1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
5 *
6 * SCSI queueing library.
7 * Initial versions: Eric Youngdale (eric@andante.org).
8 * Based upon conversations with large numbers
9 * of people at Linux Expo.
10 */
11
12 #include <linux/bio.h>
13 #include <linux/bitops.h>
14 #include <linux/blkdev.h>
15 #include <linux/completion.h>
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
23 #include <linux/blk-mq.h>
24 #include <linux/blk-integrity.h>
25 #include <linux/ratelimit.h>
26 #include <asm/unaligned.h>
27
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_driver.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
36 #include <scsi/scsi_dh.h>
37
38 #include <trace/events/scsi.h>
39
40 #include "scsi_debugfs.h"
41 #include "scsi_priv.h"
42 #include "scsi_logging.h"
43
44 /*
45 * Size of integrity metadata is usually small, 1 inline sg should
46 * cover normal cases.
47 */
48 #ifdef CONFIG_ARCH_NO_SG_CHAIN
49 #define SCSI_INLINE_PROT_SG_CNT 0
50 #define SCSI_INLINE_SG_CNT 0
51 #else
52 #define SCSI_INLINE_PROT_SG_CNT 1
53 #define SCSI_INLINE_SG_CNT 2
54 #endif
55
56 static struct kmem_cache *scsi_sense_cache;
57 static DEFINE_MUTEX(scsi_sense_cache_mutex);
58
59 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
60
scsi_init_sense_cache(struct Scsi_Host * shost)61 int scsi_init_sense_cache(struct Scsi_Host *shost)
62 {
63 int ret = 0;
64
65 mutex_lock(&scsi_sense_cache_mutex);
66 if (!scsi_sense_cache) {
67 scsi_sense_cache =
68 kmem_cache_create_usercopy("scsi_sense_cache",
69 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
70 0, SCSI_SENSE_BUFFERSIZE, NULL);
71 if (!scsi_sense_cache)
72 ret = -ENOMEM;
73 }
74 mutex_unlock(&scsi_sense_cache_mutex);
75 return ret;
76 }
77
78 /*
79 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
80 * not change behaviour from the previous unplug mechanism, experimentation
81 * may prove this needs changing.
82 */
83 #define SCSI_QUEUE_DELAY 3
84
85 static void
scsi_set_blocked(struct scsi_cmnd * cmd,int reason)86 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
87 {
88 struct Scsi_Host *host = cmd->device->host;
89 struct scsi_device *device = cmd->device;
90 struct scsi_target *starget = scsi_target(device);
91
92 /*
93 * Set the appropriate busy bit for the device/host.
94 *
95 * If the host/device isn't busy, assume that something actually
96 * completed, and that we should be able to queue a command now.
97 *
98 * Note that the prior mid-layer assumption that any host could
99 * always queue at least one command is now broken. The mid-layer
100 * will implement a user specifiable stall (see
101 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
102 * if a command is requeued with no other commands outstanding
103 * either for the device or for the host.
104 */
105 switch (reason) {
106 case SCSI_MLQUEUE_HOST_BUSY:
107 atomic_set(&host->host_blocked, host->max_host_blocked);
108 break;
109 case SCSI_MLQUEUE_DEVICE_BUSY:
110 case SCSI_MLQUEUE_EH_RETRY:
111 atomic_set(&device->device_blocked,
112 device->max_device_blocked);
113 break;
114 case SCSI_MLQUEUE_TARGET_BUSY:
115 atomic_set(&starget->target_blocked,
116 starget->max_target_blocked);
117 break;
118 }
119 }
120
scsi_mq_requeue_cmd(struct scsi_cmnd * cmd,unsigned long msecs)121 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd, unsigned long msecs)
122 {
123 struct request *rq = scsi_cmd_to_rq(cmd);
124
125 if (rq->rq_flags & RQF_DONTPREP) {
126 rq->rq_flags &= ~RQF_DONTPREP;
127 scsi_mq_uninit_cmd(cmd);
128 } else {
129 WARN_ON_ONCE(true);
130 }
131
132 if (msecs) {
133 blk_mq_requeue_request(rq, false);
134 blk_mq_delay_kick_requeue_list(rq->q, msecs);
135 } else
136 blk_mq_requeue_request(rq, true);
137 }
138
139 /**
140 * __scsi_queue_insert - private queue insertion
141 * @cmd: The SCSI command being requeued
142 * @reason: The reason for the requeue
143 * @unbusy: Whether the queue should be unbusied
144 *
145 * This is a private queue insertion. The public interface
146 * scsi_queue_insert() always assumes the queue should be unbusied
147 * because it's always called before the completion. This function is
148 * for a requeue after completion, which should only occur in this
149 * file.
150 */
__scsi_queue_insert(struct scsi_cmnd * cmd,int reason,bool unbusy)151 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
152 {
153 struct scsi_device *device = cmd->device;
154
155 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
156 "Inserting command %p into mlqueue\n", cmd));
157
158 scsi_set_blocked(cmd, reason);
159
160 /*
161 * Decrement the counters, since these commands are no longer
162 * active on the host/device.
163 */
164 if (unbusy)
165 scsi_device_unbusy(device, cmd);
166
167 /*
168 * Requeue this command. It will go before all other commands
169 * that are already in the queue. Schedule requeue work under
170 * lock such that the kblockd_schedule_work() call happens
171 * before blk_cleanup_queue() finishes.
172 */
173 cmd->result = 0;
174
175 blk_mq_requeue_request(scsi_cmd_to_rq(cmd), true);
176 }
177
178 /**
179 * scsi_queue_insert - Reinsert a command in the queue.
180 * @cmd: command that we are adding to queue.
181 * @reason: why we are inserting command to queue.
182 *
183 * We do this for one of two cases. Either the host is busy and it cannot accept
184 * any more commands for the time being, or the device returned QUEUE_FULL and
185 * can accept no more commands.
186 *
187 * Context: This could be called either from an interrupt context or a normal
188 * process context.
189 */
scsi_queue_insert(struct scsi_cmnd * cmd,int reason)190 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
191 {
192 __scsi_queue_insert(cmd, reason, true);
193 }
194
195
196 /**
197 * __scsi_execute - insert request and wait for the result
198 * @sdev: scsi device
199 * @cmd: scsi command
200 * @data_direction: data direction
201 * @buffer: data buffer
202 * @bufflen: len of buffer
203 * @sense: optional sense buffer
204 * @sshdr: optional decoded sense header
205 * @timeout: request timeout in HZ
206 * @retries: number of times to retry request
207 * @flags: flags for ->cmd_flags
208 * @rq_flags: flags for ->rq_flags
209 * @resid: optional residual length
210 *
211 * Returns the scsi_cmnd result field if a command was executed, or a negative
212 * Linux error code if we didn't get that far.
213 */
__scsi_execute(struct scsi_device * sdev,const unsigned char * cmd,int data_direction,void * buffer,unsigned bufflen,unsigned char * sense,struct scsi_sense_hdr * sshdr,int timeout,int retries,u64 flags,req_flags_t rq_flags,int * resid)214 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
215 int data_direction, void *buffer, unsigned bufflen,
216 unsigned char *sense, struct scsi_sense_hdr *sshdr,
217 int timeout, int retries, u64 flags, req_flags_t rq_flags,
218 int *resid)
219 {
220 struct request *req;
221 struct scsi_cmnd *scmd;
222 int ret;
223
224 req = scsi_alloc_request(sdev->request_queue,
225 data_direction == DMA_TO_DEVICE ?
226 REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
227 rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
228 if (IS_ERR(req))
229 return PTR_ERR(req);
230
231 if (bufflen) {
232 ret = blk_rq_map_kern(sdev->request_queue, req,
233 buffer, bufflen, GFP_NOIO);
234 if (ret)
235 goto out;
236 }
237 scmd = blk_mq_rq_to_pdu(req);
238 scmd->cmd_len = COMMAND_SIZE(cmd[0]);
239 memcpy(scmd->cmnd, cmd, scmd->cmd_len);
240 scmd->allowed = retries;
241 req->timeout = timeout;
242 req->cmd_flags |= flags;
243 req->rq_flags |= rq_flags | RQF_QUIET;
244
245 /*
246 * head injection *required* here otherwise quiesce won't work
247 */
248 blk_execute_rq(req, true);
249
250 /*
251 * Some devices (USB mass-storage in particular) may transfer
252 * garbage data together with a residue indicating that the data
253 * is invalid. Prevent the garbage from being misinterpreted
254 * and prevent security leaks by zeroing out the excess data.
255 */
256 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
257 memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
258
259 if (resid)
260 *resid = scmd->resid_len;
261 if (sense && scmd->sense_len)
262 memcpy(sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
263 if (sshdr)
264 scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
265 sshdr);
266 ret = scmd->result;
267 out:
268 blk_mq_free_request(req);
269
270 return ret;
271 }
272 EXPORT_SYMBOL(__scsi_execute);
273
274 /*
275 * Wake up the error handler if necessary. Avoid as follows that the error
276 * handler is not woken up if host in-flight requests number ==
277 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
278 * with an RCU read lock in this function to ensure that this function in
279 * its entirety either finishes before scsi_eh_scmd_add() increases the
280 * host_failed counter or that it notices the shost state change made by
281 * scsi_eh_scmd_add().
282 */
scsi_dec_host_busy(struct Scsi_Host * shost,struct scsi_cmnd * cmd)283 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
284 {
285 unsigned long flags;
286
287 rcu_read_lock();
288 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
289 if (unlikely(scsi_host_in_recovery(shost))) {
290 spin_lock_irqsave(shost->host_lock, flags);
291 if (shost->host_failed || shost->host_eh_scheduled)
292 scsi_eh_wakeup(shost);
293 spin_unlock_irqrestore(shost->host_lock, flags);
294 }
295 rcu_read_unlock();
296 }
297
scsi_device_unbusy(struct scsi_device * sdev,struct scsi_cmnd * cmd)298 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
299 {
300 struct Scsi_Host *shost = sdev->host;
301 struct scsi_target *starget = scsi_target(sdev);
302
303 scsi_dec_host_busy(shost, cmd);
304
305 if (starget->can_queue > 0)
306 atomic_dec(&starget->target_busy);
307
308 sbitmap_put(&sdev->budget_map, cmd->budget_token);
309 cmd->budget_token = -1;
310 }
311
scsi_kick_queue(struct request_queue * q)312 static void scsi_kick_queue(struct request_queue *q)
313 {
314 blk_mq_run_hw_queues(q, false);
315 }
316
317 /*
318 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
319 * and call blk_run_queue for all the scsi_devices on the target -
320 * including current_sdev first.
321 *
322 * Called with *no* scsi locks held.
323 */
scsi_single_lun_run(struct scsi_device * current_sdev)324 static void scsi_single_lun_run(struct scsi_device *current_sdev)
325 {
326 struct Scsi_Host *shost = current_sdev->host;
327 struct scsi_device *sdev, *tmp;
328 struct scsi_target *starget = scsi_target(current_sdev);
329 unsigned long flags;
330
331 spin_lock_irqsave(shost->host_lock, flags);
332 starget->starget_sdev_user = NULL;
333 spin_unlock_irqrestore(shost->host_lock, flags);
334
335 /*
336 * Call blk_run_queue for all LUNs on the target, starting with
337 * current_sdev. We race with others (to set starget_sdev_user),
338 * but in most cases, we will be first. Ideally, each LU on the
339 * target would get some limited time or requests on the target.
340 */
341 scsi_kick_queue(current_sdev->request_queue);
342
343 spin_lock_irqsave(shost->host_lock, flags);
344 if (starget->starget_sdev_user)
345 goto out;
346 list_for_each_entry_safe(sdev, tmp, &starget->devices,
347 same_target_siblings) {
348 if (sdev == current_sdev)
349 continue;
350 if (scsi_device_get(sdev))
351 continue;
352
353 spin_unlock_irqrestore(shost->host_lock, flags);
354 scsi_kick_queue(sdev->request_queue);
355 spin_lock_irqsave(shost->host_lock, flags);
356
357 scsi_device_put(sdev);
358 }
359 out:
360 spin_unlock_irqrestore(shost->host_lock, flags);
361 }
362
scsi_device_is_busy(struct scsi_device * sdev)363 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
364 {
365 if (scsi_device_busy(sdev) >= sdev->queue_depth)
366 return true;
367 if (atomic_read(&sdev->device_blocked) > 0)
368 return true;
369 return false;
370 }
371
scsi_target_is_busy(struct scsi_target * starget)372 static inline bool scsi_target_is_busy(struct scsi_target *starget)
373 {
374 if (starget->can_queue > 0) {
375 if (atomic_read(&starget->target_busy) >= starget->can_queue)
376 return true;
377 if (atomic_read(&starget->target_blocked) > 0)
378 return true;
379 }
380 return false;
381 }
382
scsi_host_is_busy(struct Scsi_Host * shost)383 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
384 {
385 if (atomic_read(&shost->host_blocked) > 0)
386 return true;
387 if (shost->host_self_blocked)
388 return true;
389 return false;
390 }
391
scsi_starved_list_run(struct Scsi_Host * shost)392 static void scsi_starved_list_run(struct Scsi_Host *shost)
393 {
394 LIST_HEAD(starved_list);
395 struct scsi_device *sdev;
396 unsigned long flags;
397
398 spin_lock_irqsave(shost->host_lock, flags);
399 list_splice_init(&shost->starved_list, &starved_list);
400
401 while (!list_empty(&starved_list)) {
402 struct request_queue *slq;
403
404 /*
405 * As long as shost is accepting commands and we have
406 * starved queues, call blk_run_queue. scsi_request_fn
407 * drops the queue_lock and can add us back to the
408 * starved_list.
