1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * scsi_error.c Copyright (C) 1997 Eric Youngdale |
4 | * |
5 | * SCSI error/timeout handling |
6 | * Initial versions: Eric Youngdale. Based upon conversations with |
7 | * Leonard Zubkoff and David Miller at Linux Expo, |
8 | * ideas originating from all over the place. |
9 | * |
10 | * Restructured scsi_unjam_host and associated functions. |
11 | * September 04, 2002 Mike Anderson (andmike@us.ibm.com) |
12 | * |
13 | * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and |
14 | * minor cleanups. |
15 | * September 30, 2002 Mike Anderson (andmike@us.ibm.com) |
16 | */ |
17 | |
18 | #include <linux/module.h> |
19 | #include <linux/sched.h> |
20 | #include <linux/gfp.h> |
21 | #include <linux/timer.h> |
22 | #include <linux/string.h> |
23 | #include <linux/kernel.h> |
24 | #include <linux/freezer.h> |
25 | #include <linux/kthread.h> |
26 | #include <linux/interrupt.h> |
27 | #include <linux/blkdev.h> |
28 | #include <linux/delay.h> |
29 | #include <linux/jiffies.h> |
30 | |
31 | #include <scsi/scsi.h> |
32 | #include <scsi/scsi_cmnd.h> |
33 | #include <scsi/scsi_dbg.h> |
34 | #include <scsi/scsi_device.h> |
35 | #include <scsi/scsi_driver.h> |
36 | #include <scsi/scsi_eh.h> |
37 | #include <scsi/scsi_common.h> |
38 | #include <scsi/scsi_transport.h> |
39 | #include <scsi/scsi_host.h> |
40 | #include <scsi/scsi_ioctl.h> |
41 | #include <scsi/scsi_dh.h> |
42 | #include <scsi/scsi_devinfo.h> |
43 | #include <scsi/sg.h> |
44 | |
45 | #include "scsi_priv.h" |
46 | #include "scsi_logging.h" |
47 | #include "scsi_transport_api.h" |
48 | |
49 | #include <trace/events/scsi.h> |
50 | |
51 | #include <asm/unaligned.h> |
52 | |
53 | /* |
54 | * These should *probably* be handled by the host itself. |
55 | * Since it is allowed to sleep, it probably should. |
56 | */ |
57 | #define BUS_RESET_SETTLE_TIME (10) |
58 | #define HOST_RESET_SETTLE_TIME (10) |
59 | |
60 | static int scsi_eh_try_stu(struct scsi_cmnd *scmd); |
61 | static enum scsi_disposition scsi_try_to_abort_cmd(const struct scsi_host_template *, |
62 | struct scsi_cmnd *); |
63 | |
64 | void scsi_eh_wakeup(struct Scsi_Host *shost) |
65 | { |
66 | lockdep_assert_held(shost->host_lock); |
67 | |
68 | if (scsi_host_busy(shost) == shost->host_failed) { |
69 | trace_scsi_eh_wakeup(shost); |
70 | wake_up_process(tsk: shost->ehandler); |
71 | SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, |
72 | "Waking error handler thread\n" )); |
73 | } |
74 | } |
75 | |
76 | /** |
77 | * scsi_schedule_eh - schedule EH for SCSI host |
78 | * @shost: SCSI host to invoke error handling on. |
79 | * |
80 | * Schedule SCSI EH without scmd. |
81 | */ |
82 | void scsi_schedule_eh(struct Scsi_Host *shost) |
83 | { |
84 | unsigned long flags; |
85 | |
86 | spin_lock_irqsave(shost->host_lock, flags); |
87 | |
88 | if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || |
89 | scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { |
90 | shost->host_eh_scheduled++; |
91 | scsi_eh_wakeup(shost); |
92 | } |
93 | |
94 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
95 | } |
96 | EXPORT_SYMBOL_GPL(scsi_schedule_eh); |
97 | |
98 | static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) |
99 | { |
100 | if (!shost->last_reset || shost->eh_deadline == -1) |
101 | return 0; |
102 | |
103 | /* |
104 | * 32bit accesses are guaranteed to be atomic |
105 | * (on all supported architectures), so instead |
106 | * of using a spinlock we can as well double check |
107 | * if eh_deadline has been set to 'off' during the |
108 | * time_before call. |
109 | */ |
110 | if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && |
111 | shost->eh_deadline > -1) |
112 | return 0; |
113 | |
114 | return 1; |
115 | } |
116 | |
117 | static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd) |
118 | { |
119 | if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) |
120 | return true; |
121 | |
122 | return ++cmd->retries <= cmd->allowed; |
123 | } |
124 | |
125 | static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd) |
126 | { |
127 | struct scsi_device *sdev = cmd->device; |
128 | struct Scsi_Host *host = sdev->host; |
129 | |
130 | if (host->hostt->eh_should_retry_cmd) |
131 | return host->hostt->eh_should_retry_cmd(cmd); |
132 | |
133 | return true; |
134 | } |
135 | |
136 | /** |
137 | * scmd_eh_abort_handler - Handle command aborts |
138 | * @work: command to be aborted. |
139 | * |
140 | * Note: this function must be called only for a command that has timed out. |
141 | * Because the block layer marks a request as complete before it calls |
142 | * scsi_timeout(), a .scsi_done() call from the LLD for a command that has |
143 | * timed out do not have any effect. Hence it is safe to call |
144 | * scsi_finish_command() from this function. |
145 | */ |
146 | void |
147 | scmd_eh_abort_handler(struct work_struct *work) |
148 | { |
149 | struct scsi_cmnd *scmd = |
150 | container_of(work, struct scsi_cmnd, abort_work.work); |
151 | struct scsi_device *sdev = scmd->device; |
152 | struct Scsi_Host *shost = sdev->host; |
153 | enum scsi_disposition rtn; |
154 | unsigned long flags; |
155 | |
156 | if (scsi_host_eh_past_deadline(shost)) { |
157 | SCSI_LOG_ERROR_RECOVERY(3, |
158 | scmd_printk(KERN_INFO, scmd, |
159 | "eh timeout, not aborting\n" )); |
160 | goto out; |
161 | } |
162 | |
163 | SCSI_LOG_ERROR_RECOVERY(3, |
164 | scmd_printk(KERN_INFO, scmd, |
165 | "aborting command\n" )); |
166 | rtn = scsi_try_to_abort_cmd(shost->hostt, scmd); |
167 | if (rtn != SUCCESS) { |
168 | SCSI_LOG_ERROR_RECOVERY(3, |
169 | scmd_printk(KERN_INFO, scmd, |
170 | "cmd abort %s\n" , |
171 | (rtn == FAST_IO_FAIL) ? |
172 | "not send" : "failed" )); |
173 | goto out; |
174 | } |
175 | set_host_byte(cmd: scmd, status: DID_TIME_OUT); |
176 | if (scsi_host_eh_past_deadline(shost)) { |
177 | SCSI_LOG_ERROR_RECOVERY(3, |
178 | scmd_printk(KERN_INFO, scmd, |
179 | "eh timeout, not retrying " |
180 | "aborted command\n" )); |
181 | goto out; |
182 | } |
183 | |
184 | spin_lock_irqsave(shost->host_lock, flags); |
185 | list_del_init(entry: &scmd->eh_entry); |
186 | |
187 | /* |
188 | * If the abort succeeds, and there is no further |
189 | * EH action, clear the ->last_reset time. |
190 | */ |
191 | if (list_empty(head: &shost->eh_abort_list) && |
192 | list_empty(head: &shost->eh_cmd_q)) |
193 | if (shost->eh_deadline != -1) |
194 | shost->last_reset = 0; |
195 | |
196 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
197 | |
198 | if (!scsi_noretry_cmd(scmd) && |
199 | scsi_cmd_retry_allowed(cmd: scmd) && |
200 | scsi_eh_should_retry_cmd(cmd: scmd)) { |
201 | SCSI_LOG_ERROR_RECOVERY(3, |
202 | scmd_printk(KERN_WARNING, scmd, |
203 | "retry aborted command\n" )); |
204 | scsi_queue_insert(cmd: scmd, SCSI_MLQUEUE_EH_RETRY); |
205 | } else { |
206 | SCSI_LOG_ERROR_RECOVERY(3, |
207 | scmd_printk(KERN_WARNING, scmd, |
208 | "finish aborted command\n" )); |
209 | scsi_finish_command(cmd: scmd); |
210 | } |
211 | return; |
212 | |
213 | out: |
214 | spin_lock_irqsave(shost->host_lock, flags); |
215 | list_del_init(entry: &scmd->eh_entry); |
216 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
217 | |
218 | scsi_eh_scmd_add(scmd); |
219 | } |
220 | |
221 | /** |
222 | * scsi_abort_command - schedule a command abort |
223 | * @scmd: scmd to abort. |
224 | * |
225 | * We only need to abort commands after a command timeout |
226 | */ |
227 | static int |
228 | scsi_abort_command(struct scsi_cmnd *scmd) |
229 | { |
230 | struct scsi_device *sdev = scmd->device; |
231 | struct Scsi_Host *shost = sdev->host; |
232 | unsigned long flags; |
233 | |
234 | if (!shost->hostt->eh_abort_handler) { |
235 | /* No abort handler, fail command directly */ |
236 | return FAILED; |
237 | } |
238 | |
239 | if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { |
240 | /* |
241 | * Retry after abort failed, escalate to next level. |
242 | */ |
243 | SCSI_LOG_ERROR_RECOVERY(3, |
244 | scmd_printk(KERN_INFO, scmd, |
245 | "previous abort failed\n" )); |
246 | BUG_ON(delayed_work_pending(&scmd->abort_work)); |
247 | return FAILED; |
248 | } |
249 | |
250 | spin_lock_irqsave(shost->host_lock, flags); |
251 | if (shost->eh_deadline != -1 && !shost->last_reset) |
252 | shost->last_reset = jiffies; |
253 | BUG_ON(!list_empty(&scmd->eh_entry)); |
254 | list_add_tail(new: &scmd->eh_entry, head: &shost->eh_abort_list); |
255 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
256 | |
257 | scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; |
258 | SCSI_LOG_ERROR_RECOVERY(3, |
259 | scmd_printk(KERN_INFO, scmd, "abort scheduled\n" )); |
260 | queue_delayed_work(wq: shost->tmf_work_q, dwork: &scmd->abort_work, HZ / 100); |
261 | return SUCCESS; |
262 | } |
263 | |
264 | /** |
265 | * scsi_eh_reset - call into ->eh_action to reset internal counters |
266 | * @scmd: scmd to run eh on. |
267 | * |
268 | * The scsi driver might be carrying internal state about the |
269 | * devices, so we need to call into the driver to reset the |
270 | * internal state once the error handler is started. |
271 | */ |
272 | static void scsi_eh_reset(struct scsi_cmnd *scmd) |
273 | { |
274 | if (!blk_rq_is_passthrough(rq: scsi_cmd_to_rq(scmd))) { |
275 | struct scsi_driver *sdrv = scsi_cmd_to_driver(cmd: scmd); |
276 | if (sdrv->eh_reset) |
277 | sdrv->eh_reset(scmd); |
278 | } |
279 | } |
280 | |
281 | static void scsi_eh_inc_host_failed(struct rcu_head *head) |
282 | { |
283 | struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu); |
284 | struct Scsi_Host *shost = scmd->device->host; |
285 | unsigned long flags; |
286 | |
287 | spin_lock_irqsave(shost->host_lock, flags); |
288 | shost->host_failed++; |
289 | scsi_eh_wakeup(shost); |
290 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
291 | } |
292 | |
293 | /** |
294 | * scsi_eh_scmd_add - add scsi cmd to error handling. |
295 | * @scmd: scmd to run eh on. |
296 | */ |
297 | void scsi_eh_scmd_add(struct scsi_cmnd *scmd) |
298 | { |
299 | struct Scsi_Host *shost = scmd->device->host; |
300 | unsigned long flags; |
301 | int ret; |
302 | |
303 | WARN_ON_ONCE(!shost->ehandler); |
304 | |
305 | spin_lock_irqsave(shost->host_lock, flags); |
306 | if (scsi_host_set_state(shost, SHOST_RECOVERY)) { |
307 | ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); |
308 | WARN_ON_ONCE(ret); |
309 | } |
310 | if (shost->eh_deadline != -1 && !shost->last_reset) |
311 | shost->last_reset = jiffies; |
312 | |
313 | scsi_eh_reset(scmd); |
314 | list_add_tail(new: &scmd->eh_entry, head: &shost->eh_cmd_q); |
315 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
316 | /* |
317 | * Ensure that all tasks observe the host state change before the |
318 | * host_failed change. |
319 | */ |
320 | call_rcu_hurry(head: &scmd->rcu, func: scsi_eh_inc_host_failed); |
321 | } |
322 | |
323 | /** |
324 | * scsi_timeout - Timeout function for normal scsi commands. |
325 | * @req: request that is timing out. |
326 | * |
327 | * Notes: |
328 | * We do not need to lock this. There is the potential for a race |
329 | * only in that the normal completion handling might run, but if the |
330 | * normal completion function determines that the timer has already |
331 | * fired, then it mustn't do anything. |
332 | */ |
333 | enum blk_eh_timer_return scsi_timeout(struct request *req) |
334 | { |
