh_stu(shost, work_q, done_q))
2096 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06002097 if (!scsi_eh_target_reset(shost, work_q, done_q))
2098 if (!scsi_eh_bus_reset(shost, work_q, done_q))
Hannes Reinecke91921e02014-06-25 16:39:59 +02002099 if (!scsi_eh_host_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06002100 scsi_eh_offline_sdevs(work_q,
2101 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08002103EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
2105/**
2106 * scsi_eh_flush_done_q - finish processed commands or retry them.
2107 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002108 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09002109void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02002111 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02002113 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2114 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 if (scsi_device_online(scmd->device) &&
Muneendra Kumar60bee272021-01-07 03:19:05 +05302116 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) &&
2117 scsi_eh_should_retry_cmd(scmd)) {
Hannes Reinecke91921e02014-06-25 16:39:59 +02002118 SCSI_LOG_ERROR_RECOVERY(3,
2119 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01002120 "%s: flush retry cmd\n",
2121 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2123 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07002124 /*
2125 * If just we got sense for the device (called
2126 * scsi_eh_get_sense), scmd->result is already
Hannes Reinecke4bd51e52021-04-27 10:30:17 +02002127 * set, do not set DID_TIME_OUT.
Patrick Mansfield 793698c2005-05-16 17:42:15 -07002128 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 if (!scmd->result)
Hannes Reinecke4bd51e52021-04-27 10:30:17 +02002130 scmd->result |= (DID_TIME_OUT << 16);
Hannes Reinecke91921e02014-06-25 16:39:59 +02002131 SCSI_LOG_ERROR_RECOVERY(3,
2132 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01002133 "%s: flush finish cmd\n",
2134 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 scsi_finish_command(scmd);
2136 }
2137 }
2138}
Tejun Heo041c5fc2006-01-23 13:09:36 +09002139EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
2141/**
2142 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2143 * @shost: Host to unjam.
2144 *
2145 * Notes:
2146 * When we come in here, we *know* that all commands on the bus have
2147 * either completed, failed or timed out. we also know that no further
2148 * commands are being sent to the host, so things are relatively quiet
2149 * and we have freedom to fiddle with things as we wish.
2150 *
2151 * This is only the *default* implementation. it is possible for
2152 * individual drivers to supply their own version of this function, and
2153 * if the maintainer wishes to do this, it is strongly suggested that
2154 * this function be taken as a template and modified. this function
2155 * was designed to correctly handle problems for about 95% of the
2156 * different cases out there, and it should always provide at least a
2157 * reasonable amount of error recovery.
2158 *
2159 * Any command marked 'failed' or 'timeout' must eventually have
2160 * scsi_finish_cmd() called for it. we do all of the retry stuff
2161 * here, so when we restart the host after we return it should have an
2162 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002163 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164static void scsi_unjam_host(struct Scsi_Host *shost)
2165{
2166 unsigned long flags;
2167 LIST_HEAD(eh_work_q);
2168 LIST_HEAD(eh_done_q);
2169
2170 spin_lock_irqsave(shost->host_lock, flags);
2171 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2172 spin_unlock_irqrestore(shost->host_lock, flags);
2173
2174 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2175
2176 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
Hannes Reineckea0658632017-04-06 15:36:35 +02002177 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
Hannes Reineckeb4562022013-10-23 10:51:21 +02002179 spin_lock_irqsave(shost->host_lock, flags);
Ren Mingxinbb3b6212013-11-11 13:44:56 +01002180 if (shost->eh_deadline != -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +02002181 shost->last_reset = 0;
2182 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 scsi_eh_flush_done_q(&eh_done_q);
2184}
2185
2186/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002187 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 * @data: Host for which we are running.
2189 *
2190 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002191 * This is the main error handling loop. This is run as a kernel thread
2192 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002193 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194int scsi_error_handler(void *data)
2195{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002196 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002199 * We use TASK_INTERRUPTIBLE so that the thread is not
2200 * counted against the load average as a running process.
2201 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01002202 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 */
Michal Hocko537b6042015-08-27 20:16:37 +02002204 while (true) {
2205 /*
2206 * The sequence in kthread_stop() sets the stop flag first
2207 * then wakes the process. To avoid missed wakeups, the task
2208 * should always be in a non running state before the stop
2209 * flag is checked
2210 */
Dan Williamsb9d5c6b2012-06-21 23:25:42 -07002211 set_current_state(TASK_INTERRUPTIBLE);
Michal Hocko537b6042015-08-27 20:16:37 +02002212 if (kthread_should_stop())
2213 break;
2214
Tejun Heoee7863b2006-05-15 20:57:20 +09002215 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
Ming Leic84b0232018-06-24 22:03:26 +08002216 shost->host_failed != scsi_host_busy(shost)) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002217 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002218 shost_printk(KERN_INFO, shost,
2219 "scsi_eh_%d: sleeping\n",
2220 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05002221 schedule();
James Bottomley3ed7a472005-09-19 09:50:04 -05002222 continue;
2223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
James Bottomley3ed7a472005-09-19 09:50:04 -05002225 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002226 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002227 shost_printk(KERN_INFO, shost,
2228 "scsi_eh_%d: waking up %d/%d/%d\n",
2229 shost->host_no, shost->host_eh_scheduled,
Christoph Hellwig74665012014-01-22 15:29:29 +01002230 shost->host_failed,
Ming Leic84b0232018-06-24 22:03:26 +08002231 scsi_host_busy(shost)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 /*
2234 * We have a host that is failing for some reason. Figure out
2235 * what we need to do to get it up and online again (if we can).