409 *
410 * host_lock protects the starved_list and starved_entry.
411 * scsi_request_fn must get the host_lock before checking
412 * or modifying starved_list or starved_entry.
413 */
414 if (scsi_host_is_busy(shost))
415 break;
416
417 sdev = list_entry(starved_list.next,
418 struct scsi_device, starved_entry);
419 list_del_init(&sdev->starved_entry);
420 if (scsi_target_is_busy(scsi_target(sdev))) {
421 list_move_tail(&sdev->starved_entry,
422 &shost->starved_list);
423 continue;
424 }
425
426 /*
427 * Once we drop the host lock, a racing scsi_remove_device()
428 * call may remove the sdev from the starved list and destroy
429 * it and the queue. Mitigate by taking a reference to the
430 * queue and never touching the sdev again after we drop the
431 * host lock. Note: if __scsi_remove_device() invokes
432 * blk_cleanup_queue() before the queue is run from this
433 * function then blk_run_queue() will return immediately since
434 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
435 */
436 slq = sdev->request_queue;
437 if (!blk_get_queue(slq))
438 continue;
439 spin_unlock_irqrestore(shost->host_lock, flags);
440
441 scsi_kick_queue(slq);
442 blk_put_queue(slq);
443
444 spin_lock_irqsave(shost->host_lock, flags);
445 }
446 /* put any unprocessed entries back */
447 list_splice(&starved_list, &shost->starved_list);
448 spin_unlock_irqrestore(shost->host_lock, flags);
449 }
450
451 /**
452 * scsi_run_queue - Select a proper request queue to serve next.
453 * @q: last request's queue
454 *
455 * The previous command was completely finished, start a new one if possible.
456 */
scsi_run_queue(struct request_queue * q)457 static void scsi_run_queue(struct request_queue *q)
458 {
459 struct scsi_device *sdev = q->queuedata;
460
461 if (scsi_target(sdev)->single_lun)
462 scsi_single_lun_run(sdev);
463 if (!list_empty(&sdev->host->starved_list))
464 scsi_starved_list_run(sdev->host);
465
466 blk_mq_run_hw_queues(q, false);
467 }
468
scsi_requeue_run_queue(struct work_struct * work)469 void scsi_requeue_run_queue(struct work_struct *work)
470 {
471 struct scsi_device *sdev;
472 struct request_queue *q;
473
474 sdev = container_of(work, struct scsi_device, requeue_work);
475 q = sdev->request_queue;
476 scsi_run_queue(q);
477 }
478
scsi_run_host_queues(struct Scsi_Host * shost)479 void scsi_run_host_queues(struct Scsi_Host *shost)
480 {
481 struct scsi_device *sdev;
482
483 shost_for_each_device(sdev, shost)
484 scsi_run_queue(sdev->request_queue);
485 }
486
scsi_uninit_cmd(struct scsi_cmnd * cmd)487 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
488 {
489 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
490 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
491
492 if (drv->uninit_command)
493 drv->uninit_command(cmd);
494 }
495 }
496
scsi_free_sgtables(struct scsi_cmnd * cmd)497 void scsi_free_sgtables(struct scsi_cmnd *cmd)
498 {
499 if (cmd->sdb.table.nents)
500 sg_free_table_chained(&cmd->sdb.table,
501 SCSI_INLINE_SG_CNT);
502 if (scsi_prot_sg_count(cmd))
503 sg_free_table_chained(&cmd->prot_sdb->table,
504 SCSI_INLINE_PROT_SG_CNT);
505 }
506 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
507
scsi_mq_uninit_cmd(struct scsi_cmnd * cmd)508 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
509 {
510 scsi_free_sgtables(cmd);
511 scsi_uninit_cmd(cmd);
512 }
513
scsi_run_queue_async(struct scsi_device * sdev)514 static void scsi_run_queue_async(struct scsi_device *sdev)
515 {
516 if (scsi_target(sdev)->single_lun ||
517 !list_empty(&sdev->host->starved_list)) {
518 kblockd_schedule_work(&sdev->requeue_work);
519 } else {
520 /*
521 * smp_mb() present in sbitmap_queue_clear() or implied in
522 * .end_io is for ordering writing .device_busy in
523 * scsi_device_unbusy() and reading sdev->restarts.
524 */
525 int old = atomic_read(&sdev->restarts);
526
527 /*
528 * ->restarts has to be kept as non-zero if new budget
529 * contention occurs.
530 *
531 * No need to run queue when either another re-run
532 * queue wins in updating ->restarts or a new budget
533 * contention occurs.
534 */
535 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
536 blk_mq_run_hw_queues(sdev->request_queue, true);
537 }
538 }
539
540 /* Returns false when no more bytes to process, true if there are more */
scsi_end_request(struct request * req,blk_status_t error,unsigned int bytes)541 static bool scsi_end_request(struct request *req, blk_status_t error,
542 unsigned int bytes)
543 {
544 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
545 struct scsi_device *sdev = cmd->device;
546 struct request_queue *q = sdev->request_queue;
547
548 if (blk_update_request(req, error, bytes))
549 return true;
550
551 // XXX:
552 if (blk_queue_add_random(q))
553 add_disk_randomness(req->q->disk);
554
555 if (!blk_rq_is_passthrough(req)) {
556 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
557 cmd->flags &= ~SCMD_INITIALIZED;
558 }
559
560 /*
561 * Calling rcu_barrier() is not necessary here because the
562 * SCSI error handler guarantees that the function called by
563 * call_rcu() has been called before scsi_end_request() is
564 * called.
565 */
566 destroy_rcu_head(&cmd->rcu);
567
568 /*
569 * In the MQ case the command gets freed by __blk_mq_end_request,
570 * so we have to do all cleanup that depends on it earlier.
571 *
572 * We also can't kick the queues from irq context, so we
573 * will have to defer it to a workqueue.
574 */
575 scsi_mq_uninit_cmd(cmd);
576
577 /*
578 * queue is still alive, so grab the ref for preventing it
579 * from being cleaned up during running queue.
580 */
581 percpu_ref_get(&q->q_usage_counter);
582
583 __blk_mq_end_request(req, error);
584
585 scsi_run_queue_async(sdev);
586
587 percpu_ref_put(&q->q_usage_counter);
588 return false;
589 }
590
591 /**
592 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
593 * @cmd: SCSI command
594 * @result: scsi error code
595 *
596 * Translate a SCSI result code into a blk_status_t value. May reset the host
597 * byte of @cmd->result.
598 */
scsi_result_to_blk_status(struct scsi_cmnd * cmd,int result)599 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
600 {
601 switch (host_byte(result)) {
602 case DID_OK:
603 if (scsi_status_is_good(result))
604 return BLK_STS_OK;
605 return BLK_STS_IOERR;
606 case DID_TRANSPORT_FAILFAST:
607 case DID_TRANSPORT_MARGINAL:
608 return BLK_STS_TRANSPORT;
609 case DID_TARGET_FAILURE:
610 set_host_byte(cmd, DID_OK);
611 return BLK_STS_TARGET;
612 case DID_NEXUS_FAILURE:
613 set_host_byte(cmd, DID_OK);
614 return BLK_STS_NEXUS;
615 case DID_ALLOC_FAILURE:
616 set_host_byte(cmd, DID_OK);
617 return BLK_STS_NOSPC;
618 case DID_MEDIUM_ERROR:
619 set_host_byte(cmd, DID_OK);
620 return BLK_STS_MEDIUM;
621 default:
622 return BLK_STS_IOERR;
623 }
624 }
625
626 /**
627 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
628 * @rq: request to examine
629 *
630 * Description:
631 * A request could be merge of IOs which require different failure
632 * handling. This function determines the number of bytes which
633 * can be failed from the beginning of the request without
634 * crossing into area which need to be retried further.
635 *
636 * Return:
637 * The number of bytes to fail.
638 */
scsi_rq_err_bytes(const struct request * rq)639 static unsigned int scsi_rq_err_bytes(const struct request *rq)
640 {
641 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
642 unsigned int bytes = 0;
643 struct bio *bio;
644
645 if (!(rq->rq_flags & RQF_MIXED_MERGE))
646 return blk_rq_bytes(rq);
647
648 /*
649 * Currently the only 'mixing' which can happen is between
650 * different fastfail types. We can safely fail portions
651 * which have all the failfast bits that the first one has -
652 * the ones which are at least as eager to fail as the first
653 * one.
654 */
655 for (bio = rq->bio; bio; bio = bio->bi_next) {
656 if ((bio->bi_opf & ff) != ff)
657 break;
658 bytes += bio->bi_iter.bi_size;
659 }
660
661 /* this could lead to infinite loop */
662 BUG_ON(blk_rq_bytes(rq) && !bytes);
663 return bytes;
664 }
665
scsi_cmd_runtime_exceeced(struct scsi_cmnd * cmd)666 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
667 {
668 struct request *req = scsi_cmd_to_rq(cmd);
669 unsigned long wait_for;
670
671 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
672 return false;
673
674 wait_for = (cmd->allowed + 1) * req->timeout;
675 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
676 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
677 wait_for/HZ);
678 return true;
679 }
680 return false;
681 }
682
683 /*
684 * When ALUA transition state is returned, reprep the cmd to
685 * use the ALUA handler's transition timeout. Delay the reprep
686 * 1 sec to avoid aggressive retries of the target in that
687 * state.
688 */
689 #define ALUA_TRANSITION_REPREP_DELAY 1000
690
691 /* Helper for scsi_io_completion() when special action required. */
scsi_io_completion_action(struct scsi_cmnd * cmd,int result)692 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
693 {
694 struct request *req = scsi_cmd_to_rq(cmd);
695 int level = 0;
696 enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
697 ACTION_RETRY, ACTION_DELAYED_RETRY} action;
698 struct scsi_sense_hdr sshdr;
699 bool sense_valid;
700 bool sense_current = true; /* false implies "deferred sense" */
701 blk_status_t blk_stat;
702
703 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
704 if (sense_valid)
705 sense_current = !scsi_sense_is_deferred(&sshdr);
706
707 blk_stat = scsi_result_to_blk_status(cmd, result);
708
709 if (host_byte(result) == DID_RESET) {
710 /* Third party bus reset or reset for error recovery
711 * reasons. Just retry the command and see what
712 * happens.
713 */
714 action = ACTION_RETRY;
715 } else if (sense_valid && sense_current) {
716 switch (sshdr.sense_key) {
717 case UNIT_ATTENTION:
718 if (cmd->device->removable) {
719 /* Detected disc change. Set a bit
720 * and quietly refuse further access.
721 */
722 cmd->device->changed = 1;
723 action = ACTION_FAIL;
724 } else {
725 /* Must have been a power glitch, or a
726 * bus reset. Could not have been a
727 * media change, so we just retry the
728 * command and see what happens.
729 */
730 action = ACTION_RETRY;
731 }
732 break;
733 case ILLEGAL_REQUEST:
734 /* If we had an ILLEGAL REQUEST returned, then
735 * we may have performed an unsupported
736 * command. The only thing this should be
737 * would be a ten byte read where only a six
738 * byte read was supported. Also, on a system
739 * where READ CAPACITY failed, we may have
740 * read past the end of the disk.
741 */
742 if ((cmd->device->use_10_for_rw &&
743 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
744 (cmd->cmnd[0] == READ_10 ||
745 cmd->cmnd[0] == WRITE_10)) {
746 /* This will issue a new 6-byte command. */
747 cmd->device->use_10_for_rw = 0;
748 action = ACTION_REPREP;
749 } else if (sshdr.asc == 0x10) /* DIX */ {
750 action = ACTION_FAIL;
751 blk_stat = BLK_STS_PROTECTION;
752 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
753 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
754 action = ACTION_FAIL;
755 blk_stat = BLK_STS_TARGET;
756 } else
757 action = ACTION_FAIL;
758 break;
759 case ABORTED_COMMAND:
760 action = ACTION_FAIL;
761 if (sshdr.asc == 0x10) /* DIF */
762 blk_stat = BLK_STS_PROTECTION;
763 break;
764 case NOT_READY:
765 /* If the device is in the process of becoming
766 * ready, or has a temporary blockage, retry.
