Loading...
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
56static struct kmem_cache *scsi_sense_cache;
57static DEFINE_MUTEX(scsi_sense_cache_mutex);
58
59static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
60
61int 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
78static void
79scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
80{
81 struct Scsi_Host *host = cmd->device->host;
82 struct scsi_device *device = cmd->device;
83 struct scsi_target *starget = scsi_target(device);
84
85 /*
86 * Set the appropriate busy bit for the device/host.
87 *
88 * If the host/device isn't busy, assume that something actually
89 * completed, and that we should be able to queue a command now.
90 *
91 * Note that the prior mid-layer assumption that any host could
92 * always queue at least one command is now broken. The mid-layer
93 * will implement a user specifiable stall (see
94 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
95 * if a command is requeued with no other commands outstanding
96 * either for the device or for the host.
97 */
98 switch (reason) {
99 case SCSI_MLQUEUE_HOST_BUSY:
100 atomic_set(&host->host_blocked, host->max_host_blocked);
101 break;
102 case SCSI_MLQUEUE_DEVICE_BUSY:
103 case SCSI_MLQUEUE_EH_RETRY:
104 atomic_set(&device->device_blocked,
105 device->max_device_blocked);
106 break;
107 case SCSI_MLQUEUE_TARGET_BUSY:
108 atomic_set(&starget->target_blocked,
109 starget->max_target_blocked);
110 break;
111 }
112}
113
114static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd, unsigned long msecs)
115{
116 struct request *rq = scsi_cmd_to_rq(cmd);
117
118 if (rq->rq_flags & RQF_DONTPREP) {
119 rq->rq_flags &= ~RQF_DONTPREP;
120 scsi_mq_uninit_cmd(cmd);
121 } else {
122 WARN_ON_ONCE(true);
123 }
124
125 blk_mq_requeue_request(rq, false);
126 if (!scsi_host_in_recovery(cmd->device->host))
127 blk_mq_delay_kick_requeue_list(rq->q, msecs);
128}
129
130/**
131 * __scsi_queue_insert - private queue insertion
132 * @cmd: The SCSI command being requeued
133 * @reason: The reason for the requeue
134 * @unbusy: Whether the queue should be unbusied
135 *
136 * This is a private queue insertion. The public interface
137 * scsi_queue_insert() always assumes the queue should be unbusied
138 * because it's always called before the completion. This function is
139 * for a requeue after completion, which should only occur in this
140 * file.
141 */
142static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
143{
144 struct scsi_device *device = cmd->device;
145
146 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
147 "Inserting command %p into mlqueue\n", cmd));
148
149 scsi_set_blocked(cmd, reason);
150
151 /*
152 * Decrement the counters, since these commands are no longer
153 * active on the host/device.
154 */
155 if (unbusy)
156 scsi_device_unbusy(device, cmd);
157
158 /*
159 * Requeue this command. It will go before all other commands
160 * that are already in the queue. Schedule requeue work under
161 * lock such that the kblockd_schedule_work() call happens
162 * before blk_mq_destroy_queue() finishes.
163 */
164 cmd->result = 0;
165
166 blk_mq_requeue_request(scsi_cmd_to_rq(cmd),
167 !scsi_host_in_recovery(cmd->device->host));
168}
169
170/**
171 * scsi_queue_insert - Reinsert a command in the queue.
172 * @cmd: command that we are adding to queue.
173 * @reason: why we are inserting command to queue.
174 *
175 * We do this for one of two cases. Either the host is busy and it cannot accept
176 * any more commands for the time being, or the device returned QUEUE_FULL and
177 * can accept no more commands.
178 *
179 * Context: This could be called either from an interrupt context or a normal
180 * process context.
181 */
182void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
183{
184 __scsi_queue_insert(cmd, reason, true);
185}
186
187/**
188 * scsi_execute_cmd - insert request and wait for the result
189 * @sdev: scsi_device
190 * @cmd: scsi command
191 * @opf: block layer request cmd_flags
192 * @buffer: data buffer
193 * @bufflen: len of buffer
194 * @timeout: request timeout in HZ
195 * @retries: number of times to retry request
196 * @args: Optional args. See struct definition for field descriptions
197 *
198 * Returns the scsi_cmnd result field if a command was executed, or a negative
199 * Linux error code if we didn't get that far.
