blob: 106884a5444e1cb6f61057c29f813c4d713760fa [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
3 *
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
Jesper Juhl0bf8c862011-03-21 20:47:31 +01006 * Leonard Zubkoff and David Miller at Linux Expo,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * ideas originating from all over the place.
8 *
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
11 *
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
Jesper Juhl0bf8c862011-03-21 20:47:31 +010013 * minor cleanups.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
15 */
16
17#include <linux/module.h>
18#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/timer.h>
21#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/kernel.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070023#include <linux/freezer.h>
Christoph Hellwigc5478de2005-09-06 14:04:26 +020024#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/interrupt.h>
26#include <linux/blkdev.h>
27#include <linux/delay.h>
Hannes Reineckefc736482013-04-25 08:10:00 +020028#include <linux/jiffies.h>
Hannes Reineckea1524f22015-03-27 16:46:37 +010029#include <asm/unaligned.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>
James Smartc829c392006-03-13 08:28:57 -050037#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <scsi/scsi_host.h>
39#include <scsi/scsi_ioctl.h>
Christoph Hellwig176aa9d2014-10-11 12:06:47 +020040#include <scsi/sg.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42#include "scsi_priv.h"
43#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070044#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Kei Tokunagabf816232010-04-01 20:41:40 +090046#include <trace/events/scsi.h>
47
James Bottomley14216562012-07-25 23:55:55 +040048static void scsi_eh_done(struct scsi_cmnd *scmd);
49
Linus Torvalds1da177e2005-04-16 15:20:36 -070050/*
51 * These should *probably* be handled by the host itself.
52 * Since it is allowed to sleep, it probably should.
53 */
54#define BUS_RESET_SETTLE_TIME (10)
55#define HOST_RESET_SETTLE_TIME (10)
56
David Jeffery3eef6252011-05-19 14:41:12 -040057static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
Hannes Reineckee494f6a2013-11-11 13:44:54 +010058static int scsi_try_to_abort_cmd(struct scsi_host_template *,
59 struct scsi_cmnd *);
David Jeffery3eef6252011-05-19 14:41:12 -040060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061/* called with shost->host_lock held */
62void scsi_eh_wakeup(struct Scsi_Host *shost)
63{
Christoph Hellwig74665012014-01-22 15:29:29 +010064 if (atomic_read(&shost->host_busy) == shost->host_failed) {
Kei Tokunagabf816232010-04-01 20:41:40 +090065 trace_scsi_eh_wakeup(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -050066 wake_up_process(shost->ehandler);
Hannes Reinecke91921e02014-06-25 16:39:59 +020067 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
68 "Waking error handler thread\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 }
70}
Tejun Heof8bbfc22006-05-19 21:07:05 +090071
72/**
73 * scsi_schedule_eh - schedule EH for SCSI host
74 * @shost: SCSI host to invoke error handling on.
75 *
76 * Schedule SCSI EH without scmd.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080077 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090078void scsi_schedule_eh(struct Scsi_Host *shost)
79{
80 unsigned long flags;
81
82 spin_lock_irqsave(shost->host_lock, flags);
83
84 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
85 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
86 shost->host_eh_scheduled++;
87 scsi_eh_wakeup(shost);
88 }
89
90 spin_unlock_irqrestore(shost->host_lock, flags);
91}
92EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Hannes Reineckeb4562022013-10-23 10:51:21 +020094static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
95{
Ren Mingxinbb3b6212013-11-11 13:44:56 +010096 if (!shost->last_reset || shost->eh_deadline == -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +020097 return 0;
98
Hannes Reinecke76ad3e52013-11-11 13:44:55 +010099 /*
100 * 32bit accesses are guaranteed to be atomic
101 * (on all supported architectures), so instead
102 * of using a spinlock we can as well double check
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100103 * if eh_deadline has been set to 'off' during the
Hannes Reinecke76ad3e52013-11-11 13:44:55 +0100104 * time_before call.
105 */
106 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100107 shost->eh_deadline > -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200108 return 0;
109
110 return 1;
111}
112
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113/**
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100114 * scmd_eh_abort_handler - Handle command aborts
115 * @work: command to be aborted.
116 */
117void
118scmd_eh_abort_handler(struct work_struct *work)
119{
120 struct scsi_cmnd *scmd =
121 container_of(work, struct scsi_cmnd, abort_work.work);
122 struct scsi_device *sdev = scmd->device;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100123 int rtn;
124
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100125 if (scsi_host_eh_past_deadline(sdev->host)) {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100126 SCSI_LOG_ERROR_RECOVERY(3,
127 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100128 "eh timeout, not aborting\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100129 } else {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100130 SCSI_LOG_ERROR_RECOVERY(3,
131 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100132 "aborting command\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100133 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
134 if (rtn == SUCCESS) {
Ulrich Obergfell8922a902014-06-04 13:34:57 +0200135 set_host_byte(scmd, DID_TIME_OUT);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100136 if (scsi_host_eh_past_deadline(sdev->host)) {
137 SCSI_LOG_ERROR_RECOVERY(3,
138 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100139 "eh timeout, not retrying "
140 "aborted command\n"));
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100141 } else if (!scsi_noretry_cmd(scmd) &&
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100142 (++scmd->retries <= scmd->allowed)) {
143 SCSI_LOG_ERROR_RECOVERY(3,
144 scmd_printk(KERN_WARNING, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100145 "retry aborted command\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100146 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100147 return;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100148 } else {
149 SCSI_LOG_ERROR_RECOVERY(3,
150 scmd_printk(KERN_WARNING, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100151 "finish aborted command\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100152 scsi_finish_command(scmd);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100153 return;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100154 }
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100155 } else {
156 SCSI_LOG_ERROR_RECOVERY(3,
157 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100158 "cmd abort %s\n",
Hannes Reinecke883a0302014-10-24 14:27:04 +0200159 (rtn == FAST_IO_FAIL) ?
160 "not send" : "failed"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100161 }
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100162 }
163
164 if (!scsi_eh_scmd_add(scmd, 0)) {
165 SCSI_LOG_ERROR_RECOVERY(3,
166 scmd_printk(KERN_WARNING, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100167 "terminate aborted command\n"));
Ulrich Obergfell8922a902014-06-04 13:34:57 +0200168 set_host_byte(scmd, DID_TIME_OUT);
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100169 scsi_finish_command(scmd);
170 }
171}
172
173/**
174 * scsi_abort_command - schedule a command abort
175 * @scmd: scmd to abort.
176 *
177 * We only need to abort commands after a command timeout
178 */
179static int
180scsi_abort_command(struct scsi_cmnd *scmd)
181{
182 struct scsi_device *sdev = scmd->device;
183 struct Scsi_Host *shost = sdev->host;
184 unsigned long flags;
185
186 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
187 /*
188 * Retry after abort failed, escalate to next level.
189 */
Hannes Reinecke7daf4802014-03-31 16:37:34 +0200190 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100191 SCSI_LOG_ERROR_RECOVERY(3,
192 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100193 "previous abort failed\n"));
Bart Van Asschefcc95a72014-06-02 11:50:52 +0200194 BUG_ON(delayed_work_pending(&scmd->abort_work));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100195 return FAILED;
196 }
197
198 /*
199 * Do not try a command abort if
200 * SCSI EH has already started.
201 */
202 spin_lock_irqsave(shost->host_lock, flags);
203 if (scsi_host_in_recovery(shost)) {
204 spin_unlock_irqrestore(shost->host_lock, flags);
205 SCSI_LOG_ERROR_RECOVERY(3,
206 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100207 "not aborting, host in recovery\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100208 return FAILED;
209 }
210
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100211 if (shost->eh_deadline != -1 && !shost->last_reset)
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100212 shost->last_reset = jiffies;
213 spin_unlock_irqrestore(shost->host_lock, flags);
214
215 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
216 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100217 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100218 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
219 return SUCCESS;
220}
221
222/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 * scsi_eh_scmd_add - add scsi cmd to error handling.
224 * @scmd: scmd to run eh on.
225 * @eh_flag: optional SCSI_EH flag.
226 *
227 * Return value:
228 * 0 on failure.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800229 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
231{
232 struct Scsi_Host *shost = scmd->device->host;
233 unsigned long flags;
James Bottomley939647e2005-09-18 15:05:20 -0500234 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
James Bottomley3ed7a472005-09-19 09:50:04 -0500236 if (!shost->ehandler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 return 0;
238
239 spin_lock_irqsave(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500240 if (scsi_host_set_state(shost, SHOST_RECOVERY))
241 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
242 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100244 if (shost->eh_deadline != -1 && !shost->last_reset)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200245 shost->last_reset = jiffies;
246
James Bottomley939647e2005-09-18 15:05:20 -0500247 ret = 1;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100248 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
249 eh_flag &= ~SCSI_EH_CANCEL_CMD;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200250 scmd->eh_eflags |= eh_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 shost->host_failed++;
253 scsi_eh_wakeup(shost);
James Bottomley939647e2005-09-18 15:05:20 -0500254 out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 spin_unlock_irqrestore(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500256 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
258
259/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 * scsi_times_out - Timeout function for normal scsi commands.
Jens Axboe242f9dc2008-09-14 05:55:09 -0700261 * @req: request that is timing out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 *
263 * Notes:
264 * We do not need to lock this. There is the potential for a race
265 * only in that the normal completion handling might run, but if the
266 * normal completion function determines that the timer has already
267 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800268 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700269enum blk_eh_timer_return scsi_times_out(struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700271 struct scsi_cmnd *scmd = req->special;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700272 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100273 struct Scsi_Host *host = scmd->device->host;
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500274
Kei Tokunagabf816232010-04-01 20:41:40 +0900275 trace_scsi_dispatch_cmd_timeout(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 scsi_log_completion(scmd, TIMEOUT_ERROR);
277
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100278 if (host->eh_deadline != -1 && !host->last_reset)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200279 host->last_reset = jiffies;
280
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100281 if (host->transportt->eh_timed_out)
282 rtn = host->transportt->eh_timed_out(scmd);
283 else if (host->hostt->eh_timed_out)
284 rtn = host->hostt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500285
Hannes Reineckea33c0702014-06-13 14:01:45 +0200286 if (rtn == BLK_EH_NOT_HANDLED) {
287 if (!host->hostt->no_async_abort &&
288 scsi_abort_command(scmd) == SUCCESS)
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100289 return BLK_EH_NOT_HANDLED;
290
Hannes Reineckea33c0702014-06-13 14:01:45 +0200291 set_host_byte(scmd, DID_TIME_OUT);
292 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))
293 rtn = BLK_EH_HANDLED;
294 }
Jens Axboe242f9dc2008-09-14 05:55:09 -0700295
Christoph Hellwigfa990782008-11-05 15:03:15 +0100296 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297}
298
299/**
300 * scsi_block_when_processing_errors - Prevent cmds from being queued.