767 */
768 if (sshdr.asc == 0x04) {
769 switch (sshdr.ascq) {
770 case 0x01: /* becoming ready */
771 case 0x04: /* format in progress */
772 case 0x05: /* rebuild in progress */
773 case 0x06: /* recalculation in progress */
774 case 0x07: /* operation in progress */
775 case 0x08: /* Long write in progress */
776 case 0x09: /* self test in progress */
777 case 0x11: /* notify (enable spinup) required */
778 case 0x14: /* space allocation in progress */
779 case 0x1a: /* start stop unit in progress */
780 case 0x1b: /* sanitize in progress */
781 case 0x1d: /* configuration in progress */
782 case 0x24: /* depopulation in progress */
783 action = ACTION_DELAYED_RETRY;
784 break;
785 case 0x0a: /* ALUA state transition */
786 action = ACTION_DELAYED_REPREP;
787 break;
788 default:
789 action = ACTION_FAIL;
790 break;
791 }
792 } else
793 action = ACTION_FAIL;
794 break;
795 case VOLUME_OVERFLOW:
796 /* See SSC3rXX or current. */
797 action = ACTION_FAIL;
798 break;
799 case DATA_PROTECT:
800 action = ACTION_FAIL;
801 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
802 (sshdr.asc == 0x55 &&
803 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
804 /* Insufficient zone resources */
805 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
806 }
807 break;
808 default:
809 action = ACTION_FAIL;
810 break;
811 }
812 } else
813 action = ACTION_FAIL;
814
815 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
816 action = ACTION_FAIL;
817
818 switch (action) {
819 case ACTION_FAIL:
820 /* Give up and fail the remainder of the request */
821 if (!(req->rq_flags & RQF_QUIET)) {
822 static DEFINE_RATELIMIT_STATE(_rs,
823 DEFAULT_RATELIMIT_INTERVAL,
824 DEFAULT_RATELIMIT_BURST);
825
826 if (unlikely(scsi_logging_level))
827 level =
828 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
829 SCSI_LOG_MLCOMPLETE_BITS);
830
831 /*
832 * if logging is enabled the failure will be printed
833 * in scsi_log_completion(), so avoid duplicate messages
834 */
835 if (!level && __ratelimit(&_rs)) {
836 scsi_print_result(cmd, NULL, FAILED);
837 if (sense_valid)
838 scsi_print_sense(cmd);
839 scsi_print_command(cmd);
840 }
841 }
842 if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
843 return;
844 fallthrough;
845 case ACTION_REPREP:
846 scsi_mq_requeue_cmd(cmd, 0);
847 break;
848 case ACTION_DELAYED_REPREP:
849 scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
850 break;
851 case ACTION_RETRY:
852 /* Retry the same command immediately */
853 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
854 break;
855 case ACTION_DELAYED_RETRY:
856 /* Retry the same command after a delay */
857 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
858 break;
859 }
860 }
861
862 /*
863 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
864 * new result that may suppress further error checking. Also modifies
865 * *blk_statp in some cases.
866 */
scsi_io_completion_nz_result(struct scsi_cmnd * cmd,int result,blk_status_t * blk_statp)867 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
868 blk_status_t *blk_statp)
869 {
870 bool sense_valid;
871 bool sense_current = true; /* false implies "deferred sense" */
872 struct request *req = scsi_cmd_to_rq(cmd);
873 struct scsi_sense_hdr sshdr;
874
875 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
876 if (sense_valid)
877 sense_current = !scsi_sense_is_deferred(&sshdr);
878
879 if (blk_rq_is_passthrough(req)) {
880 if (sense_valid) {
881 /*
882 * SG_IO wants current and deferred errors
883 */
884 cmd->sense_len = min(8 + cmd->sense_buffer[7],
885 SCSI_SENSE_BUFFERSIZE);
886 }
887 if (sense_current)
888 *blk_statp = scsi_result_to_blk_status(cmd, result);
889 } else if (blk_rq_bytes(req) == 0 && sense_current) {
890 /*
891 * Flush commands do not transfers any data, and thus cannot use
892 * good_bytes != blk_rq_bytes(req) as the signal for an error.
893 * This sets *blk_statp explicitly for the problem case.
894 */
895 *blk_statp = scsi_result_to_blk_status(cmd, result);
896 }
897 /*
898 * Recovered errors need reporting, but they're always treated as
899 * success, so fiddle the result code here. For passthrough requests
900 * we already took a copy of the original into sreq->result which
901 * is what gets returned to the user
902 */
903 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
904 bool do_print = true;
905 /*
906 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
907 * skip print since caller wants ATA registers. Only occurs
908 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
909 */
910 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
911 do_print = false;
912 else if (req->rq_flags & RQF_QUIET)
913 do_print = false;
914 if (do_print)
915 scsi_print_sense(cmd);
916 result = 0;
917 /* for passthrough, *blk_statp may be set */
918 *blk_statp = BLK_STS_OK;
919 }
920 /*
921 * Another corner case: the SCSI status byte is non-zero but 'good'.
922 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
923 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
924 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
925 * intermediate statuses (both obsolete in SAM-4) as good.
926 */
927 if ((result & 0xff) && scsi_status_is_good(result)) {
928 result = 0;
929 *blk_statp = BLK_STS_OK;
930 }
931 return result;
932 }
933
934 /**
935 * scsi_io_completion - Completion processing for SCSI commands.
936 * @cmd: command that is finished.
937 * @good_bytes: number of processed bytes.
938 *
939 * We will finish off the specified number of sectors. If we are done, the
940 * command block will be released and the queue function will be goosed. If we
941 * are not done then we have to figure out what to do next:
942 *
943 * a) We can call scsi_mq_requeue_cmd(). The request will be
944 * unprepared and put back on the queue. Then a new command will
945 * be created for it. This should be used if we made forward
946 * progress, or if we want to switch from READ(10) to READ(6) for
947 * example.
948 *
949 * b) We can call scsi_io_completion_action(). The request will be
950 * put back on the queue and retried using the same command as
951 * before, possibly after a delay.
952 *
953 * c) We can call scsi_end_request() with blk_stat other than
954 * BLK_STS_OK, to fail the remainder of the request.
955 */
scsi_io_completion(struct scsi_cmnd * cmd,unsigned int good_bytes)956 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
957 {
958 int result = cmd->result;
959 struct request *req = scsi_cmd_to_rq(cmd);
960 blk_status_t blk_stat = BLK_STS_OK;
961
962 if (unlikely(result)) /* a nz result may or may not be an error */
963 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
964
965 /*
966 * Next deal with any sectors which we were able to correctly
967 * handle.
968 */
969 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
970 "%u sectors total, %d bytes done.\n",
971 blk_rq_sectors(req), good_bytes));
972
973 /*
974 * Failed, zero length commands always need to drop down
975 * to retry code. Fast path should return in this block.
976 */
977 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
978 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
979 return; /* no bytes remaining */
980 }
981
982 /* Kill remainder if no retries. */
983 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
984 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
985 WARN_ONCE(true,
986 "Bytes remaining after failed, no-retry command");
987 return;
988 }
989
990 /*
991 * If there had been no error, but we have leftover bytes in the
992 * request just queue the command up again.
993 */
994 if (likely(result == 0))
995 scsi_mq_requeue_cmd(cmd, 0);
996 else
997 scsi_io_completion_action(cmd, result);
998 }
999
scsi_cmd_needs_dma_drain(struct scsi_device * sdev,struct request * rq)1000 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1001 struct request *rq)
1002 {
1003 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1004 !op_is_write(req_op(rq)) &&
1005 sdev->host->hostt->dma_need_drain(rq);
1006 }
1007
1008 /**
1009 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1010 * @cmd: SCSI command data structure to initialize.
1011 *
1012 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1013 * for @cmd.
1014 *
1015 * Returns:
1016 * * BLK_STS_OK - on success
1017 * * BLK_STS_RESOURCE - if the failure is retryable
1018 * * BLK_STS_IOERR - if the failure is fatal
1019 */
scsi_alloc_sgtables(struct scsi_cmnd * cmd)1020 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1021 {
1022 struct scsi_device *sdev = cmd->device;
1023 struct request *rq = scsi_cmd_to_rq(cmd);
1024 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1025 struct scatterlist *last_sg = NULL;
1026 blk_status_t ret;
1027 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1028 int count;
1029
1030 if (WARN_ON_ONCE(!nr_segs))
1031 return BLK_STS_IOERR;
1032
1033 /*
1034 * Make sure there is space for the drain. The driver must adjust
1035 * max_hw_segments to be prepared for this.
1036 */
1037 if (need_drain)
1038 nr_segs++;
1039
1040 /*
1041 * If sg table allocation fails, requeue request later.
1042 */
1043 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1044 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1045 return BLK_STS_RESOURCE;
1046
1047 /*
1048 * Next, walk the list, and fill in the addresses and sizes of
1049 * each segment.
1050 */
1051 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1052
1053 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1054 unsigned int pad_len =
1055 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1056
1057 last_sg->length += pad_len;
1058 cmd->extra_len += pad_len;
1059 }
1060
1061 if (need_drain) {
1062 sg_unmark_end(last_sg);
1063 last_sg = sg_next(last_sg);
1064 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1065 sg_mark_end(last_sg);
1066
1067 cmd->extra_len += sdev->dma_drain_len;
1068 count++;
1069 }
1070
1071 BUG_ON(count > cmd->sdb.table.nents);
1072 cmd->sdb.table.nents = count;
1073 cmd->sdb.length = blk_rq_payload_bytes(rq);
1074
1075 if (blk_integrity_rq(rq)) {
1076 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1077 int ivecs;
1078
1079 if (WARN_ON_ONCE(!prot_sdb)) {
1080 /*
1081 * This can happen if someone (e.g. multipath)
1082 * queues a command to a device on an adapter
1083 * that does not support DIX.
1084 */
1085 ret = BLK_STS_IOERR;
1086 goto out_free_sgtables;
1087 }
1088
1089 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1090
1091 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1092 prot_sdb->table.sgl,
1093 SCSI_INLINE_PROT_SG_CNT)) {
1094 ret = BLK_STS_RESOURCE;
1095 goto out_free_sgtables;
1096 }
1097
1098 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1099 prot_sdb->table.sgl);
1100 BUG_ON(count > ivecs);
1101 BUG_ON(count > queue_max_integrity_segments(rq->q));
1102
1103 cmd->prot_sdb = prot_sdb;
1104 cmd->prot_sdb->table.nents = count;
1105 }
1106
1107 return BLK_STS_OK;
1108 out_free_sgtables:
1109 scsi_free_sgtables(cmd);
1110 return ret;
1111 }
1112 EXPORT_SYMBOL(scsi_alloc_sgtables);
1113
1114 /**
1115 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1116 * @rq: Request associated with the SCSI command to be initialized.
1117 *
1118 * This function initializes the members of struct scsi_cmnd that must be
1119 * initialized before request processing starts and that won't be
1120 * reinitialized if a SCSI command is requeued.
1121 */
scsi_initialize_rq(struct request * rq)1122 static void scsi_initialize_rq(struct request *rq)
1123 {
1124 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1125
1126 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1127 cmd->cmd_len = MAX_COMMAND_SIZE;
1128 cmd->sense_len = 0;
1129 init_rcu_head(&cmd->rcu);
1130 cmd->jiffies_at_alloc = jiffies;
1131 cmd->retries = 0;
1132 }
1133
scsi_alloc_request(struct request_queue * q,unsigned int op,blk_mq_req_flags_t flags)1134 struct request *scsi_alloc_request(struct request_queue *q,
1135 unsigned int op, blk_mq_req_flags_t flags)
1136 {
1137 struct request *rq;
1138
1139 rq = blk_mq_alloc_request(q, op, flags);
1140 if (!IS_ERR(rq))
1141 scsi_initialize_rq(rq);
1142 return rq;
1143 }
1144 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1145
1146 /*
1147 * Only called when the request isn't completed by SCSI, and not freed by
1148 * SCSI
1149 */
scsi_cleanup_rq(struct request * rq)1150 static void scsi_cleanup_rq(struct request *rq)
1151 {
1152 if (rq->rq_flags & RQF_DONTPREP) {
1153 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1154 rq->rq_flags &= ~RQF_DONTPREP;
1155 }
1156 }
1157
1158 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
scsi_init_command(struct scsi_device * dev,struct scsi_cmnd * cmd)1159 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1160 {
1161 struct request *rq = scsi_cmd_to_rq(cmd);
1162
1163 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1164 cmd->flags |= SCMD_INITIALIZED;
1165 scsi_initialize_rq(rq);
1166 }
1167
1168 cmd->device = dev;
1169 INIT_LIST_HEAD(&cmd->eh_entry);
1170 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1171 }
1172
scsi_setup_scsi_cmnd(struct scsi_device * sdev,struct request * req)1173 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1174 struct request *req)
1175 {
1176 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1177
1178 /*
1179 * Passthrough requests may transfer data, in which case they must
1180 * a bio attached to them. Or they might contain a SCSI command
1181 * that does not transfer data, in which case they may optionally
1182 * submit a request without an attached bio.
1183 */
1184 if (req->bio) {
1185 blk_status_t ret = scsi_alloc_sgtables(cmd);
1186 if (unlikely(ret != BLK_STS_OK))
1187 return ret;
1188 } else {
1189 BUG_ON(blk_rq_bytes(req));
1190
1191 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1192 }
1193
1194 cmd->transfersize = blk_rq_bytes(req);
1195 return BLK_STS_OK;
1196 }
1197
1198 static blk_status_t
scsi_device_state_check(struct scsi_device * sdev,struct request * req)1199 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1200 {
1201 switch (sdev->sdev_state) {
1202 case SDEV_CREATED:
1203 return BLK_STS_OK;
1204 case SDEV_OFFLINE:
1205 case SDEV_TRANSPORT_OFFLINE:
1206 /*
1207 * If the device is offline we refuse to process any
1208 * commands. The device must be brought online
1209 * before trying any recovery commands.
1210 */
1211 if (!sdev->offline_already) {
1212 sdev->offline_already = true;
1213 sdev_printk(KERN_ERR, sdev,
1214 "rejecting I/O to offline device\n");
1215 }
1216 return BLK_STS_IOERR;
1217 case SDEV_DEL:
1218 /*
1219 * If the device is fully deleted, we refuse to
1220 * process any commands as well.
1221 */
1222 sdev_printk(KERN_ERR, sdev,
1223 "rejecting I/O to dead device\n");
1224 return BLK_STS_IOERR;
1225 case SDEV_BLOCK:
1226 case SDEV_CREATED_BLOCK:
1227 return BLK_STS_RESOURCE;
1228 case SDEV_QUIESCE:
1229 /*
1230 * If the device is blocked we only accept power management
1231 * commands.
1232 */
1233 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1234 return BLK_STS_RESOURCE;
1235 return BLK_STS_OK;
1236 default:
1237 /*
1238 * For any other not fully online state we only allow
1239 * power management commands.