335 | struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(rq: req); |
336 | struct Scsi_Host *host = scmd->device->host; |
337 | |
338 | trace_scsi_dispatch_cmd_timeout(cmd: scmd); |
339 | scsi_log_completion(cmd: scmd, disposition: TIMEOUT_ERROR); |
340 | |
341 | atomic_inc(v: &scmd->device->iotmo_cnt); |
342 | if (host->eh_deadline != -1 && !host->last_reset) |
343 | host->last_reset = jiffies; |
344 | |
345 | if (host->hostt->eh_timed_out) { |
346 | switch (host->hostt->eh_timed_out(scmd)) { |
347 | case SCSI_EH_DONE: |
348 | return BLK_EH_DONE; |
349 | case SCSI_EH_RESET_TIMER: |
350 | return BLK_EH_RESET_TIMER; |
351 | case SCSI_EH_NOT_HANDLED: |
352 | break; |
353 | } |
354 | } |
355 | |
356 | /* |
357 | * If scsi_done() has already set SCMD_STATE_COMPLETE, do not modify |
358 | * *scmd. |
359 | */ |
360 | if (test_and_set_bit(SCMD_STATE_COMPLETE, addr: &scmd->state)) |
361 | return BLK_EH_DONE; |
362 | atomic_inc(v: &scmd->device->iodone_cnt); |
363 | if (scsi_abort_command(scmd) != SUCCESS) { |
364 | set_host_byte(cmd: scmd, status: DID_TIME_OUT); |
365 | scsi_eh_scmd_add(scmd); |
366 | } |
367 | |
368 | return BLK_EH_DONE; |
369 | } |
370 | |
371 | /** |
372 | * scsi_block_when_processing_errors - Prevent cmds from being queued. |
373 | * @sdev: Device on which we are performing recovery. |
374 | * |
375 | * Description: |
376 | * We block until the host is out of error recovery, and then check to |
377 | * see whether the host or the device is offline. |
378 | * |
379 | * Return value: |
380 | * 0 when dev was taken offline by error recovery. 1 OK to proceed. |
381 | */ |
382 | int scsi_block_when_processing_errors(struct scsi_device *sdev) |
383 | { |
384 | int online; |
385 | |
386 | wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); |
387 | |
388 | online = scsi_device_online(sdev); |
389 | |
390 | return online; |
391 | } |
392 | EXPORT_SYMBOL(scsi_block_when_processing_errors); |
393 | |
394 | #ifdef CONFIG_SCSI_LOGGING |
395 | /** |
396 | * scsi_eh_prt_fail_stats - Log info on failures. |
397 | * @shost: scsi host being recovered. |
398 | * @work_q: Queue of scsi cmds to process. |
399 | */ |
400 | static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, |
401 | struct list_head *work_q) |
402 | { |
403 | struct scsi_cmnd *scmd; |
404 | struct scsi_device *sdev; |
405 | int total_failures = 0; |
406 | int cmd_failed = 0; |
407 | int cmd_cancel = 0; |
408 | int devices_failed = 0; |
409 | |
410 | shost_for_each_device(sdev, shost) { |
411 | list_for_each_entry(scmd, work_q, eh_entry) { |
412 | if (scmd->device == sdev) { |
413 | ++total_failures; |
414 | if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) |
415 | ++cmd_cancel; |
416 | else |
417 | ++cmd_failed; |
418 | } |
419 | } |
420 | |
421 | if (cmd_cancel || cmd_failed) { |
422 | SCSI_LOG_ERROR_RECOVERY(3, |
423 | shost_printk(KERN_INFO, shost, |
424 | "%s: cmds failed: %d, cancel: %d\n" , |
425 | __func__, cmd_failed, |
426 | cmd_cancel)); |
427 | cmd_cancel = 0; |
428 | cmd_failed = 0; |
429 | ++devices_failed; |
430 | } |
431 | } |
432 | |
433 | SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, |
434 | "Total of %d commands on %d" |
435 | " devices require eh work\n" , |
436 | total_failures, devices_failed)); |
437 | } |
438 | #endif |
439 | |
440 | /** |
441 | * scsi_report_lun_change - Set flag on all *other* devices on the same target |
442 | * to indicate that a UNIT ATTENTION is expected. |
443 | * @sdev: Device reporting the UNIT ATTENTION |
444 | */ |
445 | static void scsi_report_lun_change(struct scsi_device *sdev) |
446 | { |
447 | sdev->sdev_target->expecting_lun_change = 1; |
448 | } |
449 | |
450 | /** |
451 | * scsi_report_sense - Examine scsi sense information and log messages for |
452 | * certain conditions, also issue uevents for some of them. |
453 | * @sdev: Device reporting the sense code |
454 | * @sshdr: sshdr to be examined |
455 | */ |
456 | static void scsi_report_sense(struct scsi_device *sdev, |
457 | struct scsi_sense_hdr *sshdr) |
458 | { |
459 | enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ |
460 | |
461 | if (sshdr->sense_key == UNIT_ATTENTION) { |
462 | if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { |
463 | evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; |
464 | sdev_printk(KERN_WARNING, sdev, |
465 | "Inquiry data has changed" ); |
466 | } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { |
467 | evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; |
468 | scsi_report_lun_change(sdev); |
469 | sdev_printk(KERN_WARNING, sdev, |
470 | "LUN assignments on this target have " |
471 | "changed. The Linux SCSI layer does not " |
472 | "automatically remap LUN assignments.\n" ); |
473 | } else if (sshdr->asc == 0x3f) |
474 | sdev_printk(KERN_WARNING, sdev, |
475 | "Operating parameters on this target have " |
476 | "changed. The Linux SCSI layer does not " |
477 | "automatically adjust these parameters.\n" ); |
478 | |
479 | if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { |
480 | evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; |
481 | sdev_printk(KERN_WARNING, sdev, |
482 | "Warning! Received an indication that the " |
483 | "LUN reached a thin provisioning soft " |
484 | "threshold.\n" ); |
485 | } |
486 | |
487 | if (sshdr->asc == 0x29) { |
488 | evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; |
489 | /* |
490 | * Do not print message if it is an expected side-effect |
491 | * of runtime PM. |
492 | */ |
493 | if (!sdev->silence_suspend) |
494 | sdev_printk(KERN_WARNING, sdev, |
495 | "Power-on or device reset occurred\n" ); |
496 | } |
497 | |
498 | if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { |
499 | evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; |
500 | sdev_printk(KERN_WARNING, sdev, |
501 | "Mode parameters changed" ); |
502 | } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { |
503 | evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; |
504 | sdev_printk(KERN_WARNING, sdev, |
505 | "Asymmetric access state changed" ); |
506 | } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { |
507 | evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; |
508 | sdev_printk(KERN_WARNING, sdev, |
509 | "Capacity data has changed" ); |
510 | } else if (sshdr->asc == 0x2a) |
511 | sdev_printk(KERN_WARNING, sdev, |
512 | "Parameters changed" ); |
513 | } |
514 | |
515 | if (evt_type != SDEV_EVT_MAXBITS) { |
516 | set_bit(nr: evt_type, addr: sdev->pending_events); |
517 | schedule_work(work: &sdev->event_work); |
518 | } |
519 | } |
520 | |
521 | static inline void set_scsi_ml_byte(struct scsi_cmnd *cmd, u8 status) |
522 | { |
523 | cmd->result = (cmd->result & 0xffff00ff) | (status << 8); |
524 | } |
525 | |
526 | /** |
527 | * scsi_check_sense - Examine scsi cmd sense |
528 | * @scmd: Cmd to have sense checked. |
529 | * |
530 | * Return value: |
531 | * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE |
532 | * |
533 | * Notes: |
534 | * When a deferred error is detected the current command has |
535 | * not been executed and needs retrying. |
536 | */ |
537 | enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd) |
538 | { |
539 | struct request *req = scsi_cmd_to_rq(scmd); |
540 | struct scsi_device *sdev = scmd->device; |
541 | struct scsi_sense_hdr sshdr; |
542 | |
543 | if (! scsi_command_normalize_sense(cmd: scmd, sshdr: &sshdr)) |
544 | return FAILED; /* no valid sense data */ |
545 | |
546 | scsi_report_sense(sdev, sshdr: &sshdr); |
547 | |
548 | if (scsi_sense_is_deferred(sshdr: &sshdr)) |
549 | return NEEDS_RETRY; |
550 | |
551 | if (sdev->handler && sdev->handler->check_sense) { |
552 | enum scsi_disposition rc; |
553 | |
554 | rc = sdev->handler->check_sense(sdev, &sshdr); |
555 | if (rc != SCSI_RETURN_NOT_HANDLED) |
556 | return rc; |
557 | /* handler does not care. Drop down to default handling */ |
558 | } |
559 | |
560 | if (scmd->cmnd[0] == TEST_UNIT_READY && |
561 | scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER) |
562 | /* |
563 | * nasty: for mid-layer issued TURs, we need to return the |
564 | * actual sense data without any recovery attempt. For eh |
565 | * issued ones, we need to try to recover and interpret |
566 | */ |
567 | return SUCCESS; |
568 | |
569 | /* |
570 | * Previous logic looked for FILEMARK, EOM or ILI which are |
571 | * mainly associated with tapes and returned SUCCESS. |
572 | */ |
573 | if (sshdr.response_code == 0x70) { |
574 | /* fixed format */ |
575 | if (scmd->sense_buffer[2] & 0xe0) |
576 | return SUCCESS; |
577 | } else { |
578 | /* |
579 | * descriptor format: look for "stream commands sense data |
580 | * descriptor" (see SSC-3). Assume single sense data |
581 | * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. |
582 | */ |
583 | if ((sshdr.additional_length > 3) && |
584 | (scmd->sense_buffer[8] == 0x4) && |
585 | (scmd->sense_buffer[11] & 0xe0)) |
586 | return SUCCESS; |
587 | } |
588 | |
589 | switch (sshdr.sense_key) { |
590 | case NO_SENSE: |
591 | return SUCCESS; |
592 | case RECOVERED_ERROR: |
593 | return /* soft_error */ SUCCESS; |
594 | |
595 | case ABORTED_COMMAND: |
596 | if (sshdr.asc == 0x10) /* DIF */ |
597 | return SUCCESS; |
598 | |
599 | /* |
600 | * Check aborts due to command duration limit policy: |
601 | * ABORTED COMMAND additional sense code with the |
602 | * COMMAND TIMEOUT BEFORE PROCESSING or |
603 | * COMMAND TIMEOUT DURING PROCESSING or |
604 | * COMMAND TIMEOUT DURING PROCESSING DUE TO ERROR RECOVERY |
605 | * additional sense code qualifiers. |
606 | */ |
607 | if (sshdr.asc == 0x2e && |
608 | sshdr.ascq >= 0x01 && sshdr.ascq <= 0x03) { |
609 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_DL_TIMEOUT); |
610 | req->cmd_flags |= REQ_FAILFAST_DEV; |
611 | req->rq_flags |= RQF_QUIET; |
612 | return SUCCESS; |
613 | } |
614 | |
615 | if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF) |
616 | return ADD_TO_MLQUEUE; |
617 | if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 && |
618 | sdev->sdev_bflags & BLIST_RETRY_ASC_C1) |
619 | return ADD_TO_MLQUEUE; |
620 | |
621 | return NEEDS_RETRY; |
622 | case NOT_READY: |
623 | case UNIT_ATTENTION: |
624 | /* |
625 | * if we are expecting a cc/ua because of a bus reset that we |
626 | * performed, treat this just as a retry. otherwise this is |
627 | * information that we should pass up to the upper-level driver |
628 | * so that we can deal with it there. |
629 | */ |
630 | if (scmd->device->expecting_cc_ua) { |
631 | /* |
632 | * Because some device does not queue unit |
633 | * attentions correctly, we carefully check |
634 | * additional sense code and qualifier so as |
635 | * not to squash media change unit attention. |
636 | */ |
637 | if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { |
638 | scmd->device->expecting_cc_ua = 0; |
639 | return NEEDS_RETRY; |
640 | } |
641 | } |
642 | /* |
643 | * we might also expect a cc/ua if another LUN on the target |
644 | * reported a UA with an ASC/ASCQ of 3F 0E - |
645 | * REPORTED LUNS DATA HAS CHANGED. |
646 | */ |
647 | if (scmd->device->sdev_target->expecting_lun_change && |
648 | sshdr.asc == 0x3f && sshdr.ascq == 0x0e) |
649 | return NEEDS_RETRY; |
650 | /* |
651 | * if the device is in the process of becoming ready, we |
652 | * should retry. |
653 | */ |
654 | if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) |