301 * @sdev: Device on which we are performing recovery.
302 *
303 * Description:
304 * We block until the host is out of error recovery, and then check to
305 * see whether the host or the device is offline.
306 *
307 * Return value:
308 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800309 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310int scsi_block_when_processing_errors(struct scsi_device *sdev)
311{
312 int online;
313
James Bottomley939647e2005-09-18 15:05:20 -0500314 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
316 online = scsi_device_online(sdev);
317
Hannes Reinecke91921e02014-06-25 16:39:59 +0200318 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
319 "%s: rtn: %d\n", __func__, online));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
321 return online;
322}
323EXPORT_SYMBOL(scsi_block_when_processing_errors);
324
325#ifdef CONFIG_SCSI_LOGGING
326/**
327 * scsi_eh_prt_fail_stats - Log info on failures.
328 * @shost: scsi host being recovered.
329 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800330 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
332 struct list_head *work_q)
333{
334 struct scsi_cmnd *scmd;
335 struct scsi_device *sdev;
336 int total_failures = 0;
337 int cmd_failed = 0;
338 int cmd_cancel = 0;
339 int devices_failed = 0;
340
341 shost_for_each_device(sdev, shost) {
342 list_for_each_entry(scmd, work_q, eh_entry) {
343 if (scmd->device == sdev) {
344 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200345 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 ++cmd_cancel;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100347 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 ++cmd_failed;
349 }
350 }
351
352 if (cmd_cancel || cmd_failed) {
353 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea3a790d2014-10-24 14:27:03 +0200354 shost_printk(KERN_INFO, shost,
James Bottomley9ccfc752005-10-02 11:45:08 -0500355 "%s: cmds failed: %d, cancel: %d\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700356 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500357 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 cmd_cancel = 0;
359 cmd_failed = 0;
360 ++devices_failed;
361 }
362 }
363
Hannes Reinecke91921e02014-06-25 16:39:59 +0200364 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
365 "Total of %d commands on %d"
366 " devices require eh work\n",
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100367 total_failures, devices_failed));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368}
369#endif
370
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400371 /**
372 * scsi_report_lun_change - Set flag on all *other* devices on the same target
373 * to indicate that a UNIT ATTENTION is expected.
374 * @sdev: Device reporting the UNIT ATTENTION
375 */
376static void scsi_report_lun_change(struct scsi_device *sdev)
377{
378 sdev->sdev_target->expecting_lun_change = 1;
379}
380
381/**
382 * scsi_report_sense - Examine scsi sense information and log messages for
383 * certain conditions, also issue uevents for some of them.
384 * @sdev: Device reporting the sense code
385 * @sshdr: sshdr to be examined
386 */
387static void scsi_report_sense(struct scsi_device *sdev,
388 struct scsi_sense_hdr *sshdr)
389{
390 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
391
392 if (sshdr->sense_key == UNIT_ATTENTION) {
393 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
394 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
395 sdev_printk(KERN_WARNING, sdev,
396 "Inquiry data has changed");
397 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
398 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
399 scsi_report_lun_change(sdev);
400 sdev_printk(KERN_WARNING, sdev,
401 "Warning! Received an indication that the "
402 "LUN assignments on this target have "
403 "changed. The Linux SCSI layer does not "
404 "automatically remap LUN assignments.\n");
405 } else if (sshdr->asc == 0x3f)
406 sdev_printk(KERN_WARNING, sdev,
407 "Warning! Received an indication that the "
408 "operating parameters on this target have "
409 "changed. The Linux SCSI layer does not "
410 "automatically adjust these parameters.\n");
411
412 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
413 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
414 sdev_printk(KERN_WARNING, sdev,
415 "Warning! Received an indication that the "
416 "LUN reached a thin provisioning soft "
417 "threshold.\n");
418 }
419
420 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
421 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
422 sdev_printk(KERN_WARNING, sdev,
423 "Mode parameters changed");
424 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
425 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
426 sdev_printk(KERN_WARNING, sdev,
427 "Capacity data has changed");
428 } else if (sshdr->asc == 0x2a)
429 sdev_printk(KERN_WARNING, sdev,
430 "Parameters changed");
431 }
432
433 if (evt_type != SDEV_EVT_MAXBITS) {
434 set_bit(evt_type, sdev->pending_events);
435 schedule_work(&sdev->event_work);
436 }
437}
438
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439/**
440 * scsi_check_sense - Examine scsi cmd sense
441 * @scmd: Cmd to have sense checked.
442 *
443 * Return value:
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200444 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 *
446 * Notes:
447 * When a deferred error is detected the current command has
448 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800449 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450static int scsi_check_sense(struct scsi_cmnd *scmd)
451{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700452 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 struct scsi_sense_hdr sshdr;
454
455 if (! scsi_command_normalize_sense(scmd, &sshdr))
456 return FAILED; /* no valid sense data */
457
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400458 scsi_report_sense(sdev, &sshdr);
459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 if (scsi_sense_is_deferred(&sshdr))
461 return NEEDS_RETRY;
462
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700463 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
464 sdev->scsi_dh_data->scsi_dh->check_sense) {
465 int rc;
466
467 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
468 if (rc != SCSI_RETURN_NOT_HANDLED)
469 return rc;
470 /* handler does not care. Drop down to default handling */
471 }
472
Christoph Hellwige925cc42014-11-06 15:11:22 -0600473 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
474 /*
475 * nasty: for mid-layer issued TURs, we need to return the
476 * actual sense data without any recovery attempt. For eh
477 * issued ones, we need to try to recover and interpret
478 */
479 return SUCCESS;
480
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 /*
482 * Previous logic looked for FILEMARK, EOM or ILI which are
483 * mainly associated with tapes and returned SUCCESS.
484 */
485 if (sshdr.response_code == 0x70) {
486 /* fixed format */
487 if (scmd->sense_buffer[2] & 0xe0)
488 return SUCCESS;
489 } else {
490 /*
491 * descriptor format: look for "stream commands sense data
492 * descriptor" (see SSC-3). Assume single sense data
493 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
494 */
495 if ((sshdr.additional_length > 3) &&
496 (scmd->sense_buffer[8] == 0x4) &&
497 (scmd->sense_buffer[11] & 0xe0))
498 return SUCCESS;
499 }
500
501 switch (sshdr.sense_key) {
502 case NO_SENSE:
503 return SUCCESS;
504 case RECOVERED_ERROR:
505 return /* soft_error */ SUCCESS;
506
507 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400508 if (sshdr.asc == 0x10) /* DIF */
509 return SUCCESS;
510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 return NEEDS_RETRY;
512 case NOT_READY:
513 case UNIT_ATTENTION:
514 /*
515 * if we are expecting a cc/ua because of a bus reset that we
516 * performed, treat this just as a retry. otherwise this is
517 * information that we should pass up to the upper-level driver
518 * so that we can deal with it there.
519 */
520 if (scmd->device->expecting_cc_ua) {
TARUISI Hiroakidfcf7772011-08-11 20:25:20 +0900521 /*
522 * Because some device does not queue unit
523 * attentions correctly, we carefully check
524 * additional sense code and qualifier so as
525 * not to squash media change unit attention.
526 */
527 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
528 scmd->device->expecting_cc_ua = 0;
529 return NEEDS_RETRY;
530 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 }
532 /*
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400533 * we might also expect a cc/ua if another LUN on the target
534 * reported a UA with an ASC/ASCQ of 3F 0E -
535 * REPORTED LUNS DATA HAS CHANGED.
536 */
537 if (scmd->device->sdev_target->expecting_lun_change &&
538 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
539 return NEEDS_RETRY;
540 /*
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100541 * if the device is in the process of becoming ready, we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 * should retry.
543 */
544 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
545 return NEEDS_RETRY;
546 /*
547 * if the device is not started, we need to wake
548 * the error handler to start the motor
549 */
550 if (scmd->device->allow_restart &&
551 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
552 return FAILED;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100553 /*
554 * Pass the UA upwards for a determination in the completion
555 * functions.
556 */
557 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100559 /* these are not supported */
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +0200560 case DATA_PROTECT:
561 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
562 /* Thin provisioning hard threshold reached */
563 set_host_byte(scmd, DID_ALLOC_FAILURE);
564 return SUCCESS;
565 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 case COPY_ABORTED:
567 case VOLUME_OVERFLOW:
568 case MISCOMPARE:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100569 case BLANK_CHECK:
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200570 set_host_byte(scmd, DID_TARGET_FAILURE);
571 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800574 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
575 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
576 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
Hannes Reinecke7e782af2013-07-01 15:16:26 +0200577 set_host_byte(scmd, DID_MEDIUM_ERROR);
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200578 return SUCCESS;
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 return NEEDS_RETRY;
581
582 case HARDWARE_ERROR:
583 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700584 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 else
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200586 set_host_byte(scmd, DID_TARGET_FAILURE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 case ILLEGAL_REQUEST:
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500589 if (sshdr.asc == 0x20 || /* Invalid command operation code */
590 sshdr.asc == 0x21 || /* Logical block address out of range */
591 sshdr.asc == 0x24 || /* Invalid field in cdb */
592 sshdr.asc == 0x26) { /* Parameter value invalid */
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200593 set_host_byte(scmd, DID_TARGET_FAILURE);
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500594 }
595 return SUCCESS;
596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 default:
598 return SUCCESS;
599 }
600}
601
Vasu Dev4a840672009-10-22 15:46:33 -0700602static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
603{
604 struct scsi_host_template *sht = sdev->host->hostt;
605 struct scsi_device *tmp_sdev;
606
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100607 if (!sht->track_queue_depth ||
Vasu Dev4a840672009-10-22 15:46:33 -0700608 sdev->queue_depth >= sdev->max_queue_depth)
609 return;
610
611 if (time_before(jiffies,
612 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
613 return;
614
615 if (time_before(jiffies,
616 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
617 return;
618
619 /*
620 * Walk all devices of a target and do
621 * ramp up on them.
