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