blob: 78b004da288518e2a2d9279eca7d32c2fe9bbc42 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30#include <scsi/scsi.h>
Christoph Hellwigbeb40482006-06-10 18:01:03 +020031#include <scsi/scsi_cmnd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <scsi/scsi_dbg.h>
33#include <scsi/scsi_device.h>
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -050034#include <scsi/scsi_driver.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <scsi/scsi_eh.h>
James Smartc829c392006-03-13 08:28:57 -050036#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <scsi/scsi_host.h>
38#include <scsi/scsi_ioctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40#include "scsi_priv.h"
41#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070042#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Kei Tokunagabf816232010-04-01 20:41:40 +090044#include <trace/events/scsi.h>
45
James Bottomley14216562012-07-25 23:55:55 +040046static void scsi_eh_done(struct scsi_cmnd *scmd);
47
Linus Torvalds1da177e2005-04-16 15:20:36 -070048/*
49 * These should *probably* be handled by the host itself.
50 * Since it is allowed to sleep, it probably should.
51 */
52#define BUS_RESET_SETTLE_TIME (10)
53#define HOST_RESET_SETTLE_TIME (10)
54
David Jeffery3eef6252011-05-19 14:41:12 -040055static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
Hannes Reineckee494f6a2013-11-11 13:44:54 +010056static int scsi_try_to_abort_cmd(struct scsi_host_template *,
57 struct scsi_cmnd *);
David Jeffery3eef6252011-05-19 14:41:12 -040058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059/* called with shost->host_lock held */
60void scsi_eh_wakeup(struct Scsi_Host *shost)
61{
62 if (shost->host_busy == shost->host_failed) {
Kei Tokunagabf816232010-04-01 20:41:40 +090063 trace_scsi_eh_wakeup(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -050064 wake_up_process(shost->ehandler);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 SCSI_LOG_ERROR_RECOVERY(5,
66 printk("Waking error handler thread\n"));
67 }
68}
Tejun Heof8bbfc22006-05-19 21:07:05 +090069
70/**
71 * scsi_schedule_eh - schedule EH for SCSI host
72 * @shost: SCSI host to invoke error handling on.
73 *
74 * Schedule SCSI EH without scmd.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080075 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090076void scsi_schedule_eh(struct Scsi_Host *shost)
77{
78 unsigned long flags;
79
80 spin_lock_irqsave(shost->host_lock, flags);
81
82 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
83 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
84 shost->host_eh_scheduled++;
85 scsi_eh_wakeup(shost);
86 }
87
88 spin_unlock_irqrestore(shost->host_lock, flags);
89}
90EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Hannes Reineckeb4562022013-10-23 10:51:21 +020092static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
93{
Ren Mingxinbb3b6212013-11-11 13:44:56 +010094 if (!shost->last_reset || shost->eh_deadline == -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +020095 return 0;
96
Hannes Reinecke76ad3e52013-11-11 13:44:55 +010097 /*
98 * 32bit accesses are guaranteed to be atomic
99 * (on all supported architectures), so instead
100 * of using a spinlock we can as well double check
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100101 * if eh_deadline has been set to 'off' during the
Hannes Reinecke76ad3e52013-11-11 13:44:55 +0100102 * time_before call.
103 */
104 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100105 shost->eh_deadline > -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200106 return 0;
107
108 return 1;
109}
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111/**
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100112 * scmd_eh_abort_handler - Handle command aborts
113 * @work: command to be aborted.
114 */
115void
116scmd_eh_abort_handler(struct work_struct *work)
117{
118 struct scsi_cmnd *scmd =
119 container_of(work, struct scsi_cmnd, abort_work.work);
120 struct scsi_device *sdev = scmd->device;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100121 int rtn;
122
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100123 if (scsi_host_eh_past_deadline(sdev->host)) {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100124 SCSI_LOG_ERROR_RECOVERY(3,
125 scmd_printk(KERN_INFO, scmd,
126 "scmd %p eh timeout, not aborting\n",
127 scmd));
128 } else {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100129 SCSI_LOG_ERROR_RECOVERY(3,
130 scmd_printk(KERN_INFO, scmd,
131 "aborting command %p\n", scmd));
132 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
133 if (rtn == SUCCESS) {
134 scmd->result |= DID_TIME_OUT << 16;
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100135 if (scsi_host_eh_past_deadline(sdev->host)) {
136 SCSI_LOG_ERROR_RECOVERY(3,
137 scmd_printk(KERN_INFO, scmd,
138 "scmd %p eh timeout, "
139 "not retrying aborted "
140 "command\n", scmd));
141 } 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,
145 "scmd %p retry "
146 "aborted command\n", scmd));
147 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100148 return;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100149 } else {
150 SCSI_LOG_ERROR_RECOVERY(3,
151 scmd_printk(KERN_WARNING, scmd,
152 "scmd %p finish "
153 "aborted command\n", scmd));
154 scsi_finish_command(scmd);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100155 return;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100156 }
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100157 } else {
158 SCSI_LOG_ERROR_RECOVERY(3,
159 scmd_printk(KERN_INFO, scmd,
160 "scmd %p abort failed, rtn %d\n",
161 scmd, rtn));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100162 }
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100163 }
164
165 if (!scsi_eh_scmd_add(scmd, 0)) {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_WARNING, scmd,
168 "scmd %p terminate "
169 "aborted command\n", scmd));
170 scmd->result |= DID_TIME_OUT << 16;
171 scsi_finish_command(scmd);
172 }
173}
174
175/**
176 * scsi_abort_command - schedule a command abort
177 * @scmd: scmd to abort.
178 *
179 * We only need to abort commands after a command timeout
180 */
181static int
182scsi_abort_command(struct scsi_cmnd *scmd)
183{
184 struct scsi_device *sdev = scmd->device;
185 struct Scsi_Host *shost = sdev->host;
186 unsigned long flags;
187
188 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
189 /*
190 * Retry after abort failed, escalate to next level.
191 */
192 SCSI_LOG_ERROR_RECOVERY(3,
193 scmd_printk(KERN_INFO, scmd,
194 "scmd %p previous abort failed\n", scmd));
195 cancel_delayed_work(&scmd->abort_work);
196 return FAILED;
197 }
198
199 /*
200 * Do not try a command abort if
201 * SCSI EH has already started.
202 */
203 spin_lock_irqsave(shost->host_lock, flags);
204 if (scsi_host_in_recovery(shost)) {
205 spin_unlock_irqrestore(shost->host_lock, flags);
206 SCSI_LOG_ERROR_RECOVERY(3,
207 scmd_printk(KERN_INFO, scmd,
208 "scmd %p not aborting, host in recovery\n",
209 scmd));
210 return FAILED;
211 }
212
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100213 if (shost->eh_deadline != -1 && !shost->last_reset)
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100214 shost->last_reset = jiffies;
215 spin_unlock_irqrestore(shost->host_lock, flags);
216
217 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
218 SCSI_LOG_ERROR_RECOVERY(3,
219 scmd_printk(KERN_INFO, scmd,
220 "scmd %p abort scheduled\n", scmd));
221 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
222 return SUCCESS;
223}
224
225/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 * scsi_eh_scmd_add - add scsi cmd to error handling.
227 * @scmd: scmd to run eh on.
228 * @eh_flag: optional SCSI_EH flag.
229 *
230 * Return value:
231 * 0 on failure.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800232 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
234{
235 struct Scsi_Host *shost = scmd->device->host;
236 unsigned long flags;
James Bottomley939647e2005-09-18 15:05:20 -0500237 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
James Bottomley3ed7a472005-09-19 09:50:04 -0500239 if (!shost->ehandler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 return 0;
241
242 spin_lock_irqsave(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500243 if (scsi_host_set_state(shost, SHOST_RECOVERY))
244 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
245 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100247 if (shost->eh_deadline != -1 && !shost->last_reset)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200248 shost->last_reset = jiffies;
249
James Bottomley939647e2005-09-18 15:05:20 -0500250 ret = 1;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100251 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
252 eh_flag &= ~SCSI_EH_CANCEL_CMD;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200253 scmd->eh_eflags |= eh_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 shost->host_failed++;
256 scsi_eh_wakeup(shost);
James Bottomley939647e2005-09-18 15:05:20 -0500257 out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 spin_unlock_irqrestore(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500259 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260}
261
262/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 * scsi_times_out - Timeout function for normal scsi commands.
Jens Axboe242f9dc2008-09-14 05:55:09 -0700264 * @req: request that is timing out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 *
266 * Notes:
267 * We do not need to lock this. There is the potential for a race
268 * only in that the normal completion handling might run, but if the
269 * normal completion function determines that the timer has already
270 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800271 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700272enum blk_eh_timer_return scsi_times_out(struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700274 struct scsi_cmnd *scmd = req->special;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700275 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100276 struct Scsi_Host *host = scmd->device->host;
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500277
Kei Tokunagabf816232010-04-01 20:41:40 +0900278 trace_scsi_dispatch_cmd_timeout(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 scsi_log_completion(scmd, TIMEOUT_ERROR);
280
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100281 if (host->eh_deadline != -1 && !host->last_reset)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200282 host->last_reset = jiffies;
283
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100284 if (host->transportt->eh_timed_out)
285 rtn = host->transportt->eh_timed_out(scmd);
286 else if (host->hostt->eh_timed_out)
287 rtn = host->hostt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500288
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100289 if (rtn == BLK_EH_NOT_HANDLED && !host->hostt->no_async_abort)
290 if (scsi_abort_command(scmd) == SUCCESS)
291 return BLK_EH_NOT_HANDLED;
292
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -0500293 scmd->result |= DID_TIME_OUT << 16;
294
Christoph Hellwigfa990782008-11-05 15:03:15 +0100295 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -0500296 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
Christoph Hellwigfa990782008-11-05 15:03:15 +0100297 rtn = BLK_EH_HANDLED;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700298
Christoph Hellwigfa990782008-11-05 15:03:15 +0100299 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
301
302/**
303 * scsi_block_when_processing_errors - Prevent cmds from being queued.