622 */
623 shost_for_each_device(tmp_sdev, sdev->host) {
624 if (tmp_sdev->channel != sdev->channel ||
625 tmp_sdev->id != sdev->id ||
626 tmp_sdev->queue_depth == sdev->max_queue_depth)
627 continue;
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100628
Christoph Hellwigdb5ed4d2014-11-13 15:08:42 +0100629 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
Vasu Dev4a840672009-10-22 15:46:33 -0700630 sdev->last_queue_ramp_up = jiffies;
631 }
632}
633
Mike Christie42a6a912009-10-15 17:46:44 -0700634static void scsi_handle_queue_full(struct scsi_device *sdev)
635{
636 struct scsi_host_template *sht = sdev->host->hostt;
637 struct scsi_device *tmp_sdev;
638
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100639 if (!sht->track_queue_depth)
Mike Christie42a6a912009-10-15 17:46:44 -0700640 return;
641
642 shost_for_each_device(tmp_sdev, sdev->host) {
643 if (tmp_sdev->channel != sdev->channel ||
644 tmp_sdev->id != sdev->id)
645 continue;
646 /*
647 * We do not know the number of commands that were at
648 * the device when we got the queue full so we start
649 * from the highest possible value and work our way down.
650 */
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100651 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
Mike Christie42a6a912009-10-15 17:46:44 -0700652 }
653}
654
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655/**
656 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
657 * @scmd: SCSI cmd to examine.
658 *
659 * Notes:
660 * This is *only* called when we are examining the status of commands
661 * queued during error recovery. the main difference here is that we
662 * don't allow for the possibility of retries here, and we are a lot
663 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800664 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
666{
667 /*
668 * first check the host byte, to see if there is anything in there
669 * that would indicate what we need to do.
670 */
671 if (host_byte(scmd->result) == DID_RESET) {
672 /*
673 * rats. we are already in the error handler, so we now
674 * get to try and figure out what to do next. if the sense
675 * is valid, we have a pretty good idea of what to do.
676 * if not, we mark it as FAILED.
677 */
678 return scsi_check_sense(scmd);
679 }
680 if (host_byte(scmd->result) != DID_OK)
681 return FAILED;
682
683 /*
684 * next, check the message byte.
685 */
686 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
687 return FAILED;
688
689 /*
690 * now, check the status byte to see if this indicates
691 * anything special.
692 */
693 switch (status_byte(scmd->result)) {
694 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700695 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 case COMMAND_TERMINATED:
697 return SUCCESS;
698 case CHECK_CONDITION:
699 return scsi_check_sense(scmd);
700 case CONDITION_GOOD:
701 case INTERMEDIATE_GOOD:
702 case INTERMEDIATE_C_GOOD:
703 /*
704 * who knows? FIXME(eric)
705 */
706 return SUCCESS;
Michael Reed5f91bb02009-08-10 11:59:28 -0500707 case RESERVATION_CONFLICT:
James Bottomley67110df2010-08-06 15:17:24 -0500708 if (scmd->cmnd[0] == TEST_UNIT_READY)
709 /* it is a success, we probed the device and
710 * found it */
711 return SUCCESS;
712 /* otherwise, we failed to send the command */
713 return FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700715 scsi_handle_queue_full(scmd->device);
716 /* fall through */
717 case BUSY:
Hannes Reinecke3eb3a922010-07-29 10:10:16 +0200718 return NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 default:
720 return FAILED;
721 }
722 return FAILED;
723}
724
725/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 * scsi_eh_done - Completion function for error handling.
727 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800728 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729static void scsi_eh_done(struct scsi_cmnd *scmd)
730{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100731 struct completion *eh_action;
Michael Reed85631672005-12-07 21:46:27 -0600732
Hannes Reinecke91921e02014-06-25 16:39:59 +0200733 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100734 "%s result: %x\n", __func__, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600735
736 eh_action = scmd->device->host->eh_action;
737 if (eh_action)
738 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739}
740
741/**
Brian King292148f2007-01-30 17:51:17 -0600742 * scsi_try_host_reset - ask host adapter to reset itself
Geert Uytterhoevenc2b3ebd2013-05-17 13:22:29 +0200743 * @scmd: SCSI cmd to send host reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800744 */
Brian King292148f2007-01-30 17:51:17 -0600745static int scsi_try_host_reset(struct scsi_cmnd *scmd)
746{
747 unsigned long flags;
748 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100749 struct Scsi_Host *host = scmd->device->host;
750 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600751
Hannes Reinecke91921e02014-06-25 16:39:59 +0200752 SCSI_LOG_ERROR_RECOVERY(3,
753 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
Brian King292148f2007-01-30 17:51:17 -0600754
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100755 if (!hostt->eh_host_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600756 return FAILED;
757
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100758 rtn = hostt->eh_host_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600759
760 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100761 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600762 ssleep(HOST_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100763 spin_lock_irqsave(host->host_lock, flags);
764 scsi_report_bus_reset(host, scmd_channel(scmd));
765 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600766 }
767
768 return rtn;
769}
770
771/**
772 * scsi_try_bus_reset - ask host to perform a bus reset
773 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800774 */
Brian King292148f2007-01-30 17:51:17 -0600775static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
776{
777 unsigned long flags;
778 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100779 struct Scsi_Host *host = scmd->device->host;
780 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600781
Hannes Reinecke91921e02014-06-25 16:39:59 +0200782 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
783 "%s: Snd Bus RST\n", __func__));
Brian King292148f2007-01-30 17:51:17 -0600784
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100785 if (!hostt->eh_bus_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600786 return FAILED;
787
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100788 rtn = hostt->eh_bus_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600789
790 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100791 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600792 ssleep(BUS_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100793 spin_lock_irqsave(host->host_lock, flags);
794 scsi_report_bus_reset(host, scmd_channel(scmd));
795 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600796 }
797
798 return rtn;
799}
800
Mike Christie30bd7df2008-02-29 18:25:19 -0600801static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
802{
803 sdev->was_reset = 1;
804 sdev->expecting_cc_ua = 1;
805}
806
807/**
808 * scsi_try_target_reset - Ask host to perform a target reset
809 * @scmd: SCSI cmd used to send a target reset
810 *
811 * Notes:
812 * There is no timeout for this operation. if this operation is
813 * unreliable for a given host, then the host itself needs to put a
814 * timer on it, and set the host back to a consistent state prior to
815 * returning.
816 */
817static int scsi_try_target_reset(struct scsi_cmnd *scmd)
818{
819 unsigned long flags;
820 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100821 struct Scsi_Host *host = scmd->device->host;
822 struct scsi_host_template *hostt = host->hostt;
Mike Christie30bd7df2008-02-29 18:25:19 -0600823
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100824 if (!hostt->eh_target_reset_handler)
Mike Christie30bd7df2008-02-29 18:25:19 -0600825 return FAILED;
826
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100827 rtn = hostt->eh_target_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600828 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100829 spin_lock_irqsave(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600830 __starget_for_each_device(scsi_target(scmd->device), NULL,
831 __scsi_report_device_reset);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100832 spin_unlock_irqrestore(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600833 }
834
835 return rtn;
836}
837
Brian King292148f2007-01-30 17:51:17 -0600838/**
839 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
840 * @scmd: SCSI cmd used to send BDR
841 *
842 * Notes:
843 * There is no timeout for this operation. if this operation is
844 * unreliable for a given host, then the host itself needs to put a
845 * timer on it, and set the host back to a consistent state prior to
846 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800847 */
Brian King292148f2007-01-30 17:51:17 -0600848static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
849{
850 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100851 struct scsi_host_template *hostt = scmd->device->host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600852
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100853 if (!hostt->eh_device_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600854 return FAILED;
855
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100856 rtn = hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600857 if (rtn == SUCCESS)
858 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600859 return rtn;
860}
861
Hannes Reinecke883a0302014-10-24 14:27:04 +0200862/**
863 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
Randy Dunlap6583f6f2014-12-29 09:40:56 -0800864 * @hostt: SCSI driver host template
Hannes Reinecke883a0302014-10-24 14:27:04 +0200865 * @scmd: SCSI cmd used to send a target reset
866 *
867 * Return value:
868 * SUCCESS, FAILED, or FAST_IO_FAIL
869 *
870 * Notes:
871 * SUCCESS does not necessarily indicate that the command
872 * has been aborted; it only indicates that the LLDDs
873 * has cleared all references to that command.
874 * LLDDs should return FAILED only if an abort was required
875 * but could not be executed. LLDDs should return FAST_IO_FAIL
876 * if the device is temporarily unavailable (eg due to a
877 * link down on FibreChannel)
878 */
879static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
880 struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600881{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100882 if (!hostt->eh_abort_handler)
Brian King292148f2007-01-30 17:51:17 -0600883 return FAILED;
884
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100885 return hostt->eh_abort_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600886}
887
Brian King292148f2007-01-30 17:51:17 -0600888static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
889{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100890 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
Brian King292148f2007-01-30 17:51:17 -0600891 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600892 if (scsi_try_target_reset(scmd) != SUCCESS)
893 if (scsi_try_bus_reset(scmd) != SUCCESS)
894 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600895}
896
897/**
Santosh Y3b729f72012-04-08 18:47:09 +0530898 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100899 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200900 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200901 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300902 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200903 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100904 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200905 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200906 * as part of the error recovery process. If @sense_bytes is 0 the command
907 * sent must be one that does not transfer any data. If @sense_bytes != 0
908 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
909 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800910 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200911void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
912 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
f59114b2005-04-17 15:00:23 -0500914 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Christoph Hellwig631c2282006-07-08 20:42:15 +0200916 /*
917 * We need saved copies of a number of fields - this is because
918 * error handling may need to overwrite these with different values
919 * to run different commands, and once error handling is complete,
920 * we will need to restore these values prior to running the actual
921 * command.
922 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200923 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300924 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200925 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200926 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200927 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200928 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400929 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400930 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200931
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400932 scmd->prot_op = SCSI_PROT_NORMAL;
James Bottomleyc69e6f82014-04-10 13:36:11 -0700933 scmd->eh_eflags = 0;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300934 scmd->cmnd = ses->eh_cmnd;
935 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200936 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200937 scmd->request->next_rq = NULL;
Alan Stern644373a2014-03-28 10:51:15 -0700938 scmd->result = 0;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200939
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200940 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200941 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
942 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200943 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200944 scmd->sdb.length);
945 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200946 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200947 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200948 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200949 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200950 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200951 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200952 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200953 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300954 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200955 memcpy(scmd->cmnd, cmnd, cmnd_size);
956 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
957 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200958 }
959
960 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200961
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200962 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500964 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Christoph Hellwig631c2282006-07-08 20:42:15 +0200966 /*
967 * Zero the sense buffer. The scsi spec mandates that any
968 * untransferred sense data should be interpreted as being zero.