2236 * If we fail, we end up taking the thing offline.
2237 */
Lin Mingae0751f2011-12-05 09:20:24 +08002238 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
Alan Sternbc4f2402010-06-17 10:41:42 -04002239 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02002240 shost_printk(KERN_ERR, shost,
2241 "scsi_eh_%d: unable to autoresume\n",
2242 shost->host_no));
Alan Sternbc4f2402010-06-17 10:41:42 -04002243 continue;
2244 }
2245
Christoph Hellwig9227c332006-04-01 19:21:04 +02002246 if (shost->transportt->eh_strategy_handler)
2247 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 else
2249 scsi_unjam_host(shost);
2250
Wei Fang72d8c362016-06-07 14:53:56 +08002251 /* All scmds have been handled */
2252 shost->host_failed = 0;
2253
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 /*
2255 * Note - if the above fails completely, the action is to take
2256 * individual devices offline and flush the queue of any
2257 * outstanding requests that may have been pending. When we
2258 * restart, we restart any I/O to any other devices on the bus
2259 * which are still online.
2260 */
2261 scsi_restart_operations(shost);
Lin Mingae0751f2011-12-05 09:20:24 +08002262 if (!shost->eh_noresume)
2263 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04002265 __set_current_state(TASK_RUNNING);
2266
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002267 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002268 shost_printk(KERN_INFO, shost,
2269 "Error handler scsi_eh_%d exiting\n",
2270 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05002271 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 return 0;
2273}
2274
2275/*
2276 * Function: scsi_report_bus_reset()
2277 *
2278 * Purpose: Utility function used by low-level drivers to report that
2279 * they have observed a bus reset on the bus being handled.
2280 *
2281 * Arguments: shost - Host in question
2282 * channel - channel on which reset was observed.
2283 *
2284 * Returns: Nothing
2285 *
2286 * Lock status: Host lock must be held.
2287 *
2288 * Notes: This only needs to be called if the reset is one which
2289 * originates from an unknown location. Resets originated
2290 * by the mid-level itself don't need to call this, but there
2291 * should be no harm.
2292 *
2293 * The main purpose of this is to make sure that a CHECK_CONDITION
2294 * is properly treated.
2295 */
2296void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2297{
2298 struct scsi_device *sdev;
2299
2300 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06002301 if (channel == sdev_channel(sdev))
2302 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 }
2304}
2305EXPORT_SYMBOL(scsi_report_bus_reset);
2306
2307/*
2308 * Function: scsi_report_device_reset()
2309 *
2310 * Purpose: Utility function used by low-level drivers to report that
2311 * they have observed a device reset on the device being handled.
2312 *
2313 * Arguments: shost - Host in question
2314 * channel - channel on which reset was observed
2315 * target - target on which reset was observed
2316 *
2317 * Returns: Nothing
2318 *
2319 * Lock status: Host lock must be held
2320 *
2321 * Notes: This only needs to be called if the reset is one which
2322 * originates from an unknown location. Resets originated
2323 * by the mid-level itself don't need to call this, but there
2324 * should be no harm.
2325 *
2326 * The main purpose of this is to make sure that a CHECK_CONDITION
2327 * is properly treated.