304 * @sdev: Device on which we are performing recovery.
305 *
306 * Description:
307 * We block until the host is out of error recovery, and then check to
308 * see whether the host or the device is offline.
309 *
310 * Return value:
311 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800312 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313int scsi_block_when_processing_errors(struct scsi_device *sdev)
314{
315 int online;
316
James Bottomley939647e2005-09-18 15:05:20 -0500317 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319 online = scsi_device_online(sdev);
320
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700321 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 online));
323
324 return online;
325}
326EXPORT_SYMBOL(scsi_block_when_processing_errors);
327
328#ifdef CONFIG_SCSI_LOGGING
329/**
330 * scsi_eh_prt_fail_stats - Log info on failures.
331 * @shost: scsi host being recovered.
332 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800333 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
335 struct list_head *work_q)
336{
337 struct scsi_cmnd *scmd;
338 struct scsi_device *sdev;
339 int total_failures = 0;
340 int cmd_failed = 0;
341 int cmd_cancel = 0;
342 int devices_failed = 0;
343
344 shost_for_each_device(sdev, shost) {
345 list_for_each_entry(scmd, work_q, eh_entry) {
346 if (scmd->device == sdev) {
347 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200348 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 ++cmd_cancel;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100350 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 ++cmd_failed;
352 }
353 }
354
355 if (cmd_cancel || cmd_failed) {
356 SCSI_LOG_ERROR_RECOVERY(3,
James Bottomley9ccfc752005-10-02 11:45:08 -0500357 sdev_printk(KERN_INFO, sdev,
358 "%s: cmds failed: %d, cancel: %d\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700359 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500360 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 cmd_cancel = 0;
362 cmd_failed = 0;
363 ++devices_failed;
364 }
365 }
366
367 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
368 " devices require eh work\n",
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100369 total_failures, devices_failed));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
371#endif
372
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400373 /**
374 * scsi_report_lun_change - Set flag on all *other* devices on the same target
375 * to indicate that a UNIT ATTENTION is expected.
376 * @sdev: Device reporting the UNIT ATTENTION
377 */
378static void scsi_report_lun_change(struct scsi_device *sdev)
379{
380 sdev->sdev_target->expecting_lun_change = 1;
381}
382
383/**
384 * scsi_report_sense - Examine scsi sense information and log messages for
385 * certain conditions, also issue uevents for some of them.
386 * @sdev: Device reporting the sense code
387 * @sshdr: sshdr to be examined
388 */
389static void scsi_report_sense(struct scsi_device *sdev,
390 struct scsi_sense_hdr *sshdr)
391{
392 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
393
394 if (sshdr->sense_key == UNIT_ATTENTION) {
395 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
396 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
397 sdev_printk(KERN_WARNING, sdev,
398 "Inquiry data has changed");
399 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
400 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
401 scsi_report_lun_change(sdev);
402 sdev_printk(KERN_WARNING, sdev,
403 "Warning! Received an indication that the "
404 "LUN assignments on this target have "
405 "changed. The Linux SCSI layer does not "
406 "automatically remap LUN assignments.\n");
407 } else if (sshdr->asc == 0x3f)
408 sdev_printk(KERN_WARNING, sdev,
409 "Warning! Received an indication that the "
410 "operating parameters on this target have "
411 "changed. The Linux SCSI layer does not "
412 "automatically adjust these parameters.\n");
413
414 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
415 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
416 sdev_printk(KERN_WARNING, sdev,
417 "Warning! Received an indication that the "
418 "LUN reached a thin provisioning soft "
419 "threshold.\n");
420 }
421
422 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
423 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
424 sdev_printk(KERN_WARNING, sdev,
425 "Mode parameters changed");
426 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
427 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
428 sdev_printk(KERN_WARNING, sdev,
429 "Capacity data has changed");
430 } else if (sshdr->asc == 0x2a)
431 sdev_printk(KERN_WARNING, sdev,
432 "Parameters changed");
433 }
434
435 if (evt_type != SDEV_EVT_MAXBITS) {
436 set_bit(evt_type, sdev->pending_events);
437 schedule_work(&sdev->event_work);
438 }
439}
440
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441/**
442 * scsi_check_sense - Examine scsi cmd sense
443 * @scmd: Cmd to have sense checked.
444 *
445 * Return value:
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200446 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 *
448 * Notes:
449 * When a deferred error is detected the current command has
450 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800451 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452static int scsi_check_sense(struct scsi_cmnd *scmd)
453{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700454 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 struct scsi_sense_hdr sshdr;
456
457 if (! scsi_command_normalize_sense(scmd, &sshdr))
458 return FAILED; /* no valid sense data */
459
James Bottomley14216562012-07-25 23:55:55 +0400460 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
461 /*
462 * nasty: for mid-layer issued TURs, we need to return the
463 * actual sense data without any recovery attempt. For eh
464 * issued ones, we need to try to recover and interpret
465 */
466 return SUCCESS;
467
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400468 scsi_report_sense(sdev, &sshdr);
469
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 if (scsi_sense_is_deferred(&sshdr))
471 return NEEDS_RETRY;
472
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700473 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
474 sdev->scsi_dh_data->scsi_dh->check_sense) {
475 int rc;
476
477 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
478 if (rc != SCSI_RETURN_NOT_HANDLED)
479 return rc;
480 /* handler does not care. Drop down to default handling */
481 }
482
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 /*
484 * Previous logic looked for FILEMARK, EOM or ILI which are
485 * mainly associated with tapes and returned SUCCESS.
486 */
487 if (sshdr.response_code == 0x70) {
488 /* fixed format */
489 if (scmd->sense_buffer[2] & 0xe0)
490 return SUCCESS;
491 } else {
492 /*
493 * descriptor format: look for "stream commands sense data
494 * descriptor" (see SSC-3). Assume single sense data
495 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
496 */
497 if ((sshdr.additional_length > 3) &&
498 (scmd->sense_buffer[8] == 0x4) &&
499 (scmd->sense_buffer[11] & 0xe0))
500 return SUCCESS;
501 }
502
503 switch (sshdr.sense_key) {
504 case NO_SENSE:
505 return SUCCESS;
506 case RECOVERED_ERROR:
507 return /* soft_error */ SUCCESS;
508
509 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400510 if (sshdr.asc == 0x10) /* DIF */
511 return SUCCESS;
512
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 return NEEDS_RETRY;
514 case NOT_READY:
515 case UNIT_ATTENTION:
516 /*
517 * if we are expecting a cc/ua because of a bus reset that we
518 * performed, treat this just as a retry. otherwise this is
519 * information that we should pass up to the upper-level driver
520 * so that we can deal with it there.
521 */
522 if (scmd->device->expecting_cc_ua) {
TARUISI Hiroakidfcf7772011-08-11 20:25:20 +0900523 /*
524 * Because some device does not queue unit
525 * attentions correctly, we carefully check
526 * additional sense code and qualifier so as
527 * not to squash media change unit attention.
528 */
529 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
530 scmd->device->expecting_cc_ua = 0;
531 return NEEDS_RETRY;
532 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
534 /*
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400535 * we might also expect a cc/ua if another LUN on the target
536 * reported a UA with an ASC/ASCQ of 3F 0E -
537 * REPORTED LUNS DATA HAS CHANGED.
538 */
539 if (scmd->device->sdev_target->expecting_lun_change &&
540 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
541 return NEEDS_RETRY;
542 /*
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100543 * if the device is in the process of becoming ready, we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 * should retry.
545 */
546 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
547 return NEEDS_RETRY;
548 /*
549 * if the device is not started, we need to wake
550 * the error handler to start the motor
551 */
552 if (scmd->device->allow_restart &&
553 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
554 return FAILED;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100555 /*
556 * Pass the UA upwards for a determination in the completion
557 * functions.
558 */
559 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100561 /* these are not supported */
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +0200562 case DATA_PROTECT:
563 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
564 /* Thin provisioning hard threshold reached */
565 set_host_byte(scmd, DID_ALLOC_FAILURE);
566 return SUCCESS;
567 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 case COPY_ABORTED:
569 case VOLUME_OVERFLOW:
570 case MISCOMPARE:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100571 case BLANK_CHECK:
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200572 set_host_byte(scmd, DID_TARGET_FAILURE);
573 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800576 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
577 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
578 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
Hannes Reinecke7e782af2013-07-01 15:16:26 +0200579 set_host_byte(scmd, DID_MEDIUM_ERROR);
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200580 return SUCCESS;
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 return NEEDS_RETRY;
583
584 case HARDWARE_ERROR:
585 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700586 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 else
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200588 set_host_byte(scmd, DID_TARGET_FAILURE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 case ILLEGAL_REQUEST:
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500591 if (sshdr.asc == 0x20 || /* Invalid command operation code */
592 sshdr.asc == 0x21 || /* Logical block address out of range */
593 sshdr.asc == 0x24 || /* Invalid field in cdb */
594 sshdr.asc == 0x26) { /* Parameter value invalid */
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200595 set_host_byte(scmd, DID_TARGET_FAILURE);
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500596 }
597 return SUCCESS;
598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 default:
600 return SUCCESS;
601 }
602}
603
Vasu Dev4a840672009-10-22 15:46:33 -0700604static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
605{
606 struct scsi_host_template *sht = sdev->host->hostt;
607 struct scsi_device *tmp_sdev;
608
609 if (!sht->change_queue_depth ||
610 sdev->queue_depth >= sdev->max_queue_depth)
611 return;
612
613 if (time_before(jiffies,
614 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
615 return;
616
617 if (time_before(jiffies,
618 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
619 return;
620
621 /*
622 * Walk all devices of a target and do
623 * ramp up on them.