969 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900970 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200971}
972EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200973
Boaz Harroshe1c23462007-10-08 16:36:45 +0200974/**
Santosh Y3b729f72012-04-08 18:47:09 +0530975 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
Boaz Harroshe1c23462007-10-08 16:36:45 +0200976 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200977 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +0200978 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200979 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800980 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200981void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
982{
983 /*
984 * Restore original data
985 */
986 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300987 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200988 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200989 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200990 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200991 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400992 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400993 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200994}
995EXPORT_SYMBOL(scsi_eh_restore_cmnd);
996
997/**
Santosh Y3b729f72012-04-08 18:47:09 +0530998 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
Boaz Harroshe1c23462007-10-08 16:36:45 +0200999 * @scmd: SCSI command structure to hijack
1000 * @cmnd: CDB to send
1001 * @cmnd_size: size in bytes of @cmnd
1002 * @timeout: timeout for this request
1003 * @sense_bytes: size of sense data to copy or 0
1004 *
1005 * This function is used to send a scsi command down to a target device
1006 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1007 *
1008 * Return value:
1009 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001010 */
Boaz Harroshe1c23462007-10-08 16:36:45 +02001011static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1012 int cmnd_size, int timeout, unsigned sense_bytes)
1013{
1014 struct scsi_device *sdev = scmd->device;
1015 struct Scsi_Host *shost = sdev->host;
1016 DECLARE_COMPLETION_ONSTACK(done);
Hannes Reineckefc736482013-04-25 08:10:00 +02001017 unsigned long timeleft = timeout;
Boaz Harroshe1c23462007-10-08 16:36:45 +02001018 struct scsi_eh_save ses;
Hannes Reineckefc736482013-04-25 08:10:00 +02001019 const unsigned long stall_for = msecs_to_jiffies(100);
Boaz Harroshe1c23462007-10-08 16:36:45 +02001020 int rtn;
1021
Hannes Reineckefc736482013-04-25 08:10:00 +02001022retry:
Boaz Harroshe1c23462007-10-08 16:36:45 +02001023 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001024 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 scsi_log_send(scmd);
Jeff Garzikf2812332010-11-16 02:10:29 -05001027 scmd->scsi_done = scsi_eh_done;
Hannes Reineckefc736482013-04-25 08:10:00 +02001028 rtn = shost->hostt->queuecommand(shost, scmd);
1029 if (rtn) {
1030 if (timeleft > stall_for) {
1031 scsi_eh_restore_cmnd(scmd, &ses);
1032 timeleft -= stall_for;
1033 msleep(jiffies_to_msecs(stall_for));
1034 goto retry;
1035 }
1036 /* signal not to enter either branch of the if () below */
1037 timeleft = 0;
Alan Stern511833a2014-11-21 10:44:49 -05001038 rtn = FAILED;
Hannes Reineckefc736482013-04-25 08:10:00 +02001039 } else {
1040 timeleft = wait_for_completion_timeout(&done, timeout);
Hannes Reineckeac61d192014-05-08 08:09:56 +02001041 rtn = SUCCESS;
Hannes Reineckefc736482013-04-25 08:10:00 +02001042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
f59114b2005-04-17 15:00:23 -05001044 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Hannes Reineckefc736482013-04-25 08:10:00 +02001046 scsi_log_completion(scmd, rtn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Hannes Reinecke91921e02014-06-25 16:39:59 +02001048 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01001049 "%s timeleft: %ld\n",
1050 __func__, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
1052 /*
Hannes Reineckefc736482013-04-25 08:10:00 +02001053 * If there is time left scsi_eh_done got called, and we will examine
1054 * the actual status codes to see whether the command actually did
1055 * complete normally, else if we have a zero return and no time left,
1056 * the command must still be pending, so abort it and return FAILED.
1057 * If we never actually managed to issue the command, because
1058 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1059 * value above (so don't execute either branch of the if)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001061 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 rtn = scsi_eh_completed_normally(scmd);
Hannes Reinecke91921e02014-06-25 16:39:59 +02001063 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1064 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001065
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 switch (rtn) {
1067 case SUCCESS:
1068 case NEEDS_RETRY:
1069 case FAILED:
1070 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +02001071 case ADD_TO_MLQUEUE:
1072 rtn = NEEDS_RETRY;
1073 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 default:
1075 rtn = FAILED;
1076 break;
1077 }
Alan Stern511833a2014-11-21 10:44:49 -05001078 } else if (rtn != FAILED) {
Brian King292148f2007-01-30 17:51:17 -06001079 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001080 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
1082
Boaz Harroshe1c23462007-10-08 16:36:45 +02001083 scsi_eh_restore_cmnd(scmd, &ses);
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -05001084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 return rtn;
1086}
1087
1088/**
1089 * scsi_request_sense - Request sense data from a particular target.
1090 * @scmd: SCSI cmd for request sense.
1091 *
1092 * Notes:
1093 * Some hosts automatically obtain this information, others require
1094 * that we obtain it on our own. This function will *not* return until
1095 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001096 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097static int scsi_request_sense(struct scsi_cmnd *scmd)
1098{
Martin K. Petersen0816c922013-05-10 10:36:04 -04001099 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100}
1101
James Bottomley24510792013-11-11 13:44:53 +01001102static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1103{
1104 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1105 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1106 if (sdrv->eh_action)
1107 rtn = sdrv->eh_action(scmd, rtn);
1108 }
1109 return rtn;
1110}
1111
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112/**
1113 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1114 * @scmd: Original SCSI cmd that eh has finished.
1115 * @done_q: Queue for processed commands.
1116 *
1117 * Notes:
1118 * We don't want to use the normal command completion while we are are
1119 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -05001120 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 * keep a list of pending commands for final completion, and once we
1122 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001123 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001124void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125{
1126 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001127 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 list_move_tail(&scmd->eh_entry, done_q);
1129}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001130EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
1132/**
1133 * scsi_eh_get_sense - Get device sense data.
1134 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -05001135 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 *
1137 * Description:
1138 * See if we need to request sense information. if so, then get it
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001139 * now, so we have a better idea of what to do.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 *
1141 * Notes:
1142 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -05001143 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 * it down before we request it.
1145 *
1146 * All drivers should request sense information internally these days,
1147 * so for now all I have to say is tough noogies if you end up in here.
1148 *
1149 * XXX: Long term this code should go away, but that needs an audit of
1150 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001151 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001152int scsi_eh_get_sense(struct list_head *work_q,
1153 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001155 struct scsi_cmnd *scmd, *next;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001156 struct Scsi_Host *shost;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 int rtn;
1158
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001159 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001160 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 SCSI_SENSE_VALID(scmd))
1162 continue;
1163
Hannes Reineckeb4562022013-10-23 10:51:21 +02001164 shost = scmd->device->host;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001165 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001166 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001167 scmd_printk(KERN_INFO, scmd,
1168 "%s: skip request sense, past eh deadline\n",
1169 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001170 break;
1171 }
James Bottomleyd555a2a2014-03-28 10:50:17 -07001172 if (status_byte(scmd->result) != CHECK_CONDITION)
1173 /*
1174 * don't request sense if there's no check condition
1175 * status because the error we're processing isn't one
1176 * that has a sense code (and some devices get
1177 * confused by sense requests out of the blue)
1178 */
1179 continue;
1180
Jeff Garzik3bf743e2005-10-24 18:04:06 -04001181 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1182 "%s: requesting sense\n",
1183 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 rtn = scsi_request_sense(scmd);
1185 if (rtn != SUCCESS)
1186 continue;
1187
Hannes Reinecke91921e02014-06-25 16:39:59 +02001188 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01001189 "sense requested, result %x\n", scmd->result));
Hannes Reinecked811b842014-10-24 14:26:45 +02001190 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
1192 rtn = scsi_decide_disposition(scmd);
1193
1194 /*
1195 * if the result was normal, then just pass it along to the
1196 * upper level.
1197 */
1198 if (rtn == SUCCESS)
1199 /* we don't want this command reissued, just
1200 * finished with the sense data, so set
1201 * retries to the max allowed to ensure it
1202 * won't get reissued */
1203 scmd->retries = scmd->allowed;
1204 else if (rtn != NEEDS_RETRY)
1205 continue;
1206
1207 scsi_eh_finish_cmd(scmd, done_q);
1208 }
1209
1210 return list_empty(work_q);
1211}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001212EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
1214/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001216 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 *
1218 * Return value:
1219 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001220 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221static int scsi_eh_tur(struct scsi_cmnd *scmd)
1222{
1223 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1224 int retry_cnt = 1, rtn;
1225
1226retry_tur:
Martin K. Petersen0816c922013-05-10 10:36:04 -04001227 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1228 scmd->device->eh_timeout, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
Hannes Reinecke91921e02014-06-25 16:39:59 +02001230 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01001231 "%s return: %x\n", __func__, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +02001232
1233 switch (rtn) {
1234 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 if (retry_cnt--)
1236 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001237 /*FALLTHRU*/
1238 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -05001239 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001240 default:
1241 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -05001242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243}
1244
1245/**
David Jeffery3eef6252011-05-19 14:41:12 -04001246 * scsi_eh_test_devices - check if devices are responding from error recovery.
1247 * @cmd_list: scsi commands in error recovery.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001248 * @work_q: queue for commands which still need more error recovery
1249 * @done_q: queue for commands which are finished
1250 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
David Jeffery3eef6252011-05-19 14:41:12 -04001251 *
1252 * Decription:
1253 * Tests if devices are in a working state. Commands to devices now in
1254 * a working state are sent to the done_q while commands to devices which
1255 * are still failing to respond are returned to the work_q for more
1256 * processing.