655 | return NEEDS_RETRY; |
656 | /* |
657 | * if the device is not started, we need to wake |
658 | * the error handler to start the motor |
659 | */ |
660 | if (scmd->device->allow_restart && |
661 | (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) |
662 | return FAILED; |
663 | /* |
664 | * Pass the UA upwards for a determination in the completion |
665 | * functions. |
666 | */ |
667 | return SUCCESS; |
668 | |
669 | /* these are not supported */ |
670 | case DATA_PROTECT: |
671 | if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { |
672 | /* Thin provisioning hard threshold reached */ |
673 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_NOSPC); |
674 | return SUCCESS; |
675 | } |
676 | fallthrough; |
677 | case COPY_ABORTED: |
678 | case VOLUME_OVERFLOW: |
679 | case MISCOMPARE: |
680 | case BLANK_CHECK: |
681 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_TGT_FAILURE); |
682 | return SUCCESS; |
683 | |
684 | case MEDIUM_ERROR: |
685 | if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ |
686 | sshdr.asc == 0x13 || /* AMNF DATA FIELD */ |
687 | sshdr.asc == 0x14) { /* RECORD NOT FOUND */ |
688 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_MED_ERROR); |
689 | return SUCCESS; |
690 | } |
691 | return NEEDS_RETRY; |
692 | |
693 | case HARDWARE_ERROR: |
694 | if (scmd->device->retry_hwerror) |
695 | return ADD_TO_MLQUEUE; |
696 | else |
697 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_TGT_FAILURE); |
698 | fallthrough; |
699 | |
700 | case ILLEGAL_REQUEST: |
701 | if (sshdr.asc == 0x20 || /* Invalid command operation code */ |
702 | sshdr.asc == 0x21 || /* Logical block address out of range */ |
703 | sshdr.asc == 0x22 || /* Invalid function */ |
704 | sshdr.asc == 0x24 || /* Invalid field in cdb */ |
705 | sshdr.asc == 0x26 || /* Parameter value invalid */ |
706 | sshdr.asc == 0x27) { /* Write protected */ |
707 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_TGT_FAILURE); |
708 | } |
709 | return SUCCESS; |
710 | |
711 | case COMPLETED: |
712 | if (sshdr.asc == 0x55 && sshdr.ascq == 0x0a) { |
713 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_DL_TIMEOUT); |
714 | req->cmd_flags |= REQ_FAILFAST_DEV; |
715 | req->rq_flags |= RQF_QUIET; |
716 | } |
717 | return SUCCESS; |
718 | |
719 | default: |
720 | return SUCCESS; |
721 | } |
722 | } |
723 | EXPORT_SYMBOL_GPL(scsi_check_sense); |
724 | |
725 | static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) |
726 | { |
727 | const struct scsi_host_template *sht = sdev->host->hostt; |
728 | struct scsi_device *tmp_sdev; |
729 | |
730 | if (!sht->track_queue_depth || |
731 | sdev->queue_depth >= sdev->max_queue_depth) |
732 | return; |
733 | |
734 | if (time_before(jiffies, |
735 | sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) |
736 | return; |
737 | |
738 | if (time_before(jiffies, |
739 | sdev->last_queue_full_time + sdev->queue_ramp_up_period)) |
740 | return; |
741 | |
742 | /* |
743 | * Walk all devices of a target and do |
744 | * ramp up on them. |
745 | */ |
746 | shost_for_each_device(tmp_sdev, sdev->host) { |
747 | if (tmp_sdev->channel != sdev->channel || |
748 | tmp_sdev->id != sdev->id || |
749 | tmp_sdev->queue_depth == sdev->max_queue_depth) |
750 | continue; |
751 | |
752 | scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); |
753 | sdev->last_queue_ramp_up = jiffies; |
754 | } |
755 | } |
756 | |
757 | static void scsi_handle_queue_full(struct scsi_device *sdev) |
758 | { |
759 | const struct scsi_host_template *sht = sdev->host->hostt; |
760 | struct scsi_device *tmp_sdev; |
761 | |
762 | if (!sht->track_queue_depth) |
763 | return; |
764 | |
765 | shost_for_each_device(tmp_sdev, sdev->host) { |
766 | if (tmp_sdev->channel != sdev->channel || |
767 | tmp_sdev->id != sdev->id) |
768 | continue; |
769 | /* |
770 | * We do not know the number of commands that were at |
771 | * the device when we got the queue full so we start |
772 | * from the highest possible value and work our way down. |
773 | */ |
774 | scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); |
775 | } |
776 | } |
777 | |
778 | /** |
779 | * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. |
780 | * @scmd: SCSI cmd to examine. |
781 | * |
782 | * Notes: |
783 | * This is *only* called when we are examining the status of commands |
784 | * queued during error recovery. the main difference here is that we |
785 | * don't allow for the possibility of retries here, and we are a lot |
786 | * more restrictive about what we consider acceptable. |
787 | */ |
788 | static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd) |
789 | { |
790 | /* |
791 | * first check the host byte, to see if there is anything in there |
792 | * that would indicate what we need to do. |
793 | */ |
794 | if (host_byte(scmd->result) == DID_RESET) { |
795 | /* |
796 | * rats. we are already in the error handler, so we now |
797 | * get to try and figure out what to do next. if the sense |
798 | * is valid, we have a pretty good idea of what to do. |
799 | * if not, we mark it as FAILED. |
800 | */ |
801 | return scsi_check_sense(scmd); |
802 | } |
803 | if (host_byte(scmd->result) != DID_OK) |
804 | return FAILED; |
805 | |
806 | /* |
807 | * now, check the status byte to see if this indicates |
808 | * anything special. |
809 | */ |
810 | switch (get_status_byte(cmd: scmd)) { |
811 | case SAM_STAT_GOOD: |
812 | scsi_handle_queue_ramp_up(sdev: scmd->device); |
813 | if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd)) |
814 | /* |
815 | * If we have sense data, call scsi_check_sense() in |
816 | * order to set the correct SCSI ML byte (if any). |
817 | * No point in checking the return value, since the |
818 | * command has already completed successfully. |
819 | */ |
820 | scsi_check_sense(scmd); |
821 | fallthrough; |
822 | case SAM_STAT_COMMAND_TERMINATED: |
823 | return SUCCESS; |
824 | case SAM_STAT_CHECK_CONDITION: |
825 | return scsi_check_sense(scmd); |
826 | case SAM_STAT_CONDITION_MET: |
827 | case SAM_STAT_INTERMEDIATE: |
828 | case SAM_STAT_INTERMEDIATE_CONDITION_MET: |
829 | /* |
830 | * who knows? FIXME(eric) |
831 | */ |
832 | return SUCCESS; |
833 | case SAM_STAT_RESERVATION_CONFLICT: |
834 | if (scmd->cmnd[0] == TEST_UNIT_READY) |
835 | /* it is a success, we probed the device and |
836 | * found it */ |
837 | return SUCCESS; |
838 | /* otherwise, we failed to send the command */ |
839 | return FAILED; |
840 | case SAM_STAT_TASK_SET_FULL: |
841 | scsi_handle_queue_full(sdev: scmd->device); |
842 | fallthrough; |
843 | case SAM_STAT_BUSY: |
844 | return NEEDS_RETRY; |
845 | default: |
846 | return FAILED; |
847 | } |
848 | return FAILED; |
849 | } |
850 | |
851 | /** |
852 | * scsi_eh_done - Completion function for error handling. |
853 | * @scmd: Cmd that is done. |
854 | */ |
855 | void scsi_eh_done(struct scsi_cmnd *scmd) |
856 | { |
857 | struct completion *eh_action; |
858 | |
859 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
860 | "%s result: %x\n" , __func__, scmd->result)); |
861 | |
862 | eh_action = scmd->device->host->eh_action; |
863 | if (eh_action) |
864 | complete(eh_action); |
865 | } |
866 | |
867 | /** |
868 | * scsi_try_host_reset - ask host adapter to reset itself |
869 | * @scmd: SCSI cmd to send host reset. |
870 | */ |
871 | static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd) |
872 | { |
873 | unsigned long flags; |
874 | enum scsi_disposition rtn; |
875 | struct Scsi_Host *host = scmd->device->host; |
876 | const struct scsi_host_template *hostt = host->hostt; |
877 | |
878 | SCSI_LOG_ERROR_RECOVERY(3, |
879 | shost_printk(KERN_INFO, host, "Snd Host RST\n" )); |
880 | |
881 | if (!hostt->eh_host_reset_handler) |
882 | return FAILED; |
883 | |
884 | rtn = hostt->eh_host_reset_handler(scmd); |
885 | |
886 | if (rtn == SUCCESS) { |
887 | if (!hostt->skip_settle_delay) |
888 | ssleep(HOST_RESET_SETTLE_TIME); |
889 | spin_lock_irqsave(host->host_lock, flags); |
890 | scsi_report_bus_reset(host, scmd_channel(scmd)); |
891 | spin_unlock_irqrestore(lock: host->host_lock, flags); |
892 | } |
893 | |
894 | return rtn; |
895 | } |
896 | |
897 | /** |
898 | * scsi_try_bus_reset - ask host to perform a bus reset |
899 | * @scmd: SCSI cmd to send bus reset. |
900 | */ |
901 | static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd) |
902 | { |
903 | unsigned long flags; |
904 | enum scsi_disposition rtn; |
905 | struct Scsi_Host *host = scmd->device->host; |
906 | const struct scsi_host_template *hostt = host->hostt; |
907 | |
908 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
909 | "%s: Snd Bus RST\n" , __func__)); |
910 | |
911 | if (!hostt->eh_bus_reset_handler) |
912 | return FAILED; |
913 | |
914 | rtn = hostt->eh_bus_reset_handler(scmd); |
915 | |
916 | if (rtn == SUCCESS) { |
917 | if (!hostt->skip_settle_delay) |
918 | ssleep(BUS_RESET_SETTLE_TIME); |
919 | spin_lock_irqsave(host->host_lock, flags); |
920 | scsi_report_bus_reset(host, scmd_channel(scmd)); |
921 | spin_unlock_irqrestore(lock: host->host_lock, flags); |
922 | } |
923 | |
924 | return rtn; |
925 | } |
926 | |
927 | static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) |
928 | { |
929 | sdev->was_reset = 1; |
930 | sdev->expecting_cc_ua = 1; |
931 | } |
932 | |
933 | /** |
934 | * scsi_try_target_reset - Ask host to perform a target reset |
935 | * @scmd: SCSI cmd used to send a target reset |
936 | * |
937 | * Notes: |
938 | * There is no timeout for this operation. if this operation is |
939 | * unreliable for a given host, then the host itself needs to put a |
940 | * timer on it, and set the host back to a consistent state prior to |
941 | * returning. |
942 | */ |
943 | static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd) |
944 | { |
945 | unsigned long flags; |
946 | enum scsi_disposition rtn; |
947 | struct Scsi_Host *host = scmd->device->host; |
948 | const struct scsi_host_template *hostt = host->hostt; |
949 | |
950 | if (!hostt->eh_target_reset_handler) |
951 | return FAILED; |
952 | |
953 | rtn = hostt->eh_target_reset_handler(scmd); |
954 | if (rtn == SUCCESS) { |
955 | spin_lock_irqsave(host->host_lock, flags); |
956 | __starget_for_each_device(scsi_target(sdev: scmd->device), NULL, |
957 | fn: __scsi_report_device_reset); |
958 | spin_unlock_irqrestore(lock: host->host_lock, flags); |
959 | } |
960 | |
961 | return rtn; |
962 | } |
963 | |
964 | /** |
965 | * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev |
966 | * @scmd: SCSI cmd used to send BDR |
967 | * |
968 | * Notes: |
969 | * There is no timeout for this operation. if this operation is |
970 | * unreliable for a given host, then the host itself needs to put a |
971 | * timer on it, and set the host back to a consistent state prior to |
972 | * returning. |
973 | */ |
974 | static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd) |
975 | { |
976 | enum scsi_disposition rtn; |
977 | const struct scsi_host_template *hostt = scmd->device->host->hostt; |
978 | |
979 | if (!hostt->eh_device_reset_handler) |
980 | return FAILED; |
981 | |
982 | rtn = hostt->eh_device_reset_handler(scmd); |
983 | if (rtn == SUCCESS) |
984 | __scsi_report_device_reset(sdev: scmd->device, NULL); |
985 | return rtn; |
986 | } |
987 | |
988 | /** |
989 | * scsi_try_to_abort_cmd - Ask host to abort a SCSI command |
990 | * @hostt: SCSI driver host template |