624 */
625 shost_for_each_device(tmp_sdev, sdev->host) {
626 if (tmp_sdev->channel != sdev->channel ||
627 tmp_sdev->id != sdev->id ||
628 tmp_sdev->queue_depth == sdev->max_queue_depth)
629 continue;
630 /*
631 * call back into LLD to increase queue_depth by one
632 * with ramp up reason code.
633 */
634 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
635 SCSI_QDEPTH_RAMP_UP);
636 sdev->last_queue_ramp_up = jiffies;
637 }
638}
639
Mike Christie42a6a912009-10-15 17:46:44 -0700640static void scsi_handle_queue_full(struct scsi_device *sdev)
641{
642 struct scsi_host_template *sht = sdev->host->hostt;
643 struct scsi_device *tmp_sdev;
644
645 if (!sht->change_queue_depth)
646 return;
647
648 shost_for_each_device(tmp_sdev, sdev->host) {
649 if (tmp_sdev->channel != sdev->channel ||
650 tmp_sdev->id != sdev->id)
651 continue;
652 /*
653 * We do not know the number of commands that were at
654 * the device when we got the queue full so we start
655 * from the highest possible value and work our way down.
656 */
657 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
658 SCSI_QDEPTH_QFULL);
659 }
660}
661
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662/**
663 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
664 * @scmd: SCSI cmd to examine.
665 *
666 * Notes:
667 * This is *only* called when we are examining the status of commands
668 * queued during error recovery. the main difference here is that we
669 * don't allow for the possibility of retries here, and we are a lot
670 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800671 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
673{
674 /*
675 * first check the host byte, to see if there is anything in there
676 * that would indicate what we need to do.
677 */
678 if (host_byte(scmd->result) == DID_RESET) {
679 /*
680 * rats. we are already in the error handler, so we now
681 * get to try and figure out what to do next. if the sense
682 * is valid, we have a pretty good idea of what to do.
683 * if not, we mark it as FAILED.
684 */
685 return scsi_check_sense(scmd);
686 }
687 if (host_byte(scmd->result) != DID_OK)
688 return FAILED;
689
690 /*
691 * next, check the message byte.
692 */
693 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
694 return FAILED;
695
696 /*
697 * now, check the status byte to see if this indicates
698 * anything special.
699 */
700 switch (status_byte(scmd->result)) {
701 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700702 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 case COMMAND_TERMINATED:
704 return SUCCESS;
705 case CHECK_CONDITION:
706 return scsi_check_sense(scmd);
707 case CONDITION_GOOD:
708 case INTERMEDIATE_GOOD:
709 case INTERMEDIATE_C_GOOD:
710 /*
711 * who knows? FIXME(eric)
712 */
713 return SUCCESS;
Michael Reed5f91bb02009-08-10 11:59:28 -0500714 case RESERVATION_CONFLICT:
James Bottomley67110df2010-08-06 15:17:24 -0500715 if (scmd->cmnd[0] == TEST_UNIT_READY)
716 /* it is a success, we probed the device and
717 * found it */
718 return SUCCESS;
719 /* otherwise, we failed to send the command */
720 return FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700722 scsi_handle_queue_full(scmd->device);
723 /* fall through */
724 case BUSY:
Hannes Reinecke3eb3a922010-07-29 10:10:16 +0200725 return NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 default:
727 return FAILED;
728 }
729 return FAILED;
730}
731
732/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 * scsi_eh_done - Completion function for error handling.
734 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800735 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736static void scsi_eh_done(struct scsi_cmnd *scmd)
737{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100738 struct completion *eh_action;
Michael Reed85631672005-12-07 21:46:27 -0600739
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100740 SCSI_LOG_ERROR_RECOVERY(3,
741 printk("%s scmd: %p result: %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700742 __func__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600743
744 eh_action = scmd->device->host->eh_action;
745 if (eh_action)
746 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747}
748
749/**
Brian King292148f2007-01-30 17:51:17 -0600750 * scsi_try_host_reset - ask host adapter to reset itself
Geert Uytterhoevenc2b3ebd2013-05-17 13:22:29 +0200751 * @scmd: SCSI cmd to send host reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800752 */
Brian King292148f2007-01-30 17:51:17 -0600753static int scsi_try_host_reset(struct scsi_cmnd *scmd)
754{
755 unsigned long flags;
756 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100757 struct Scsi_Host *host = scmd->device->host;
758 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600759
760 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700761 __func__));
Brian King292148f2007-01-30 17:51:17 -0600762
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100763 if (!hostt->eh_host_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600764 return FAILED;
765
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100766 rtn = hostt->eh_host_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600767
768 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100769 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600770 ssleep(HOST_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100771 spin_lock_irqsave(host->host_lock, flags);
772 scsi_report_bus_reset(host, scmd_channel(scmd));
773 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600774 }
775
776 return rtn;
777}
778
779/**
780 * scsi_try_bus_reset - ask host to perform a bus reset
781 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800782 */
Brian King292148f2007-01-30 17:51:17 -0600783static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
784{
785 unsigned long flags;
786 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100787 struct Scsi_Host *host = scmd->device->host;
788 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600789
790 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700791 __func__));
Brian King292148f2007-01-30 17:51:17 -0600792
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100793 if (!hostt->eh_bus_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600794 return FAILED;
795
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100796 rtn = hostt->eh_bus_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600797
798 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100799 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600800 ssleep(BUS_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100801 spin_lock_irqsave(host->host_lock, flags);
802 scsi_report_bus_reset(host, scmd_channel(scmd));
803 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600804 }
805
806 return rtn;
807}
808
Mike Christie30bd7df2008-02-29 18:25:19 -0600809static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
810{
811 sdev->was_reset = 1;
812 sdev->expecting_cc_ua = 1;
813}
814
815/**
816 * scsi_try_target_reset - Ask host to perform a target reset
817 * @scmd: SCSI cmd used to send a target reset
818 *
819 * Notes:
820 * There is no timeout for this operation. if this operation is
821 * unreliable for a given host, then the host itself needs to put a
822 * timer on it, and set the host back to a consistent state prior to
823 * returning.
824 */
825static int scsi_try_target_reset(struct scsi_cmnd *scmd)
826{
827 unsigned long flags;
828 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100829 struct Scsi_Host *host = scmd->device->host;
830 struct scsi_host_template *hostt = host->hostt;
Mike Christie30bd7df2008-02-29 18:25:19 -0600831
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100832 if (!hostt->eh_target_reset_handler)
Mike Christie30bd7df2008-02-29 18:25:19 -0600833 return FAILED;
834
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100835 rtn = hostt->eh_target_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600836 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100837 spin_lock_irqsave(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600838 __starget_for_each_device(scsi_target(scmd->device), NULL,
839 __scsi_report_device_reset);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100840 spin_unlock_irqrestore(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600841 }
842
843 return rtn;
844}
845
Brian King292148f2007-01-30 17:51:17 -0600846/**
847 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
848 * @scmd: SCSI cmd used to send BDR
849 *
850 * Notes:
851 * There is no timeout for this operation. if this operation is
852 * unreliable for a given host, then the host itself needs to put a
853 * timer on it, and set the host back to a consistent state prior to
854 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800855 */
Brian King292148f2007-01-30 17:51:17 -0600856static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
857{
858 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100859 struct scsi_host_template *hostt = scmd->device->host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600860
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100861 if (!hostt->eh_device_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600862 return FAILED;
863
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100864 rtn = hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600865 if (rtn == SUCCESS)
866 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600867 return rtn;
868}
869
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100870static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600871{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100872 if (!hostt->eh_abort_handler)
Brian King292148f2007-01-30 17:51:17 -0600873 return FAILED;
874
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100875 return hostt->eh_abort_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600876}
877
Brian King292148f2007-01-30 17:51:17 -0600878static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
879{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100880 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
Brian King292148f2007-01-30 17:51:17 -0600881 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600882 if (scsi_try_target_reset(scmd) != SUCCESS)
883 if (scsi_try_bus_reset(scmd) != SUCCESS)
884 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600885}
886
887/**
Santosh Y3b729f72012-04-08 18:47:09 +0530888 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100889 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200890 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200891 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300892 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200893 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100894 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200895 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200896 * as part of the error recovery process. If @sense_bytes is 0 the command
897 * sent must be one that does not transfer any data. If @sense_bytes != 0
898 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
899 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800900 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200901void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
902 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
f59114b2005-04-17 15:00:23 -0500904 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
Christoph Hellwig631c2282006-07-08 20:42:15 +0200906 /*
907 * We need saved copies of a number of fields - this is because
908 * error handling may need to overwrite these with different values
909 * to run different commands, and once error handling is complete,
910 * we will need to restore these values prior to running the actual
911 * command.
912 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200913 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300914 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200915 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200916 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200917 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200918 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400919 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400920 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200921
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400922 scmd->prot_op = SCSI_PROT_NORMAL;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300923 scmd->cmnd = ses->eh_cmnd;
924 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200925 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200926 scmd->request->next_rq = NULL;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200927
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200928 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200929 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
930 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200931 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200932 scmd->sdb.length);
933 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200934 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200935 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200936 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200937 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200938 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200939 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200940 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200941 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300942 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200943 memcpy(scmd->cmnd, cmnd, cmnd_size);
944 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
945 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200946 }
947
948 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200949
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200950 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500952 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
Christoph Hellwig631c2282006-07-08 20:42:15 +0200954 /*
955 * Zero the sense buffer. The scsi spec mandates that any
956 * untransferred sense data should be interpreted as being zero.