1257 **/
1258static int scsi_eh_test_devices(struct list_head *cmd_list,
1259 struct list_head *work_q,
1260 struct list_head *done_q, int try_stu)
1261{
1262 struct scsi_cmnd *scmd, *next;
1263 struct scsi_device *sdev;
1264 int finish_cmds;
1265
1266 while (!list_empty(cmd_list)) {
1267 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1268 sdev = scmd->device;
1269
Hannes Reineckeb4562022013-10-23 10:51:21 +02001270 if (!try_stu) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001271 if (scsi_host_eh_past_deadline(sdev->host)) {
1272 /* Push items back onto work_q */
1273 list_splice_init(cmd_list, work_q);
Hannes Reineckeb4562022013-10-23 10:51:21 +02001274 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001275 sdev_printk(KERN_INFO, sdev,
1276 "%s: skip test device, past eh deadline",
1277 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001278 break;
1279 }
Hannes Reineckeb4562022013-10-23 10:51:21 +02001280 }
1281
David Jeffery3eef6252011-05-19 14:41:12 -04001282 finish_cmds = !scsi_device_online(scmd->device) ||
1283 (try_stu && !scsi_eh_try_stu(scmd) &&
1284 !scsi_eh_tur(scmd)) ||
1285 !scsi_eh_tur(scmd);
1286
1287 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1288 if (scmd->device == sdev) {
James Bottomley24510792013-11-11 13:44:53 +01001289 if (finish_cmds &&
1290 (try_stu ||
1291 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
David Jeffery3eef6252011-05-19 14:41:12 -04001292 scsi_eh_finish_cmd(scmd, done_q);
1293 else
1294 list_move_tail(&scmd->eh_entry, work_q);
1295 }
1296 }
1297 return list_empty(work_q);
1298}
1299
1300
1301/**
Rob Landleyeb448202007-11-03 13:30:39 -05001302 * scsi_eh_abort_cmds - abort pending commands.
1303 * @work_q: &list_head for pending commands.
1304 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 *
1306 * Decription:
1307 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -05001308 * running command. This only works out to be the case if we have one
1309 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 * no sense to try and abort the command, since as far as the shost
1311 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001312 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313static int scsi_eh_abort_cmds(struct list_head *work_q,
1314 struct list_head *done_q)
1315{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001316 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001317 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 int rtn;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001319 struct Scsi_Host *shost;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001321 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001322 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 continue;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001324 shost = scmd->device->host;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001325 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001326 list_splice_init(&check_list, work_q);
1327 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001328 scmd_printk(KERN_INFO, scmd,
1329 "%s: skip aborting cmd, past eh deadline\n",
1330 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001331 return list_empty(work_q);
1332 }
Hannes Reinecke91921e02014-06-25 16:39:59 +02001333 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001334 scmd_printk(KERN_INFO, scmd,
1335 "%s: aborting cmd\n", current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001336 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
Hannes Reinecke6fd046f2013-10-23 10:51:22 +02001337 if (rtn == FAILED) {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001338 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001339 scmd_printk(KERN_INFO, scmd,
1340 "%s: aborting cmd failed\n",
1341 current->comm));
Hannes Reinecke6fd046f2013-10-23 10:51:22 +02001342 list_splice_init(&check_list, work_q);
1343 return list_empty(work_q);
1344 }
1345 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1346 if (rtn == FAST_IO_FAIL)
1347 scsi_eh_finish_cmd(scmd, done_q);
1348 else
1349 list_move_tail(&scmd->eh_entry, &check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 }
1351
David Jeffery3eef6252011-05-19 14:41:12 -04001352 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353}
1354
1355/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001357 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 *
1359 * Return value:
1360 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001361 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1363{
1364 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
Christoph Hellwig631c2282006-07-08 20:42:15 +02001366 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -05001367 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Brian Kinged773e62007-03-29 15:25:52 -05001369 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -06001370 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -05001371
Christoph Hellwig631c2282006-07-08 20:42:15 +02001372 if (rtn == SUCCESS)
1373 return 0;
1374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 return 1;
1377}
1378
1379 /**
1380 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001381 * @shost: &scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001382 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05001383 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 *
1385 * Notes:
1386 * If commands are failing due to not ready, initializing command required,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001387 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001388 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389static int scsi_eh_stu(struct Scsi_Host *shost,
1390 struct list_head *work_q,
1391 struct list_head *done_q)
1392{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001393 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 struct scsi_device *sdev;
1395
1396 shost_for_each_device(sdev, shost) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001397 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001398 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001399 sdev_printk(KERN_INFO, sdev,
1400 "%s: skip START_UNIT, past eh deadline\n",
1401 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001402 break;
1403 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 stu_scmd = NULL;
1405 list_for_each_entry(scmd, work_q, eh_entry)
1406 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1407 scsi_check_sense(scmd) == FAILED ) {
1408 stu_scmd = scmd;
1409 break;
1410 }
1411
1412 if (!stu_scmd)
1413 continue;
1414
Hannes Reinecke91921e02014-06-25 16:39:59 +02001415 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001416 sdev_printk(KERN_INFO, sdev,
1417 "%s: Sending START_UNIT\n",
1418 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
1420 if (!scsi_eh_try_stu(stu_scmd)) {
1421 if (!scsi_device_online(sdev) ||
1422 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001423 list_for_each_entry_safe(scmd, next,
1424 work_q, eh_entry) {
James Bottomley24510792013-11-11 13:44:53 +01001425 if (scmd->device == sdev &&
1426 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 scsi_eh_finish_cmd(scmd, done_q);
1428 }
1429 }
1430 } else {
1431 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001432 sdev_printk(KERN_INFO, sdev,
1433 "%s: START_UNIT failed\n",
1434 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 }
1436 }
1437
1438 return list_empty(work_q);
1439}
1440
1441
1442/**
1443 * scsi_eh_bus_device_reset - send bdr if needed
1444 * @shost: scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001445 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05001446 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 *
1448 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001449 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 * devices that are jammed or not - if we have multiple devices, it
1451 * makes no sense to try bus_device_reset - we really would need to try
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001452 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001453 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1455 struct list_head *work_q,
1456 struct list_head *done_q)
1457{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001458 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 struct scsi_device *sdev;
1460 int rtn;
1461
1462 shost_for_each_device(sdev, shost) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001463 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001464 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001465 sdev_printk(KERN_INFO, sdev,
1466 "%s: skip BDR, past eh deadline\n",
1467 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001468 break;
1469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 bdr_scmd = NULL;
1471 list_for_each_entry(scmd, work_q, eh_entry)
1472 if (scmd->device == sdev) {
1473 bdr_scmd = scmd;
1474 break;
1475 }
1476
1477 if (!bdr_scmd)
1478 continue;
1479
Hannes Reinecke91921e02014-06-25 16:39:59 +02001480 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001481 sdev_printk(KERN_INFO, sdev,
1482 "%s: Sending BDR\n", current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 rtn = scsi_try_bus_device_reset(bdr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001484 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 if (!scsi_device_online(sdev) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001486 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001488 list_for_each_entry_safe(scmd, next,
1489 work_q, eh_entry) {
James Bottomley24510792013-11-11 13:44:53 +01001490 if (scmd->device == sdev &&
1491 scsi_eh_action(scmd, rtn) != FAILED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 scsi_eh_finish_cmd(scmd,
1493 done_q);
1494 }
1495 }
1496 } else {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001497 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001498 sdev_printk(KERN_INFO, sdev,
1499 "%s: BDR failed\n", current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 }
1501 }
1502
1503 return list_empty(work_q);
1504}
1505
1506/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001507 * scsi_eh_target_reset - send target reset if needed
1508 * @shost: scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001509 * @work_q: &list_head for pending commands.
Mike Christie30bd7df2008-02-29 18:25:19 -06001510 * @done_q: &list_head for processed commands.
1511 *
1512 * Notes:
1513 * Try a target reset.
1514 */
1515static int scsi_eh_target_reset(struct Scsi_Host *shost,
1516 struct list_head *work_q,
1517 struct list_head *done_q)
1518{
James Bottomley98db5192010-10-25 15:53:41 -05001519 LIST_HEAD(tmp_list);
David Jeffery3eef6252011-05-19 14:41:12 -04001520 LIST_HEAD(check_list);
Mike Christie30bd7df2008-02-29 18:25:19 -06001521
James Bottomley98db5192010-10-25 15:53:41 -05001522 list_splice_init(work_q, &tmp_list);
1523
1524 while (!list_empty(&tmp_list)) {
1525 struct scsi_cmnd *next, *scmd;
1526 int rtn;
1527 unsigned int id;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001528
Hannes Reineckeb4562022013-10-23 10:51:21 +02001529 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001530 /* push back on work queue for further processing */
1531 list_splice_init(&check_list, work_q);
1532 list_splice_init(&tmp_list, work_q);
1533 SCSI_LOG_ERROR_RECOVERY(3,
1534 shost_printk(KERN_INFO, shost,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001535 "%s: Skip target reset, past eh deadline\n",
1536 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001537 return list_empty(work_q);
1538 }
James Bottomley98db5192010-10-25 15:53:41 -05001539
1540 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1541 id = scmd_id(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -06001542
Hannes Reinecke91921e02014-06-25 16:39:59 +02001543 SCSI_LOG_ERROR_RECOVERY(3,
1544 shost_printk(KERN_INFO, shost,
1545 "%s: Sending target reset to target %d\n",
1546 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001547 rtn = scsi_try_target_reset(scmd);
1548 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
Hannes Reinecke91921e02014-06-25 16:39:59 +02001549 SCSI_LOG_ERROR_RECOVERY(3,
1550 shost_printk(KERN_INFO, shost,
1551 "%s: Target reset failed"
1552 " target: %d\n",
1553 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001554 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1555 if (scmd_id(scmd) != id)
1556 continue;
1557
David Jeffery3eef6252011-05-19 14:41:12 -04001558 if (rtn == SUCCESS)
1559 list_move_tail(&scmd->eh_entry, &check_list);
1560 else if (rtn == FAST_IO_FAIL)
James Bottomley98db5192010-10-25 15:53:41 -05001561 scsi_eh_finish_cmd(scmd, done_q);
1562 else
1563 /* push back on work queue for further processing */
1564 list_move(&scmd->eh_entry, work_q);
1565 }
1566 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001567
David Jeffery3eef6252011-05-19 14:41:12 -04001568 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001569}
1570
1571/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001572 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001573 * @shost: &scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001574 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05001575 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001576 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1578 struct list_head *work_q,
1579 struct list_head *done_q)
1580{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001581 struct scsi_cmnd *scmd, *chan_scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001582 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 unsigned int channel;
1584 int rtn;
1585
1586 /*
1587 * we really want to loop over the various channels, and do this on
1588 * a channel by channel basis. we should also check to see if any
1589 * of the failed commands are on soft_reset devices, and if so, skip
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001590 * the reset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 */
1592
1593 for (channel = 0; channel <= shost->max_channel; channel++) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001594 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001595 list_splice_init(&check_list, work_q);
1596 SCSI_LOG_ERROR_RECOVERY(3,
1597 shost_printk(KERN_INFO, shost,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001598 "%s: skip BRST, past eh deadline\n",
1599 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001600 return list_empty(work_q);
1601 }
Hannes Reineckeb4562022013-10-23 10:51:21 +02001602
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 chan_scmd = NULL;
1604 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001605 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 chan_scmd = scmd;
1607 break;
1608 /*
1609 * FIXME add back in some support for
1610 * soft_reset devices.