991 | * @scmd: SCSI cmd used to send a target reset |
992 | * |
993 | * Return value: |
994 | * SUCCESS, FAILED, or FAST_IO_FAIL |
995 | * |
996 | * Notes: |
997 | * SUCCESS does not necessarily indicate that the command |
998 | * has been aborted; it only indicates that the LLDDs |
999 | * has cleared all references to that command. |
1000 | * LLDDs should return FAILED only if an abort was required |
1001 | * but could not be executed. LLDDs should return FAST_IO_FAIL |
1002 | * if the device is temporarily unavailable (eg due to a |
1003 | * link down on FibreChannel) |
1004 | */ |
1005 | static enum scsi_disposition |
1006 | scsi_try_to_abort_cmd(const struct scsi_host_template *hostt, struct scsi_cmnd *scmd) |
1007 | { |
1008 | if (!hostt->eh_abort_handler) |
1009 | return FAILED; |
1010 | |
1011 | return hostt->eh_abort_handler(scmd); |
1012 | } |
1013 | |
1014 | static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) |
1015 | { |
1016 | if (scsi_try_to_abort_cmd(hostt: scmd->device->host->hostt, scmd) != SUCCESS) |
1017 | if (scsi_try_bus_device_reset(scmd) != SUCCESS) |
1018 | if (scsi_try_target_reset(scmd) != SUCCESS) |
1019 | if (scsi_try_bus_reset(scmd) != SUCCESS) |
1020 | scsi_try_host_reset(scmd); |
1021 | } |
1022 | |
1023 | /** |
1024 | * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery |
1025 | * @scmd: SCSI command structure to hijack |
1026 | * @ses: structure to save restore information |
1027 | * @cmnd: CDB to send. Can be NULL if no new cmnd is needed |
1028 | * @cmnd_size: size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE) |
1029 | * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) |
1030 | * |
1031 | * This function is used to save a scsi command information before re-execution |
1032 | * as part of the error recovery process. If @sense_bytes is 0 the command |
1033 | * sent must be one that does not transfer any data. If @sense_bytes != 0 |
1034 | * @cmnd is ignored and this functions sets up a REQUEST_SENSE command |
1035 | * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. |
1036 | */ |
1037 | void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, |
1038 | unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) |
1039 | { |
1040 | struct scsi_device *sdev = scmd->device; |
1041 | |
1042 | /* |
1043 | * We need saved copies of a number of fields - this is because |
1044 | * error handling may need to overwrite these with different values |
1045 | * to run different commands, and once error handling is complete, |
1046 | * we will need to restore these values prior to running the actual |
1047 | * command. |
1048 | */ |
1049 | ses->cmd_len = scmd->cmd_len; |
1050 | ses->data_direction = scmd->sc_data_direction; |
1051 | ses->sdb = scmd->sdb; |
1052 | ses->result = scmd->result; |
1053 | ses->resid_len = scmd->resid_len; |
1054 | ses->underflow = scmd->underflow; |
1055 | ses->prot_op = scmd->prot_op; |
1056 | ses->eh_eflags = scmd->eh_eflags; |
1057 | |
1058 | scmd->prot_op = SCSI_PROT_NORMAL; |
1059 | scmd->eh_eflags = 0; |
1060 | memcpy(ses->cmnd, scmd->cmnd, sizeof(ses->cmnd)); |
1061 | memset(scmd->cmnd, 0, sizeof(scmd->cmnd)); |
1062 | memset(&scmd->sdb, 0, sizeof(scmd->sdb)); |
1063 | scmd->result = 0; |
1064 | scmd->resid_len = 0; |
1065 | |
1066 | if (sense_bytes) { |
1067 | scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, |
1068 | sense_bytes); |
1069 | sg_init_one(&ses->sense_sgl, scmd->sense_buffer, |
1070 | scmd->sdb.length); |
1071 | scmd->sdb.table.sgl = &ses->sense_sgl; |
1072 | scmd->sc_data_direction = DMA_FROM_DEVICE; |
1073 | scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; |
1074 | scmd->cmnd[0] = REQUEST_SENSE; |
1075 | scmd->cmnd[4] = scmd->sdb.length; |
1076 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
1077 | } else { |
1078 | scmd->sc_data_direction = DMA_NONE; |
1079 | if (cmnd) { |
1080 | BUG_ON(cmnd_size > sizeof(scmd->cmnd)); |
1081 | memcpy(scmd->cmnd, cmnd, cmnd_size); |
1082 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
1083 | } |
1084 | } |
1085 | |
1086 | scmd->underflow = 0; |
1087 | |
1088 | if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) |
1089 | scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | |
1090 | (sdev->lun << 5 & 0xe0); |
1091 | |
1092 | /* |
1093 | * Zero the sense buffer. The scsi spec mandates that any |
1094 | * untransferred sense data should be interpreted as being zero. |
1095 | */ |
1096 | memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); |
1097 | } |
1098 | EXPORT_SYMBOL(scsi_eh_prep_cmnd); |
1099 | |
1100 | /** |
1101 | * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery |
1102 | * @scmd: SCSI command structure to restore |
1103 | * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd |
1104 | * |
1105 | * Undo any damage done by above scsi_eh_prep_cmnd(). |
1106 | */ |
1107 | void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) |
1108 | { |
1109 | /* |
1110 | * Restore original data |
1111 | */ |
1112 | scmd->cmd_len = ses->cmd_len; |
1113 | memcpy(scmd->cmnd, ses->cmnd, sizeof(ses->cmnd)); |
1114 | scmd->sc_data_direction = ses->data_direction; |
1115 | scmd->sdb = ses->sdb; |
1116 | scmd->result = ses->result; |
1117 | scmd->resid_len = ses->resid_len; |
1118 | scmd->underflow = ses->underflow; |
1119 | scmd->prot_op = ses->prot_op; |
1120 | scmd->eh_eflags = ses->eh_eflags; |
1121 | } |
1122 | EXPORT_SYMBOL(scsi_eh_restore_cmnd); |
1123 | |
1124 | /** |
1125 | * scsi_send_eh_cmnd - submit a scsi command as part of error recovery |
1126 | * @scmd: SCSI command structure to hijack |
1127 | * @cmnd: CDB to send |
1128 | * @cmnd_size: size in bytes of @cmnd |
1129 | * @timeout: timeout for this request |
1130 | * @sense_bytes: size of sense data to copy or 0 |
1131 | * |
1132 | * This function is used to send a scsi command down to a target device |
1133 | * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. |
1134 | * |
1135 | * Return value: |
1136 | * SUCCESS or FAILED or NEEDS_RETRY |
1137 | */ |
1138 | static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd, |
1139 | unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes) |
1140 | { |
1141 | struct scsi_device *sdev = scmd->device; |
1142 | struct Scsi_Host *shost = sdev->host; |
1143 | DECLARE_COMPLETION_ONSTACK(done); |
1144 | unsigned long timeleft = timeout, delay; |
1145 | struct scsi_eh_save ses; |
1146 | const unsigned long stall_for = msecs_to_jiffies(m: 100); |
1147 | int rtn; |
1148 | |
1149 | retry: |
1150 | scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); |
1151 | shost->eh_action = &done; |
1152 | |
1153 | scsi_log_send(cmd: scmd); |
1154 | scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER; |
1155 | |
1156 | /* |
1157 | * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can |
1158 | * change the SCSI device state after we have examined it and before |
1159 | * .queuecommand() is called. |
1160 | */ |
1161 | mutex_lock(&sdev->state_mutex); |
1162 | while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) { |
1163 | mutex_unlock(lock: &sdev->state_mutex); |
1164 | SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev, |
1165 | "%s: state %d <> %d\n" , __func__, sdev->sdev_state, |
1166 | SDEV_BLOCK)); |
1167 | delay = min(timeleft, stall_for); |
1168 | timeleft -= delay; |
1169 | msleep(msecs: jiffies_to_msecs(j: delay)); |
1170 | mutex_lock(&sdev->state_mutex); |
1171 | } |
1172 | if (sdev->sdev_state != SDEV_BLOCK) |
1173 | rtn = shost->hostt->queuecommand(shost, scmd); |
1174 | else |
1175 | rtn = FAILED; |
1176 | mutex_unlock(lock: &sdev->state_mutex); |
1177 | |
1178 | if (rtn) { |
1179 | if (timeleft > stall_for) { |
1180 | scsi_eh_restore_cmnd(scmd, &ses); |
1181 | |
1182 | timeleft -= stall_for; |
1183 | msleep(msecs: jiffies_to_msecs(j: stall_for)); |
1184 | goto retry; |
1185 | } |
1186 | /* signal not to enter either branch of the if () below */ |
1187 | timeleft = 0; |
1188 | rtn = FAILED; |
1189 | } else { |
1190 | timeleft = wait_for_completion_timeout(x: &done, timeout); |
1191 | rtn = SUCCESS; |
1192 | } |
1193 | |
1194 | shost->eh_action = NULL; |
1195 | |
1196 | scsi_log_completion(cmd: scmd, disposition: rtn); |
1197 | |
1198 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
1199 | "%s timeleft: %ld\n" , |
1200 | __func__, timeleft)); |
1201 | |
1202 | /* |
1203 | * If there is time left scsi_eh_done got called, and we will examine |
1204 | * the actual status codes to see whether the command actually did |
1205 | * complete normally, else if we have a zero return and no time left, |
1206 | * the command must still be pending, so abort it and return FAILED. |
1207 | * If we never actually managed to issue the command, because |
1208 | * ->queuecommand() kept returning non zero, use the rtn = FAILED |
1209 | * value above (so don't execute either branch of the if) |
1210 | */ |
1211 | if (timeleft) { |
1212 | rtn = scsi_eh_completed_normally(scmd); |
1213 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
1214 | "%s: scsi_eh_completed_normally %x\n" , __func__, rtn)); |
1215 | |
1216 | switch (rtn) { |
1217 | case SUCCESS: |
1218 | case NEEDS_RETRY: |
1219 | case FAILED: |
1220 | break; |
1221 | case ADD_TO_MLQUEUE: |
1222 | rtn = NEEDS_RETRY; |
1223 | break; |
1224 | default: |
1225 | rtn = FAILED; |
1226 | break; |
1227 | } |
1228 | } else if (rtn != FAILED) { |
1229 | scsi_abort_eh_cmnd(scmd); |
1230 | rtn = FAILED; |
1231 | } |
1232 | |
1233 | scsi_eh_restore_cmnd(scmd, &ses); |
1234 | |
1235 | return rtn; |
1236 | } |
1237 | |
1238 | /** |
1239 | * scsi_request_sense - Request sense data from a particular target. |
1240 | * @scmd: SCSI cmd for request sense. |
1241 | * |
1242 | * Notes: |
1243 | * Some hosts automatically obtain this information, others require |
1244 | * that we obtain it on our own. This function will *not* return until |
1245 | * the command either times out, or it completes. |
1246 | */ |
1247 | static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd) |
1248 | { |
1249 | return scsi_send_eh_cmnd(scmd, NULL, cmnd_size: 0, timeout: scmd->device->eh_timeout, sense_bytes: ~0); |
1250 | } |
1251 | |
1252 | static enum scsi_disposition |
1253 | scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn) |
1254 | { |
1255 | if (!blk_rq_is_passthrough(rq: scsi_cmd_to_rq(scmd))) { |
1256 | struct scsi_driver *sdrv = scsi_cmd_to_driver(cmd: scmd); |
1257 | if (sdrv->eh_action) |
1258 | rtn = sdrv->eh_action(scmd, rtn); |
1259 | } |
1260 | return rtn; |
1261 | } |
1262 | |
1263 | /** |
1264 | * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. |
1265 | * @scmd: Original SCSI cmd that eh has finished. |
1266 | * @done_q: Queue for processed commands. |
1267 | * |
1268 | * Notes: |
1269 | * We don't want to use the normal command completion while we are are |
1270 | * still handling errors - it may cause other commands to be queued, |
1271 | * and that would disturb what we are doing. Thus we really want to |
1272 | * keep a list of pending commands for final completion, and once we |
1273 | * are ready to leave error handling we handle completion for real. |
1274 | */ |
1275 | void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) |
1276 | { |
1277 | list_move_tail(list: &scmd->eh_entry, head: done_q); |
1278 | } |
1279 | EXPORT_SYMBOL(scsi_eh_finish_cmd); |
1280 | |
1281 | /** |