957 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900958 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200959}
960EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200961
Boaz Harroshe1c23462007-10-08 16:36:45 +0200962/**
Santosh Y3b729f72012-04-08 18:47:09 +0530963 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
Boaz Harroshe1c23462007-10-08 16:36:45 +0200964 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200965 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +0200966 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200967 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800968 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200969void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
970{
971 /*
972 * Restore original data
973 */
974 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300975 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200976 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200977 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200978 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200979 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400980 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400981 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200982}
983EXPORT_SYMBOL(scsi_eh_restore_cmnd);
984
985/**
Santosh Y3b729f72012-04-08 18:47:09 +0530986 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
Boaz Harroshe1c23462007-10-08 16:36:45 +0200987 * @scmd: SCSI command structure to hijack
988 * @cmnd: CDB to send
989 * @cmnd_size: size in bytes of @cmnd
990 * @timeout: timeout for this request
991 * @sense_bytes: size of sense data to copy or 0
992 *
993 * This function is used to send a scsi command down to a target device
994 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
995 *
996 * Return value:
997 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800998 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200999static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1000 int cmnd_size, int timeout, unsigned sense_bytes)
1001{
1002 struct scsi_device *sdev = scmd->device;
1003 struct Scsi_Host *shost = sdev->host;
1004 DECLARE_COMPLETION_ONSTACK(done);
Hannes Reineckefc736482013-04-25 08:10:00 +02001005 unsigned long timeleft = timeout;
Boaz Harroshe1c23462007-10-08 16:36:45 +02001006 struct scsi_eh_save ses;
Hannes Reineckefc736482013-04-25 08:10:00 +02001007 const unsigned long stall_for = msecs_to_jiffies(100);
Boaz Harroshe1c23462007-10-08 16:36:45 +02001008 int rtn;
1009
Hannes Reineckefc736482013-04-25 08:10:00 +02001010retry:
Boaz Harroshe1c23462007-10-08 16:36:45 +02001011 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001012 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 scsi_log_send(scmd);
Jeff Garzikf2812332010-11-16 02:10:29 -05001015 scmd->scsi_done = scsi_eh_done;
Hannes Reineckefc736482013-04-25 08:10:00 +02001016 rtn = shost->hostt->queuecommand(shost, scmd);
1017 if (rtn) {
1018 if (timeleft > stall_for) {
1019 scsi_eh_restore_cmnd(scmd, &ses);
1020 timeleft -= stall_for;
1021 msleep(jiffies_to_msecs(stall_for));
1022 goto retry;
1023 }
1024 /* signal not to enter either branch of the if () below */
1025 timeleft = 0;
1026 rtn = NEEDS_RETRY;
1027 } else {
1028 timeleft = wait_for_completion_timeout(&done, timeout);
1029 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
f59114b2005-04-17 15:00:23 -05001031 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
Hannes Reineckefc736482013-04-25 08:10:00 +02001033 scsi_log_completion(scmd, rtn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001035 SCSI_LOG_ERROR_RECOVERY(3,
1036 printk("%s: scmd: %p, timeleft: %ld\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001037 __func__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
1039 /*
Hannes Reineckefc736482013-04-25 08:10:00 +02001040 * If there is time left scsi_eh_done got called, and we will examine
1041 * the actual status codes to see whether the command actually did
1042 * complete normally, else if we have a zero return and no time left,
1043 * the command must still be pending, so abort it and return FAILED.
1044 * If we never actually managed to issue the command, because
1045 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1046 * value above (so don't execute either branch of the if)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001048 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 rtn = scsi_eh_completed_normally(scmd);
1050 SCSI_LOG_ERROR_RECOVERY(3,
1051 printk("%s: scsi_eh_completed_normally %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001052 __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001053
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 switch (rtn) {
1055 case SUCCESS:
1056 case NEEDS_RETRY:
1057 case FAILED:
1058 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +02001059 case ADD_TO_MLQUEUE:
1060 rtn = NEEDS_RETRY;
1061 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 default:
1063 rtn = FAILED;
1064 break;
1065 }
Hannes Reineckefc736482013-04-25 08:10:00 +02001066 } else if (!rtn) {
Brian King292148f2007-01-30 17:51:17 -06001067 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001068 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 }
1070
Boaz Harroshe1c23462007-10-08 16:36:45 +02001071 scsi_eh_restore_cmnd(scmd, &ses);
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -05001072
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 return rtn;
1074}
1075
1076/**
1077 * scsi_request_sense - Request sense data from a particular target.
1078 * @scmd: SCSI cmd for request sense.
1079 *
1080 * Notes:
1081 * Some hosts automatically obtain this information, others require
1082 * that we obtain it on our own. This function will *not* return until
1083 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001084 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085static int scsi_request_sense(struct scsi_cmnd *scmd)
1086{
Martin K. Petersen0816c922013-05-10 10:36:04 -04001087 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088}
1089
James Bottomley24510792013-11-11 13:44:53 +01001090static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1091{
1092 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1093 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1094 if (sdrv->eh_action)
1095 rtn = sdrv->eh_action(scmd, rtn);
1096 }
1097 return rtn;
1098}
1099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100/**
1101 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1102 * @scmd: Original SCSI cmd that eh has finished.
1103 * @done_q: Queue for processed commands.
1104 *
1105 * Notes:
1106 * We don't want to use the normal command completion while we are are
1107 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -05001108 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 * keep a list of pending commands for final completion, and once we
1110 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001111 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001112void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113{
1114 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001115 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 list_move_tail(&scmd->eh_entry, done_q);
1117}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001118EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
1120/**
1121 * scsi_eh_get_sense - Get device sense data.
1122 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -05001123 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 *
1125 * Description:
1126 * See if we need to request sense information. if so, then get it
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001127 * now, so we have a better idea of what to do.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 *
1129 * Notes:
1130 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -05001131 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 * it down before we request it.
1133 *
1134 * All drivers should request sense information internally these days,
1135 * so for now all I have to say is tough noogies if you end up in here.
1136 *
1137 * XXX: Long term this code should go away, but that needs an audit of
1138 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001139 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001140int scsi_eh_get_sense(struct list_head *work_q,
1141 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001143 struct scsi_cmnd *scmd, *next;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001144 struct Scsi_Host *shost;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 int rtn;
1146
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001147 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001148 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 SCSI_SENSE_VALID(scmd))
1150 continue;
1151
Hannes Reineckeb4562022013-10-23 10:51:21 +02001152 shost = scmd->device->host;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001153 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001154 SCSI_LOG_ERROR_RECOVERY(3,
1155 shost_printk(KERN_INFO, shost,
1156 "skip %s, past eh deadline\n",
1157 __func__));
1158 break;
1159 }
Jeff Garzik3bf743e2005-10-24 18:04:06 -04001160 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1161 "%s: requesting sense\n",
1162 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 rtn = scsi_request_sense(scmd);
1164 if (rtn != SUCCESS)
1165 continue;
1166
1167 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
1168 " result %x\n", scmd,
1169 scmd->result));
1170 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
1171
1172 rtn = scsi_decide_disposition(scmd);
1173
1174 /*
1175 * if the result was normal, then just pass it along to the
1176 * upper level.
1177 */
1178 if (rtn == SUCCESS)
1179 /* we don't want this command reissued, just
1180 * finished with the sense data, so set
1181 * retries to the max allowed to ensure it
1182 * won't get reissued */
1183 scmd->retries = scmd->allowed;
1184 else if (rtn != NEEDS_RETRY)
1185 continue;
1186
1187 scsi_eh_finish_cmd(scmd, done_q);
1188 }
1189
1190 return list_empty(work_q);
1191}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001192EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
1194/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001196 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 *
1198 * Return value:
1199 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001200 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201static int scsi_eh_tur(struct scsi_cmnd *scmd)
1202{
1203 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1204 int retry_cnt = 1, rtn;
1205
1206retry_tur:
Martin K. Petersen0816c922013-05-10 10:36:04 -04001207 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1208 scmd->device->eh_timeout, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001211 __func__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +02001212
1213 switch (rtn) {
1214 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (retry_cnt--)
1216 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001217 /*FALLTHRU*/
1218 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -05001219 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001220 default:
1221 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -05001222 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223}
1224
1225/**
David Jeffery3eef6252011-05-19 14:41:12 -04001226 * scsi_eh_test_devices - check if devices are responding from error recovery.
1227 * @cmd_list: scsi commands in error recovery.
1228 * @work_q: queue for commands which still need more error recovery
1229 * @done_q: queue for commands which are finished
1230 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1231 *
1232 * Decription:
1233 * Tests if devices are in a working state. Commands to devices now in
1234 * a working state are sent to the done_q while commands to devices which
1235 * are still failing to respond are returned to the work_q for more
1236 * processing.
1237 **/
1238static int scsi_eh_test_devices(struct list_head *cmd_list,
1239 struct list_head *work_q,
1240 struct list_head *done_q, int try_stu)
1241{
1242 struct scsi_cmnd *scmd, *next;
1243 struct scsi_device *sdev;
1244 int finish_cmds;
1245
1246 while (!list_empty(cmd_list)) {
1247 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1248 sdev = scmd->device;
1249
Hannes Reineckeb4562022013-10-23 10:51:21 +02001250 if (!try_stu) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001251 if (scsi_host_eh_past_deadline(sdev->host)) {
1252 /* Push items back onto work_q */
1253 list_splice_init(cmd_list, work_q);
Hannes Reineckeb4562022013-10-23 10:51:21 +02001254 SCSI_LOG_ERROR_RECOVERY(3,
1255 shost_printk(KERN_INFO, sdev->host,
1256 "skip %s, past eh deadline",
1257 __func__));
1258 break;
1259 }
Hannes Reineckeb4562022013-10-23 10:51:21 +02001260 }
1261
David Jeffery3eef6252011-05-19 14:41:12 -04001262 finish_cmds = !scsi_device_online(scmd->device) ||
1263 (try_stu && !scsi_eh_try_stu(scmd) &&
1264 !scsi_eh_tur(scmd)) ||
1265 !scsi_eh_tur(scmd);
1266
1267 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1268 if (scmd->device == sdev) {
James Bottomley24510792013-11-11 13:44:53 +01001269 if (finish_cmds &&
1270 (try_stu ||
1271 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
David Jeffery3eef6252011-05-19 14:41:12 -04001272 scsi_eh_finish_cmd(scmd, done_q);
1273 else
1274 list_move_tail(&scmd->eh_entry, work_q);
1275 }
1276 }
1277 return list_empty(work_q);
1278}
1279
1280
1281/**
Rob Landleyeb448202007-11-03 13:30:39 -05001282 * scsi_eh_abort_cmds - abort pending commands.