1611 */
1612 }
1613 }
1614
1615 if (!chan_scmd)
1616 continue;
Hannes Reinecke91921e02014-06-25 16:39:59 +02001617 SCSI_LOG_ERROR_RECOVERY(3,
1618 shost_printk(KERN_INFO, shost,
1619 "%s: Sending BRST chan: %d\n",
1620 current->comm, channel));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 rtn = scsi_try_bus_reset(chan_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001622 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001623 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
David Jeffery3eef6252011-05-19 14:41:12 -04001624 if (channel == scmd_channel(scmd)) {
1625 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 scsi_eh_finish_cmd(scmd,
1627 done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001628 else
1629 list_move_tail(&scmd->eh_entry,
1630 &check_list);
1631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 }
1633 } else {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001634 SCSI_LOG_ERROR_RECOVERY(3,
1635 shost_printk(KERN_INFO, shost,
1636 "%s: BRST failed chan: %d\n",
1637 current->comm, channel));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
1639 }
David Jeffery3eef6252011-05-19 14:41:12 -04001640 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641}
1642
1643/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001644 * scsi_eh_host_reset - send a host reset
Randy Dunlap74cf2982014-08-16 14:15:11 -07001645 * @shost: host to be reset.
1646 * @work_q: &list_head for pending commands.
1647 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001648 */
Hannes Reinecke91921e02014-06-25 16:39:59 +02001649static int scsi_eh_host_reset(struct Scsi_Host *shost,
1650 struct list_head *work_q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 struct list_head *done_q)
1652{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001653 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001654 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
1657 if (!list_empty(work_q)) {
1658 scmd = list_entry(work_q->next,
1659 struct scsi_cmnd, eh_entry);
1660
Hannes Reinecke91921e02014-06-25 16:39:59 +02001661 SCSI_LOG_ERROR_RECOVERY(3,
1662 shost_printk(KERN_INFO, shost,
1663 "%s: Sending HRST\n",
1664 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
1666 rtn = scsi_try_host_reset(scmd);
David Jeffery3eef6252011-05-19 14:41:12 -04001667 if (rtn == SUCCESS) {
1668 list_splice_init(work_q, &check_list);
1669 } else if (rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001670 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 scsi_eh_finish_cmd(scmd, done_q);
1672 }
1673 } else {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001674 SCSI_LOG_ERROR_RECOVERY(3,
1675 shost_printk(KERN_INFO, shost,
1676 "%s: HRST failed\n",
1677 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 }
1679 }
David Jeffery3eef6252011-05-19 14:41:12 -04001680 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681}
1682
1683/**
1684 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
Randy Dunlap74cf2982014-08-16 14:15:11 -07001685 * @work_q: &list_head for pending commands.
1686 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001687 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688static void scsi_eh_offline_sdevs(struct list_head *work_q,
1689 struct list_head *done_q)
1690{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001691 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001693 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001694 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1695 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001697 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 /*
1699 * FIXME: Handle lost cmds.
1700 */
1701 }
1702 scsi_eh_finish_cmd(scmd, done_q);
1703 }
1704 return;
1705}
1706
1707/**
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001708 * scsi_noretry_cmd - determine if command should be failed fast
Mike Christie4a274462008-08-19 18:45:31 -05001709 * @scmd: SCSI cmd to examine.
1710 */
1711int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1712{
1713 switch (host_byte(scmd->result)) {
1714 case DID_OK:
1715 break;
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001716 case DID_TIME_OUT:
1717 goto check_type;
Mike Christie4a274462008-08-19 18:45:31 -05001718 case DID_BUS_BUSY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001719 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
Mike Christie4a274462008-08-19 18:45:31 -05001720 case DID_PARITY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001721 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
Mike Christie4a274462008-08-19 18:45:31 -05001722 case DID_ERROR:
1723 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1724 status_byte(scmd->result) == RESERVATION_CONFLICT)
1725 return 0;
1726 /* fall through */
1727 case DID_SOFT_ERROR:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001728 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
Mike Christie4a274462008-08-19 18:45:31 -05001729 }
1730
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001731 if (status_byte(scmd->result) != CHECK_CONDITION)
1732 return 0;
Mike Christie4a274462008-08-19 18:45:31 -05001733
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001734check_type:
1735 /*
1736 * assume caller has checked sense and determined
1737 * the check condition was retryable.
1738 */
1739 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1740 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1741 return 1;
1742 else
1743 return 0;
Mike Christie4a274462008-08-19 18:45:31 -05001744}
1745
1746/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1748 * @scmd: SCSI cmd to examine.
1749 *
1750 * Notes:
1751 * This is *only* called when we are examining the status after sending
1752 * out the actual data command. any commands that are queued for error
1753 * recovery (e.g. test_unit_ready) do *not* come through here.
1754 *
1755 * When this routine returns failed, it means the error handler thread
1756 * is woken. In cases where the error code indicates an error that
1757 * doesn't require the error handler read (i.e. we don't need to
1758 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001759 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760int scsi_decide_disposition(struct scsi_cmnd *scmd)
1761{
1762 int rtn;
1763
1764 /*
1765 * if the device is offline, then we clearly just pass the result back
1766 * up to the top level.
1767 */
1768 if (!scsi_device_online(scmd->device)) {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001769 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1770 "%s: device offline - report as SUCCESS\n", __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 return SUCCESS;
1772 }
1773
1774 /*
1775 * first check the host byte, to see if there is anything in there
1776 * that would indicate what we need to do.
1777 */
1778 switch (host_byte(scmd->result)) {
1779 case DID_PASSTHROUGH:
1780 /*
1781 * no matter what, pass this through to the upper layer.
1782 * nuke this special code so that it looks like we are saying
1783 * did_ok.
1784 */
1785 scmd->result &= 0xff00ffff;
1786 return SUCCESS;
1787 case DID_OK:
1788 /*
1789 * looks good. drop through, and check the next byte.
1790 */
1791 break;
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001792 case DID_ABORT:
1793 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
Ulrich Obergfell8922a902014-06-04 13:34:57 +02001794 set_host_byte(scmd, DID_TIME_OUT);
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001795 return SUCCESS;
1796 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 case DID_NO_CONNECT:
1798 case DID_BAD_TARGET:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 /*
1800 * note - this means that we just report the status back
1801 * to the top level driver, not that we actually think
1802 * that it indicates SUCCESS.
1803 */
1804 return SUCCESS;
1805 /*
1806 * when the low level driver returns did_soft_error,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001807 * it is responsible for keeping an internal retry counter
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 * in order to avoid endless loops (db)
1809 *
1810 * actually this is a bug in this function here. we should
1811 * be mindful of the maximum number of retries specified
1812 * and not get stuck in a loop.
1813 */
1814 case DID_SOFT_ERROR:
1815 goto maybe_retry;
1816 case DID_IMM_RETRY:
1817 return NEEDS_RETRY;
1818
bf341912005-04-12 17:49:09 -05001819 case DID_REQUEUE:
1820 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001821 case DID_TRANSPORT_DISRUPTED:
1822 /*
1823 * LLD/transport was disrupted during processing of the IO.
1824 * The transport class is now blocked/blocking,
1825 * and the transport will decide what to do with the IO
Mike Christie939c2282008-11-04 19:47:19 -06001826 * based on its timers and recovery capablilities if
1827 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001828 */
Mike Christie939c2282008-11-04 19:47:19 -06001829 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001830 case DID_TRANSPORT_FAILFAST:
1831 /*
1832 * The transport decided to failfast the IO (most likely
1833 * the fast io fail tmo fired), so send IO directly upwards.
1834 */
1835 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 case DID_ERROR:
1837 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1838 status_byte(scmd->result) == RESERVATION_CONFLICT)
1839 /*
1840 * execute reservation conflict processing code
1841 * lower down
1842 */
1843 break;
1844 /* fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 case DID_BUS_BUSY:
1846 case DID_PARITY:
1847 goto maybe_retry;
1848 case DID_TIME_OUT:
1849 /*
1850 * when we scan the bus, we get timeout messages for
1851 * these commands if there is no device available.
1852 * other hosts report did_no_connect for the same thing.
1853 */
1854 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1855 scmd->cmnd[0] == INQUIRY)) {
1856 return SUCCESS;
1857 } else {
1858 return FAILED;
1859 }
1860 case DID_RESET:
1861 return SUCCESS;
1862 default:
1863 return FAILED;
1864 }
1865
1866 /*
1867 * next, check the message byte.
1868 */
1869 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1870 return FAILED;
1871
1872 /*
1873 * check the status byte to see if this indicates anything special.
1874 */
1875 switch (status_byte(scmd->result)) {
1876 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001877 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 /*
1879 * the case of trying to send too many commands to a
1880 * tagged queueing device.
1881 */
1882 case BUSY:
1883 /*
1884 * device can't talk to us at the moment. Should only
1885 * occur (SAM-3) when the task queue is empty, so will cause
1886 * the empty queue handling to trigger a stall in the
1887 * device.
1888 */
1889 return ADD_TO_MLQUEUE;
1890 case GOOD:
Ewan D. Milne279afdf2013-08-08 15:07:48 -04001891 if (scmd->cmnd[0] == REPORT_LUNS)
1892 scmd->device->sdev_target->expecting_lun_change = 0;
Vasu Dev4a840672009-10-22 15:46:33 -07001893 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001896 case TASK_ABORTED:
1897 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 case CHECK_CONDITION:
1899 rtn = scsi_check_sense(scmd);
1900 if (rtn == NEEDS_RETRY)
1901 goto maybe_retry;
1902 /* if rtn == FAILED, we have no sense information;
1903 * returning FAILED will wake the error handler thread
1904 * to collect the sense and redo the decide
1905 * disposition */
1906 return rtn;
1907 case CONDITION_GOOD:
1908 case INTERMEDIATE_GOOD:
1909 case INTERMEDIATE_C_GOOD:
1910 case ACA_ACTIVE:
1911 /*
1912 * who knows? FIXME(eric)
1913 */
1914 return SUCCESS;
1915
1916 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001917 sdev_printk(KERN_INFO, scmd->device,
1918 "reservation conflict\n");
Moger, Babu2082ebc2012-01-24 20:38:46 +00001919 set_host_byte(scmd, DID_NEXUS_FAILURE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 return SUCCESS; /* causes immediate i/o error */
1921 default:
1922 return FAILED;
1923 }
1924 return FAILED;
1925
1926 maybe_retry:
1927
1928 /* we requeue for retry because the error was retryable, and
1929 * the request was not marked fast fail. Note that above,
1930 * even if the request is marked fast fail, we still requeue
1931 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001932 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001933 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 return NEEDS_RETRY;
1935 } else {
1936 /*
1937 * no more retries - report this one back to upper level.