1282 | * scsi_eh_get_sense - Get device sense data. |
1283 | * @work_q: Queue of commands to process. |
1284 | * @done_q: Queue of processed commands. |
1285 | * |
1286 | * Description: |
1287 | * See if we need to request sense information. if so, then get it |
1288 | * now, so we have a better idea of what to do. |
1289 | * |
1290 | * Notes: |
1291 | * This has the unfortunate side effect that if a shost adapter does |
1292 | * not automatically request sense information, we end up shutting |
1293 | * it down before we request it. |
1294 | * |
1295 | * All drivers should request sense information internally these days, |
1296 | * so for now all I have to say is tough noogies if you end up in here. |
1297 | * |
1298 | * XXX: Long term this code should go away, but that needs an audit of |
1299 | * all LLDDs first. |
1300 | */ |
1301 | int scsi_eh_get_sense(struct list_head *work_q, |
1302 | struct list_head *done_q) |
1303 | { |
1304 | struct scsi_cmnd *scmd, *next; |
1305 | struct Scsi_Host *shost; |
1306 | enum scsi_disposition rtn; |
1307 | |
1308 | /* |
1309 | * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, |
1310 | * should not get sense. |
1311 | */ |
1312 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
1313 | if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || |
1314 | SCSI_SENSE_VALID(scmd)) |
1315 | continue; |
1316 | |
1317 | shost = scmd->device->host; |
1318 | if (scsi_host_eh_past_deadline(shost)) { |
1319 | SCSI_LOG_ERROR_RECOVERY(3, |
1320 | scmd_printk(KERN_INFO, scmd, |
1321 | "%s: skip request sense, past eh deadline\n" , |
1322 | current->comm)); |
1323 | break; |
1324 | } |
1325 | if (!scsi_status_is_check_condition(status: scmd->result)) |
1326 | /* |
1327 | * don't request sense if there's no check condition |
1328 | * status because the error we're processing isn't one |
1329 | * that has a sense code (and some devices get |
1330 | * confused by sense requests out of the blue) |
1331 | */ |
1332 | continue; |
1333 | |
1334 | SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, |
1335 | "%s: requesting sense\n" , |
1336 | current->comm)); |
1337 | rtn = scsi_request_sense(scmd); |
1338 | if (rtn != SUCCESS) |
1339 | continue; |
1340 | |
1341 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
1342 | "sense requested, result %x\n" , scmd->result)); |
1343 | SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); |
1344 | |
1345 | rtn = scsi_decide_disposition(cmd: scmd); |
1346 | |
1347 | /* |
1348 | * if the result was normal, then just pass it along to the |
1349 | * upper level. |
1350 | */ |
1351 | if (rtn == SUCCESS) |
1352 | /* |
1353 | * We don't want this command reissued, just finished |
1354 | * with the sense data, so set retries to the max |
1355 | * allowed to ensure it won't get reissued. If the user |
1356 | * has requested infinite retries, we also want to |
1357 | * finish this command, so force completion by setting |
1358 | * retries and allowed to the same value. |
1359 | */ |
1360 | if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) |
1361 | scmd->retries = scmd->allowed = 1; |
1362 | else |
1363 | scmd->retries = scmd->allowed; |
1364 | else if (rtn != NEEDS_RETRY) |
1365 | continue; |
1366 | |
1367 | scsi_eh_finish_cmd(scmd, done_q); |
1368 | } |
1369 | |
1370 | return list_empty(head: work_q); |
1371 | } |
1372 | EXPORT_SYMBOL_GPL(scsi_eh_get_sense); |
1373 | |
1374 | /** |
1375 | * scsi_eh_tur - Send TUR to device. |
1376 | * @scmd: &scsi_cmnd to send TUR |
1377 | * |
1378 | * Return value: |
1379 | * 0 - Device is ready. 1 - Device NOT ready. |
1380 | */ |
1381 | static int scsi_eh_tur(struct scsi_cmnd *scmd) |
1382 | { |
1383 | static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; |
1384 | int retry_cnt = 1; |
1385 | enum scsi_disposition rtn; |
1386 | |
1387 | retry_tur: |
1388 | rtn = scsi_send_eh_cmnd(scmd, cmnd: tur_command, cmnd_size: 6, |
1389 | timeout: scmd->device->eh_timeout, sense_bytes: 0); |
1390 | |
1391 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
1392 | "%s return: %x\n" , __func__, rtn)); |
1393 | |
1394 | switch (rtn) { |
1395 | case NEEDS_RETRY: |
1396 | if (retry_cnt--) |
1397 | goto retry_tur; |
1398 | fallthrough; |
1399 | case SUCCESS: |
1400 | return 0; |
1401 | default: |
1402 | return 1; |
1403 | } |
1404 | } |
1405 | |
1406 | /** |
1407 | * scsi_eh_test_devices - check if devices are responding from error recovery. |
1408 | * @cmd_list: scsi commands in error recovery. |
1409 | * @work_q: queue for commands which still need more error recovery |
1410 | * @done_q: queue for commands which are finished |
1411 | * @try_stu: boolean on if a STU command should be tried in addition to TUR. |
1412 | * |
1413 | * Decription: |
1414 | * Tests if devices are in a working state. Commands to devices now in |
1415 | * a working state are sent to the done_q while commands to devices which |
1416 | * are still failing to respond are returned to the work_q for more |
1417 | * processing. |
1418 | **/ |
1419 | static int scsi_eh_test_devices(struct list_head *cmd_list, |
1420 | struct list_head *work_q, |
1421 | struct list_head *done_q, int try_stu) |
1422 | { |
1423 | struct scsi_cmnd *scmd, *next; |
1424 | struct scsi_device *sdev; |
1425 | int finish_cmds; |
1426 | |
1427 | while (!list_empty(head: cmd_list)) { |
1428 | scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); |
1429 | sdev = scmd->device; |
1430 | |
1431 | if (!try_stu) { |
1432 | if (scsi_host_eh_past_deadline(shost: sdev->host)) { |
1433 | /* Push items back onto work_q */ |
1434 | list_splice_init(list: cmd_list, head: work_q); |
1435 | SCSI_LOG_ERROR_RECOVERY(3, |
1436 | sdev_printk(KERN_INFO, sdev, |
1437 | "%s: skip test device, past eh deadline" , |
1438 | current->comm)); |
1439 | break; |
1440 | } |
1441 | } |
1442 | |
1443 | finish_cmds = !scsi_device_online(sdev: scmd->device) || |
1444 | (try_stu && !scsi_eh_try_stu(scmd) && |
1445 | !scsi_eh_tur(scmd)) || |
1446 | !scsi_eh_tur(scmd); |
1447 | |
1448 | list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) |
1449 | if (scmd->device == sdev) { |
1450 | if (finish_cmds && |
1451 | (try_stu || |
1452 | scsi_eh_action(scmd, rtn: SUCCESS) == SUCCESS)) |
1453 | scsi_eh_finish_cmd(scmd, done_q); |
1454 | else |
1455 | list_move_tail(list: &scmd->eh_entry, head: work_q); |
1456 | } |
1457 | } |
1458 | return list_empty(head: work_q); |
1459 | } |
1460 | |
1461 | /** |
1462 | * scsi_eh_try_stu - Send START_UNIT to device. |
1463 | * @scmd: &scsi_cmnd to send START_UNIT |
1464 | * |
1465 | * Return value: |
1466 | * 0 - Device is ready. 1 - Device NOT ready. |
1467 | */ |
1468 | static int scsi_eh_try_stu(struct scsi_cmnd *scmd) |
1469 | { |
1470 | static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; |
1471 | |
1472 | if (scmd->device->allow_restart) { |
1473 | int i; |
1474 | enum scsi_disposition rtn = NEEDS_RETRY; |
1475 | |
1476 | for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) |
1477 | rtn = scsi_send_eh_cmnd(scmd, cmnd: stu_command, cmnd_size: 6, |
1478 | timeout: scmd->device->eh_timeout, sense_bytes: 0); |
1479 | |
1480 | if (rtn == SUCCESS) |
1481 | return 0; |
1482 | } |
1483 | |
1484 | return 1; |
1485 | } |
1486 | |
1487 | /** |
1488 | * scsi_eh_stu - send START_UNIT if needed |
1489 | * @shost: &scsi host being recovered. |
1490 | * @work_q: &list_head for pending commands. |
1491 | * @done_q: &list_head for processed commands. |
1492 | * |
1493 | * Notes: |
1494 | * If commands are failing due to not ready, initializing command required, |
1495 | * try revalidating the device, which will end up sending a start unit. |
1496 | */ |
1497 | static int scsi_eh_stu(struct Scsi_Host *shost, |
1498 | struct list_head *work_q, |
1499 | struct list_head *done_q) |
1500 | { |
1501 | struct scsi_cmnd *scmd, *stu_scmd, *next; |
1502 | struct scsi_device *sdev; |
1503 | |
1504 | shost_for_each_device(sdev, shost) { |
1505 | if (scsi_host_eh_past_deadline(shost)) { |
1506 | SCSI_LOG_ERROR_RECOVERY(3, |
1507 | sdev_printk(KERN_INFO, sdev, |
1508 | "%s: skip START_UNIT, past eh deadline\n" , |
1509 | current->comm)); |
1510 | scsi_device_put(sdev); |
1511 | break; |
1512 | } |
1513 | stu_scmd = NULL; |
1514 | list_for_each_entry(scmd, work_q, eh_entry) |
1515 | if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && |
1516 | scsi_check_sense(scmd) == FAILED ) { |
1517 | stu_scmd = scmd; |
1518 | break; |
1519 | } |
1520 | |
1521 | if (!stu_scmd) |
1522 | continue; |
1523 | |
1524 | SCSI_LOG_ERROR_RECOVERY(3, |
1525 | sdev_printk(KERN_INFO, sdev, |
1526 | "%s: Sending START_UNIT\n" , |
1527 | current->comm)); |
1528 | |
1529 | if (!scsi_eh_try_stu(scmd: stu_scmd)) { |
1530 | if (!scsi_device_online(sdev) || |
1531 | !scsi_eh_tur(scmd: stu_scmd)) { |
1532 | list_for_each_entry_safe(scmd, next, |
1533 | work_q, eh_entry) { |
1534 | if (scmd->device == sdev && |
1535 | scsi_eh_action(scmd, rtn: SUCCESS) == SUCCESS) |
1536 | scsi_eh_finish_cmd(scmd, done_q); |
1537 | } |
1538 | } |
1539 | } else { |
1540 | SCSI_LOG_ERROR_RECOVERY(3, |
1541 | sdev_printk(KERN_INFO, sdev, |
1542 | "%s: START_UNIT failed\n" , |
1543 | current->comm)); |
1544 | } |
1545 | } |
1546 | |
1547 | return list_empty(head: work_q); |
1548 | } |
1549 | |
1550 | |
1551 | /** |
1552 | * scsi_eh_bus_device_reset - send bdr if needed |
1553 | * @shost: scsi host being recovered. |
1554 | * @work_q: &list_head for pending commands. |
1555 | * @done_q: &list_head for processed commands. |
1556 | * |
1557 | * Notes: |
1558 | * Try a bus device reset. Still, look to see whether we have multiple |
1559 | * devices that are jammed or not - if we have multiple devices, it |
1560 | * makes no sense to try bus_device_reset - we really would need to try |
1561 | * a bus_reset instead. |
1562 | */ |
1563 | static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, |
1564 | struct list_head *work_q, |
1565 | struct list_head *done_q) |
1566 | { |
1567 | struct scsi_cmnd *scmd, *bdr_scmd, *next; |
1568 | struct scsi_device *sdev; |
1569 | enum scsi_disposition rtn; |
1570 | |
1571 | shost_for_each_device(sdev, shost) { |
1572 | if (scsi_host_eh_past_deadline(shost)) { |
1573 | SCSI_LOG_ERROR_RECOVERY(3, |
1574 | sdev_printk(KERN_INFO, sdev, |
1575 | "%s: skip BDR, past eh deadline\n" , |
1576 | current->comm)); |
1577 | scsi_device_put(sdev); |
1578 | break; |
1579 | } |
1580 | bdr_scmd = NULL; |
1581 | list_for_each_entry(scmd, work_q, eh_entry) |
1582 | if (scmd->device == sdev) { |
1583 | bdr_scmd = scmd; |
1584 | break; |
1585 | } |
1586 | |
1587 | if (!bdr_scmd) |
1588 | continue; |
1589 | |
1590 | SCSI_LOG_ERROR_RECOVERY(3, |
1591 | sdev_printk(KERN_INFO, sdev, |
1592 | "%s: Sending BDR\n" , current->comm)); |
1593 | rtn = scsi_try_bus_device_reset(scmd: bdr_scmd); |
1594 | if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { |
1595 | if (!scsi_device_online(sdev) || |
1596 | rtn == FAST_IO_FAIL || |
1597 | !scsi_eh_tur(scmd: bdr_scmd)) { |
1598 | list_for_each_entry_safe(scmd, next, |
1599 | work_q, eh_entry) { |
1600 | if (scmd->device == sdev && |
1601 | scsi_eh_action(scmd, rtn) != FAILED) |
1602 | scsi_eh_finish_cmd(scmd, |
1603 | done_q); |
1604 | } |
1605 | } |
1606 | } else { |
1607 | SCSI_LOG_ERROR_RECOVERY(3, |
1608 | sdev_printk(KERN_INFO, sdev, |
1609 | "%s: BDR failed\n" , current->comm)); |
1610 | } |
1611 | } |
1612 | |