1283 * @work_q: &list_head for pending commands.
1284 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 *
1286 * Decription:
1287 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -05001288 * running command. This only works out to be the case if we have one
1289 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 * no sense to try and abort the command, since as far as the shost
1291 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001292 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293static int scsi_eh_abort_cmds(struct list_head *work_q,
1294 struct list_head *done_q)
1295{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001296 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001297 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 int rtn;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001299 struct Scsi_Host *shost;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001301 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001302 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 continue;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001304 shost = scmd->device->host;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001305 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001306 list_splice_init(&check_list, work_q);
1307 SCSI_LOG_ERROR_RECOVERY(3,
1308 shost_printk(KERN_INFO, shost,
1309 "skip %s, past eh deadline\n",
1310 __func__));
1311 return list_empty(work_q);
1312 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1314 "0x%p\n", current->comm,
1315 scmd));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001316 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
Hannes Reinecke6fd046f2013-10-23 10:51:22 +02001317 if (rtn == FAILED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1319 " cmd failed:"
1320 "0x%p\n",
1321 current->comm,
1322 scmd));
Hannes Reinecke6fd046f2013-10-23 10:51:22 +02001323 list_splice_init(&check_list, work_q);
1324 return list_empty(work_q);
1325 }
1326 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1327 if (rtn == FAST_IO_FAIL)
1328 scsi_eh_finish_cmd(scmd, done_q);
1329 else
1330 list_move_tail(&scmd->eh_entry, &check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 }
1332
David Jeffery3eef6252011-05-19 14:41:12 -04001333 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334}
1335
1336/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001338 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 *
1340 * Return value:
1341 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001342 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1344{
1345 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
Christoph Hellwig631c2282006-07-08 20:42:15 +02001347 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -05001348 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
Brian Kinged773e62007-03-29 15:25:52 -05001350 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -06001351 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -05001352
Christoph Hellwig631c2282006-07-08 20:42:15 +02001353 if (rtn == SUCCESS)
1354 return 0;
1355 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 return 1;
1358}
1359
1360 /**
1361 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001362 * @shost: &scsi host being recovered.
1363 * @work_q: &list_head for pending commands.
1364 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 *
1366 * Notes:
1367 * If commands are failing due to not ready, initializing command required,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001368 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001369 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370static int scsi_eh_stu(struct Scsi_Host *shost,
1371 struct list_head *work_q,
1372 struct list_head *done_q)
1373{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001374 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 struct scsi_device *sdev;
1376
1377 shost_for_each_device(sdev, shost) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001378 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001379 SCSI_LOG_ERROR_RECOVERY(3,
1380 shost_printk(KERN_INFO, shost,
1381 "skip %s, past eh deadline\n",
1382 __func__));
1383 break;
1384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 stu_scmd = NULL;
1386 list_for_each_entry(scmd, work_q, eh_entry)
1387 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1388 scsi_check_sense(scmd) == FAILED ) {
1389 stu_scmd = scmd;
1390 break;
1391 }
1392
1393 if (!stu_scmd)
1394 continue;
1395
1396 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1397 " 0x%p\n", current->comm, sdev));
1398
1399 if (!scsi_eh_try_stu(stu_scmd)) {
1400 if (!scsi_device_online(sdev) ||
1401 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001402 list_for_each_entry_safe(scmd, next,
1403 work_q, eh_entry) {
James Bottomley24510792013-11-11 13:44:53 +01001404 if (scmd->device == sdev &&
1405 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 scsi_eh_finish_cmd(scmd, done_q);
1407 }
1408 }
1409 } else {
1410 SCSI_LOG_ERROR_RECOVERY(3,
1411 printk("%s: START_UNIT failed to sdev:"
1412 " 0x%p\n", current->comm, sdev));
1413 }
1414 }
1415
1416 return list_empty(work_q);
1417}
1418
1419
1420/**
1421 * scsi_eh_bus_device_reset - send bdr if needed
1422 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001423 * @work_q: &list_head for pending commands.
1424 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 *
1426 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001427 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 * devices that are jammed or not - if we have multiple devices, it
1429 * makes no sense to try bus_device_reset - we really would need to try
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001430 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001431 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1433 struct list_head *work_q,
1434 struct list_head *done_q)
1435{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001436 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 struct scsi_device *sdev;
1438 int rtn;
1439
1440 shost_for_each_device(sdev, shost) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001441 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001442 SCSI_LOG_ERROR_RECOVERY(3,
1443 shost_printk(KERN_INFO, shost,
1444 "skip %s, past eh deadline\n",
1445 __func__));
1446 break;
1447 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 bdr_scmd = NULL;
1449 list_for_each_entry(scmd, work_q, eh_entry)
1450 if (scmd->device == sdev) {
1451 bdr_scmd = scmd;
1452 break;
1453 }
1454
1455 if (!bdr_scmd)
1456 continue;
1457
1458 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1459 " 0x%p\n", current->comm,
1460 sdev));
1461 rtn = scsi_try_bus_device_reset(bdr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001462 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 if (!scsi_device_online(sdev) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001464 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001466 list_for_each_entry_safe(scmd, next,
1467 work_q, eh_entry) {
James Bottomley24510792013-11-11 13:44:53 +01001468 if (scmd->device == sdev &&
1469 scsi_eh_action(scmd, rtn) != FAILED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 scsi_eh_finish_cmd(scmd,
1471 done_q);
1472 }
1473 }
1474 } else {
1475 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1476 " failed sdev:"
1477 "0x%p\n",
1478 current->comm,
1479 sdev));
1480 }
1481 }
1482
1483 return list_empty(work_q);
1484}
1485
1486/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001487 * scsi_eh_target_reset - send target reset if needed
1488 * @shost: scsi host being recovered.
1489 * @work_q: &list_head for pending commands.
1490 * @done_q: &list_head for processed commands.
1491 *
1492 * Notes:
1493 * Try a target reset.
1494 */
1495static int scsi_eh_target_reset(struct Scsi_Host *shost,
1496 struct list_head *work_q,
1497 struct list_head *done_q)
1498{
James Bottomley98db5192010-10-25 15:53:41 -05001499 LIST_HEAD(tmp_list);
David Jeffery3eef6252011-05-19 14:41:12 -04001500 LIST_HEAD(check_list);
Mike Christie30bd7df2008-02-29 18:25:19 -06001501
James Bottomley98db5192010-10-25 15:53:41 -05001502 list_splice_init(work_q, &tmp_list);
1503
1504 while (!list_empty(&tmp_list)) {
1505 struct scsi_cmnd *next, *scmd;
1506 int rtn;
1507 unsigned int id;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001508
Hannes Reineckeb4562022013-10-23 10:51:21 +02001509 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001510 /* push back on work queue for further processing */
1511 list_splice_init(&check_list, work_q);
1512 list_splice_init(&tmp_list, work_q);
1513 SCSI_LOG_ERROR_RECOVERY(3,
1514 shost_printk(KERN_INFO, shost,
1515 "skip %s, past eh deadline\n",
1516 __func__));
1517 return list_empty(work_q);
1518 }
James Bottomley98db5192010-10-25 15:53:41 -05001519
1520 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1521 id = scmd_id(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -06001522
1523 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1524 "to target %d\n",
1525 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001526 rtn = scsi_try_target_reset(scmd);
1527 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
Mike Christie30bd7df2008-02-29 18:25:19 -06001528 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1529 " failed target: "
1530 "%d\n",
1531 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001532 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1533 if (scmd_id(scmd) != id)
1534 continue;
1535
David Jeffery3eef6252011-05-19 14:41:12 -04001536 if (rtn == SUCCESS)
1537 list_move_tail(&scmd->eh_entry, &check_list);
1538 else if (rtn == FAST_IO_FAIL)
James Bottomley98db5192010-10-25 15:53:41 -05001539 scsi_eh_finish_cmd(scmd, done_q);
1540 else
1541 /* push back on work queue for further processing */
1542 list_move(&scmd->eh_entry, work_q);
1543 }
1544 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001545
David Jeffery3eef6252011-05-19 14:41:12 -04001546 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001547}
1548
1549/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001550 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001551 * @shost: &scsi host being recovered.
1552 * @work_q: &list_head for pending commands.
1553 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001554 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1556 struct list_head *work_q,
1557 struct list_head *done_q)
1558{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001559 struct scsi_cmnd *scmd, *chan_scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001560 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 unsigned int channel;
1562 int rtn;
1563
1564 /*
1565 * we really want to loop over the various channels, and do this on
1566 * a channel by channel basis. we should also check to see if any
1567 * of the failed commands are on soft_reset devices, and if so, skip
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001568 * the reset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 */
1570
1571 for (channel = 0; channel <= shost->max_channel; channel++) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001572 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001573 list_splice_init(&check_list, work_q);
1574 SCSI_LOG_ERROR_RECOVERY(3,
1575 shost_printk(KERN_INFO, shost,
1576 "skip %s, past eh deadline\n",
1577 __func__));
1578 return list_empty(work_q);
1579 }
Hannes Reineckeb4562022013-10-23 10:51:21 +02001580
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 chan_scmd = NULL;
1582 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001583 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 chan_scmd = scmd;
1585 break;
1586 /*
1587 * FIXME add back in some support for
1588 * soft_reset devices.