1938 */
1939 return SUCCESS;
1940 }
1941}
1942
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001943static void eh_lock_door_done(struct request *req, int uptodate)
1944{
1945 __blk_put_request(req->q, req);
1946}
1947
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 * scsi_eh_lock_door - Prevent medium removal for the specified device
1950 * @sdev: SCSI device to prevent medium removal
1951 *
1952 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001953 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 *
1955 * Notes:
1956 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1957 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001958 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959static void scsi_eh_lock_door(struct scsi_device *sdev)
1960{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001961 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
James Bottomley91bc31f2009-05-17 09:30:48 -05001963 /*
1964 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1965 * request becomes available
1966 */
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001967 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
Joe Lawrencea492f072014-08-28 08:15:21 -06001968 if (IS_ERR(req))
Joe Lawrenceeb571ee2014-07-02 15:35:16 -04001969 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
Jens Axboef27b0872014-06-06 07:57:37 -06001971 blk_rq_set_block_pc(req);
1972
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001973 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1974 req->cmd[1] = 0;
1975 req->cmd[2] = 0;
1976 req->cmd[3] = 0;
1977 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1978 req->cmd[5] = 0;
1979
1980 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1981
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001982 req->cmd_flags |= REQ_QUIET;
1983 req->timeout = 10 * HZ;
1984 req->retries = 5;
1985
1986 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987}
1988
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989/**
1990 * scsi_restart_operations - restart io operations to the specified host.
1991 * @shost: Host we are restarting.
1992 *
1993 * Notes:
1994 * When we entered the error handler, we blocked all further i/o to
1995 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001996 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997static void scsi_restart_operations(struct Scsi_Host *shost)
1998{
1999 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05002000 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001
2002 /*
2003 * If the door was locked, we need to insert a door lock request
2004 * onto the head of the SCSI request queue for the device. There
2005 * is no point trying to lock the door of an off-line device.
2006 */
2007 shost_for_each_device(sdev, shost) {
Christoph Hellwig48379272014-11-03 19:36:40 +01002008 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 scsi_eh_lock_door(sdev);
Christoph Hellwig48379272014-11-03 19:36:40 +01002010 sdev->was_reset = 0;
2011 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 }
2013
2014 /*
2015 * next free up anything directly waiting upon the host. this
2016 * will be requests for character device operations, and also for
2017 * ioctls to queued block devices.
2018 */
Hannes Reineckeb4562022013-10-23 10:51:21 +02002019 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002020 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
James Bottomley939647e2005-09-18 15:05:20 -05002022 spin_lock_irqsave(shost->host_lock, flags);
2023 if (scsi_host_set_state(shost, SHOST_RUNNING))
2024 if (scsi_host_set_state(shost, SHOST_CANCEL))
2025 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2026 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
2028 wake_up(&shost->host_wait);
2029
2030 /*
2031 * finally we need to re-initiate requests that may be pending. we will
2032 * have had everything blocked while error handling is taking place, and
2033 * now that error recovery is done, we will need to ensure that these
2034 * requests are started.
2035 */
2036 scsi_run_host_queues(shost);
Dan Williams57fc2e32012-06-21 23:25:32 -07002037
2038 /*
2039 * if eh is active and host_eh_scheduled is pending we need to re-run
2040 * recovery. we do this check after scsi_run_host_queues() to allow
2041 * everything pent up since the last eh run a chance to make forward
2042 * progress before we sync again. Either we'll immediately re-run
2043 * recovery or scsi_device_unbusy() will wake us again when these
2044 * pending commands complete.
2045 */
2046 spin_lock_irqsave(shost->host_lock, flags);
2047 if (shost->host_eh_scheduled)
2048 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2049 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2050 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051}
2052
2053/**
2054 * scsi_eh_ready_devs - check device ready state and recover if not.
Randy Dunlap74cf2982014-08-16 14:15:11 -07002055 * @shost: host to be recovered.
2056 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05002057 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002058 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08002059void scsi_eh_ready_devs(struct Scsi_Host *shost,
2060 struct list_head *work_q,
2061 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062{
2063 if (!scsi_eh_stu(shost, work_q, done_q))
2064 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06002065 if (!scsi_eh_target_reset(shost, work_q, done_q))
2066 if (!scsi_eh_bus_reset(shost, work_q, done_q))
Hannes Reinecke91921e02014-06-25 16:39:59 +02002067 if (!scsi_eh_host_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06002068 scsi_eh_offline_sdevs(work_q,
2069 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08002071EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
2073/**
2074 * scsi_eh_flush_done_q - finish processed commands or retry them.
2075 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002076 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09002077void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02002079 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02002081 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2082 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05002084 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06002085 (++scmd->retries <= scmd->allowed)) {
Hannes Reinecke91921e02014-06-25 16:39:59 +02002086 SCSI_LOG_ERROR_RECOVERY(3,
2087 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01002088 "%s: flush retry cmd\n",
2089 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2091 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07002092 /*
2093 * If just we got sense for the device (called
2094 * scsi_eh_get_sense), scmd->result is already
2095 * set, do not set DRIVER_TIMEOUT.
2096 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 if (!scmd->result)
2098 scmd->result |= (DRIVER_TIMEOUT << 24);
Hannes Reinecke91921e02014-06-25 16:39:59 +02002099 SCSI_LOG_ERROR_RECOVERY(3,
2100 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01002101 "%s: flush finish cmd\n",
2102 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 scsi_finish_command(scmd);
2104 }
2105 }
2106}
Tejun Heo041c5fc2006-01-23 13:09:36 +09002107EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
2109/**
2110 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2111 * @shost: Host to unjam.
2112 *
2113 * Notes:
2114 * When we come in here, we *know* that all commands on the bus have
2115 * either completed, failed or timed out. we also know that no further
2116 * commands are being sent to the host, so things are relatively quiet
2117 * and we have freedom to fiddle with things as we wish.
2118 *
2119 * This is only the *default* implementation. it is possible for
2120 * individual drivers to supply their own version of this function, and
2121 * if the maintainer wishes to do this, it is strongly suggested that
2122 * this function be taken as a template and modified. this function
2123 * was designed to correctly handle problems for about 95% of the
2124 * different cases out there, and it should always provide at least a
2125 * reasonable amount of error recovery.
2126 *
2127 * Any command marked 'failed' or 'timeout' must eventually have
2128 * scsi_finish_cmd() called for it. we do all of the retry stuff
2129 * here, so when we restart the host after we return it should have an
2130 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002131 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132static void scsi_unjam_host(struct Scsi_Host *shost)
2133{
2134 unsigned long flags;
2135 LIST_HEAD(eh_work_q);
2136 LIST_HEAD(eh_done_q);
2137
2138 spin_lock_irqsave(shost->host_lock, flags);
2139 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2140 spin_unlock_irqrestore(shost->host_lock, flags);
2141
2142 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2143
2144 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2145 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2146 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2147
Hannes Reineckeb4562022013-10-23 10:51:21 +02002148 spin_lock_irqsave(shost->host_lock, flags);
Ren Mingxinbb3b6212013-11-11 13:44:56 +01002149 if (shost->eh_deadline != -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +02002150 shost->last_reset = 0;
2151 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 scsi_eh_flush_done_q(&eh_done_q);
2153}
2154
2155/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002156 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 * @data: Host for which we are running.
2158 *
2159 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002160 * This is the main error handling loop. This is run as a kernel thread
2161 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002162 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163int scsi_error_handler(void *data)
2164{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002165 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002168 * We use TASK_INTERRUPTIBLE so that the thread is not
2169 * counted against the load average as a running process.
2170 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01002171 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 */
James Bottomley3ed7a472005-09-19 09:50:04 -05002173 while (!kthread_should_stop()) {
Dan Williamsb9d5c6b2012-06-21 23:25:42 -07002174 set_current_state(TASK_INTERRUPTIBLE);
Tejun Heoee7863b2006-05-15 20:57:20 +09002175 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
Christoph Hellwig74665012014-01-22 15:29:29 +01002176 shost->host_failed != atomic_read(&shost->host_busy)) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002177 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002178 shost_printk(KERN_INFO, shost,
2179 "scsi_eh_%d: sleeping\n",
2180 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05002181 schedule();
James Bottomley3ed7a472005-09-19 09:50:04 -05002182 continue;
2183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
James Bottomley3ed7a472005-09-19 09:50:04 -05002185 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002186 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002187 shost_printk(KERN_INFO, shost,
2188 "scsi_eh_%d: waking up %d/%d/%d\n",
2189 shost->host_no, shost->host_eh_scheduled,
Christoph Hellwig74665012014-01-22 15:29:29 +01002190 shost->host_failed,
2191 atomic_read(&shost->host_busy)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 /*
2194 * We have a host that is failing for some reason. Figure out
2195 * what we need to do to get it up and online again (if we can).
2196 * If we fail, we end up taking the thing offline.
2197 */
Lin Mingae0751f2011-12-05 09:20:24 +08002198 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
Alan Sternbc4f2402010-06-17 10:41:42 -04002199 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02002200 shost_printk(KERN_ERR, shost,
2201 "scsi_eh_%d: unable to autoresume\n",
2202 shost->host_no));
Alan Sternbc4f2402010-06-17 10:41:42 -04002203 continue;
2204 }
2205
Christoph Hellwig9227c332006-04-01 19:21:04 +02002206 if (shost->transportt->eh_strategy_handler)
2207 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 else
2209 scsi_unjam_host(shost);
2210
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 /*
2212 * Note - if the above fails completely, the action is to take
2213 * individual devices offline and flush the queue of any
2214 * outstanding requests that may have been pending. When we
2215 * restart, we restart any I/O to any other devices on the bus
2216 * which are still online.
2217 */
2218 scsi_restart_operations(shost);
Lin Mingae0751f2011-12-05 09:20:24 +08002219 if (!shost->eh_noresume)
2220 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04002222 __set_current_state(TASK_RUNNING);
2223
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002224 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002225 shost_printk(KERN_INFO, shost,
2226 "Error handler scsi_eh_%d exiting\n",
2227 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05002228 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 return 0;
2230}
2231
2232/*
2233 * Function: scsi_report_bus_reset()
2234 *
2235 * Purpose: Utility function used by low-level drivers to report that
2236 * they have observed a bus reset on the bus being handled.