1240 */
1241 if (req && !(req->rq_flags & RQF_PM))
1242 return BLK_STS_OFFLINE;
1243 return BLK_STS_OK;
1244 }
1245 }
1246
1247 /*
1248 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1249 * and return the token else return -1.
1250 */
scsi_dev_queue_ready(struct request_queue * q,struct scsi_device * sdev)1251 static inline int scsi_dev_queue_ready(struct request_queue *q,
1252 struct scsi_device *sdev)
1253 {
1254 int token;
1255
1256 token = sbitmap_get(&sdev->budget_map);
1257 if (atomic_read(&sdev->device_blocked)) {
1258 if (token < 0)
1259 goto out;
1260
1261 if (scsi_device_busy(sdev) > 1)
1262 goto out_dec;
1263
1264 /*
1265 * unblock after device_blocked iterates to zero
1266 */
1267 if (atomic_dec_return(&sdev->device_blocked) > 0)
1268 goto out_dec;
1269 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1270 "unblocking device at zero depth\n"));
1271 }
1272
1273 return token;
1274 out_dec:
1275 if (token >= 0)
1276 sbitmap_put(&sdev->budget_map, token);
1277 out:
1278 return -1;
1279 }
1280
1281 /*
1282 * scsi_target_queue_ready: checks if there we can send commands to target
1283 * @sdev: scsi device on starget to check.
1284 */
scsi_target_queue_ready(struct Scsi_Host * shost,struct scsi_device * sdev)1285 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1286 struct scsi_device *sdev)
1287 {
1288 struct scsi_target *starget = scsi_target(sdev);
1289 unsigned int busy;
1290
1291 if (starget->single_lun) {
1292 spin_lock_irq(shost->host_lock);
1293 if (starget->starget_sdev_user &&
1294 starget->starget_sdev_user != sdev) {
1295 spin_unlock_irq(shost->host_lock);
1296 return 0;
1297 }
1298 starget->starget_sdev_user = sdev;
1299 spin_unlock_irq(shost->host_lock);
1300 }
1301
1302 if (starget->can_queue <= 0)
1303 return 1;
1304
1305 busy = atomic_inc_return(&starget->target_busy) - 1;
1306 if (atomic_read(&starget->target_blocked) > 0) {
1307 if (busy)
1308 goto starved;
1309
1310 /*
1311 * unblock after target_blocked iterates to zero
1312 */
1313 if (atomic_dec_return(&starget->target_blocked) > 0)
1314 goto out_dec;
1315
1316 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1317 "unblocking target at zero depth\n"));
1318 }
1319
1320 if (busy >= starget->can_queue)
1321 goto starved;
1322
1323 return 1;
1324
1325 starved:
1326 spin_lock_irq(shost->host_lock);
1327 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1328 spin_unlock_irq(shost->host_lock);
1329 out_dec:
1330 if (starget->can_queue > 0)
1331 atomic_dec(&starget->target_busy);
1332 return 0;
1333 }
1334
1335 /*
1336 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1337 * return 0. We must end up running the queue again whenever 0 is
1338 * returned, else IO can hang.
1339 */
scsi_host_queue_ready(struct request_queue * q,struct Scsi_Host * shost,struct scsi_device * sdev,struct scsi_cmnd * cmd)1340 static inline int scsi_host_queue_ready(struct request_queue *q,
1341 struct Scsi_Host *shost,
1342 struct scsi_device *sdev,
1343 struct scsi_cmnd *cmd)
1344 {
1345 if (scsi_host_in_recovery(shost))
1346 return 0;
1347
1348 if (atomic_read(&shost->host_blocked) > 0) {
1349 if (scsi_host_busy(shost) > 0)
1350 goto starved;
1351
1352 /*
1353 * unblock after host_blocked iterates to zero
1354 */
1355 if (atomic_dec_return(&shost->host_blocked) > 0)
1356 goto out_dec;
1357
1358 SCSI_LOG_MLQUEUE(3,
1359 shost_printk(KERN_INFO, shost,
1360 "unblocking host at zero depth\n"));
1361 }
1362
1363 if (shost->host_self_blocked)
1364 goto starved;
1365
1366 /* We're OK to process the command, so we can't be starved */
1367 if (!list_empty(&sdev->starved_entry)) {
1368 spin_lock_irq(shost->host_lock);
1369 if (!list_empty(&sdev->starved_entry))
1370 list_del_init(&sdev->starved_entry);
1371 spin_unlock_irq(shost->host_lock);
1372 }
1373
1374 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1375
1376 return 1;
1377
1378 starved:
1379 spin_lock_irq(shost->host_lock);
1380 if (list_empty(&sdev->starved_entry))
1381 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1382 spin_unlock_irq(shost->host_lock);
1383 out_dec:
1384 scsi_dec_host_busy(shost, cmd);
1385 return 0;
1386 }
1387
1388 /*
1389 * Busy state exporting function for request stacking drivers.
1390 *
1391 * For efficiency, no lock is taken to check the busy state of
1392 * shost/starget/sdev, since the returned value is not guaranteed and
1393 * may be changed after request stacking drivers call the function,
1394 * regardless of taking lock or not.
1395 *
1396 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1397 * needs to return 'not busy'. Otherwise, request stacking drivers
1398 * may hold requests forever.
1399 */
scsi_mq_lld_busy(struct request_queue * q)1400 static bool scsi_mq_lld_busy(struct request_queue *q)
1401 {
1402 struct scsi_device *sdev = q->queuedata;
1403 struct Scsi_Host *shost;
1404
1405 if (blk_queue_dying(q))
1406 return false;
1407
1408 shost = sdev->host;
1409
1410 /*
1411 * Ignore host/starget busy state.
1412 * Since block layer does not have a concept of fairness across
1413 * multiple queues, congestion of host/starget needs to be handled
1414 * in SCSI layer.
1415 */
1416 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1417 return true;
1418
1419 return false;
1420 }
1421
1422 /*
1423 * Block layer request completion callback. May be called from interrupt
1424 * context.
1425 */
scsi_complete(struct request * rq)1426 static void scsi_complete(struct request *rq)
1427 {
1428 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1429 enum scsi_disposition disposition;
1430
1431 INIT_LIST_HEAD(&cmd->eh_entry);
1432
1433 atomic_inc(&cmd->device->iodone_cnt);
1434 if (cmd->result)
1435 atomic_inc(&cmd->device->ioerr_cnt);
1436
1437 disposition = scsi_decide_disposition(cmd);
1438 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1439 disposition = SUCCESS;
1440
1441 scsi_log_completion(cmd, disposition);
1442
1443 switch (disposition) {
1444 case SUCCESS:
1445 scsi_finish_command(cmd);
1446 break;
1447 case NEEDS_RETRY:
1448 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1449 break;
1450 case ADD_TO_MLQUEUE:
1451 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1452 break;
1453 default:
1454 scsi_eh_scmd_add(cmd);
1455 break;
1456 }
1457 }
1458
1459 /**
1460 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1461 * @cmd: command block we are dispatching.
1462 *
1463 * Return: nonzero return request was rejected and device's queue needs to be
1464 * plugged.
1465 */
scsi_dispatch_cmd(struct scsi_cmnd * cmd)1466 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1467 {
1468 struct Scsi_Host *host = cmd->device->host;
1469 int rtn = 0;
1470
1471 atomic_inc(&cmd->device->iorequest_cnt);
1472
1473 /* check if the device is still usable */
1474 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1475 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1476 * returns an immediate error upwards, and signals
1477 * that the device is no longer present */
1478 cmd->result = DID_NO_CONNECT << 16;
1479 goto done;
1480 }
1481
1482 /* Check to see if the scsi lld made this device blocked. */
1483 if (unlikely(scsi_device_blocked(cmd->device))) {
1484 /*
1485 * in blocked state, the command is just put back on
1486 * the device queue. The suspend state has already
1487 * blocked the queue so future requests should not
1488 * occur until the device transitions out of the
1489 * suspend state.
1490 */
1491 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1492 "queuecommand : device blocked\n"));
1493 return SCSI_MLQUEUE_DEVICE_BUSY;
1494 }
1495
1496 /* Store the LUN value in cmnd, if needed. */
1497 if (cmd->device->lun_in_cdb)
1498 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1499 (cmd->device->lun << 5 & 0xe0);
1500
1501 scsi_log_send(cmd);
1502
1503 /*
1504 * Before we queue this command, check if the command
1505 * length exceeds what the host adapter can handle.
1506 */
1507 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1508 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1509 "queuecommand : command too long. "
1510 "cdb_size=%d host->max_cmd_len=%d\n",
1511 cmd->cmd_len, cmd->device->host->max_cmd_len));
1512 cmd->result = (DID_ABORT << 16);
1513 goto done;
1514 }
1515
1516 if (unlikely(host->shost_state == SHOST_DEL)) {
1517 cmd->result = (DID_NO_CONNECT << 16);
1518 goto done;
1519
1520 }
1521
1522 trace_scsi_dispatch_cmd_start(cmd);
1523 rtn = host->hostt->queuecommand(host, cmd);
1524 if (rtn) {
1525 trace_scsi_dispatch_cmd_error(cmd, rtn);
1526 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1527 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1528 rtn = SCSI_MLQUEUE_HOST_BUSY;
1529
1530 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1531 "queuecommand : request rejected\n"));
1532 }
1533
1534 return rtn;
1535 done:
1536 scsi_done(cmd);
1537 return 0;
1538 }
1539
1540 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
scsi_mq_inline_sgl_size(struct Scsi_Host * shost)1541 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1542 {
1543 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1544 sizeof(struct scatterlist);
1545 }
1546
scsi_prepare_cmd(struct request * req)1547 static blk_status_t scsi_prepare_cmd(struct request *req)
1548 {
1549 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1550 struct scsi_device *sdev = req->q->queuedata;
1551 struct Scsi_Host *shost = sdev->host;
1552 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1553 struct scatterlist *sg;
1554
1555 scsi_init_command(sdev, cmd);
1556
1557 cmd->eh_eflags = 0;
1558 cmd->prot_type = 0;
1559 cmd->prot_flags = 0;
1560 cmd->submitter = 0;
1561 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1562 cmd->underflow = 0;
1563 cmd->transfersize = 0;
1564 cmd->host_scribble = NULL;
1565 cmd->result = 0;
1566 cmd->extra_len = 0;
1567 cmd->state = 0;
1568 if (in_flight)
1569 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1570
1571 /*
1572 * Only clear the driver-private command data if the LLD does not supply
1573 * a function to initialize that data.
1574 */
1575 if (!shost->hostt->init_cmd_priv)
1576 memset(cmd + 1, 0, shost->hostt->cmd_size);
1577
1578 cmd->prot_op = SCSI_PROT_NORMAL;
1579 if (blk_rq_bytes(req))
1580 cmd->sc_data_direction = rq_dma_dir(req);
1581 else
1582 cmd->sc_data_direction = DMA_NONE;
1583
1584 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1585 cmd->sdb.table.sgl = sg;
1586
1587 if (scsi_host_get_prot(shost)) {
1588 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1589
1590 cmd->prot_sdb->table.sgl =
1591 (struct scatterlist *)(cmd->prot_sdb + 1);
1592 }
1593
1594 /*
1595 * Special handling for passthrough commands, which don't go to the ULP
1596 * at all:
1597 */
1598 if (blk_rq_is_passthrough(req))
1599 return scsi_setup_scsi_cmnd(sdev, req);
1600
1601 if (sdev->handler && sdev->handler->prep_fn) {
1602 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1603
1604 if (ret != BLK_STS_OK)
1605 return ret;
1606 }
1607
1608 /* Usually overridden by the ULP */
1609 cmd->allowed = 0;
1610 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1611 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1612 }
1613
scsi_done_internal(struct scsi_cmnd * cmd,bool complete_directly)1614 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1615 {
1616 struct request *req = scsi_cmd_to_rq(cmd);
1617
1618 switch (cmd->submitter) {
1619 case SUBMITTED_BY_BLOCK_LAYER:
1620 break;
1621 case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1622 return scsi_eh_done(cmd);
1623 case SUBMITTED_BY_SCSI_RESET_IOCTL:
1624 return;
1625 }
1626
1627 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1628 return;
1629 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1630 return;
1631 trace_scsi_dispatch_cmd_done(cmd);
1632
1633 if (complete_directly)
1634 blk_mq_complete_request_direct(req, scsi_complete);
1635 else
1636 blk_mq_complete_request(req);
1637 }
1638
scsi_done(struct scsi_cmnd * cmd)1639 void scsi_done(struct scsi_cmnd *cmd)
1640 {
1641 scsi_done_internal(cmd, false);
1642 }
1643 EXPORT_SYMBOL(scsi_done);
1644
scsi_done_direct(struct scsi_cmnd * cmd)1645 void scsi_done_direct(struct scsi_cmnd *cmd)
1646 {
1647 scsi_done_internal(cmd, true);
1648 }
1649 EXPORT_SYMBOL(scsi_done_direct);
1650
scsi_mq_put_budget(struct request_queue * q,int budget_token)1651 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1652 {
1653 struct scsi_device *sdev = q->queuedata;
1654
1655 sbitmap_put(&sdev->budget_map, budget_token);
1656 }
1657
scsi_mq_get_budget(struct request_queue * q)1658 static int scsi_mq_get_budget(struct request_queue *q)
1659 {
1660 struct scsi_device *sdev = q->queuedata;
1661 int token = scsi_dev_queue_ready(q, sdev);
1662
1663 if (token >= 0)
1664 return token;
1665
1666 atomic_inc(&sdev->restarts);
1667
1668 /*
1669 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1670 * .restarts must be incremented before .device_busy is read because the
1671 * code in scsi_run_queue_async() depends on the order of these operations.