1589 */
1590 }
1591 }
1592
1593 if (!chan_scmd)
1594 continue;
1595 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1596 " %d\n", current->comm,
1597 channel));
1598 rtn = scsi_try_bus_reset(chan_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001599 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001600 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
David Jeffery3eef6252011-05-19 14:41:12 -04001601 if (channel == scmd_channel(scmd)) {
1602 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 scsi_eh_finish_cmd(scmd,
1604 done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001605 else
1606 list_move_tail(&scmd->eh_entry,
1607 &check_list);
1608 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 }
1610 } else {
1611 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1612 " failed chan: %d\n",
1613 current->comm,
1614 channel));
1615 }
1616 }
David Jeffery3eef6252011-05-19 14:41:12 -04001617 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618}
1619
1620/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001621 * scsi_eh_host_reset - send a host reset
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 * @work_q: list_head for processed commands.
1623 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001624 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625static int scsi_eh_host_reset(struct list_head *work_q,
1626 struct list_head *done_q)
1627{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001628 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001629 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
1632 if (!list_empty(work_q)) {
1633 scmd = list_entry(work_q->next,
1634 struct scsi_cmnd, eh_entry);
1635
1636 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1637 , current->comm));
1638
1639 rtn = scsi_try_host_reset(scmd);
David Jeffery3eef6252011-05-19 14:41:12 -04001640 if (rtn == SUCCESS) {
1641 list_splice_init(work_q, &check_list);
1642 } else if (rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001643 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 scsi_eh_finish_cmd(scmd, done_q);
1645 }
1646 } else {
1647 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1648 " failed\n",
1649 current->comm));
1650 }
1651 }
David Jeffery3eef6252011-05-19 14:41:12 -04001652 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653}
1654
1655/**
1656 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1657 * @work_q: list_head for processed commands.
1658 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001659 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660static void scsi_eh_offline_sdevs(struct list_head *work_q,
1661 struct list_head *done_q)
1662{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001663 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001665 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001666 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1667 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001669 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 /*
1671 * FIXME: Handle lost cmds.
1672 */
1673 }
1674 scsi_eh_finish_cmd(scmd, done_q);
1675 }
1676 return;
1677}
1678
1679/**
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001680 * scsi_noretry_cmd - determine if command should be failed fast
Mike Christie4a274462008-08-19 18:45:31 -05001681 * @scmd: SCSI cmd to examine.
1682 */
1683int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1684{
1685 switch (host_byte(scmd->result)) {
1686 case DID_OK:
1687 break;
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001688 case DID_TIME_OUT:
1689 goto check_type;
Mike Christie4a274462008-08-19 18:45:31 -05001690 case DID_BUS_BUSY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001691 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
Mike Christie4a274462008-08-19 18:45:31 -05001692 case DID_PARITY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001693 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
Mike Christie4a274462008-08-19 18:45:31 -05001694 case DID_ERROR:
1695 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1696 status_byte(scmd->result) == RESERVATION_CONFLICT)
1697 return 0;
1698 /* fall through */
1699 case DID_SOFT_ERROR:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001700 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
Mike Christie4a274462008-08-19 18:45:31 -05001701 }
1702
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001703 if (status_byte(scmd->result) != CHECK_CONDITION)
1704 return 0;
Mike Christie4a274462008-08-19 18:45:31 -05001705
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001706check_type:
1707 /*
1708 * assume caller has checked sense and determined
1709 * the check condition was retryable.
1710 */
1711 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1712 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1713 return 1;
1714 else
1715 return 0;
Mike Christie4a274462008-08-19 18:45:31 -05001716}
1717
1718/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1720 * @scmd: SCSI cmd to examine.
1721 *
1722 * Notes:
1723 * This is *only* called when we are examining the status after sending
1724 * out the actual data command. any commands that are queued for error
1725 * recovery (e.g. test_unit_ready) do *not* come through here.
1726 *
1727 * When this routine returns failed, it means the error handler thread
1728 * is woken. In cases where the error code indicates an error that
1729 * doesn't require the error handler read (i.e. we don't need to
1730 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001731 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732int scsi_decide_disposition(struct scsi_cmnd *scmd)
1733{
1734 int rtn;
1735
1736 /*
1737 * if the device is offline, then we clearly just pass the result back
1738 * up to the top level.
1739 */
1740 if (!scsi_device_online(scmd->device)) {
1741 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1742 " as SUCCESS\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001743 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 return SUCCESS;
1745 }
1746
1747 /*
1748 * first check the host byte, to see if there is anything in there
1749 * that would indicate what we need to do.
1750 */
1751 switch (host_byte(scmd->result)) {
1752 case DID_PASSTHROUGH:
1753 /*
1754 * no matter what, pass this through to the upper layer.
1755 * nuke this special code so that it looks like we are saying
1756 * did_ok.
1757 */
1758 scmd->result &= 0xff00ffff;
1759 return SUCCESS;
1760 case DID_OK:
1761 /*
1762 * looks good. drop through, and check the next byte.
1763 */
1764 break;
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001765 case DID_ABORT:
1766 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1767 scmd->result |= DID_TIME_OUT << 16;
1768 return SUCCESS;
1769 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 case DID_NO_CONNECT:
1771 case DID_BAD_TARGET:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 /*
1773 * note - this means that we just report the status back
1774 * to the top level driver, not that we actually think
1775 * that it indicates SUCCESS.
1776 */
1777 return SUCCESS;
1778 /*
1779 * when the low level driver returns did_soft_error,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001780 * it is responsible for keeping an internal retry counter
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 * in order to avoid endless loops (db)
1782 *
1783 * actually this is a bug in this function here. we should
1784 * be mindful of the maximum number of retries specified
1785 * and not get stuck in a loop.
1786 */
1787 case DID_SOFT_ERROR:
1788 goto maybe_retry;
1789 case DID_IMM_RETRY:
1790 return NEEDS_RETRY;
1791
bf341912005-04-12 17:49:09 -05001792 case DID_REQUEUE:
1793 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001794 case DID_TRANSPORT_DISRUPTED:
1795 /*
1796 * LLD/transport was disrupted during processing of the IO.
1797 * The transport class is now blocked/blocking,
1798 * and the transport will decide what to do with the IO
Mike Christie939c2282008-11-04 19:47:19 -06001799 * based on its timers and recovery capablilities if
1800 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001801 */
Mike Christie939c2282008-11-04 19:47:19 -06001802 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001803 case DID_TRANSPORT_FAILFAST:
1804 /*
1805 * The transport decided to failfast the IO (most likely
1806 * the fast io fail tmo fired), so send IO directly upwards.
1807 */
1808 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 case DID_ERROR:
1810 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1811 status_byte(scmd->result) == RESERVATION_CONFLICT)
1812 /*
1813 * execute reservation conflict processing code
1814 * lower down
1815 */
1816 break;
1817 /* fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 case DID_BUS_BUSY:
1819 case DID_PARITY:
1820 goto maybe_retry;
1821 case DID_TIME_OUT:
1822 /*
1823 * when we scan the bus, we get timeout messages for
1824 * these commands if there is no device available.
1825 * other hosts report did_no_connect for the same thing.
1826 */
1827 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1828 scmd->cmnd[0] == INQUIRY)) {
1829 return SUCCESS;
1830 } else {
1831 return FAILED;
1832 }
1833 case DID_RESET:
1834 return SUCCESS;
1835 default:
1836 return FAILED;
1837 }
1838
1839 /*
1840 * next, check the message byte.
1841 */
1842 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1843 return FAILED;
1844
1845 /*
1846 * check the status byte to see if this indicates anything special.
1847 */
1848 switch (status_byte(scmd->result)) {
1849 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001850 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 /*
1852 * the case of trying to send too many commands to a
1853 * tagged queueing device.
1854 */
1855 case BUSY:
1856 /*
1857 * device can't talk to us at the moment. Should only
1858 * occur (SAM-3) when the task queue is empty, so will cause
1859 * the empty queue handling to trigger a stall in the
1860 * device.
1861 */
1862 return ADD_TO_MLQUEUE;
1863 case GOOD:
Ewan D. Milne279afdf2013-08-08 15:07:48 -04001864 if (scmd->cmnd[0] == REPORT_LUNS)
1865 scmd->device->sdev_target->expecting_lun_change = 0;
Vasu Dev4a840672009-10-22 15:46:33 -07001866 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001869 case TASK_ABORTED:
1870 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 case CHECK_CONDITION:
1872 rtn = scsi_check_sense(scmd);
1873 if (rtn == NEEDS_RETRY)
1874 goto maybe_retry;
1875 /* if rtn == FAILED, we have no sense information;
1876 * returning FAILED will wake the error handler thread
1877 * to collect the sense and redo the decide
1878 * disposition */
1879 return rtn;
1880 case CONDITION_GOOD:
1881 case INTERMEDIATE_GOOD:
1882 case INTERMEDIATE_C_GOOD:
1883 case ACA_ACTIVE:
1884 /*
1885 * who knows? FIXME(eric)
1886 */
1887 return SUCCESS;
1888
1889 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001890 sdev_printk(KERN_INFO, scmd->device,
1891 "reservation conflict\n");
Moger, Babu2082ebc2012-01-24 20:38:46 +00001892 set_host_byte(scmd, DID_NEXUS_FAILURE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 return SUCCESS; /* causes immediate i/o error */
1894 default:
1895 return FAILED;
1896 }
1897 return FAILED;
1898
1899 maybe_retry:
1900
1901 /* we requeue for retry because the error was retryable, and
1902 * the request was not marked fast fail. Note that above,
1903 * even if the request is marked fast fail, we still requeue
1904 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001905 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001906 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 return NEEDS_RETRY;
1908 } else {
1909 /*
1910 * no more retries - report this one back to upper level.