2237 *
2238 * Arguments: shost - Host in question
2239 * channel - channel on which reset was observed.
2240 *
2241 * Returns: Nothing
2242 *
2243 * Lock status: Host lock must be held.
2244 *
2245 * Notes: This only needs to be called if the reset is one which
2246 * originates from an unknown location. Resets originated
2247 * by the mid-level itself don't need to call this, but there
2248 * should be no harm.
2249 *
2250 * The main purpose of this is to make sure that a CHECK_CONDITION
2251 * is properly treated.
2252 */
2253void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2254{
2255 struct scsi_device *sdev;
2256
2257 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06002258 if (channel == sdev_channel(sdev))
2259 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 }
2261}
2262EXPORT_SYMBOL(scsi_report_bus_reset);
2263
2264/*
2265 * Function: scsi_report_device_reset()
2266 *
2267 * Purpose: Utility function used by low-level drivers to report that
2268 * they have observed a device reset on the device being handled.
2269 *
2270 * Arguments: shost - Host in question
2271 * channel - channel on which reset was observed
2272 * target - target on which reset was observed
2273 *
2274 * Returns: Nothing
2275 *
2276 * Lock status: Host lock must be held
2277 *
2278 * Notes: This only needs to be called if the reset is one which
2279 * originates from an unknown location. Resets originated
2280 * by the mid-level itself don't need to call this, but there
2281 * should be no harm.
2282 *
2283 * The main purpose of this is to make sure that a CHECK_CONDITION
2284 * is properly treated.
2285 */
2286void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2287{
2288 struct scsi_device *sdev;
2289
2290 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04002291 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06002292 target == sdev_id(sdev))
2293 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 }
2295}
2296EXPORT_SYMBOL(scsi_report_device_reset);
2297
2298static void
2299scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2300{
2301}
2302
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002303/**
2304 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2305 * @dev: scsi_device to operate on
2306 * @arg: reset type (see sg.h)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 */
2308int
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002309scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310{
Alan Sternbc4f2402010-06-17 10:41:42 -04002311 struct scsi_cmnd *scmd;
James Smartd7a1bb02006-03-08 14:50:12 -05002312 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05002314 unsigned long flags;
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002315 int error = 0, rtn, val;
2316
2317 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2318 return -EACCES;
2319
2320 error = get_user(val, arg);
2321 if (error)
2322 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323
Alan Sternbc4f2402010-06-17 10:41:42 -04002324 if (scsi_autopm_get_host(shost) < 0)
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002325 return -EIO;
Alan Sternbc4f2402010-06-17 10:41:42 -04002326
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002327 error = -EIO;
Alan Sternbc4f2402010-06-17 10:41:42 -04002328 scmd = scsi_get_command(dev, GFP_KERNEL);
Christoph Hellwig0f121dd2014-09-05 18:00:05 -07002329 if (!scmd)
Christoph Hellwig95eeb5f2014-05-01 16:51:05 +02002330 goto out_put_autopm_host;
Christoph Hellwig95eeb5f2014-05-01 16:51:05 +02002331
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02002332 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02002334
Boaz Harrosh64a87b22008-04-30 11:19:47 +03002335 scmd->cmnd = req.cmd;
2336
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02002338 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
2340 scmd->cmd_len = 0;
2341
2342 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
James Smartd7a1bb02006-03-08 14:50:12 -05002344 spin_lock_irqsave(shost->host_lock, flags);
2345 shost->tmf_in_progress = 1;
2346 spin_unlock_irqrestore(shost->host_lock, flags);
2347
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002348 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2349 case SG_SCSI_RESET_NOTHING:
2350 rtn = SUCCESS;
2351 break;
2352 case SG_SCSI_RESET_DEVICE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 rtn = scsi_try_bus_device_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002354 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 break;
2356 /* FALLTHROUGH */
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002357 case SG_SCSI_RESET_TARGET:
Mike Christie30bd7df2008-02-29 18:25:19 -06002358 rtn = scsi_try_target_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002359 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
Mike Christie30bd7df2008-02-29 18:25:19 -06002360 break;
2361 /* FALLTHROUGH */
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002362 case SG_SCSI_RESET_BUS:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 rtn = scsi_try_bus_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002364 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 break;
2366 /* FALLTHROUGH */
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002367 case SG_SCSI_RESET_HOST:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 rtn = scsi_try_host_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002369 if (rtn == SUCCESS)
2370 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 default:
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002372 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 rtn = FAILED;
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002374 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 }
2376
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002377 error = (rtn == SUCCESS) ? 0 : -EIO;
2378
James Smartd7a1bb02006-03-08 14:50:12 -05002379 spin_lock_irqsave(shost->host_lock, flags);
2380 shost->tmf_in_progress = 0;
2381 spin_unlock_irqrestore(shost->host_lock, flags);
2382
2383 /*
2384 * be sure to wake up anyone who was sleeping or had their queue
2385 * suspended while we performed the TMF.
2386 */
2387 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002388 shost_printk(KERN_INFO, shost,
2389 "waking up host to restart after TMF\n"));
James Smartd7a1bb02006-03-08 14:50:12 -05002390
2391 wake_up(&shost->host_wait);
James Smartd7a1bb02006-03-08 14:50:12 -05002392 scsi_run_host_queues(shost);
2393
Christoph Hellwig0f121dd2014-09-05 18:00:05 -07002394 scsi_put_command(scmd);
2395
Christoph Hellwig04796332014-02-20 14:20:55 -08002396out_put_autopm_host:
Alan Sternbc4f2402010-06-17 10:41:42 -04002397 scsi_autopm_put_host(shost);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002398 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399}
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002400EXPORT_SYMBOL(scsi_ioctl_reset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401
Hannes Reinecke4753cbc2014-10-24 14:26:52 +02002402bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2403 struct scsi_sense_hdr *sshdr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404{
2405 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09002406 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407}
2408EXPORT_SYMBOL(scsi_command_normalize_sense);
2409
2410/**
Rob Landleyeb448202007-11-03 13:30:39 -05002411 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 * @sense_buffer: byte array of descriptor format sense data
2413 * @sb_len: number of valid bytes in sense_buffer
2414 * @desc_type: value of descriptor type to find
2415 * (e.g. 0 -> information)
2416 *
2417 * Notes:
2418 * only valid when sense data is in descriptor format
2419 *
2420 * Return value:
2421 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002422 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2424 int desc_type)
2425{
2426 int add_sen_len, add_len, desc_len, k;
2427 const u8 * descp;
2428
2429 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2430 return NULL;
2431 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2432 return NULL;
2433 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2434 add_sen_len : (sb_len - 8);
2435 descp = &sense_buffer[8];
2436 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2437 descp += desc_len;
2438 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2439 desc_len = add_len + 2;
2440 if (descp[0] == desc_type)
2441 return descp;
2442 if (add_len < 0) // short descriptor ??
2443 break;
2444 }
2445 return NULL;
2446}
2447EXPORT_SYMBOL(scsi_sense_desc_find);
2448
2449/**
Rob Landleyeb448202007-11-03 13:30:39 -05002450 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 * @sense_buffer: byte array of sense data
2452 * @sb_len: number of valid bytes in sense_buffer
2453 * @info_out: pointer to 64 integer where 8 or 4 byte information
2454 * field will be placed if found.
2455 *
2456 * Return value:
2457 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002458 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2460 u64 * info_out)
2461{
2462 int j;
2463 const u8 * ucp;
2464 u64 ull;
2465
2466 if (sb_len < 7)
2467 return 0;
2468 switch (sense_buffer[0] & 0x7f) {
2469 case 0x70:
2470 case 0x71:
2471 if (sense_buffer[0] & 0x80) {
2472 *info_out = (sense_buffer[3] << 24) +
2473 (sense_buffer[4] << 16) +
2474 (sense_buffer[5] << 8) + sense_buffer[6];
2475 return 1;
2476 } else
2477 return 0;
2478 case 0x72:
2479 case 0x73:
2480 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2481 0 /* info desc */);
2482 if (ucp && (0xa == ucp[1])) {
2483 ull = 0;
2484 for (j = 0; j < 8; ++j) {
2485 if (j > 0)
2486 ull <<= 8;
2487 ull |= ucp[4 + j];
2488 }
2489 *info_out = ull;
2490 return 1;
2491 } else
2492 return 0;
2493 default:
2494 return 0;
2495 }
2496}
2497EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002498
2499/**
2500 * scsi_build_sense_buffer - build sense data in a buffer
2501 * @desc: Sense format (non zero == descriptor format,
2502 * 0 == fixed format)
2503 * @buf: Where to build sense data
2504 * @key: Sense key
2505 * @asc: Additional sense code
2506 * @ascq: Additional sense code qualifier
2507 *
2508 **/
2509void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2510{
2511 if (desc) {
2512 buf[0] = 0x72; /* descriptor, current */
2513 buf[1] = key;
2514 buf[2] = asc;
2515 buf[3] = ascq;
2516 buf[7] = 0;
2517 } else {
2518 buf[0] = 0x70; /* fixed, current */
2519 buf[2] = key;
2520 buf[7] = 0xa;
2521 buf[12] = asc;
2522 buf[13] = ascq;
2523 }
2524}
2525EXPORT_SYMBOL(scsi_build_sense_buffer);
Hannes Reineckea1524f22015-03-27 16:46:37 +01002526
2527/**
2528 * scsi_set_sense_information - set the information field in a
2529 * formatted sense data buffer
2530 * @buf: Where to build sense data
2531 * @info: 64-bit information value to be set
2532 *
2533 **/
2534void scsi_set_sense_information(u8 *buf, u64 info)
2535{
2536 if ((buf[0] & 0x7f) == 0x72) {
2537 u8 *ucp, len;
2538
2539 len = buf[7];
2540 ucp = (char *)scsi_sense_desc_find(buf, len + 8, 0);
2541 if (!ucp) {
2542 buf[7] = len + 0xa;
2543 ucp = buf + 8 + len;
2544 }
2545 ucp[0] = 0;
2546 ucp[1] = 0xa;
2547 ucp[2] = 0x80; /* Valid bit */
2548 ucp[3] = 0;
2549 put_unaligned_be64(info, &ucp[4]);
2550 } else if ((buf[0] & 0x7f) == 0x70) {
2551 buf[0] |= 0x80;
2552 put_unaligned_be64(info, &buf[3]);
2553 }
2554}
2555EXPORT_SYMBOL(scsi_set_sense_information);