200 */
201int scsi_execute_cmd(struct scsi_device *sdev, const unsigned char *cmd,
202 blk_opf_t opf, void *buffer, unsigned int bufflen,
203 int timeout, int retries,
204 const struct scsi_exec_args *args)
205{
206 static const struct scsi_exec_args default_args;
207 struct request *req;
208 struct scsi_cmnd *scmd;
209 int ret;
210
211 if (!args)
212 args = &default_args;
213 else if (WARN_ON_ONCE(args->sense &&
214 args->sense_len != SCSI_SENSE_BUFFERSIZE))
215 return -EINVAL;
216
217 req = scsi_alloc_request(sdev->request_queue, opf, args->req_flags);
218 if (IS_ERR(req))
219 return PTR_ERR(req);
220
221 if (bufflen) {
222 ret = blk_rq_map_kern(sdev->request_queue, req,
223 buffer, bufflen, GFP_NOIO);
224 if (ret)
225 goto out;
226 }
227 scmd = blk_mq_rq_to_pdu(req);
228 scmd->cmd_len = COMMAND_SIZE(cmd[0]);
229 memcpy(scmd->cmnd, cmd, scmd->cmd_len);
230 scmd->allowed = retries;
231 scmd->flags |= args->scmd_flags;
232 req->timeout = timeout;
233 req->rq_flags |= RQF_QUIET;
234
235 /*
236 * head injection *required* here otherwise quiesce won't work
237 */
238 blk_execute_rq(req, true);
239
240 /*
241 * Some devices (USB mass-storage in particular) may transfer
242 * garbage data together with a residue indicating that the data
243 * is invalid. Prevent the garbage from being misinterpreted
244 * and prevent security leaks by zeroing out the excess data.
245 */
246 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
247 memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
248
249 if (args->resid)
250 *args->resid = scmd->resid_len;
251 if (args->sense)
252 memcpy(args->sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
253 if (args->sshdr)
254 scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
255 args->sshdr);
256
257 ret = scmd->result;
258 out:
259 blk_mq_free_request(req);
260
261 return ret;
262}
263EXPORT_SYMBOL(scsi_execute_cmd);
264
265/*
266 * Wake up the error handler if necessary. Avoid as follows that the error
267 * handler is not woken up if host in-flight requests number ==
268 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
269 * with an RCU read lock in this function to ensure that this function in
270 * its entirety either finishes before scsi_eh_scmd_add() increases the
271 * host_failed counter or that it notices the shost state change made by
272 * scsi_eh_scmd_add().
273 */
274static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
275{
276 unsigned long flags;
277
278 rcu_read_lock();
279 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
280 if (unlikely(scsi_host_in_recovery(shost))) {
281 unsigned int busy = scsi_host_busy(shost);
282
283 spin_lock_irqsave(shost->host_lock, flags);
284 if (shost->host_failed || shost->host_eh_scheduled)
285 scsi_eh_wakeup(shost, busy);
286 spin_unlock_irqrestore(shost->host_lock, flags);
287 }
288 rcu_read_unlock();
289}
290
291void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
292{
293 struct Scsi_Host *shost = sdev->host;
294 struct scsi_target *starget = scsi_target(sdev);
295
296 scsi_dec_host_busy(shost, cmd);
297
298 if (starget->can_queue > 0)
299 atomic_dec(&starget->target_busy);
300
301 sbitmap_put(&sdev->budget_map, cmd->budget_token);
302 cmd->budget_token = -1;
303}
304
305/*
306 * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
307 * interrupts disabled.
308 */
309static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)
310{
311 struct scsi_device *current_sdev = data;
312
313 if (sdev != current_sdev)
314 blk_mq_run_hw_queues(sdev->request_queue, true);
315}
316
317/*
318 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
319 * and call blk_run_queue for all the scsi_devices on the target -
320 * including current_sdev first.
321 *
322 * Called with *no* scsi locks held.