1672 */
1673 smp_mb__after_atomic();
1674
1675 /*
1676 * If all in-flight requests originated from this LUN are completed
1677 * before reading .device_busy, sdev->device_busy will be observed as
1678 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1679 * soon. Otherwise, completion of one of these requests will observe
1680 * the .restarts flag, and the request queue will be run for handling
1681 * this request, see scsi_end_request().
1682 */
1683 if (unlikely(scsi_device_busy(sdev) == 0 &&
1684 !scsi_device_blocked(sdev)))
1685 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1686 return -1;
1687 }
1688
scsi_mq_set_rq_budget_token(struct request * req,int token)1689 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1690 {
1691 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1692
1693 cmd->budget_token = token;
1694 }
1695
scsi_mq_get_rq_budget_token(struct request * req)1696 static int scsi_mq_get_rq_budget_token(struct request *req)
1697 {
1698 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1699
1700 return cmd->budget_token;
1701 }
1702
scsi_queue_rq(struct blk_mq_hw_ctx * hctx,const struct blk_mq_queue_data * bd)1703 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1704 const struct blk_mq_queue_data *bd)
1705 {
1706 struct request *req = bd->rq;
1707 struct request_queue *q = req->q;
1708 struct scsi_device *sdev = q->queuedata;
1709 struct Scsi_Host *shost = sdev->host;
1710 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1711 blk_status_t ret;
1712 int reason;
1713
1714 WARN_ON_ONCE(cmd->budget_token < 0);
1715
1716 /*
1717 * If the device is not in running state we will reject some or all
1718 * commands.
1719 */
1720 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1721 ret = scsi_device_state_check(sdev, req);
1722 if (ret != BLK_STS_OK)
1723 goto out_put_budget;
1724 }
1725
1726 ret = BLK_STS_RESOURCE;
1727 if (!scsi_target_queue_ready(shost, sdev))
1728 goto out_put_budget;
1729 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1730 goto out_dec_target_busy;
1731
1732 if (!(req->rq_flags & RQF_DONTPREP)) {
1733 ret = scsi_prepare_cmd(req);
1734 if (ret != BLK_STS_OK)
1735 goto out_dec_host_busy;
1736 req->rq_flags |= RQF_DONTPREP;
1737 } else {
1738 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1739 }
1740
1741 cmd->flags &= SCMD_PRESERVED_FLAGS;
1742 if (sdev->simple_tags)
1743 cmd->flags |= SCMD_TAGGED;
1744 if (bd->last)
1745 cmd->flags |= SCMD_LAST;
1746
1747 scsi_set_resid(cmd, 0);
1748 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1749 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1750
1751 blk_mq_start_request(req);
1752 reason = scsi_dispatch_cmd(cmd);
1753 if (reason) {
1754 scsi_set_blocked(cmd, reason);
1755 ret = BLK_STS_RESOURCE;
1756 goto out_dec_host_busy;
1757 }
1758
1759 return BLK_STS_OK;
1760
1761 out_dec_host_busy:
1762 scsi_dec_host_busy(shost, cmd);
1763 out_dec_target_busy:
1764 if (scsi_target(sdev)->can_queue > 0)
1765 atomic_dec(&scsi_target(sdev)->target_busy);
1766 out_put_budget:
1767 scsi_mq_put_budget(q, cmd->budget_token);
1768 cmd->budget_token = -1;
1769 switch (ret) {
1770 case BLK_STS_OK:
1771 break;
1772 case BLK_STS_RESOURCE:
1773 case BLK_STS_ZONE_RESOURCE:
1774 if (scsi_device_blocked(sdev))
1775 ret = BLK_STS_DEV_RESOURCE;
1776 break;
1777 case BLK_STS_AGAIN:
1778 cmd->result = DID_BUS_BUSY << 16;
1779 if (req->rq_flags & RQF_DONTPREP)
1780 scsi_mq_uninit_cmd(cmd);
1781 break;
1782 default:
1783 if (unlikely(!scsi_device_online(sdev)))
1784 cmd->result = DID_NO_CONNECT << 16;
1785 else
1786 cmd->result = DID_ERROR << 16;
1787 /*
1788 * Make sure to release all allocated resources when
1789 * we hit an error, as we will never see this command
1790 * again.
1791 */
1792 if (req->rq_flags & RQF_DONTPREP)
1793 scsi_mq_uninit_cmd(cmd);
1794 scsi_run_queue_async(sdev);
1795 break;
1796 }
1797 return ret;
1798 }
1799
scsi_timeout(struct request * req,bool reserved)1800 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1801 bool reserved)
1802 {
1803 if (reserved)
1804 return BLK_EH_RESET_TIMER;
1805 return scsi_times_out(req);
1806 }
1807
scsi_mq_init_request(struct blk_mq_tag_set * set,struct request * rq,unsigned int hctx_idx,unsigned int numa_node)1808 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1809 unsigned int hctx_idx, unsigned int numa_node)
1810 {
1811 struct Scsi_Host *shost = set->driver_data;
1812 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1813 struct scatterlist *sg;
1814 int ret = 0;
1815
1816 cmd->sense_buffer =
1817 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1818 if (!cmd->sense_buffer)
1819 return -ENOMEM;
1820
1821 if (scsi_host_get_prot(shost)) {
1822 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1823 shost->hostt->cmd_size;
1824 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1825 }
1826
1827 if (shost->hostt->init_cmd_priv) {
1828 ret = shost->hostt->init_cmd_priv(shost, cmd);
1829 if (ret < 0)
1830 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1831 }
1832
1833 return ret;
1834 }
1835
scsi_mq_exit_request(struct blk_mq_tag_set * set,struct request * rq,unsigned int hctx_idx)1836 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1837 unsigned int hctx_idx)
1838 {
1839 struct Scsi_Host *shost = set->driver_data;
1840 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1841
1842 if (shost->hostt->exit_cmd_priv)
1843 shost->hostt->exit_cmd_priv(shost, cmd);
1844 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1845 }
1846
1847
scsi_mq_poll(struct blk_mq_hw_ctx * hctx,struct io_comp_batch * iob)1848 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1849 {
1850 struct Scsi_Host *shost = hctx->driver_data;
1851
1852 if (shost->hostt->mq_poll)
1853 return shost->hostt->mq_poll(shost, hctx->queue_num);
1854
1855 return 0;
1856 }
1857
scsi_init_hctx(struct blk_mq_hw_ctx * hctx,void * data,unsigned int hctx_idx)1858 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1859 unsigned int hctx_idx)
1860 {
1861 struct Scsi_Host *shost = data;
1862
1863 hctx->driver_data = shost;
1864 return 0;
1865 }
1866
scsi_map_queues(struct blk_mq_tag_set * set)1867 static int scsi_map_queues(struct blk_mq_tag_set *set)
1868 {
1869 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1870
1871 if (shost->hostt->map_queues)
1872 return shost->hostt->map_queues(shost);
1873 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1874 }
1875
__scsi_init_queue(struct Scsi_Host * shost,struct request_queue * q)1876 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1877 {
1878 struct device *dev = shost->dma_dev;
1879
1880 /*
1881 * this limit is imposed by hardware restrictions
1882 */
1883 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1884 SG_MAX_SEGMENTS));
1885
1886 if (scsi_host_prot_dma(shost)) {
1887 shost->sg_prot_tablesize =
1888 min_not_zero(shost->sg_prot_tablesize,
1889 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1890 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1891 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1892 }
1893
1894 if (dev->dma_mask) {
1895 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1896 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1897 }
1898 blk_queue_max_hw_sectors(q, shost->max_sectors);
1899 blk_queue_segment_boundary(q, shost->dma_boundary);
1900 dma_set_seg_boundary(dev, shost->dma_boundary);
1901
1902 blk_queue_max_segment_size(q, shost->max_segment_size);
1903 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1904 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1905
1906 /*
1907 * Set a reasonable default alignment: The larger of 32-byte (dword),
1908 * which is a common minimum for HBAs, and the minimum DMA alignment,
1909 * which is set by the platform.
1910 *
1911 * Devices that require a bigger alignment can increase it later.
1912 */
1913 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1914 }
1915 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1916
1917 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1918 .get_budget = scsi_mq_get_budget,
1919 .put_budget = scsi_mq_put_budget,
1920 .queue_rq = scsi_queue_rq,
1921 .complete = scsi_complete,
1922 .timeout = scsi_timeout,
1923 #ifdef CONFIG_BLK_DEBUG_FS
1924 .show_rq = scsi_show_rq,
1925 #endif
1926 .init_request = scsi_mq_init_request,
1927 .exit_request = scsi_mq_exit_request,
1928 .cleanup_rq = scsi_cleanup_rq,
1929 .busy = scsi_mq_lld_busy,
1930 .map_queues = scsi_map_queues,
1931 .init_hctx = scsi_init_hctx,
1932 .poll = scsi_mq_poll,
1933 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1934 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1935 };
1936
1937
scsi_commit_rqs(struct blk_mq_hw_ctx * hctx)1938 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1939 {
1940 struct Scsi_Host *shost = hctx->driver_data;
1941
1942 shost->hostt->commit_rqs(shost, hctx->queue_num);
1943 }
1944
1945 static const struct blk_mq_ops scsi_mq_ops = {
1946 .get_budget = scsi_mq_get_budget,
1947 .put_budget = scsi_mq_put_budget,
1948 .queue_rq = scsi_queue_rq,
1949 .commit_rqs = scsi_commit_rqs,
1950 .complete = scsi_complete,
1951 .timeout = scsi_timeout,
1952 #ifdef CONFIG_BLK_DEBUG_FS
1953 .show_rq = scsi_show_rq,
1954 #endif
1955 .init_request = scsi_mq_init_request,
1956 .exit_request = scsi_mq_exit_request,
1957 .cleanup_rq = scsi_cleanup_rq,
1958 .busy = scsi_mq_lld_busy,
1959 .map_queues = scsi_map_queues,
1960 .init_hctx = scsi_init_hctx,
1961 .poll = scsi_mq_poll,
1962 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1963 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1964 };
1965
scsi_mq_setup_tags(struct Scsi_Host * shost)1966 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1967 {
1968 unsigned int cmd_size, sgl_size;
1969 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1970
1971 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1972 scsi_mq_inline_sgl_size(shost));
1973 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1974 if (scsi_host_get_prot(shost))
1975 cmd_size += sizeof(struct scsi_data_buffer) +
1976 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1977
1978 memset(tag_set, 0, sizeof(*tag_set));
1979 if (shost->hostt->commit_rqs)
1980 tag_set->ops = &scsi_mq_ops;
1981 else
1982 tag_set->ops = &scsi_mq_ops_no_commit;
1983 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1984 tag_set->nr_maps = shost->nr_maps ? : 1;
1985 tag_set->queue_depth = shost->can_queue;
1986 tag_set->cmd_size = cmd_size;
1987 tag_set->numa_node = dev_to_node(shost->dma_dev);
1988 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1989 tag_set->flags |=
1990 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1991 tag_set->driver_data = shost;
1992 if (shost->host_tagset)
1993 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1994
1995 return blk_mq_alloc_tag_set(tag_set);
1996 }
1997
scsi_mq_destroy_tags(struct Scsi_Host * shost)1998 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1999 {
2000 blk_mq_free_tag_set(&shost->tag_set);
2001 }
2002
2003 /**
2004 * scsi_device_from_queue - return sdev associated with a request_queue
2005 * @q: The request queue to return the sdev from
2006 *
2007 * Return the sdev associated with a request queue or NULL if the
2008 * request_queue does not reference a SCSI device.
2009 */
scsi_device_from_queue(struct request_queue * q)2010 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2011 {
2012 struct scsi_device *sdev = NULL;
2013
2014 if (q->mq_ops == &scsi_mq_ops_no_commit ||
2015 q->mq_ops == &scsi_mq_ops)
2016 sdev = q->queuedata;
2017 if (!sdev || !get_device(&sdev->sdev_gendev))
2018 sdev = NULL;
2019
2020 return sdev;
2021 }
2022 /*
2023 * pktcdvd should have been integrated into the SCSI layers, but for historical
2024 * reasons like the old IDE driver it isn't. This export allows it to safely
2025 * probe if a given device is a SCSI one and only attach to that.
2026 */
2027 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2028 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2029 #endif
2030
2031 /**
2032 * scsi_block_requests - Utility function used by low-level drivers to prevent
2033 * further commands from being queued to the device.
2034 * @shost: host in question
2035 *
2036 * There is no timer nor any other means by which the requests get unblocked
2037 * other than the low-level driver calling scsi_unblock_requests().
2038 */
scsi_block_requests(struct Scsi_Host * shost)2039 void scsi_block_requests(struct Scsi_Host *shost)
2040 {
2041 shost->host_self_blocked = 1;
2042 }
2043 EXPORT_SYMBOL(scsi_block_requests);
2044
2045 /**
2046 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2047 * further commands to be queued to the device.
2048 * @shost: host in question
2049 *
2050 * There is no timer nor any other means by which the requests get unblocked
2051 * other than the low-level driver calling scsi_unblock_requests(). This is done
2052 * as an API function so that changes to the internals of the scsi mid-layer
2053 * won't require wholesale changes to drivers that use this feature.
2054 */
scsi_unblock_requests(struct Scsi_Host * shost)2055 void scsi_unblock_requests(struct Scsi_Host *shost)
2056 {
2057 shost->host_self_blocked = 0;
2058 scsi_run_host_queues(shost);
2059 }
2060 EXPORT_SYMBOL(scsi_unblock_requests);
2061
scsi_exit_queue(void)2062 void scsi_exit_queue(void)
2063 {
2064 kmem_cache_destroy(scsi_sense_cache);
2065 }
2066
2067 /**
2068 * scsi_mode_select - issue a mode select
2069 * @sdev: SCSI device to be queried
2070 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2071 * @sp: Save page bit (0 == don't save, 1 == save)
2072 * @buffer: request buffer (may not be smaller than eight bytes)
2073 * @len: length of request buffer.