1613 | return list_empty(head: work_q); |
1614 | } |
1615 | |
1616 | /** |
1617 | * scsi_eh_target_reset - send target reset if needed |
1618 | * @shost: scsi host being recovered. |
1619 | * @work_q: &list_head for pending commands. |
1620 | * @done_q: &list_head for processed commands. |
1621 | * |
1622 | * Notes: |
1623 | * Try a target reset. |
1624 | */ |
1625 | static int scsi_eh_target_reset(struct Scsi_Host *shost, |
1626 | struct list_head *work_q, |
1627 | struct list_head *done_q) |
1628 | { |
1629 | LIST_HEAD(tmp_list); |
1630 | LIST_HEAD(check_list); |
1631 | |
1632 | list_splice_init(list: work_q, head: &tmp_list); |
1633 | |
1634 | while (!list_empty(head: &tmp_list)) { |
1635 | struct scsi_cmnd *next, *scmd; |
1636 | enum scsi_disposition rtn; |
1637 | unsigned int id; |
1638 | |
1639 | if (scsi_host_eh_past_deadline(shost)) { |
1640 | /* push back on work queue for further processing */ |
1641 | list_splice_init(list: &check_list, head: work_q); |
1642 | list_splice_init(list: &tmp_list, head: work_q); |
1643 | SCSI_LOG_ERROR_RECOVERY(3, |
1644 | shost_printk(KERN_INFO, shost, |
1645 | "%s: Skip target reset, past eh deadline\n" , |
1646 | current->comm)); |
1647 | return list_empty(head: work_q); |
1648 | } |
1649 | |
1650 | scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); |
1651 | id = scmd_id(scmd); |
1652 | |
1653 | SCSI_LOG_ERROR_RECOVERY(3, |
1654 | shost_printk(KERN_INFO, shost, |
1655 | "%s: Sending target reset to target %d\n" , |
1656 | current->comm, id)); |
1657 | rtn = scsi_try_target_reset(scmd); |
1658 | if (rtn != SUCCESS && rtn != FAST_IO_FAIL) |
1659 | SCSI_LOG_ERROR_RECOVERY(3, |
1660 | shost_printk(KERN_INFO, shost, |
1661 | "%s: Target reset failed" |
1662 | " target: %d\n" , |
1663 | current->comm, id)); |
1664 | list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { |
1665 | if (scmd_id(scmd) != id) |
1666 | continue; |
1667 | |
1668 | if (rtn == SUCCESS) |
1669 | list_move_tail(list: &scmd->eh_entry, head: &check_list); |
1670 | else if (rtn == FAST_IO_FAIL) |
1671 | scsi_eh_finish_cmd(scmd, done_q); |
1672 | else |
1673 | /* push back on work queue for further processing */ |
1674 | list_move(list: &scmd->eh_entry, head: work_q); |
1675 | } |
1676 | } |
1677 | |
1678 | return scsi_eh_test_devices(cmd_list: &check_list, work_q, done_q, try_stu: 0); |
1679 | } |
1680 | |
1681 | /** |
1682 | * scsi_eh_bus_reset - send a bus reset |
1683 | * @shost: &scsi host being recovered. |
1684 | * @work_q: &list_head for pending commands. |
1685 | * @done_q: &list_head for processed commands. |
1686 | */ |
1687 | static int scsi_eh_bus_reset(struct Scsi_Host *shost, |
1688 | struct list_head *work_q, |
1689 | struct list_head *done_q) |
1690 | { |
1691 | struct scsi_cmnd *scmd, *chan_scmd, *next; |
1692 | LIST_HEAD(check_list); |
1693 | unsigned int channel; |
1694 | enum scsi_disposition rtn; |
1695 | |
1696 | /* |
1697 | * we really want to loop over the various channels, and do this on |
1698 | * a channel by channel basis. we should also check to see if any |
1699 | * of the failed commands are on soft_reset devices, and if so, skip |
1700 | * the reset. |
1701 | */ |
1702 | |
1703 | for (channel = 0; channel <= shost->max_channel; channel++) { |
1704 | if (scsi_host_eh_past_deadline(shost)) { |
1705 | list_splice_init(list: &check_list, head: work_q); |
1706 | SCSI_LOG_ERROR_RECOVERY(3, |
1707 | shost_printk(KERN_INFO, shost, |
1708 | "%s: skip BRST, past eh deadline\n" , |
1709 | current->comm)); |
1710 | return list_empty(head: work_q); |
1711 | } |
1712 | |
1713 | chan_scmd = NULL; |
1714 | list_for_each_entry(scmd, work_q, eh_entry) { |
1715 | if (channel == scmd_channel(scmd)) { |
1716 | chan_scmd = scmd; |
1717 | break; |
1718 | /* |
1719 | * FIXME add back in some support for |
1720 | * soft_reset devices. |
1721 | */ |
1722 | } |
1723 | } |
1724 | |
1725 | if (!chan_scmd) |
1726 | continue; |
1727 | SCSI_LOG_ERROR_RECOVERY(3, |
1728 | shost_printk(KERN_INFO, shost, |
1729 | "%s: Sending BRST chan: %d\n" , |
1730 | current->comm, channel)); |
1731 | rtn = scsi_try_bus_reset(scmd: chan_scmd); |
1732 | if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { |
1733 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
1734 | if (channel == scmd_channel(scmd)) { |
1735 | if (rtn == FAST_IO_FAIL) |
1736 | scsi_eh_finish_cmd(scmd, |
1737 | done_q); |
1738 | else |
1739 | list_move_tail(list: &scmd->eh_entry, |
1740 | head: &check_list); |
1741 | } |
1742 | } |
1743 | } else { |
1744 | SCSI_LOG_ERROR_RECOVERY(3, |
1745 | shost_printk(KERN_INFO, shost, |
1746 | "%s: BRST failed chan: %d\n" , |
1747 | current->comm, channel)); |
1748 | } |
1749 | } |
1750 | return scsi_eh_test_devices(cmd_list: &check_list, work_q, done_q, try_stu: 0); |
1751 | } |
1752 | |
1753 | /** |
1754 | * scsi_eh_host_reset - send a host reset |
1755 | * @shost: host to be reset. |
1756 | * @work_q: &list_head for pending commands. |
1757 | * @done_q: &list_head for processed commands. |
1758 | */ |
1759 | static int scsi_eh_host_reset(struct Scsi_Host *shost, |
1760 | struct list_head *work_q, |
1761 | struct list_head *done_q) |
1762 | { |
1763 | struct scsi_cmnd *scmd, *next; |
1764 | LIST_HEAD(check_list); |
1765 | enum scsi_disposition rtn; |
1766 | |
1767 | if (!list_empty(head: work_q)) { |
1768 | scmd = list_entry(work_q->next, |
1769 | struct scsi_cmnd, eh_entry); |
1770 | |
1771 | SCSI_LOG_ERROR_RECOVERY(3, |
1772 | shost_printk(KERN_INFO, shost, |
1773 | "%s: Sending HRST\n" , |
1774 | current->comm)); |
1775 | |
1776 | rtn = scsi_try_host_reset(scmd); |
1777 | if (rtn == SUCCESS) { |
1778 | list_splice_init(list: work_q, head: &check_list); |
1779 | } else if (rtn == FAST_IO_FAIL) { |
1780 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
1781 | scsi_eh_finish_cmd(scmd, done_q); |
1782 | } |
1783 | } else { |
1784 | SCSI_LOG_ERROR_RECOVERY(3, |
1785 | shost_printk(KERN_INFO, shost, |
1786 | "%s: HRST failed\n" , |
1787 | current->comm)); |
1788 | } |
1789 | } |
1790 | return scsi_eh_test_devices(cmd_list: &check_list, work_q, done_q, try_stu: 1); |
1791 | } |
1792 | |
1793 | /** |
1794 | * scsi_eh_offline_sdevs - offline scsi devices that fail to recover |
1795 | * @work_q: &list_head for pending commands. |
1796 | * @done_q: &list_head for processed commands. |
1797 | */ |
1798 | static void scsi_eh_offline_sdevs(struct list_head *work_q, |
1799 | struct list_head *done_q) |
1800 | { |
1801 | struct scsi_cmnd *scmd, *next; |
1802 | struct scsi_device *sdev; |
1803 | |
1804 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
1805 | sdev_printk(KERN_INFO, scmd->device, "Device offlined - " |
1806 | "not ready after error recovery\n" ); |
1807 | sdev = scmd->device; |
1808 | |
1809 | mutex_lock(&sdev->state_mutex); |
1810 | scsi_device_set_state(sdev, state: SDEV_OFFLINE); |
1811 | mutex_unlock(lock: &sdev->state_mutex); |
1812 | |
1813 | scsi_eh_finish_cmd(scmd, done_q); |
1814 | } |
1815 | return; |
1816 | } |
1817 | |
1818 | /** |
1819 | * scsi_noretry_cmd - determine if command should be failed fast |
1820 | * @scmd: SCSI cmd to examine. |
1821 | */ |
1822 | bool scsi_noretry_cmd(struct scsi_cmnd *scmd) |
1823 | { |
1824 | struct request *req = scsi_cmd_to_rq(scmd); |
1825 | |
1826 | switch (host_byte(scmd->result)) { |
1827 | case DID_OK: |
1828 | break; |
1829 | case DID_TIME_OUT: |
1830 | goto check_type; |
1831 | case DID_BUS_BUSY: |
1832 | return !!(req->cmd_flags & REQ_FAILFAST_TRANSPORT); |
1833 | case DID_PARITY: |
1834 | return !!(req->cmd_flags & REQ_FAILFAST_DEV); |
1835 | case DID_ERROR: |
1836 | if (get_status_byte(cmd: scmd) == SAM_STAT_RESERVATION_CONFLICT) |
1837 | return false; |
1838 | fallthrough; |
1839 | case DID_SOFT_ERROR: |
1840 | return !!(req->cmd_flags & REQ_FAILFAST_DRIVER); |
1841 | } |
1842 | |
1843 | /* Never retry commands aborted due to a duration limit timeout */ |
1844 | if (scsi_ml_byte(result: scmd->result) == SCSIML_STAT_DL_TIMEOUT) |
1845 | return true; |
1846 | |
1847 | if (!scsi_status_is_check_condition(status: scmd->result)) |
1848 | return false; |
1849 | |
1850 | check_type: |
1851 | /* |
1852 | * assume caller has checked sense and determined |
1853 | * the check condition was retryable. |
1854 | */ |
1855 | if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(rq: req)) |
1856 | return true; |
1857 | |
1858 | return false; |
1859 | } |
1860 | |
1861 | /** |
1862 | * scsi_decide_disposition - Disposition a cmd on return from LLD. |
1863 | * @scmd: SCSI cmd to examine. |
1864 | * |
1865 | * Notes: |
1866 | * This is *only* called when we are examining the status after sending |
1867 | * out the actual data command. any commands that are queued for error |
1868 | * recovery (e.g. test_unit_ready) do *not* come through here. |
1869 | * |
1870 | * When this routine returns failed, it means the error handler thread |
1871 | * is woken. In cases where the error code indicates an error that |
1872 | * doesn't require the error handler read (i.e. we don't need to |
1873 | * abort/reset), this function should return SUCCESS. |
1874 | */ |
1875 | enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd) |
1876 | { |
1877 | enum scsi_disposition rtn; |
1878 | |
1879 | /* |
1880 | * if the device is offline, then we clearly just pass the result back |
1881 | * up to the top level. |
1882 | */ |
1883 | if (!scsi_device_online(sdev: scmd->device)) { |
1884 | SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, |
1885 | "%s: device offline - report as SUCCESS\n" , __func__)); |
1886 | return SUCCESS; |
1887 | } |
1888 | |
1889 | /* |
1890 | * first check the host byte, to see if there is anything in there |
1891 | * that would indicate what we need to do. |
1892 | */ |
1893 | switch (host_byte(scmd->result)) { |
1894 | case DID_PASSTHROUGH: |
1895 | /* |
1896 | * no matter what, pass this through to the upper layer. |
1897 | * nuke this special code so that it looks like we are saying |
1898 | * did_ok. |
1899 | */ |
1900 | scmd->result &= 0xff00ffff; |
1901 | return SUCCESS; |
1902 | case DID_OK: |
1903 | /* |
1904 | * looks good. drop through, and check the next byte. |
1905 | */ |
1906 | break; |
1907 | case DID_ABORT: |
1908 | if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { |
1909 | set_host_byte(cmd: scmd, status: DID_TIME_OUT); |
1910 | return SUCCESS; |
1911 | } |
1912 | fallthrough; |
1913 | case DID_NO_CONNECT: |
1914 | case DID_BAD_TARGET: |
1915 | /* |
1916 | * note - this means that we just report the status back |
1917 | * to the top level driver, not that we actually think |
1918 | * that it indicates SUCCESS. |
1919 | */ |
1920 | return SUCCESS; |
1921 | case DID_SOFT_ERROR: |
1922 | /* |
1923 | * when the low level driver returns did_soft_error, |
1924 | * it is responsible for keeping an internal retry counter |
1925 | * in order to avoid endless loops (db) |
1926 | */ |
1927 | goto maybe_retry; |
1928 | case DID_IMM_RETRY: |
1929 | return NEEDS_RETRY; |
1930 | |
1931 | case DID_REQUEUE: |
1932 | return ADD_TO_MLQUEUE; |
1933 | case DID_TRANSPORT_DISRUPTED: |
1934 | /* |
1935 | * LLD/transport was disrupted during processing of the IO. |
1936 | * The transport class is now blocked/blocking, |
1937 | * and the transport will decide what to do with the IO |
1938 | * based on its timers and recovery capablilities if |
1939 | * there are enough retries. |