323 */
324static void scsi_single_lun_run(struct scsi_device *current_sdev)
325{
326 struct Scsi_Host *shost = current_sdev->host;
327 struct scsi_target *starget = scsi_target(current_sdev);
328 unsigned long flags;
329
330 spin_lock_irqsave(shost->host_lock, flags);
331 starget->starget_sdev_user = NULL;
332 spin_unlock_irqrestore(shost->host_lock, flags);
333
334 /*
335 * Call blk_run_queue for all LUNs on the target, starting with
336 * current_sdev. We race with others (to set starget_sdev_user),
337 * but in most cases, we will be first. Ideally, each LU on the
338 * target would get some limited time or requests on the target.
339 */
340 blk_mq_run_hw_queues(current_sdev->request_queue,
341 shost->queuecommand_may_block);
342
343 spin_lock_irqsave(shost->host_lock, flags);
344 if (!starget->starget_sdev_user)
345 __starget_for_each_device(starget, current_sdev,
346 scsi_kick_sdev_queue);
347 spin_unlock_irqrestore(shost->host_lock, flags);
348}
349
350static inline bool scsi_device_is_busy(struct scsi_device *sdev)
351{
352 if (scsi_device_busy(sdev) >= sdev->queue_depth)
353 return true;
354 if (atomic_read(&sdev->device_blocked) > 0)
355 return true;
356 return false;
357}
358
359static inline bool scsi_target_is_busy(struct scsi_target *starget)
360{
361 if (starget->can_queue > 0) {
362 if (atomic_read(&starget->target_busy) >= starget->can_queue)
363 return true;
364 if (atomic_read(&starget->target_blocked) > 0)
365 return true;
366 }
367 return false;
368}
369
370static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
371{
372 if (atomic_read(&shost->host_blocked) > 0)
373 return true;
374 if (shost->host_self_blocked)
375 return true;
376 return false;
377}
378
379static void scsi_starved_list_run(struct Scsi_Host *shost)
380{
381 LIST_HEAD(starved_list);
382 struct scsi_device *sdev;
383 unsigned long flags;
384
385 spin_lock_irqsave(shost->host_lock, flags);
386 list_splice_init(&shost->starved_list, &starved_list);
387
388 while (!list_empty(&starved_list)) {
389 struct request_queue *slq;
390
391 /*
392 * As long as shost is accepting commands and we have
393 * starved queues, call blk_run_queue. scsi_request_fn
394 * drops the queue_lock and can add us back to the
395 * starved_list.
396 *
397 * host_lock protects the starved_list and starved_entry.
398 * scsi_request_fn must get the host_lock before checking
399 * or modifying starved_list or starved_entry.
400 */
401 if (scsi_host_is_busy(shost))
402 break;
403
404 sdev = list_entry(starved_list.next,
405 struct scsi_device, starved_entry);
406 list_del_init(&sdev->starved_entry);
407 if (scsi_target_is_busy(scsi_target(sdev))) {
408 list_move_tail(&sdev->starved_entry,
409 &shost->starved_list);
410 continue;
411 }
412
413 /*
414 * Once we drop the host lock, a racing scsi_remove_device()
415 * call may remove the sdev from the starved list and destroy
416 * it and the queue. Mitigate by taking a reference to the
417 * queue and never touching the sdev again after we drop the
418 * host lock. Note: if __scsi_remove_device() invokes
419 * blk_mq_destroy_queue() before the queue is run from this
420 * function then blk_run_queue() will return immediately since
421 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
422 */
423 slq = sdev->request_queue;
424 if (!blk_get_queue(slq))
425 continue;
426 spin_unlock_irqrestore(shost->host_lock, flags);
427
428 blk_mq_run_hw_queues(slq, false);
429 blk_put_queue(slq);
430
431 spin_lock_irqsave(shost->host_lock, flags);
432 }
433 /* put any unprocessed entries back */
434 list_splice(&starved_list, &shost->starved_list);
435 spin_unlock_irqrestore(shost->host_lock, flags);
436}
437
438/**
439 * scsi_run_queue - Select a proper request queue to serve next.
440 * @q: last request's queue
441 *
442 * The previous command was completely finished, start a new one if possible.