2074 * @timeout: command timeout
2075 * @retries: number of retries before failing
2076 * @data: returns a structure abstracting the mode header data
2077 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2078 * must be SCSI_SENSE_BUFFERSIZE big.
2079 *
2080 * Returns zero if successful; negative error number or scsi
2081 * status on error
2082 *
2083 */
scsi_mode_select(struct scsi_device * sdev,int pf,int sp,unsigned char * buffer,int len,int timeout,int retries,struct scsi_mode_data * data,struct scsi_sense_hdr * sshdr)2084 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2085 unsigned char *buffer, int len, int timeout, int retries,
2086 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2087 {
2088 unsigned char cmd[10];
2089 unsigned char *real_buffer;
2090 int ret;
2091
2092 memset(cmd, 0, sizeof(cmd));
2093 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2094
2095 /*
2096 * Use MODE SELECT(10) if the device asked for it or if the mode page
2097 * and the mode select header cannot fit within the maximumm 255 bytes
2098 * of the MODE SELECT(6) command.
2099 */
2100 if (sdev->use_10_for_ms ||
2101 len + 4 > 255 ||
2102 data->block_descriptor_length > 255) {
2103 if (len > 65535 - 8)
2104 return -EINVAL;
2105 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2106 if (!real_buffer)
2107 return -ENOMEM;
2108 memcpy(real_buffer + 8, buffer, len);
2109 len += 8;
2110 real_buffer[0] = 0;
2111 real_buffer[1] = 0;
2112 real_buffer[2] = data->medium_type;
2113 real_buffer[3] = data->device_specific;
2114 real_buffer[4] = data->longlba ? 0x01 : 0;
2115 real_buffer[5] = 0;
2116 put_unaligned_be16(data->block_descriptor_length,
2117 &real_buffer[6]);
2118
2119 cmd[0] = MODE_SELECT_10;
2120 put_unaligned_be16(len, &cmd[7]);
2121 } else {
2122 if (data->longlba)
2123 return -EINVAL;
2124
2125 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2126 if (!real_buffer)
2127 return -ENOMEM;
2128 memcpy(real_buffer + 4, buffer, len);
2129 len += 4;
2130 real_buffer[0] = 0;
2131 real_buffer[1] = data->medium_type;
2132 real_buffer[2] = data->device_specific;
2133 real_buffer[3] = data->block_descriptor_length;
2134
2135 cmd[0] = MODE_SELECT;
2136 cmd[4] = len;
2137 }
2138
2139 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2140 sshdr, timeout, retries, NULL);
2141 kfree(real_buffer);
2142 return ret;
2143 }
2144 EXPORT_SYMBOL_GPL(scsi_mode_select);
2145
2146 /**
2147 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2148 * @sdev: SCSI device to be queried
2149 * @dbd: set to prevent mode sense from returning block descriptors
2150 * @modepage: mode page being requested
2151 * @buffer: request buffer (may not be smaller than eight bytes)
2152 * @len: length of request buffer.
2153 * @timeout: command timeout
2154 * @retries: number of retries before failing
2155 * @data: returns a structure abstracting the mode header data
2156 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2157 * must be SCSI_SENSE_BUFFERSIZE big.
2158 *
2159 * Returns zero if successful, or a negative error number on failure
2160 */
2161 int
scsi_mode_sense(struct scsi_device * sdev,int dbd,int modepage,unsigned char * buffer,int len,int timeout,int retries,struct scsi_mode_data * data,struct scsi_sense_hdr * sshdr)2162 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2163 unsigned char *buffer, int len, int timeout, int retries,
2164 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2165 {
2166 unsigned char cmd[12];
2167 int use_10_for_ms;
2168 int header_length;
2169 int result, retry_count = retries;
2170 struct scsi_sense_hdr my_sshdr;
2171
2172 memset(data, 0, sizeof(*data));
2173 memset(&cmd[0], 0, 12);
2174
2175 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2176 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2177 cmd[2] = modepage;
2178
2179 /* caller might not be interested in sense, but we need it */
2180 if (!sshdr)
2181 sshdr = &my_sshdr;
2182
2183 retry:
2184 use_10_for_ms = sdev->use_10_for_ms || len > 255;
2185
2186 if (use_10_for_ms) {
2187 if (len < 8 || len > 65535)
2188 return -EINVAL;
2189
2190 cmd[0] = MODE_SENSE_10;
2191 put_unaligned_be16(len, &cmd[7]);
2192 header_length = 8;
2193 } else {
2194 if (len < 4)
2195 return -EINVAL;
2196
2197 cmd[0] = MODE_SENSE;
2198 cmd[4] = len;
2199 header_length = 4;
2200 }
2201
2202 memset(buffer, 0, len);
2203
2204 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2205 sshdr, timeout, retries, NULL);
2206 if (result < 0)
2207 return result;
2208
2209 /* This code looks awful: what it's doing is making sure an
2210 * ILLEGAL REQUEST sense return identifies the actual command
2211 * byte as the problem. MODE_SENSE commands can return
2212 * ILLEGAL REQUEST if the code page isn't supported */
2213
2214 if (!scsi_status_is_good(result)) {
2215 if (scsi_sense_valid(sshdr)) {
2216 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2217 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2218 /*
2219 * Invalid command operation code: retry using
2220 * MODE SENSE(6) if this was a MODE SENSE(10)
2221 * request, except if the request mode page is
2222 * too large for MODE SENSE single byte
2223 * allocation length field.
2224 */
2225 if (use_10_for_ms) {
2226 if (len > 255)
2227 return -EIO;
2228 sdev->use_10_for_ms = 0;
2229 goto retry;
2230 }
2231 }
2232 if (scsi_status_is_check_condition(result) &&
2233 sshdr->sense_key == UNIT_ATTENTION &&
2234 retry_count) {
2235 retry_count--;
2236 goto retry;
2237 }
2238 }
2239 return -EIO;
2240 }
2241 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2242 (modepage == 6 || modepage == 8))) {
2243 /* Initio breakage? */
2244 header_length = 0;
2245 data->length = 13;
2246 data->medium_type = 0;
2247 data->device_specific = 0;
2248 data->longlba = 0;
2249 data->block_descriptor_length = 0;
2250 } else if (use_10_for_ms) {
2251 data->length = get_unaligned_be16(&buffer[0]) + 2;
2252 data->medium_type = buffer[2];
2253 data->device_specific = buffer[3];
2254 data->longlba = buffer[4] & 0x01;
2255 data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2256 } else {
2257 data->length = buffer[0] + 1;
2258 data->medium_type = buffer[1];
2259 data->device_specific = buffer[2];
2260 data->block_descriptor_length = buffer[3];
2261 }
2262 data->header_length = header_length;
2263
2264 return 0;
2265 }
2266 EXPORT_SYMBOL(scsi_mode_sense);
2267
2268 /**
2269 * scsi_test_unit_ready - test if unit is ready
2270 * @sdev: scsi device to change the state of.
2271 * @timeout: command timeout
2272 * @retries: number of retries before failing
2273 * @sshdr: outpout pointer for decoded sense information.
2274 *
2275 * Returns zero if unsuccessful or an error if TUR failed. For
2276 * removable media, UNIT_ATTENTION sets ->changed flag.
2277 **/
2278 int
scsi_test_unit_ready(struct scsi_device * sdev,int timeout,int retries,struct scsi_sense_hdr * sshdr)2279 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2280 struct scsi_sense_hdr *sshdr)
2281 {
2282 char cmd[] = {
2283 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2284 };
2285 int result;
2286
2287 /* try to eat the UNIT_ATTENTION if there are enough retries */
2288 do {
2289 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2290 timeout, 1, NULL);
2291 if (sdev->removable && scsi_sense_valid(sshdr) &&
2292 sshdr->sense_key == UNIT_ATTENTION)
2293 sdev->changed = 1;
2294 } while (scsi_sense_valid(sshdr) &&
2295 sshdr->sense_key == UNIT_ATTENTION && --retries);
2296
2297 return result;
2298 }
2299 EXPORT_SYMBOL(scsi_test_unit_ready);
2300
2301 /**
2302 * scsi_device_set_state - Take the given device through the device state model.
2303 * @sdev: scsi device to change the state of.
2304 * @state: state to change to.
2305 *
2306 * Returns zero if successful or an error if the requested
2307 * transition is illegal.
2308 */
2309 int
scsi_device_set_state(struct scsi_device * sdev,enum scsi_device_state state)2310 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2311 {
2312 enum scsi_device_state oldstate = sdev->sdev_state;
2313
2314 if (state == oldstate)
2315 return 0;
2316
2317 switch (state) {
2318 case SDEV_CREATED:
2319 switch (oldstate) {
2320 case SDEV_CREATED_BLOCK:
2321 break;
2322 default:
2323 goto illegal;
2324 }
2325 break;
2326
2327 case SDEV_RUNNING:
2328 switch (oldstate) {
2329 case SDEV_CREATED:
2330 case SDEV_OFFLINE:
2331 case SDEV_TRANSPORT_OFFLINE:
2332 case SDEV_QUIESCE:
2333 case SDEV_BLOCK:
2334 break;
2335 default:
2336 goto illegal;
2337 }
2338 break;
2339
2340 case SDEV_QUIESCE:
2341 switch (oldstate) {
2342 case SDEV_RUNNING:
2343 case SDEV_OFFLINE:
2344 case SDEV_TRANSPORT_OFFLINE:
2345 break;
2346 default:
2347 goto illegal;
2348 }
2349 break;
2350
2351 case SDEV_OFFLINE:
2352 case SDEV_TRANSPORT_OFFLINE:
2353 switch (oldstate) {
2354 case SDEV_CREATED:
2355 case SDEV_RUNNING:
2356 case SDEV_QUIESCE:
2357 case SDEV_BLOCK:
2358 break;
2359 default:
2360 goto illegal;
2361 }
2362 break;
2363
2364 case SDEV_BLOCK:
2365 switch (oldstate) {
2366 case SDEV_RUNNING:
2367 case SDEV_CREATED_BLOCK:
2368 case SDEV_QUIESCE:
2369 case SDEV_OFFLINE:
2370 break;
2371 default:
2372 goto illegal;
2373 }
2374 break;
2375
2376 case SDEV_CREATED_BLOCK:
2377 switch (oldstate) {
2378 case SDEV_CREATED:
2379 break;
2380 default:
2381 goto illegal;
2382 }
2383 break;
2384
2385 case SDEV_CANCEL:
2386 switch (oldstate) {
2387 case SDEV_CREATED:
2388 case SDEV_RUNNING:
2389 case SDEV_QUIESCE:
2390 case SDEV_OFFLINE:
2391 case SDEV_TRANSPORT_OFFLINE:
2392 break;
2393 default:
2394 goto illegal;
2395 }
2396 break;
2397
2398 case SDEV_DEL:
2399 switch (oldstate) {
2400 case SDEV_CREATED:
2401 case SDEV_RUNNING:
2402 case SDEV_OFFLINE:
2403 case SDEV_TRANSPORT_OFFLINE:
2404 case SDEV_CANCEL:
2405 case SDEV_BLOCK:
2406 case SDEV_CREATED_BLOCK:
2407 break;
2408 default:
2409 goto illegal;
2410 }
2411 break;
2412
2413 }
2414 sdev->offline_already = false;
2415 sdev->sdev_state = state;
2416 return 0;
2417
2418 illegal:
2419 SCSI_LOG_ERROR_RECOVERY(1,
2420 sdev_printk(KERN_ERR, sdev,
2421 "Illegal state transition %s->%s",
2422 scsi_device_state_name(oldstate),
2423 scsi_device_state_name(state))
2424 );
2425 return -EINVAL;
2426 }
2427 EXPORT_SYMBOL(scsi_device_set_state);
2428
2429 /**
2430 * scsi_evt_emit - emit a single SCSI device uevent
2431 * @sdev: associated SCSI device
2432 * @evt: event to emit
2433 *
2434 * Send a single uevent (scsi_event) to the associated scsi_device.
2435 */
scsi_evt_emit(struct scsi_device * sdev,struct scsi_event * evt)2436 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2437 {
2438 int idx = 0;
2439 char *envp[3];
2440
2441 switch (evt->evt_type) {
2442 case SDEV_EVT_MEDIA_CHANGE:
2443 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2444 break;
2445 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2446 scsi_rescan_device(&sdev->sdev_gendev);
2447 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2448 break;
2449 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2450 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2451 break;
2452 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2453 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2454 break;
2455 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2456 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2457 break;
2458 case SDEV_EVT_LUN_CHANGE_REPORTED:
2459 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2460 break;
2461 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2462 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2463 break;
2464 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2465 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2466 break;
2467 default:
2468 /* do nothing */
2469 break;
2470 }
2471
2472 envp[idx++] = NULL;
2473
2474 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2475 }
2476
2477 /**
2478 * scsi_evt_thread - send a uevent for each scsi event
2479 * @work: work struct for scsi_device
2480 *
2481 * Dispatch queued events to their associated scsi_device kobjects
2482 * as uevents.