1940 | */ |
1941 | goto maybe_retry; |
1942 | case DID_TRANSPORT_FAILFAST: |
1943 | /* |
1944 | * The transport decided to failfast the IO (most likely |
1945 | * the fast io fail tmo fired), so send IO directly upwards. |
1946 | */ |
1947 | return SUCCESS; |
1948 | case DID_TRANSPORT_MARGINAL: |
1949 | /* |
1950 | * caller has decided not to do retries on |
1951 | * abort success, so send IO directly upwards |
1952 | */ |
1953 | return SUCCESS; |
1954 | case DID_ERROR: |
1955 | if (get_status_byte(cmd: scmd) == SAM_STAT_RESERVATION_CONFLICT) |
1956 | /* |
1957 | * execute reservation conflict processing code |
1958 | * lower down |
1959 | */ |
1960 | break; |
1961 | fallthrough; |
1962 | case DID_BUS_BUSY: |
1963 | case DID_PARITY: |
1964 | goto maybe_retry; |
1965 | case DID_TIME_OUT: |
1966 | /* |
1967 | * when we scan the bus, we get timeout messages for |
1968 | * these commands if there is no device available. |
1969 | * other hosts report did_no_connect for the same thing. |
1970 | */ |
1971 | if ((scmd->cmnd[0] == TEST_UNIT_READY || |
1972 | scmd->cmnd[0] == INQUIRY)) { |
1973 | return SUCCESS; |
1974 | } else { |
1975 | return FAILED; |
1976 | } |
1977 | case DID_RESET: |
1978 | return SUCCESS; |
1979 | default: |
1980 | return FAILED; |
1981 | } |
1982 | |
1983 | /* |
1984 | * check the status byte to see if this indicates anything special. |
1985 | */ |
1986 | switch (get_status_byte(cmd: scmd)) { |
1987 | case SAM_STAT_TASK_SET_FULL: |
1988 | scsi_handle_queue_full(sdev: scmd->device); |
1989 | /* |
1990 | * the case of trying to send too many commands to a |
1991 | * tagged queueing device. |
1992 | */ |
1993 | fallthrough; |
1994 | case SAM_STAT_BUSY: |
1995 | /* |
1996 | * device can't talk to us at the moment. Should only |
1997 | * occur (SAM-3) when the task queue is empty, so will cause |
1998 | * the empty queue handling to trigger a stall in the |
1999 | * device. |
2000 | */ |
2001 | return ADD_TO_MLQUEUE; |
2002 | case SAM_STAT_GOOD: |
2003 | if (scmd->cmnd[0] == REPORT_LUNS) |
2004 | scmd->device->sdev_target->expecting_lun_change = 0; |
2005 | scsi_handle_queue_ramp_up(sdev: scmd->device); |
2006 | if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd)) |
2007 | /* |
2008 | * If we have sense data, call scsi_check_sense() in |
2009 | * order to set the correct SCSI ML byte (if any). |
2010 | * No point in checking the return value, since the |
2011 | * command has already completed successfully. |
2012 | */ |
2013 | scsi_check_sense(scmd); |
2014 | fallthrough; |
2015 | case SAM_STAT_COMMAND_TERMINATED: |
2016 | return SUCCESS; |
2017 | case SAM_STAT_TASK_ABORTED: |
2018 | goto maybe_retry; |
2019 | case SAM_STAT_CHECK_CONDITION: |
2020 | rtn = scsi_check_sense(scmd); |
2021 | if (rtn == NEEDS_RETRY) |
2022 | goto maybe_retry; |
2023 | /* if rtn == FAILED, we have no sense information; |
2024 | * returning FAILED will wake the error handler thread |
2025 | * to collect the sense and redo the decide |
2026 | * disposition */ |
2027 | return rtn; |
2028 | case SAM_STAT_CONDITION_MET: |
2029 | case SAM_STAT_INTERMEDIATE: |
2030 | case SAM_STAT_INTERMEDIATE_CONDITION_MET: |
2031 | case SAM_STAT_ACA_ACTIVE: |
2032 | /* |
2033 | * who knows? FIXME(eric) |
2034 | */ |
2035 | return SUCCESS; |
2036 | |
2037 | case SAM_STAT_RESERVATION_CONFLICT: |
2038 | sdev_printk(KERN_INFO, scmd->device, |
2039 | "reservation conflict\n" ); |
2040 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_RESV_CONFLICT); |
2041 | return SUCCESS; /* causes immediate i/o error */ |
2042 | } |
2043 | return FAILED; |
2044 | |
2045 | maybe_retry: |
2046 | |
2047 | /* we requeue for retry because the error was retryable, and |
2048 | * the request was not marked fast fail. Note that above, |
2049 | * even if the request is marked fast fail, we still requeue |
2050 | * for queue congestion conditions (QUEUE_FULL or BUSY) */ |
2051 | if (scsi_cmd_retry_allowed(cmd: scmd) && !scsi_noretry_cmd(scmd)) { |
2052 | return NEEDS_RETRY; |
2053 | } else { |
2054 | /* |
2055 | * no more retries - report this one back to upper level. |
2056 | */ |
2057 | return SUCCESS; |
2058 | } |
2059 | } |
2060 | |
2061 | static enum rq_end_io_ret eh_lock_door_done(struct request *req, |
2062 | blk_status_t status) |
2063 | { |
2064 | blk_mq_free_request(rq: req); |
2065 | return RQ_END_IO_NONE; |
2066 | } |
2067 | |
2068 | /** |
2069 | * scsi_eh_lock_door - Prevent medium removal for the specified device |
2070 | * @sdev: SCSI device to prevent medium removal |
2071 | * |
2072 | * Locking: |
2073 | * We must be called from process context. |
2074 | * |
2075 | * Notes: |
2076 | * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the |
2077 | * head of the devices request queue, and continue. |
2078 | */ |
2079 | static void scsi_eh_lock_door(struct scsi_device *sdev) |
2080 | { |
2081 | struct scsi_cmnd *scmd; |
2082 | struct request *req; |
2083 | |
2084 | req = scsi_alloc_request(q: sdev->request_queue, opf: REQ_OP_DRV_IN, flags: 0); |
2085 | if (IS_ERR(ptr: req)) |
2086 | return; |
2087 | scmd = blk_mq_rq_to_pdu(rq: req); |
2088 | |
2089 | scmd->cmnd[0] = ALLOW_MEDIUM_REMOVAL; |
2090 | scmd->cmnd[1] = 0; |
2091 | scmd->cmnd[2] = 0; |
2092 | scmd->cmnd[3] = 0; |
2093 | scmd->cmnd[4] = SCSI_REMOVAL_PREVENT; |
2094 | scmd->cmnd[5] = 0; |
2095 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
2096 | scmd->allowed = 5; |
2097 | |
2098 | req->rq_flags |= RQF_QUIET; |
2099 | req->timeout = 10 * HZ; |
2100 | req->end_io = eh_lock_door_done; |
2101 | |
2102 | blk_execute_rq_nowait(rq: req, at_head: true); |
2103 | } |
2104 | |
2105 | /** |
2106 | * scsi_restart_operations - restart io operations to the specified host. |
2107 | * @shost: Host we are restarting. |
2108 | * |
2109 | * Notes: |
2110 | * When we entered the error handler, we blocked all further i/o to |
2111 | * this device. we need to 'reverse' this process. |
2112 | */ |
2113 | static void scsi_restart_operations(struct Scsi_Host *shost) |
2114 | { |
2115 | struct scsi_device *sdev; |
2116 | unsigned long flags; |
2117 | |
2118 | /* |
2119 | * If the door was locked, we need to insert a door lock request |
2120 | * onto the head of the SCSI request queue for the device. There |
2121 | * is no point trying to lock the door of an off-line device. |
2122 | */ |
2123 | shost_for_each_device(sdev, shost) { |
2124 | if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { |
2125 | scsi_eh_lock_door(sdev); |
2126 | sdev->was_reset = 0; |
2127 | } |
2128 | } |
2129 | |
2130 | /* |
2131 | * next free up anything directly waiting upon the host. this |
2132 | * will be requests for character device operations, and also for |
2133 | * ioctls to queued block devices. |
2134 | */ |
2135 | SCSI_LOG_ERROR_RECOVERY(3, |
2136 | shost_printk(KERN_INFO, shost, "waking up host to restart\n" )); |
2137 | |
2138 | spin_lock_irqsave(shost->host_lock, flags); |
2139 | if (scsi_host_set_state(shost, SHOST_RUNNING)) |
2140 | if (scsi_host_set_state(shost, SHOST_CANCEL)) |
2141 | BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); |
2142 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
2143 | |
2144 | wake_up(&shost->host_wait); |
2145 | |
2146 | /* |
2147 | * finally we need to re-initiate requests that may be pending. we will |
2148 | * have had everything blocked while error handling is taking place, and |
2149 | * now that error recovery is done, we will need to ensure that these |
2150 | * requests are started. |
2151 | */ |
2152 | scsi_run_host_queues(shost); |
2153 | |
2154 | /* |
2155 | * if eh is active and host_eh_scheduled is pending we need to re-run |
2156 | * recovery. we do this check after scsi_run_host_queues() to allow |
2157 | * everything pent up since the last eh run a chance to make forward |
2158 | * progress before we sync again. Either we'll immediately re-run |
2159 | * recovery or scsi_device_unbusy() will wake us again when these |
2160 | * pending commands complete. |
2161 | */ |
2162 | spin_lock_irqsave(shost->host_lock, flags); |
2163 | if (shost->host_eh_scheduled) |
2164 | if (scsi_host_set_state(shost, SHOST_RECOVERY)) |
2165 | WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); |
2166 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
2167 | } |
2168 | |
2169 | /** |
2170 | * scsi_eh_ready_devs - check device ready state and recover if not. |
2171 | * @shost: host to be recovered. |
2172 | * @work_q: &list_head for pending commands. |
2173 | * @done_q: &list_head for processed commands. |
2174 | */ |
2175 | void scsi_eh_ready_devs(struct Scsi_Host *shost, |
2176 | struct list_head *work_q, |
2177 | struct list_head *done_q) |
2178 | { |
2179 | if (!scsi_eh_stu(shost, work_q, done_q)) |
2180 | if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) |
2181 | if (!scsi_eh_target_reset(shost, work_q, done_q)) |
2182 | if (!scsi_eh_bus_reset(shost, work_q, done_q)) |
2183 | if (!scsi_eh_host_reset(shost, work_q, done_q)) |
2184 | scsi_eh_offline_sdevs(work_q, |
2185 | done_q); |
2186 | } |
2187 | EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); |
2188 | |
2189 | /** |
2190 | * scsi_eh_flush_done_q - finish processed commands or retry them. |
2191 | * @done_q: list_head of processed commands. |
2192 | */ |
2193 | void scsi_eh_flush_done_q(struct list_head *done_q) |
2194 | { |
2195 | struct scsi_cmnd *scmd, *next; |
2196 | |
2197 | list_for_each_entry_safe(scmd, next, done_q, eh_entry) { |
2198 | list_del_init(entry: &scmd->eh_entry); |
2199 | if (scsi_device_online(sdev: scmd->device) && |
2200 | !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(cmd: scmd) && |
2201 | scsi_eh_should_retry_cmd(cmd: scmd)) { |
2202 | SCSI_LOG_ERROR_RECOVERY(3, |
2203 | scmd_printk(KERN_INFO, scmd, |
2204 | "%s: flush retry cmd\n" , |
2205 | current->comm)); |
2206 | scsi_queue_insert(cmd: scmd, SCSI_MLQUEUE_EH_RETRY); |
2207 | } else { |
2208 | /* |
2209 | * If just we got sense for the device (called |
2210 | * scsi_eh_get_sense), scmd->result is already |
2211 | * set, do not set DID_TIME_OUT. |
2212 | */ |
2213 | if (!scmd->result && |
2214 | !(scmd->flags & SCMD_FORCE_EH_SUCCESS)) |
2215 | scmd->result |= (DID_TIME_OUT << 16); |
2216 | SCSI_LOG_ERROR_RECOVERY(3, |
2217 | scmd_printk(KERN_INFO, scmd, |
2218 | "%s: flush finish cmd\n" , |
2219 | current->comm)); |
2220 | scsi_finish_command(cmd: scmd); |
2221 | } |
2222 | } |
2223 | } |
2224 | EXPORT_SYMBOL(scsi_eh_flush_done_q); |
2225 | |
2226 | /** |
2227 | * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. |
2228 | * @shost: Host to unjam. |
2229 | * |
2230 | * Notes: |
2231 | * When we come in here, we *know* that all commands on the bus have |
2232 | * either completed, failed or timed out. we also know that no further |
2233 | * commands are being sent to the host, so things are relatively quiet |
2234 | * and we have freedom to fiddle with things as we wish. |
2235 | * |
2236 | * This is only the *default* implementation. it is possible for |
2237 | * individual drivers to supply their own version of this function, and |
2238 | * if the maintainer wishes to do this, it is strongly suggested that |
2239 | * this function be taken as a template and modified. this function |
2240 | * was designed to correctly handle problems for about 95% of the |
2241 | * different cases out there, and it should always provide at least a |