2328 */
2329void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2330{
2331 struct scsi_device *sdev;
2332
2333 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04002334 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06002335 target == sdev_id(sdev))
2336 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 }
2338}
2339EXPORT_SYMBOL(scsi_report_device_reset);
2340
2341static void
2342scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2343{
2344}
2345
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002346/**
2347 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2348 * @dev: scsi_device to operate on
2349 * @arg: reset type (see sg.h)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 */
2351int
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002352scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353{
Alan Sternbc4f2402010-06-17 10:41:42 -04002354 struct scsi_cmnd *scmd;
James Smartd7a1bb02006-03-08 14:50:12 -05002355 struct Scsi_Host *shost = dev->host;
Christoph Hellwige9c787e2017-01-02 21:55:26 +03002356 struct request *rq;
James Smartd7a1bb02006-03-08 14:50:12 -05002357 unsigned long flags;
Bart Van Asscheb8e162f2021-04-15 15:08:11 -07002358 int error = 0, val;
2359 enum scsi_disposition rtn;
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002360
2361 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2362 return -EACCES;
2363
2364 error = get_user(val, arg);
2365 if (error)
2366 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367
Alan Sternbc4f2402010-06-17 10:41:42 -04002368 if (scsi_autopm_get_host(shost) < 0)
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002369 return -EIO;
Alan Sternbc4f2402010-06-17 10:41:42 -04002370
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002371 error = -EIO;
Christoph Hellwige9c787e2017-01-02 21:55:26 +03002372 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2373 shost->hostt->cmd_size, GFP_KERNEL);
2374 if (!rq)
Christoph Hellwig95eeb5f2014-05-01 16:51:05 +02002375 goto out_put_autopm_host;
Christoph Hellwige9c787e2017-01-02 21:55:26 +03002376 blk_rq_init(NULL, rq);
Christoph Hellwig95eeb5f2014-05-01 16:51:05 +02002377
Christoph Hellwige9c787e2017-01-02 21:55:26 +03002378 scmd = (struct scsi_cmnd *)(rq + 1);
2379 scsi_init_command(dev, scmd);
Christoph Hellwig82ed4db2017-01-27 09:46:29 +01002380 scmd->cmnd = scsi_req(rq)->cmd;
Boaz Harrosh64a87b22008-04-30 11:19:47 +03002381
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02002383 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384
2385 scmd->cmd_len = 0;
2386
2387 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388
James Smartd7a1bb02006-03-08 14:50:12 -05002389 spin_lock_irqsave(shost->host_lock, flags);
2390 shost->tmf_in_progress = 1;
2391 spin_unlock_irqrestore(shost->host_lock, flags);
2392
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002393 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2394 case SG_SCSI_RESET_NOTHING:
2395 rtn = SUCCESS;
2396 break;
2397 case SG_SCSI_RESET_DEVICE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 rtn = scsi_try_bus_device_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002399 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 break;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002401 fallthrough;
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002402 case SG_SCSI_RESET_TARGET:
Mike Christie30bd7df2008-02-29 18:25:19 -06002403 rtn = scsi_try_target_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002404 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
Mike Christie30bd7df2008-02-29 18:25:19 -06002405 break;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002406 fallthrough;
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002407 case SG_SCSI_RESET_BUS:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 rtn = scsi_try_bus_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002409 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 break;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002411 fallthrough;
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002412 case SG_SCSI_RESET_HOST:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 rtn = scsi_try_host_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002414 if (rtn == SUCCESS)
2415 break;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002416 fallthrough;
Bart Van Assche3bf2ff62017-08-25 13:46:29 -07002417 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 rtn = FAILED;
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002419 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 }
2421
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002422 error = (rtn == SUCCESS) ? 0 : -EIO;
2423
James Smartd7a1bb02006-03-08 14:50:12 -05002424 spin_lock_irqsave(shost->host_lock, flags);
2425 shost->tmf_in_progress = 0;
2426 spin_unlock_irqrestore(shost->host_lock, flags);
2427
2428 /*
2429 * be sure to wake up anyone who was sleeping or had their queue
2430 * suspended while we performed the TMF.
2431 */
2432 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002433 shost_printk(KERN_INFO, shost,
2434 "waking up host to restart after TMF\n"));
James Smartd7a1bb02006-03-08 14:50:12 -05002435
2436 wake_up(&shost->host_wait);
James Smartd7a1bb02006-03-08 14:50:12 -05002437 scsi_run_host_queues(shost);
2438
Christoph Hellwige9c787e2017-01-02 21:55:26 +03002439 kfree(rq);
Christoph Hellwig0f121dd2014-09-05 18:00:05 -07002440
Christoph Hellwig04796332014-02-20 14:20:55 -08002441out_put_autopm_host:
Alan Sternbc4f2402010-06-17 10:41:42 -04002442 scsi_autopm_put_host(shost);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002443 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
Hannes Reinecke4753cbc2014-10-24 14:26:52 +02002446bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2447 struct scsi_sense_hdr *sshdr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448{
2449 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09002450 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451}
2452EXPORT_SYMBOL(scsi_command_normalize_sense);
2453
2454/**
Rob Landleyeb448202007-11-03 13:30:39 -05002455 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 * @sense_buffer: byte array of sense data
2457 * @sb_len: number of valid bytes in sense_buffer
2458 * @info_out: pointer to 64 integer where 8 or 4 byte information
2459 * field will be placed if found.
2460 *
2461 * Return value:
Damien Le Moal29087692017-04-24 16:51:12 +09002462 * true if information field found, false if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002463 */
Damien Le Moal29087692017-04-24 16:51:12 +09002464bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2465 u64 *info_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 const u8 * ucp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468
2469 if (sb_len < 7)
Damien Le Moal29087692017-04-24 16:51:12 +09002470 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 switch (sense_buffer[0] & 0x7f) {
2472 case 0x70:
2473 case 0x71:
2474 if (sense_buffer[0] & 0x80) {
Damien Le Moal29087692017-04-24 16:51:12 +09002475 *info_out = get_unaligned_be32(&sense_buffer[3]);
2476 return true;
2477 }
2478 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 case 0x72:
2480 case 0x73:
2481 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2482 0 /* info desc */);
2483 if (ucp && (0xa == ucp[1])) {
Damien Le Moal29087692017-04-24 16:51:12 +09002484 *info_out = get_unaligned_be64(&ucp[4]);
2485 return true;
2486 }
2487 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 default:
Damien Le Moal29087692017-04-24 16:51:12 +09002489 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 }
2491}
2492EXPORT_SYMBOL(scsi_get_sense_info_fld);