2483 */
scsi_evt_thread(struct work_struct * work)2484 void scsi_evt_thread(struct work_struct *work)
2485 {
2486 struct scsi_device *sdev;
2487 enum scsi_device_event evt_type;
2488 LIST_HEAD(event_list);
2489
2490 sdev = container_of(work, struct scsi_device, event_work);
2491
2492 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2493 if (test_and_clear_bit(evt_type, sdev->pending_events))
2494 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2495
2496 while (1) {
2497 struct scsi_event *evt;
2498 struct list_head *this, *tmp;
2499 unsigned long flags;
2500
2501 spin_lock_irqsave(&sdev->list_lock, flags);
2502 list_splice_init(&sdev->event_list, &event_list);
2503 spin_unlock_irqrestore(&sdev->list_lock, flags);
2504
2505 if (list_empty(&event_list))
2506 break;
2507
2508 list_for_each_safe(this, tmp, &event_list) {
2509 evt = list_entry(this, struct scsi_event, node);
2510 list_del(&evt->node);
2511 scsi_evt_emit(sdev, evt);
2512 kfree(evt);
2513 }
2514 }
2515 }
2516
2517 /**
2518 * sdev_evt_send - send asserted event to uevent thread
2519 * @sdev: scsi_device event occurred on
2520 * @evt: event to send
2521 *
2522 * Assert scsi device event asynchronously.
2523 */
sdev_evt_send(struct scsi_device * sdev,struct scsi_event * evt)2524 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2525 {
2526 unsigned long flags;
2527
2528 #if 0
2529 /* FIXME: currently this check eliminates all media change events
2530 * for polled devices. Need to update to discriminate between AN
2531 * and polled events */
2532 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2533 kfree(evt);
2534 return;
2535 }
2536 #endif
2537
2538 spin_lock_irqsave(&sdev->list_lock, flags);
2539 list_add_tail(&evt->node, &sdev->event_list);
2540 schedule_work(&sdev->event_work);
2541 spin_unlock_irqrestore(&sdev->list_lock, flags);
2542 }
2543 EXPORT_SYMBOL_GPL(sdev_evt_send);
2544
2545 /**
2546 * sdev_evt_alloc - allocate a new scsi event
2547 * @evt_type: type of event to allocate
2548 * @gfpflags: GFP flags for allocation
2549 *
2550 * Allocates and returns a new scsi_event.
2551 */
sdev_evt_alloc(enum scsi_device_event evt_type,gfp_t gfpflags)2552 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2553 gfp_t gfpflags)
2554 {
2555 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2556 if (!evt)
2557 return NULL;
2558
2559 evt->evt_type = evt_type;
2560 INIT_LIST_HEAD(&evt->node);
2561
2562 /* evt_type-specific initialization, if any */
2563 switch (evt_type) {
2564 case SDEV_EVT_MEDIA_CHANGE:
2565 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2566 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2567 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2568 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2569 case SDEV_EVT_LUN_CHANGE_REPORTED:
2570 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2571 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2572 default:
2573 /* do nothing */
2574 break;
2575 }
2576
2577 return evt;
2578 }
2579 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2580
2581 /**
2582 * sdev_evt_send_simple - send asserted event to uevent thread
2583 * @sdev: scsi_device event occurred on
2584 * @evt_type: type of event to send
2585 * @gfpflags: GFP flags for allocation
2586 *
2587 * Assert scsi device event asynchronously, given an event type.
2588 */
sdev_evt_send_simple(struct scsi_device * sdev,enum scsi_device_event evt_type,gfp_t gfpflags)2589 void sdev_evt_send_simple(struct scsi_device *sdev,
2590 enum scsi_device_event evt_type, gfp_t gfpflags)
2591 {
2592 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2593 if (!evt) {
2594 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2595 evt_type);
2596 return;
2597 }
2598
2599 sdev_evt_send(sdev, evt);
2600 }
2601 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2602
2603 /**
2604 * scsi_device_quiesce - Block all commands except power management.
2605 * @sdev: scsi device to quiesce.
2606 *
2607 * This works by trying to transition to the SDEV_QUIESCE state
2608 * (which must be a legal transition). When the device is in this
2609 * state, only power management requests will be accepted, all others will
2610 * be deferred.
2611 *
2612 * Must be called with user context, may sleep.
2613 *
2614 * Returns zero if unsuccessful or an error if not.
2615 */
2616 int
scsi_device_quiesce(struct scsi_device * sdev)2617 scsi_device_quiesce(struct scsi_device *sdev)
2618 {
2619 struct request_queue *q = sdev->request_queue;
2620 int err;
2621
2622 /*
2623 * It is allowed to call scsi_device_quiesce() multiple times from
2624 * the same context but concurrent scsi_device_quiesce() calls are
2625 * not allowed.
2626 */
2627 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2628
2629 if (sdev->quiesced_by == current)
2630 return 0;
2631
2632 blk_set_pm_only(q);
2633
2634 blk_mq_freeze_queue(q);
2635 /*
2636 * Ensure that the effect of blk_set_pm_only() will be visible
2637 * for percpu_ref_tryget() callers that occur after the queue
2638 * unfreeze even if the queue was already frozen before this function
2639 * was called. See also https://lwn.net/Articles/573497/.
2640 */
2641 synchronize_rcu();
2642 blk_mq_unfreeze_queue(q);
2643
2644 mutex_lock(&sdev->state_mutex);
2645 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2646 if (err == 0)
2647 sdev->quiesced_by = current;
2648 else
2649 blk_clear_pm_only(q);
2650 mutex_unlock(&sdev->state_mutex);
2651
2652 return err;
2653 }
2654 EXPORT_SYMBOL(scsi_device_quiesce);
2655
2656 /**
2657 * scsi_device_resume - Restart user issued commands to a quiesced device.
2658 * @sdev: scsi device to resume.
2659 *
2660 * Moves the device from quiesced back to running and restarts the
2661 * queues.
2662 *
2663 * Must be called with user context, may sleep.
2664 */
scsi_device_resume(struct scsi_device * sdev)2665 void scsi_device_resume(struct scsi_device *sdev)
2666 {
2667 /* check if the device state was mutated prior to resume, and if
2668 * so assume the state is being managed elsewhere (for example
2669 * device deleted during suspend)
2670 */
2671 mutex_lock(&sdev->state_mutex);
2672 if (sdev->sdev_state == SDEV_QUIESCE)
2673 scsi_device_set_state(sdev, SDEV_RUNNING);
2674 if (sdev->quiesced_by) {
2675 sdev->quiesced_by = NULL;
2676 blk_clear_pm_only(sdev->request_queue);
2677 }
2678 mutex_unlock(&sdev->state_mutex);
2679 }
2680 EXPORT_SYMBOL(scsi_device_resume);
2681
2682 static void
device_quiesce_fn(struct scsi_device * sdev,void * data)2683 device_quiesce_fn(struct scsi_device *sdev, void *data)
2684 {
2685 scsi_device_quiesce(sdev);
2686 }
2687
2688 void
scsi_target_quiesce(struct scsi_target * starget)2689 scsi_target_quiesce(struct scsi_target *starget)
2690 {
2691 starget_for_each_device(starget, NULL, device_quiesce_fn);
2692 }
2693 EXPORT_SYMBOL(scsi_target_quiesce);
2694
2695 static void
device_resume_fn(struct scsi_device * sdev,void * data)2696 device_resume_fn(struct scsi_device *sdev, void *data)
2697 {
2698 scsi_device_resume(sdev);
2699 }
2700
2701 void
scsi_target_resume(struct scsi_target * starget)2702 scsi_target_resume(struct scsi_target *starget)
2703 {
2704 starget_for_each_device(starget, NULL, device_resume_fn);
2705 }
2706 EXPORT_SYMBOL(scsi_target_resume);
2707
__scsi_internal_device_block_nowait(struct scsi_device * sdev)2708 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2709 {
2710 if (scsi_device_set_state(sdev, SDEV_BLOCK))
2711 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2712
2713 return 0;
2714 }
2715
scsi_start_queue(struct scsi_device * sdev)2716 void scsi_start_queue(struct scsi_device *sdev)
2717 {
2718 if (cmpxchg(&sdev->queue_stopped, 1, 0))
2719 blk_mq_unquiesce_queue(sdev->request_queue);
2720 }
2721
scsi_stop_queue(struct scsi_device * sdev,bool nowait)2722 static void scsi_stop_queue(struct scsi_device *sdev, bool nowait)
2723 {
2724 /*
2725 * The atomic variable of ->queue_stopped covers that
2726 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2727 *
2728 * However, we still need to wait until quiesce is done
2729 * in case that queue has been stopped.
2730 */
2731 if (!cmpxchg(&sdev->queue_stopped, 0, 1)) {
2732 if (nowait)
2733 blk_mq_quiesce_queue_nowait(sdev->request_queue);
2734 else
2735 blk_mq_quiesce_queue(sdev->request_queue);
2736 } else {
2737 if (!nowait)
2738 blk_mq_wait_quiesce_done(sdev->request_queue);
2739 }
2740 }
2741
2742 /**
2743 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2744 * @sdev: device to block
2745 *
2746 * Pause SCSI command processing on the specified device. Does not sleep.
2747 *
2748 * Returns zero if successful or a negative error code upon failure.
2749 *
2750 * Notes:
2751 * This routine transitions the device to the SDEV_BLOCK state (which must be
2752 * a legal transition). When the device is in this state, command processing
2753 * is paused until the device leaves the SDEV_BLOCK state. See also
2754 * scsi_internal_device_unblock_nowait().
2755 */
scsi_internal_device_block_nowait(struct scsi_device * sdev)2756 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2757 {
2758 int ret = __scsi_internal_device_block_nowait(sdev);
2759
2760 /*
2761 * The device has transitioned to SDEV_BLOCK. Stop the
2762 * block layer from calling the midlayer with this device's
2763 * request queue.
2764 */
2765 if (!ret)
2766 scsi_stop_queue(sdev, true);
2767 return ret;
2768 }
2769 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2770
2771 /**
2772 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2773 * @sdev: device to block
2774 *
2775 * Pause SCSI command processing on the specified device and wait until all
2776 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2777 *
2778 * Returns zero if successful or a negative error code upon failure.
2779 *
2780 * Note:
2781 * This routine transitions the device to the SDEV_BLOCK state (which must be
2782 * a legal transition). When the device is in this state, command processing
2783 * is paused until the device leaves the SDEV_BLOCK state. See also
2784 * scsi_internal_device_unblock().
2785 */
scsi_internal_device_block(struct scsi_device * sdev)2786 static int scsi_internal_device_block(struct scsi_device *sdev)
2787 {
2788 int err;
2789
2790 mutex_lock(&sdev->state_mutex);
2791 err = __scsi_internal_device_block_nowait(sdev);
2792 if (err == 0)
2793 scsi_stop_queue(sdev, false);
2794 mutex_unlock(&sdev->state_mutex);
2795
2796 return err;
2797 }
2798
2799 /**
2800 * scsi_internal_device_unblock_nowait - resume a device after a block request
2801 * @sdev: device to resume
2802 * @new_state: state to set the device to after unblocking
2803 *
2804 * Restart the device queue for a previously suspended SCSI device. Does not
2805 * sleep.
2806 *
2807 * Returns zero if successful or a negative error code upon failure.
2808 *
2809 * Notes:
2810 * This routine transitions the device to the SDEV_RUNNING state or to one of
2811 * the offline states (which must be a legal transition) allowing the midlayer
2812 * to goose the queue for this device.
2813 */
scsi_internal_device_unblock_nowait(struct scsi_device * sdev,enum scsi_device_state new_state)2814 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2815 enum scsi_device_state new_state)
2816 {
2817 switch (new_state) {
2818 case SDEV_RUNNING:
2819 case SDEV_TRANSPORT_OFFLINE:
2820 break;
2821 default:
2822 return -EINVAL;
2823 }
2824
2825 /*
2826 * Try to transition the scsi device to SDEV_RUNNING or one of the
2827 * offlined states and goose the device queue if successful.
2828 */
2829 switch (sdev->sdev_state) {
2830 case SDEV_BLOCK:
2831 case SDEV_TRANSPORT_OFFLINE:
2832 sdev->sdev_state = new_state;
2833 break;
2834 case SDEV_CREATED_BLOCK:
2835 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2836 new_state == SDEV_OFFLINE)
2837 sdev->sdev_state = new_state;
2838 else
2839 sdev->sdev_state = SDEV_CREATED;
2840 break;
2841 case SDEV_CANCEL:
2842 case SDEV_OFFLINE:
2843 break;
2844 default:
2845 return -EINVAL;
2846 }
2847 scsi_start_queue(sdev);
2848
2849 return 0;
2850 }
2851 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2852
2853 /**
2854 * scsi_internal_device_unblock - resume a device after a block request
2855 * @sdev: device to resume
2856 * @new_state: state to set the device to after unblocking
2857 *
2858 * Restart the device queue for a previously suspended SCSI device. May sleep.
2859 *
2860 * Returns zero if successful or a negative error code upon failure.
2861 *
2862 * Notes:
2863 * This routine transitions the device to the SDEV_RUNNING state or to one of
2864 * the offline states (which must be a legal transition) allowing the midlayer
2865 * to goose the queue for this device.