443 */
444static void scsi_run_queue(struct request_queue *q)
445{
446 struct scsi_device *sdev = q->queuedata;
447
448 if (scsi_target(sdev)->single_lun)
449 scsi_single_lun_run(sdev);
450 if (!list_empty(&sdev->host->starved_list))
451 scsi_starved_list_run(sdev->host);
452
453 /* Note: blk_mq_kick_requeue_list() runs the queue asynchronously. */
454 blk_mq_kick_requeue_list(q);
455}
456
457void scsi_requeue_run_queue(struct work_struct *work)
458{
459 struct scsi_device *sdev;
460 struct request_queue *q;
461
462 sdev = container_of(work, struct scsi_device, requeue_work);
463 q = sdev->request_queue;
464 scsi_run_queue(q);
465}
466
467void scsi_run_host_queues(struct Scsi_Host *shost)
468{
469 struct scsi_device *sdev;
470
471 shost_for_each_device(sdev, shost)
472 scsi_run_queue(sdev->request_queue);
473}
474
475static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
476{
477 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
478 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
479
480 if (drv->uninit_command)
481 drv->uninit_command(cmd);
482 }
483}
484
485void scsi_free_sgtables(struct scsi_cmnd *cmd)
486{
487 if (cmd->sdb.table.nents)
488 sg_free_table_chained(&cmd->sdb.table,
489 SCSI_INLINE_SG_CNT);
490 if (scsi_prot_sg_count(cmd))
491 sg_free_table_chained(&cmd->prot_sdb->table,
492 SCSI_INLINE_PROT_SG_CNT);
493}
494EXPORT_SYMBOL_GPL(scsi_free_sgtables);
495
496static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
497{
498 scsi_free_sgtables(cmd);
499 scsi_uninit_cmd(cmd);
500}
501
502static void scsi_run_queue_async(struct scsi_device *sdev)
503{
504 if (scsi_host_in_recovery(sdev->host))
505 return;
506
507 if (scsi_target(sdev)->single_lun ||
508 !list_empty(&sdev->host->starved_list)) {
509 kblockd_schedule_work(&sdev->requeue_work);
510 } else {
511 /*
512 * smp_mb() present in sbitmap_queue_clear() or implied in
513 * .end_io is for ordering writing .device_busy in
514 * scsi_device_unbusy() and reading sdev->restarts.
515 */
516 int old = atomic_read(&sdev->restarts);
517
518 /*
519 * ->restarts has to be kept as non-zero if new budget
520 * contention occurs.
521 *
522 * No need to run queue when either another re-run
523 * queue wins in updating ->restarts or a new budget
524 * contention occurs.
525 */
526 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
527 blk_mq_run_hw_queues(sdev->request_queue, true);
528 }
529}
530
531/* Returns false when no more bytes to process, true if there are more */
532static bool scsi_end_request(struct request *req, blk_status_t error,
533 unsigned int bytes)
534{
535 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
536 struct scsi_device *sdev = cmd->device;
537 struct request_queue *q = sdev->request_queue;
538
539 if (blk_update_request(req, error, bytes))
540 return true;
541
542 // XXX:
543 if (blk_queue_add_random(q))
544 add_disk_randomness(req->q->disk);
545
546 if (!blk_rq_is_passthrough(req)) {
547 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
548 cmd->flags &= ~SCMD_INITIALIZED;
549 }
550
551 /*
552 * Calling rcu_barrier() is not necessary here because the
553 * SCSI error handler guarantees that the function called by
554 * call_rcu() has been called before scsi_end_request() is
555 * called.
556 */
557 destroy_rcu_head(&cmd->rcu);
558
559 /*
560 * In the MQ case the command gets freed by __blk_mq_end_request,
561 * so we have to do all cleanup that depends on it earlier.
562 *
563 * We also can't kick the queues from irq context, so we
564 * will have to defer it to a workqueue.
565 */
566 scsi_mq_uninit_cmd(cmd);
567
568 /*
569 * queue is still alive, so grab the ref for preventing it
570 * from being cleaned up during running queue.