1911 */
1912 return SUCCESS;
1913 }
1914}
1915
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001916static void eh_lock_door_done(struct request *req, int uptodate)
1917{
1918 __blk_put_request(req->q, req);
1919}
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 * scsi_eh_lock_door - Prevent medium removal for the specified device
1923 * @sdev: SCSI device to prevent medium removal
1924 *
1925 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001926 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 *
1928 * Notes:
1929 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1930 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001931 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932static void scsi_eh_lock_door(struct scsi_device *sdev)
1933{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001934 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
James Bottomley91bc31f2009-05-17 09:30:48 -05001936 /*
1937 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1938 * request becomes available
1939 */
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001940 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001942 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1943 req->cmd[1] = 0;
1944 req->cmd[2] = 0;
1945 req->cmd[3] = 0;
1946 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1947 req->cmd[5] = 0;
1948
1949 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1950
1951 req->cmd_type = REQ_TYPE_BLOCK_PC;
1952 req->cmd_flags |= REQ_QUIET;
1953 req->timeout = 10 * HZ;
1954 req->retries = 5;
1955
1956 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957}
1958
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959/**
1960 * scsi_restart_operations - restart io operations to the specified host.
1961 * @shost: Host we are restarting.
1962 *
1963 * Notes:
1964 * When we entered the error handler, we blocked all further i/o to
1965 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001966 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967static void scsi_restart_operations(struct Scsi_Host *shost)
1968{
1969 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001970 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971
1972 /*
1973 * If the door was locked, we need to insert a door lock request
1974 * onto the head of the SCSI request queue for the device. There
1975 * is no point trying to lock the door of an off-line device.
1976 */
1977 shost_for_each_device(sdev, shost) {
1978 if (scsi_device_online(sdev) && sdev->locked)
1979 scsi_eh_lock_door(sdev);
1980 }
1981
1982 /*
1983 * next free up anything directly waiting upon the host. this
1984 * will be requests for character device operations, and also for
1985 * ioctls to queued block devices.
1986 */
Hannes Reineckeb4562022013-10-23 10:51:21 +02001987 SCSI_LOG_ERROR_RECOVERY(3,
1988 printk("scsi_eh_%d waking up host to restart\n",
1989 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
James Bottomley939647e2005-09-18 15:05:20 -05001991 spin_lock_irqsave(shost->host_lock, flags);
1992 if (scsi_host_set_state(shost, SHOST_RUNNING))
1993 if (scsi_host_set_state(shost, SHOST_CANCEL))
1994 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1995 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996
1997 wake_up(&shost->host_wait);
1998
1999 /*
2000 * finally we need to re-initiate requests that may be pending. we will
2001 * have had everything blocked while error handling is taking place, and
2002 * now that error recovery is done, we will need to ensure that these
2003 * requests are started.
2004 */
2005 scsi_run_host_queues(shost);
Dan Williams57fc2e32012-06-21 23:25:32 -07002006
2007 /*
2008 * if eh is active and host_eh_scheduled is pending we need to re-run
2009 * recovery. we do this check after scsi_run_host_queues() to allow
2010 * everything pent up since the last eh run a chance to make forward
2011 * progress before we sync again. Either we'll immediately re-run
2012 * recovery or scsi_device_unbusy() will wake us again when these
2013 * pending commands complete.
2014 */
2015 spin_lock_irqsave(shost->host_lock, flags);
2016 if (shost->host_eh_scheduled)
2017 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2018 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2019 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020}
2021
2022/**
2023 * scsi_eh_ready_devs - check device ready state and recover if not.
2024 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05002025 * @work_q: &list_head for pending commands.
2026 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002027 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08002028void scsi_eh_ready_devs(struct Scsi_Host *shost,
2029 struct list_head *work_q,
2030 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031{
2032 if (!scsi_eh_stu(shost, work_q, done_q))
2033 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06002034 if (!scsi_eh_target_reset(shost, work_q, done_q))
2035 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2036 if (!scsi_eh_host_reset(work_q, done_q))
2037 scsi_eh_offline_sdevs(work_q,
2038 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08002040EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
2042/**
2043 * scsi_eh_flush_done_q - finish processed commands or retry them.
2044 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002045 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09002046void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02002048 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02002050 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2051 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05002053 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06002054 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
2056 " retry cmd: %p\n",
2057 current->comm,
2058 scmd));
2059 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2060 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07002061 /*
2062 * If just we got sense for the device (called
2063 * scsi_eh_get_sense), scmd->result is already
2064 * set, do not set DRIVER_TIMEOUT.
2065 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 if (!scmd->result)
2067 scmd->result |= (DRIVER_TIMEOUT << 24);
2068 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
2069 " cmd: %p\n",
2070 current->comm, scmd));
2071 scsi_finish_command(scmd);
2072 }
2073 }
2074}
Tejun Heo041c5fc2006-01-23 13:09:36 +09002075EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076
2077/**
2078 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2079 * @shost: Host to unjam.
2080 *
2081 * Notes:
2082 * When we come in here, we *know* that all commands on the bus have
2083 * either completed, failed or timed out. we also know that no further
2084 * commands are being sent to the host, so things are relatively quiet
2085 * and we have freedom to fiddle with things as we wish.
2086 *
2087 * This is only the *default* implementation. it is possible for
2088 * individual drivers to supply their own version of this function, and
2089 * if the maintainer wishes to do this, it is strongly suggested that
2090 * this function be taken as a template and modified. this function
2091 * was designed to correctly handle problems for about 95% of the
2092 * different cases out there, and it should always provide at least a
2093 * reasonable amount of error recovery.
2094 *
2095 * Any command marked 'failed' or 'timeout' must eventually have
2096 * scsi_finish_cmd() called for it. we do all of the retry stuff
2097 * here, so when we restart the host after we return it should have an
2098 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002099 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100static void scsi_unjam_host(struct Scsi_Host *shost)
2101{
2102 unsigned long flags;
2103 LIST_HEAD(eh_work_q);
2104 LIST_HEAD(eh_done_q);
2105
2106 spin_lock_irqsave(shost->host_lock, flags);
2107 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2108 spin_unlock_irqrestore(shost->host_lock, flags);
2109
2110 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2111
2112 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2113 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2114 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2115
Hannes Reineckeb4562022013-10-23 10:51:21 +02002116 spin_lock_irqsave(shost->host_lock, flags);
Ren Mingxinbb3b6212013-11-11 13:44:56 +01002117 if (shost->eh_deadline != -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +02002118 shost->last_reset = 0;
2119 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 scsi_eh_flush_done_q(&eh_done_q);
2121}
2122
2123/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002124 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 * @data: Host for which we are running.
2126 *
2127 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002128 * This is the main error handling loop. This is run as a kernel thread
2129 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002130 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131int scsi_error_handler(void *data)
2132{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002133 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002136 * We use TASK_INTERRUPTIBLE so that the thread is not
2137 * counted against the load average as a running process.
2138 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01002139 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 */
James Bottomley3ed7a472005-09-19 09:50:04 -05002141 while (!kthread_should_stop()) {
Dan Williamsb9d5c6b2012-06-21 23:25:42 -07002142 set_current_state(TASK_INTERRUPTIBLE);
Tejun Heoee7863b2006-05-15 20:57:20 +09002143 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05002144 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002145 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reineckeb4562022013-10-23 10:51:21 +02002146 printk("scsi_eh_%d: sleeping\n",
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002147 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05002148 schedule();
James Bottomley3ed7a472005-09-19 09:50:04 -05002149 continue;
2150 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151
James Bottomley3ed7a472005-09-19 09:50:04 -05002152 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002153 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reineckeb4562022013-10-23 10:51:21 +02002154 printk("scsi_eh_%d: waking up %d/%d/%d\n",
2155 shost->host_no, shost->host_eh_scheduled,
2156 shost->host_failed, shost->host_busy));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 /*
2159 * We have a host that is failing for some reason. Figure out
2160 * what we need to do to get it up and online again (if we can).
2161 * If we fail, we end up taking the thing offline.
2162 */
Lin Mingae0751f2011-12-05 09:20:24 +08002163 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
Alan Sternbc4f2402010-06-17 10:41:42 -04002164 SCSI_LOG_ERROR_RECOVERY(1,
2165 printk(KERN_ERR "Error handler scsi_eh_%d "
2166 "unable to autoresume\n",
2167 shost->host_no));
2168 continue;
2169 }
2170
Christoph Hellwig9227c332006-04-01 19:21:04 +02002171 if (shost->transportt->eh_strategy_handler)
2172 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 else
2174 scsi_unjam_host(shost);
2175
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 /*
2177 * Note - if the above fails completely, the action is to take
2178 * individual devices offline and flush the queue of any
2179 * outstanding requests that may have been pending. When we
2180 * restart, we restart any I/O to any other devices on the bus
2181 * which are still online.
2182 */
2183 scsi_restart_operations(shost);
Lin Mingae0751f2011-12-05 09:20:24 +08002184 if (!shost->eh_noresume)
2185 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04002187 __set_current_state(TASK_RUNNING);
2188
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002189 SCSI_LOG_ERROR_RECOVERY(1,
2190 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05002191 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 return 0;
2193}
2194
2195/*
2196 * Function: scsi_report_bus_reset()
2197 *
2198 * Purpose: Utility function used by low-level drivers to report that
2199 * they have observed a bus reset on the bus being handled.
2200 *
2201 * Arguments: shost - Host in question
2202 * channel - channel on which reset was observed.
2203 *
2204 * Returns: Nothing
2205 *
2206 * Lock status: Host lock must be held.
2207 *
2208 * Notes: This only needs to be called if the reset is one which
2209 * originates from an unknown location. Resets originated
2210 * by the mid-level itself don't need to call this, but there
2211 * should be no harm.
2212 *
2213 * The main purpose of this is to make sure that a CHECK_CONDITION
2214 * is properly treated.
2215 */
2216void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2217{
2218 struct scsi_device *sdev;
2219
2220 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06002221 if (channel == sdev_channel(sdev))
2222 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 }
2224}
2225EXPORT_SYMBOL(scsi_report_bus_reset);
2226
2227/*
2228 * Function: scsi_report_device_reset()
2229 *
2230 * Purpose: Utility function used by low-level drivers to report that
2231 * they have observed a device reset on the device being handled.