2242 | * reasonable amount of error recovery. |
2243 | * |
2244 | * Any command marked 'failed' or 'timeout' must eventually have |
2245 | * scsi_finish_cmd() called for it. we do all of the retry stuff |
2246 | * here, so when we restart the host after we return it should have an |
2247 | * empty queue. |
2248 | */ |
2249 | static void scsi_unjam_host(struct Scsi_Host *shost) |
2250 | { |
2251 | unsigned long flags; |
2252 | LIST_HEAD(eh_work_q); |
2253 | LIST_HEAD(eh_done_q); |
2254 | |
2255 | spin_lock_irqsave(shost->host_lock, flags); |
2256 | list_splice_init(list: &shost->eh_cmd_q, head: &eh_work_q); |
2257 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
2258 | |
2259 | SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); |
2260 | |
2261 | if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) |
2262 | scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); |
2263 | |
2264 | spin_lock_irqsave(shost->host_lock, flags); |
2265 | if (shost->eh_deadline != -1) |
2266 | shost->last_reset = 0; |
2267 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
2268 | scsi_eh_flush_done_q(&eh_done_q); |
2269 | } |
2270 | |
2271 | /** |
2272 | * scsi_error_handler - SCSI error handler thread |
2273 | * @data: Host for which we are running. |
2274 | * |
2275 | * Notes: |
2276 | * This is the main error handling loop. This is run as a kernel thread |
2277 | * for every SCSI host and handles all error handling activity. |
2278 | */ |
2279 | int scsi_error_handler(void *data) |
2280 | { |
2281 | struct Scsi_Host *shost = data; |
2282 | |
2283 | /* |
2284 | * We use TASK_INTERRUPTIBLE so that the thread is not |
2285 | * counted against the load average as a running process. |
2286 | * We never actually get interrupted because kthread_run |
2287 | * disables signal delivery for the created thread. |
2288 | */ |
2289 | while (true) { |
2290 | /* |
2291 | * The sequence in kthread_stop() sets the stop flag first |
2292 | * then wakes the process. To avoid missed wakeups, the task |
2293 | * should always be in a non running state before the stop |
2294 | * flag is checked |
2295 | */ |
2296 | set_current_state(TASK_INTERRUPTIBLE); |
2297 | if (kthread_should_stop()) |
2298 | break; |
2299 | |
2300 | if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || |
2301 | shost->host_failed != scsi_host_busy(shost)) { |
2302 | SCSI_LOG_ERROR_RECOVERY(1, |
2303 | shost_printk(KERN_INFO, shost, |
2304 | "scsi_eh_%d: sleeping\n" , |
2305 | shost->host_no)); |
2306 | schedule(); |
2307 | continue; |
2308 | } |
2309 | |
2310 | __set_current_state(TASK_RUNNING); |
2311 | SCSI_LOG_ERROR_RECOVERY(1, |
2312 | shost_printk(KERN_INFO, shost, |
2313 | "scsi_eh_%d: waking up %d/%d/%d\n" , |
2314 | shost->host_no, shost->host_eh_scheduled, |
2315 | shost->host_failed, |
2316 | scsi_host_busy(shost))); |
2317 | |
2318 | /* |
2319 | * We have a host that is failing for some reason. Figure out |
2320 | * what we need to do to get it up and online again (if we can). |
2321 | * If we fail, we end up taking the thing offline. |
2322 | */ |
2323 | if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { |
2324 | SCSI_LOG_ERROR_RECOVERY(1, |
2325 | shost_printk(KERN_ERR, shost, |
2326 | "scsi_eh_%d: unable to autoresume\n" , |
2327 | shost->host_no)); |
2328 | continue; |
2329 | } |
2330 | |
2331 | if (shost->transportt->eh_strategy_handler) |
2332 | shost->transportt->eh_strategy_handler(shost); |
2333 | else |
2334 | scsi_unjam_host(shost); |
2335 | |
2336 | /* All scmds have been handled */ |
2337 | shost->host_failed = 0; |
2338 | |
2339 | /* |
2340 | * Note - if the above fails completely, the action is to take |
2341 | * individual devices offline and flush the queue of any |
2342 | * outstanding requests that may have been pending. When we |
2343 | * restart, we restart any I/O to any other devices on the bus |
2344 | * which are still online. |
2345 | */ |
2346 | scsi_restart_operations(shost); |
2347 | if (!shost->eh_noresume) |
2348 | scsi_autopm_put_host(shost); |
2349 | } |
2350 | __set_current_state(TASK_RUNNING); |
2351 | |
2352 | SCSI_LOG_ERROR_RECOVERY(1, |
2353 | shost_printk(KERN_INFO, shost, |
2354 | "Error handler scsi_eh_%d exiting\n" , |
2355 | shost->host_no)); |
2356 | shost->ehandler = NULL; |
2357 | return 0; |
2358 | } |
2359 | |
2360 | /* |
2361 | * Function: scsi_report_bus_reset() |
2362 | * |
2363 | * Purpose: Utility function used by low-level drivers to report that |
2364 | * they have observed a bus reset on the bus being handled. |
2365 | * |
2366 | * Arguments: shost - Host in question |
2367 | * channel - channel on which reset was observed. |
2368 | * |
2369 | * Returns: Nothing |
2370 | * |
2371 | * Lock status: Host lock must be held. |
2372 | * |
2373 | * Notes: This only needs to be called if the reset is one which |
2374 | * originates from an unknown location. Resets originated |
2375 | * by the mid-level itself don't need to call this, but there |
2376 | * should be no harm. |
2377 | * |
2378 | * The main purpose of this is to make sure that a CHECK_CONDITION |
2379 | * is properly treated. |
2380 | */ |
2381 | void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) |
2382 | { |
2383 | struct scsi_device *sdev; |
2384 | |
2385 | __shost_for_each_device(sdev, shost) { |
2386 | if (channel == sdev_channel(sdev)) |
2387 | __scsi_report_device_reset(sdev, NULL); |
2388 | } |
2389 | } |
2390 | EXPORT_SYMBOL(scsi_report_bus_reset); |
2391 | |
2392 | /* |
2393 | * Function: scsi_report_device_reset() |
2394 | * |
2395 | * Purpose: Utility function used by low-level drivers to report that |
2396 | * they have observed a device reset on the device being handled. |
2397 | * |
2398 | * Arguments: shost - Host in question |
2399 | * channel - channel on which reset was observed |
2400 | * target - target on which reset was observed |
2401 | * |
2402 | * Returns: Nothing |
2403 | * |
2404 | * Lock status: Host lock must be held |
2405 | * |
2406 | * Notes: This only needs to be called if the reset is one which |
2407 | * originates from an unknown location. Resets originated |
2408 | * by the mid-level itself don't need to call this, but there |
2409 | * should be no harm. |
2410 | * |
2411 | * The main purpose of this is to make sure that a CHECK_CONDITION |
2412 | * is properly treated. |
2413 | */ |
2414 | void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) |
2415 | { |
2416 | struct scsi_device *sdev; |
2417 | |
2418 | __shost_for_each_device(sdev, shost) { |
2419 | if (channel == sdev_channel(sdev) && |
2420 | target == sdev_id(sdev)) |
2421 | __scsi_report_device_reset(sdev, NULL); |
2422 | } |
2423 | } |
2424 | EXPORT_SYMBOL(scsi_report_device_reset); |
2425 | |
2426 | /** |
2427 | * scsi_ioctl_reset: explicitly reset a host/bus/target/device |
2428 | * @dev: scsi_device to operate on |
2429 | * @arg: reset type (see sg.h) |
2430 | */ |
2431 | int |
2432 | scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) |
2433 | { |
2434 | struct scsi_cmnd *scmd; |
2435 | struct Scsi_Host *shost = dev->host; |
2436 | struct request *rq; |
2437 | unsigned long flags; |
2438 | int error = 0, val; |
2439 | enum scsi_disposition rtn; |
2440 | |
2441 | if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) |
2442 | return -EACCES; |
2443 | |
2444 | error = get_user(val, arg); |
2445 | if (error) |
2446 | return error; |
2447 | |
2448 | if (scsi_autopm_get_host(shost) < 0) |
2449 | return -EIO; |
2450 | |
2451 | error = -EIO; |
2452 | rq = kzalloc(size: sizeof(struct request) + sizeof(struct scsi_cmnd) + |
2453 | shost->hostt->cmd_size, GFP_KERNEL); |
2454 | if (!rq) |
2455 | goto out_put_autopm_host; |
2456 | blk_rq_init(NULL, rq); |
2457 | |
2458 | scmd = (struct scsi_cmnd *)(rq + 1); |
2459 | scsi_init_command(dev, cmd: scmd); |
2460 | |
2461 | scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL; |
2462 | memset(&scmd->sdb, 0, sizeof(scmd->sdb)); |
2463 | |
2464 | scmd->cmd_len = 0; |
2465 | |
2466 | scmd->sc_data_direction = DMA_BIDIRECTIONAL; |
2467 | |
2468 | spin_lock_irqsave(shost->host_lock, flags); |
2469 | shost->tmf_in_progress = 1; |
2470 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
2471 | |
2472 | switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { |
2473 | case SG_SCSI_RESET_NOTHING: |
2474 | rtn = SUCCESS; |
2475 | break; |
2476 | case SG_SCSI_RESET_DEVICE: |
2477 | rtn = scsi_try_bus_device_reset(scmd); |
2478 | if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) |
2479 | break; |
2480 | fallthrough; |
2481 | case SG_SCSI_RESET_TARGET: |
2482 | rtn = scsi_try_target_reset(scmd); |
2483 | if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) |
2484 | break; |
2485 | fallthrough; |
2486 | case SG_SCSI_RESET_BUS: |
2487 | rtn = scsi_try_bus_reset(scmd); |
2488 | if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) |
2489 | break; |
2490 | fallthrough; |
2491 | case SG_SCSI_RESET_HOST: |
2492 | rtn = scsi_try_host_reset(scmd); |
2493 | if (rtn == SUCCESS) |
2494 | break; |
2495 | fallthrough; |
2496 | default: |
2497 | rtn = FAILED; |
2498 | break; |
2499 | } |
2500 | |
2501 | error = (rtn == SUCCESS) ? 0 : -EIO; |
2502 | |
2503 | spin_lock_irqsave(shost->host_lock, flags); |
2504 | shost->tmf_in_progress = 0; |
2505 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
2506 | |
2507 | /* |
2508 | * be sure to wake up anyone who was sleeping or had their queue |
2509 | * suspended while we performed the TMF. |
2510 | */ |
2511 | SCSI_LOG_ERROR_RECOVERY(3, |
2512 | shost_printk(KERN_INFO, shost, |
2513 | "waking up host to restart after TMF\n" )); |
2514 | |
2515 | wake_up(&shost->host_wait); |
2516 | scsi_run_host_queues(shost); |
2517 | |
2518 | kfree(objp: rq); |
2519 | |
2520 | out_put_autopm_host: |
2521 | scsi_autopm_put_host(shost); |
2522 | return error; |
2523 | } |
2524 | |
2525 | bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, |
2526 | struct scsi_sense_hdr *sshdr) |
2527 | { |
2528 | return scsi_normalize_sense(sense_buffer: cmd->sense_buffer, |
2529 | SCSI_SENSE_BUFFERSIZE, sshdr); |
2530 | } |
2531 | EXPORT_SYMBOL(scsi_command_normalize_sense); |
2532 | |
2533 | /** |
2534 | * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) |
2535 | * @sense_buffer: byte array of sense data |
2536 | * @sb_len: number of valid bytes in sense_buffer |
2537 | * @info_out: pointer to 64 integer where 8 or 4 byte information |
2538 | * field will be placed if found. |
2539 | * |
2540 | * Return value: |
2541 | * true if information field found, false if not found. |
2542 | */ |
2543 | bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, |
2544 | u64 *info_out) |
2545 | { |
2546 | const u8 * ucp; |
2547 | |
2548 | if (sb_len < 7) |
2549 | return false; |
2550 | switch (sense_buffer[0] & 0x7f) { |
2551 | case 0x70: |
2552 | case 0x71: |
2553 | if (sense_buffer[0] & 0x80) { |
2554 | *info_out = get_unaligned_be32(p: &sense_buffer[3]); |
2555 | return true; |
2556 | } |
2557 | return false; |
2558 | case 0x72: |
2559 | case 0x73: |
2560 | ucp = scsi_sense_desc_find(sense_buffer, sb_len, |
2561 | desc_type: 0 /* info desc */); |
2562 | if (ucp && (0xa == ucp[1])) { |
2563 | *info_out = get_unaligned_be64(p: &ucp[4]); |
2564 | return true; |
2565 | } |
2566 | return false; |
2567 | default: |
2568 | return false; |
2569 | } |
2570 | } |
2571 | EXPORT_SYMBOL(scsi_get_sense_info_fld); |
2572 | |