571 */
572 percpu_ref_get(&q->q_usage_counter);
573
574 __blk_mq_end_request(req, error);
575
576 scsi_run_queue_async(sdev);
577
578 percpu_ref_put(&q->q_usage_counter);
579 return false;
580}
581
582/**
583 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
584 * @result: scsi error code
585 *
586 * Translate a SCSI result code into a blk_status_t value.
587 */
588static blk_status_t scsi_result_to_blk_status(int result)
589{
590 /*
591 * Check the scsi-ml byte first in case we converted a host or status
592 * byte.
593 */
594 switch (scsi_ml_byte(result)) {
595 case SCSIML_STAT_OK:
596 break;
597 case SCSIML_STAT_RESV_CONFLICT:
598 return BLK_STS_RESV_CONFLICT;
599 case SCSIML_STAT_NOSPC:
600 return BLK_STS_NOSPC;
601 case SCSIML_STAT_MED_ERROR:
602 return BLK_STS_MEDIUM;
603 case SCSIML_STAT_TGT_FAILURE:
604 return BLK_STS_TARGET;
605 case SCSIML_STAT_DL_TIMEOUT:
606 return BLK_STS_DURATION_LIMIT;
607 }
608
609 switch (host_byte(result)) {
610 case DID_OK:
611 if (scsi_status_is_good(result))
612 return BLK_STS_OK;
613 return BLK_STS_IOERR;
614 case DID_TRANSPORT_FAILFAST:
615 case DID_TRANSPORT_MARGINAL:
616 return BLK_STS_TRANSPORT;
617 default:
618 return BLK_STS_IOERR;
619 }
620}
621
622/**
623 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
624 * @rq: request to examine
625 *
626 * Description:
627 * A request could be merge of IOs which require different failure
628 * handling. This function determines the number of bytes which
629 * can be failed from the beginning of the request without
630 * crossing into area which need to be retried further.
631 *
632 * Return:
633 * The number of bytes to fail.
634 */
635static unsigned int scsi_rq_err_bytes(const struct request *rq)
636{
637 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
638 unsigned int bytes = 0;
639 struct bio *bio;
640
641 if (!(rq->rq_flags & RQF_MIXED_MERGE))
642 return blk_rq_bytes(rq);
643
644 /*
645 * Currently the only 'mixing' which can happen is between
646 * different fastfail types. We can safely fail portions
647 * which have all the failfast bits that the first one has -
648 * the ones which are at least as eager to fail as the first
649 * one.
650 */
651 for (bio = rq->bio; bio; bio = bio->bi_next) {
652 if ((bio->bi_opf & ff) != ff)
653 break;
654 bytes += bio->bi_iter.bi_size;
655 }
656
657 /* this could lead to infinite loop */
658 BUG_ON(blk_rq_bytes(rq) && !bytes);
659 return bytes;
660}
661
662static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
663{
664 struct request *req = scsi_cmd_to_rq(cmd);
665 unsigned long wait_for;
666
667 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
668 return false;
669
670 wait_for = (cmd->allowed + 1) * req->timeout;
671 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
672 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
673 wait_for/HZ);
674 return true;
675 }
676 return false;
677}
678
679/*
680 * When ALUA transition state is returned, reprep the cmd to
681 * use the ALUA handler's transition timeout. Delay the reprep
682 * 1 sec to avoid aggressive retries of the target in that
683 * state.
684 */
685#define ALUA_TRANSITION_REPREP_DELAY 1000
686
687/* Helper for scsi_io_completion() when special action required. */
688static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
689{
690 struct request *req = scsi_cmd_to_rq(cmd);
691 int level = 0;
692 enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
693 ACTION_RETRY, ACTION_DELAYED_RETRY} action;
694 struct scsi_sense_hdr sshdr;
695 bool sense_valid;
696 bool sense_current = true; /* false implies "deferred sense" */
697 blk_status_t blk_stat;
698
699 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
700 if (sense_valid)
701 sense_current = !scsi_sense_is_deferred(&sshdr);
702
703 blk_stat = scsi_result_to_blk_status(result);
704
705 if (host_byte(result) == DID_RESET) {
706 /* Third party bus reset or reset for error recovery
707 * reasons. Just retry the command and see what
708 * happens.