2866 */
scsi_internal_device_unblock(struct scsi_device * sdev,enum scsi_device_state new_state)2867 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2868 enum scsi_device_state new_state)
2869 {
2870 int ret;
2871
2872 mutex_lock(&sdev->state_mutex);
2873 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2874 mutex_unlock(&sdev->state_mutex);
2875
2876 return ret;
2877 }
2878
2879 static void
device_block(struct scsi_device * sdev,void * data)2880 device_block(struct scsi_device *sdev, void *data)
2881 {
2882 int ret;
2883
2884 ret = scsi_internal_device_block(sdev);
2885
2886 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2887 dev_name(&sdev->sdev_gendev), ret);
2888 }
2889
2890 static int
target_block(struct device * dev,void * data)2891 target_block(struct device *dev, void *data)
2892 {
2893 if (scsi_is_target_device(dev))
2894 starget_for_each_device(to_scsi_target(dev), NULL,
2895 device_block);
2896 return 0;
2897 }
2898
2899 void
scsi_target_block(struct device * dev)2900 scsi_target_block(struct device *dev)
2901 {
2902 if (scsi_is_target_device(dev))
2903 starget_for_each_device(to_scsi_target(dev), NULL,
2904 device_block);
2905 else
2906 device_for_each_child(dev, NULL, target_block);
2907 }
2908 EXPORT_SYMBOL_GPL(scsi_target_block);
2909
2910 static void
device_unblock(struct scsi_device * sdev,void * data)2911 device_unblock(struct scsi_device *sdev, void *data)
2912 {
2913 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2914 }
2915
2916 static int
target_unblock(struct device * dev,void * data)2917 target_unblock(struct device *dev, void *data)
2918 {
2919 if (scsi_is_target_device(dev))
2920 starget_for_each_device(to_scsi_target(dev), data,
2921 device_unblock);
2922 return 0;
2923 }
2924
2925 void
scsi_target_unblock(struct device * dev,enum scsi_device_state new_state)2926 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2927 {
2928 if (scsi_is_target_device(dev))
2929 starget_for_each_device(to_scsi_target(dev), &new_state,
2930 device_unblock);
2931 else
2932 device_for_each_child(dev, &new_state, target_unblock);
2933 }
2934 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2935
2936 int
scsi_host_block(struct Scsi_Host * shost)2937 scsi_host_block(struct Scsi_Host *shost)
2938 {
2939 struct scsi_device *sdev;
2940 int ret = 0;
2941
2942 /*
2943 * Call scsi_internal_device_block_nowait so we can avoid
2944 * calling synchronize_rcu() for each LUN.
2945 */
2946 shost_for_each_device(sdev, shost) {
2947 mutex_lock(&sdev->state_mutex);
2948 ret = scsi_internal_device_block_nowait(sdev);
2949 mutex_unlock(&sdev->state_mutex);
2950 if (ret) {
2951 scsi_device_put(sdev);
2952 break;
2953 }
2954 }
2955
2956 /*
2957 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2958 * calling synchronize_rcu() once is enough.
2959 */
2960 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2961
2962 if (!ret)
2963 synchronize_rcu();
2964
2965 return ret;
2966 }
2967 EXPORT_SYMBOL_GPL(scsi_host_block);
2968
2969 int
scsi_host_unblock(struct Scsi_Host * shost,int new_state)2970 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2971 {
2972 struct scsi_device *sdev;
2973 int ret = 0;
2974
2975 shost_for_each_device(sdev, shost) {
2976 ret = scsi_internal_device_unblock(sdev, new_state);
2977 if (ret) {
2978 scsi_device_put(sdev);
2979 break;
2980 }
2981 }
2982 return ret;
2983 }
2984 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2985
2986 /**
2987 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2988 * @sgl: scatter-gather list
2989 * @sg_count: number of segments in sg
2990 * @offset: offset in bytes into sg, on return offset into the mapped area
2991 * @len: bytes to map, on return number of bytes mapped
2992 *
2993 * Returns virtual address of the start of the mapped page
2994 */
scsi_kmap_atomic_sg(struct scatterlist * sgl,int sg_count,size_t * offset,size_t * len)2995 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2996 size_t *offset, size_t *len)
2997 {
2998 int i;
2999 size_t sg_len = 0, len_complete = 0;
3000 struct scatterlist *sg;
3001 struct page *page;
3002
3003 WARN_ON(!irqs_disabled());
3004
3005 for_each_sg(sgl, sg, sg_count, i) {
3006 len_complete = sg_len; /* Complete sg-entries */
3007 sg_len += sg->length;
3008 if (sg_len > *offset)
3009 break;
3010 }
3011
3012 if (unlikely(i == sg_count)) {
3013 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3014 "elements %d\n",
3015 __func__, sg_len, *offset, sg_count);
3016 WARN_ON(1);
3017 return NULL;
3018 }
3019
3020 /* Offset starting from the beginning of first page in this sg-entry */
3021 *offset = *offset - len_complete + sg->offset;
3022
3023 /* Assumption: contiguous pages can be accessed as "page + i" */
3024 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3025 *offset &= ~PAGE_MASK;
3026
3027 /* Bytes in this sg-entry from *offset to the end of the page */
3028 sg_len = PAGE_SIZE - *offset;
3029 if (*len > sg_len)
3030 *len = sg_len;
3031
3032 return kmap_atomic(page);
3033 }
3034 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3035
3036 /**
3037 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3038 * @virt: virtual address to be unmapped
3039 */
scsi_kunmap_atomic_sg(void * virt)3040 void scsi_kunmap_atomic_sg(void *virt)
3041 {
3042 kunmap_atomic(virt);
3043 }
3044 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3045
sdev_disable_disk_events(struct scsi_device * sdev)3046 void sdev_disable_disk_events(struct scsi_device *sdev)
3047 {
3048 atomic_inc(&sdev->disk_events_disable_depth);
3049 }
3050 EXPORT_SYMBOL(sdev_disable_disk_events);
3051
sdev_enable_disk_events(struct scsi_device * sdev)3052 void sdev_enable_disk_events(struct scsi_device *sdev)
3053 {
3054 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3055 return;
3056 atomic_dec(&sdev->disk_events_disable_depth);
3057 }
3058 EXPORT_SYMBOL(sdev_enable_disk_events);
3059
designator_prio(const unsigned char * d)3060 static unsigned char designator_prio(const unsigned char *d)
3061 {
3062 if (d[1] & 0x30)
3063 /* not associated with LUN */
3064 return 0;
3065
3066 if (d[3] == 0)
3067 /* invalid length */
3068 return 0;
3069
3070 /*
3071 * Order of preference for lun descriptor:
3072 * - SCSI name string
3073 * - NAA IEEE Registered Extended
3074 * - EUI-64 based 16-byte
3075 * - EUI-64 based 12-byte
3076 * - NAA IEEE Registered
3077 * - NAA IEEE Extended
3078 * - EUI-64 based 8-byte
3079 * - SCSI name string (truncated)
3080 * - T10 Vendor ID
3081 * as longer descriptors reduce the likelyhood
3082 * of identification clashes.
3083 */
3084
3085 switch (d[1] & 0xf) {
3086 case 8:
3087 /* SCSI name string, variable-length UTF-8 */
3088 return 9;
3089 case 3:
3090 switch (d[4] >> 4) {
3091 case 6:
3092 /* NAA registered extended */
3093 return 8;
3094 case 5:
3095 /* NAA registered */
3096 return 5;
3097 case 4:
3098 /* NAA extended */
3099 return 4;
3100 case 3:
3101 /* NAA locally assigned */
3102 return 1;
3103 default:
3104 break;
3105 }
3106 break;
3107 case 2:
3108 switch (d[3]) {
3109 case 16:
3110 /* EUI64-based, 16 byte */
3111 return 7;
3112 case 12:
3113 /* EUI64-based, 12 byte */
3114 return 6;
3115 case 8:
3116 /* EUI64-based, 8 byte */
3117 return 3;
3118 default:
3119 break;
3120 }
3121 break;
3122 case 1:
3123 /* T10 vendor ID */
3124 return 1;
3125 default:
3126 break;
3127 }
3128
3129 return 0;
3130 }
3131
3132 /**
3133 * scsi_vpd_lun_id - return a unique device identification
3134 * @sdev: SCSI device
3135 * @id: buffer for the identification
3136 * @id_len: length of the buffer
3137 *
3138 * Copies a unique device identification into @id based
3139 * on the information in the VPD page 0x83 of the device.
3140 * The string will be formatted as a SCSI name string.
3141 *
3142 * Returns the length of the identification or error on failure.
3143 * If the identifier is longer than the supplied buffer the actual
3144 * identifier length is returned and the buffer is not zero-padded.
3145 */
scsi_vpd_lun_id(struct scsi_device * sdev,char * id,size_t id_len)3146 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3147 {
3148 u8 cur_id_prio = 0;
3149 u8 cur_id_size = 0;
3150 const unsigned char *d, *cur_id_str;
3151 const struct scsi_vpd *vpd_pg83;
3152 int id_size = -EINVAL;
3153
3154 rcu_read_lock();
3155 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3156 if (!vpd_pg83) {
3157 rcu_read_unlock();
3158 return -ENXIO;
3159 }
3160
3161 /* The id string must be at least 20 bytes + terminating NULL byte */
3162 if (id_len < 21) {
3163 rcu_read_unlock();
3164 return -EINVAL;
3165 }
3166
3167 memset(id, 0, id_len);
3168 for (d = vpd_pg83->data + 4;
3169 d < vpd_pg83->data + vpd_pg83->len;
3170 d += d[3] + 4) {
3171 u8 prio = designator_prio(d);
3172
3173 if (prio == 0 || cur_id_prio > prio)
3174 continue;
3175
3176 switch (d[1] & 0xf) {
3177 case 0x1:
3178 /* T10 Vendor ID */
3179 if (cur_id_size > d[3])
3180 break;
3181 cur_id_prio = prio;
3182 cur_id_size = d[3];
3183 if (cur_id_size + 4 > id_len)
3184 cur_id_size = id_len - 4;
3185 cur_id_str = d + 4;
3186 id_size = snprintf(id, id_len, "t10.%*pE",
3187 cur_id_size, cur_id_str);
3188 break;
3189 case 0x2:
3190 /* EUI-64 */
3191 cur_id_prio = prio;
3192 cur_id_size = d[3];
3193 cur_id_str = d + 4;
3194 switch (cur_id_size) {
3195 case 8:
3196 id_size = snprintf(id, id_len,
3197 "eui.%8phN",
3198 cur_id_str);
3199 break;
3200 case 12:
3201 id_size = snprintf(id, id_len,
3202 "eui.%12phN",
3203 cur_id_str);
3204 break;
3205 case 16:
3206 id_size = snprintf(id, id_len,
3207 "eui.%16phN",
3208 cur_id_str);
3209 break;
3210 default:
3211 break;
3212 }
3213 break;
3214 case 0x3:
3215 /* NAA */
3216 cur_id_prio = prio;
3217 cur_id_size = d[3];
3218 cur_id_str = d + 4;
3219 switch (cur_id_size) {
3220 case 8:
3221 id_size = snprintf(id, id_len,
3222 "naa.%8phN",
3223 cur_id_str);
3224 break;
3225 case 16:
3226 id_size = snprintf(id, id_len,
3227 "naa.%16phN",
3228 cur_id_str);
3229 break;
3230 default:
3231 break;
3232 }
3233 break;
3234 case 0x8:
3235 /* SCSI name string */
3236 if (cur_id_size > d[3])
3237 break;
3238 /* Prefer others for truncated descriptor */
3239 if (d[3] > id_len) {
3240 prio = 2;
3241 if (cur_id_prio > prio)
3242 break;
3243 }
3244 cur_id_prio = prio;
3245 cur_id_size = id_size = d[3];
3246 cur_id_str = d + 4;
3247 if (cur_id_size >= id_len)
3248 cur_id_size = id_len - 1;
3249 memcpy(id, cur_id_str, cur_id_size);
3250 break;
3251 default:
3252 break;
3253 }
3254 }
3255 rcu_read_unlock();
3256
3257 return id_size;
3258 }
3259 EXPORT_SYMBOL(scsi_vpd_lun_id);
3260
3261 /*
3262 * scsi_vpd_tpg_id - return a target port group identifier
3263 * @sdev: SCSI device
3264 *
3265 * Returns the Target Port Group identifier from the information
3266 * froom VPD page 0x83 of the device.
3267 *
3268 * Returns the identifier or error on failure.
3269 */
scsi_vpd_tpg_id(struct scsi_device * sdev,int * rel_id)3270 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3271 {
3272 const unsigned char *d;
3273 const struct scsi_vpd *vpd_pg83;
3274 int group_id = -EAGAIN, rel_port = -1;
3275
3276 rcu_read_lock();
3277 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3278 if (!vpd_pg83) {
3279 rcu_read_unlock();
3280 return -ENXIO;
3281 }
3282
3283 d = vpd_pg83->data + 4;
3284 while (d < vpd_pg83->data + vpd_pg83->len) {
3285 switch (d[1] & 0xf) {
3286 case 0x4:
3287 /* Relative target port */
3288 rel_port = get_unaligned_be16(&d[6]);
3289 break;
3290 case 0x5:
3291 /* Target port group */
3292 group_id = get_unaligned_be16(&d[6]);
3293 break;
3294 default:
3295 break;
3296 }
3297 d += d[3] + 4;
3298 }
3299 rcu_read_unlock();
3300
3301 if (group_id >= 0 && rel_id && rel_port != -1)
3302 *rel_id = rel_port;
3303
3304 return group_id;
3305 }
3306 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3307
3308 /**
3309 * scsi_build_sense - build sense data for a command
3310 * @scmd: scsi command for which the sense should be formatted
3311 * @desc: Sense format (non-zero == descriptor format,
3312 * 0 == fixed format)
3313 * @key: Sense key
3314 * @asc: Additional sense code
3315 * @ascq: Additional sense code qualifier
3316 *
3317 **/
scsi_build_sense(struct scsi_cmnd * scmd,int desc,u8 key,u8 asc,u8 ascq)3318 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3319 {
3320 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3321 scmd->result = SAM_STAT_CHECK_CONDITION;
3322 }
3323 EXPORT_SYMBOL_GPL(scsi_build_sense);
3324