2232 *
2233 * Arguments: shost - Host in question
2234 * channel - channel on which reset was observed
2235 * target - target on which reset was observed
2236 *
2237 * Returns: Nothing
2238 *
2239 * Lock status: Host lock must be held
2240 *
2241 * Notes: This only needs to be called if the reset is one which
2242 * originates from an unknown location. Resets originated
2243 * by the mid-level itself don't need to call this, but there
2244 * should be no harm.
2245 *
2246 * The main purpose of this is to make sure that a CHECK_CONDITION
2247 * is properly treated.
2248 */
2249void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2250{
2251 struct scsi_device *sdev;
2252
2253 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04002254 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06002255 target == sdev_id(sdev))
2256 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 }
2258}
2259EXPORT_SYMBOL(scsi_report_device_reset);
2260
2261static void
2262scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2263{
2264}
2265
2266/*
2267 * Function: scsi_reset_provider
2268 *
2269 * Purpose: Send requested reset to a bus or device at any phase.
2270 *
2271 * Arguments: device - device to send reset to
2272 * flag - reset type (see scsi.h)
2273 *
2274 * Returns: SUCCESS/FAILURE.
2275 *
2276 * Notes: This is used by the SCSI Generic driver to provide
2277 * Bus/Device reset capability.
2278 */
2279int
2280scsi_reset_provider(struct scsi_device *dev, int flag)
2281{
Alan Sternbc4f2402010-06-17 10:41:42 -04002282 struct scsi_cmnd *scmd;
James Smartd7a1bb02006-03-08 14:50:12 -05002283 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05002285 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 int rtn;
2287
Alan Sternbc4f2402010-06-17 10:41:42 -04002288 if (scsi_autopm_get_host(shost) < 0)
2289 return FAILED;
2290
2291 scmd = scsi_get_command(dev, GFP_KERNEL);
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02002292 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02002294
Boaz Harrosh64a87b22008-04-30 11:19:47 +03002295 scmd->cmnd = req.cmd;
2296
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02002298 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299
2300 scmd->cmd_len = 0;
2301
2302 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303
James Smartd7a1bb02006-03-08 14:50:12 -05002304 spin_lock_irqsave(shost->host_lock, flags);
2305 shost->tmf_in_progress = 1;
2306 spin_unlock_irqrestore(shost->host_lock, flags);
2307
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 switch (flag) {
2309 case SCSI_TRY_RESET_DEVICE:
2310 rtn = scsi_try_bus_device_reset(scmd);
2311 if (rtn == SUCCESS)
2312 break;
2313 /* FALLTHROUGH */
Mike Christie30bd7df2008-02-29 18:25:19 -06002314 case SCSI_TRY_RESET_TARGET:
2315 rtn = scsi_try_target_reset(scmd);
2316 if (rtn == SUCCESS)
2317 break;
2318 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 case SCSI_TRY_RESET_BUS:
2320 rtn = scsi_try_bus_reset(scmd);
2321 if (rtn == SUCCESS)
2322 break;
2323 /* FALLTHROUGH */
2324 case SCSI_TRY_RESET_HOST:
2325 rtn = scsi_try_host_reset(scmd);
2326 break;
2327 default:
2328 rtn = FAILED;
2329 }
2330
James Smartd7a1bb02006-03-08 14:50:12 -05002331 spin_lock_irqsave(shost->host_lock, flags);
2332 shost->tmf_in_progress = 0;
2333 spin_unlock_irqrestore(shost->host_lock, flags);
2334
2335 /*
2336 * be sure to wake up anyone who was sleeping or had their queue
2337 * suspended while we performed the TMF.
2338 */
2339 SCSI_LOG_ERROR_RECOVERY(3,
2340 printk("%s: waking up host to restart after TMF\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07002341 __func__));
James Smartd7a1bb02006-03-08 14:50:12 -05002342
2343 wake_up(&shost->host_wait);
2344
2345 scsi_run_host_queues(shost);
2346
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 scsi_next_command(scmd);
Alan Sternbc4f2402010-06-17 10:41:42 -04002348 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 return rtn;
2350}
2351EXPORT_SYMBOL(scsi_reset_provider);
2352
2353/**
2354 * scsi_normalize_sense - normalize main elements from either fixed or
2355 * descriptor sense data format into a common format.
2356 *
2357 * @sense_buffer: byte array containing sense data returned by device
2358 * @sb_len: number of valid bytes in sense_buffer
2359 * @sshdr: pointer to instance of structure that common
2360 * elements are written to.
2361 *
2362 * Notes:
2363 * The "main elements" from sense data are: response_code, sense_key,
2364 * asc, ascq and additional_length (only for descriptor format).
2365 *
2366 * Typically this function can be called after a device has
2367 * responded to a SCSI command with the CHECK_CONDITION status.
2368 *
2369 * Return value:
2370 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002371 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2373 struct scsi_sense_hdr *sshdr)
2374{
James Bottomley33aa6872005-08-28 11:31:14 -05002375 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 return 0;
2377
2378 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2379
2380 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05002381
2382 if (!scsi_sense_valid(sshdr))
2383 return 0;
2384
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 if (sshdr->response_code >= 0x72) {
2386 /*
2387 * descriptor format
2388 */
2389 if (sb_len > 1)
2390 sshdr->sense_key = (sense_buffer[1] & 0xf);
2391 if (sb_len > 2)
2392 sshdr->asc = sense_buffer[2];
2393 if (sb_len > 3)
2394 sshdr->ascq = sense_buffer[3];
2395 if (sb_len > 7)
2396 sshdr->additional_length = sense_buffer[7];
2397 } else {
Jesper Juhl0bf8c862011-03-21 20:47:31 +01002398 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 * fixed format
2400 */
2401 if (sb_len > 2)
2402 sshdr->sense_key = (sense_buffer[2] & 0xf);
2403 if (sb_len > 7) {
2404 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2405 sb_len : (sense_buffer[7] + 8);
2406 if (sb_len > 12)
2407 sshdr->asc = sense_buffer[12];
2408 if (sb_len > 13)
2409 sshdr->ascq = sense_buffer[13];
2410 }
2411 }
2412
2413 return 1;
2414}
2415EXPORT_SYMBOL(scsi_normalize_sense);
2416
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2418 struct scsi_sense_hdr *sshdr)
2419{
2420 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09002421 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422}
2423EXPORT_SYMBOL(scsi_command_normalize_sense);
2424
2425/**
Rob Landleyeb448202007-11-03 13:30:39 -05002426 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 * @sense_buffer: byte array of descriptor format sense data
2428 * @sb_len: number of valid bytes in sense_buffer
2429 * @desc_type: value of descriptor type to find
2430 * (e.g. 0 -> information)
2431 *
2432 * Notes:
2433 * only valid when sense data is in descriptor format
2434 *
2435 * Return value:
2436 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002437 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2439 int desc_type)
2440{
2441 int add_sen_len, add_len, desc_len, k;
2442 const u8 * descp;
2443
2444 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2445 return NULL;
2446 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2447 return NULL;
2448 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2449 add_sen_len : (sb_len - 8);
2450 descp = &sense_buffer[8];
2451 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2452 descp += desc_len;
2453 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2454 desc_len = add_len + 2;
2455 if (descp[0] == desc_type)
2456 return descp;
2457 if (add_len < 0) // short descriptor ??
2458 break;
2459 }
2460 return NULL;
2461}
2462EXPORT_SYMBOL(scsi_sense_desc_find);
2463
2464/**
Rob Landleyeb448202007-11-03 13:30:39 -05002465 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 * @sense_buffer: byte array of sense data
2467 * @sb_len: number of valid bytes in sense_buffer
2468 * @info_out: pointer to 64 integer where 8 or 4 byte information
2469 * field will be placed if found.
2470 *
2471 * Return value:
2472 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002473 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2475 u64 * info_out)
2476{
2477 int j;
2478 const u8 * ucp;
2479 u64 ull;
2480
2481 if (sb_len < 7)
2482 return 0;
2483 switch (sense_buffer[0] & 0x7f) {
2484 case 0x70:
2485 case 0x71:
2486 if (sense_buffer[0] & 0x80) {
2487 *info_out = (sense_buffer[3] << 24) +
2488 (sense_buffer[4] << 16) +
2489 (sense_buffer[5] << 8) + sense_buffer[6];
2490 return 1;
2491 } else
2492 return 0;
2493 case 0x72:
2494 case 0x73:
2495 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2496 0 /* info desc */);
2497 if (ucp && (0xa == ucp[1])) {
2498 ull = 0;
2499 for (j = 0; j < 8; ++j) {
2500 if (j > 0)
2501 ull <<= 8;
2502 ull |= ucp[4 + j];
2503 }
2504 *info_out = ull;
2505 return 1;
2506 } else
2507 return 0;
2508 default:
2509 return 0;
2510 }
2511}
2512EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002513
2514/**
2515 * scsi_build_sense_buffer - build sense data in a buffer
2516 * @desc: Sense format (non zero == descriptor format,
2517 * 0 == fixed format)
2518 * @buf: Where to build sense data
2519 * @key: Sense key
2520 * @asc: Additional sense code
2521 * @ascq: Additional sense code qualifier
2522 *
2523 **/
2524void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2525{
2526 if (desc) {
2527 buf[0] = 0x72; /* descriptor, current */
2528 buf[1] = key;
2529 buf[2] = asc;
2530 buf[3] = ascq;
2531 buf[7] = 0;
2532 } else {
2533 buf[0] = 0x70; /* fixed, current */
2534 buf[2] = key;
2535 buf[7] = 0xa;
2536 buf[12] = asc;
2537 buf[13] = ascq;
2538 }
2539}
2540EXPORT_SYMBOL(scsi_build_sense_buffer);