709 */
710 action = ACTION_RETRY;
711 } else if (sense_valid && sense_current) {
712 switch (sshdr.sense_key) {
713 case UNIT_ATTENTION:
714 if (cmd->device->removable) {
715 /* Detected disc change. Set a bit
716 * and quietly refuse further access.
717 */
718 cmd->device->changed = 1;
719 action = ACTION_FAIL;
720 } else {
721 /* Must have been a power glitch, or a
722 * bus reset. Could not have been a
723 * media change, so we just retry the
724 * command and see what happens.
725 */
726 action = ACTION_RETRY;
727 }
728 break;
729 case ILLEGAL_REQUEST:
730 /* If we had an ILLEGAL REQUEST returned, then
731 * we may have performed an unsupported
732 * command. The only thing this should be
733 * would be a ten byte read where only a six
734 * byte read was supported. Also, on a system
735 * where READ CAPACITY failed, we may have
736 * read past the end of the disk.
737 */
738 if ((cmd->device->use_10_for_rw &&
739 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
740 (cmd->cmnd[0] == READ_10 ||
741 cmd->cmnd[0] == WRITE_10)) {
742 /* This will issue a new 6-byte command. */
743 cmd->device->use_10_for_rw = 0;
744 action = ACTION_REPREP;
745 } else if (sshdr.asc == 0x10) /* DIX */ {
746 action = ACTION_FAIL;
747 blk_stat = BLK_STS_PROTECTION;
748 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
749 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
750 action = ACTION_FAIL;
751 blk_stat = BLK_STS_TARGET;
752 } else
753 action = ACTION_FAIL;
754 break;
755 case ABORTED_COMMAND:
756 action = ACTION_FAIL;
757 if (sshdr.asc == 0x10) /* DIF */
758 blk_stat = BLK_STS_PROTECTION;
759 break;
760 case NOT_READY:
761 /* If the device is in the process of becoming
762 * ready, or has a temporary blockage, retry.
763 */
764 if (sshdr.asc == 0x04) {
765 switch (sshdr.ascq) {
766 case 0x01: /* becoming ready */
767 case 0x04: /* format in progress */
768 case 0x05: /* rebuild in progress */
769 case 0x06: /* recalculation in progress */
770 case 0x07: /* operation in progress */
771 case 0x08: /* Long write in progress */
772 case 0x09: /* self test in progress */
773 case 0x11: /* notify (enable spinup) required */
774 case 0x14: /* space allocation in progress */
775 case 0x1a: /* start stop unit in progress */
776 case 0x1b: /* sanitize in progress */
777 case 0x1d: /* configuration in progress */
778 case 0x24: /* depopulation in progress */
779 case 0x25: /* depopulation restore in progress */
780 action = ACTION_DELAYED_RETRY;
781 break;
782 case 0x0a: /* ALUA state transition */
783 action = ACTION_DELAYED_REPREP;
784 break;
785 default:
786 action = ACTION_FAIL;
787 break;
788 }
789 } else
790 action = ACTION_FAIL;
791 break;
792 case VOLUME_OVERFLOW:
793 /* See SSC3rXX or current. */
794 action = ACTION_FAIL;
795 break;
796 case DATA_PROTECT:
797 action = ACTION_FAIL;
798 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
799 (sshdr.asc == 0x55 &&
800 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
801 /* Insufficient zone resources */
802 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
803 }
804 break;
805 case COMPLETED:
806 fallthrough;
807 default:
808 action = ACTION_FAIL;
809 break;
810 }
811 } else
812 action = ACTION_FAIL;
813
814 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
815 action = ACTION_FAIL;
816
817 switch (action) {
818 case ACTION_FAIL:
819 /* Give up and fail the remainder of the request */
820 if (!(req->rq_flags & RQF_QUIET)) {
821 static DEFINE_RATELIMIT_STATE(_rs,
822 DEFAULT_RATELIMIT_INTERVAL,
823 DEFAULT_RATELIMIT_BURST);
824
825 if (unlikely(scsi_logging_level))
826 level =
827 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
828 SCSI_LOG_MLCOMPLETE_BITS);
829
830 /*
831 * if logging is enabled the failure will be printed
832 * in scsi_log_completion(), so avoid duplicate messages
833 */
834 if (!level && __ratelimit(&_rs)) {
835 scsi_print_result(cmd, NULL, FAILED);
836 if (sense_valid)
837 scsi_print_sense(cmd);
838 scsi_print_command(cmd);
839 }
840 }
841 if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
842 return;
843 fallthrough;
844 case ACTION_REPREP:
845 scsi_mq_requeue_cmd(cmd, 0);
846 break;
847 case ACTION_DELAYED_REPREP:
848 scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
849 break;
850 case ACTION_RETRY:
851 /* Retry the same command immediately */
852 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
853 break;
854 case ACTION_DELAYED_RETRY:
855 /* Retry the same command after a delay */
856 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
857 break;
858 }
859}
860
861/*
862 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
863 * new result that may suppress further error checking. Also modifies
864 * *blk_statp in some cases.
865 */
866static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
867 blk_status_t *blk_statp)
868{
869 bool sense_valid;
870 bool sense_current = true; /* false implies "deferred sense" */
871 struct request *req = scsi_cmd_to_rq(cmd);
872 struct scsi_sense_hdr sshdr;
873
874 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
875 if (sense_valid)
876 sense_current = !scsi_sense_is_deferred(&sshdr);
877
878 if (blk_rq_is_passthrough(req)) {
879 if (sense_valid) {
880 /*
881 * SG_IO wants current and deferred errors
882 */
883 cmd->sense_len = min(8 + cmd->sense_buffer[7],
884 SCSI_SENSE_BUFFERSIZE);
885 }
886 if (sense_current)
887 *blk_statp = scsi_result_to_blk_status(result);
888 } else if (blk_rq_bytes(req) == 0 && sense_current) {
889 /*
890 * Flush commands do not transfers any data, and thus cannot use
891 * good_bytes != blk_rq_bytes(req) as the signal for an error.
892 * This sets *blk_statp explicitly for the problem case.
893 */
894 *blk_statp = scsi_result_to_blk_status(result);
895 }
896 /*
897 * Recovered errors need reporting, but they're always treated as
898 * success, so fiddle the result code here. For passthrough requests
899 * we already took a copy of the original into sreq->result which
900 * is what gets returned to the user
901 */
902 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
903 bool do_print = true;
904 /*
905 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
906 * skip print since caller wants ATA registers. Only occurs
907 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
908 */
909 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
910 do_print = false;
911 else if (req->rq_flags & RQF_QUIET)
912 do_print = false;
913 if (do_print)
914 scsi_print_sense(cmd);
915 result = 0;
916 /* for passthrough, *blk_statp may be set */
917 *blk_statp = BLK_STS_OK;
918 }
919 /*
920 * Another corner case: the SCSI status byte is non-zero but 'good'.
921 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
922 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
923 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
924 * intermediate statuses (both obsolete in SAM-4) as good.
925 */
926 if ((result & 0xff) && scsi_status_is_good(result)) {
927 result = 0;
928 *blk_statp = BLK_STS_OK;
929 }
930 return result;
931}
932
933/**
934 * scsi_io_completion - Completion processing for SCSI commands.
935 * @cmd: command that is finished.
936 * @good_bytes: number of processed bytes.
937 *
938 * We will finish off the specified number of sectors. If we are done, the
939 * command block will be released and the queue function will be goosed. If we
940 * are not done then we have to figure out what to do next:
941 *
942 * a) We can call scsi_mq_requeue_cmd(). The request will be
943 * unprepared and put back on the queue. Then a new command will
944 * be created for it. This should be used if we made forward
945 * progress, or if we want to switch from READ(10) to READ(6) for
946 * example.
947 *
948 * b) We can call scsi_io_completion_action(). The request will be
949 * put back on the queue and retried using the same command as
950 * before, possibly after a delay.
951 *
952 * c) We can call scsi_end_request() with blk_stat other than
953 * BLK_STS_OK, to fail the remainder of the request.
954 */
955void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
956{
957 int result = cmd->result;
958 struct request *req = scsi_cmd_to_rq(cmd);
959 blk_status_t blk_stat = BLK_STS_OK;
960
961 if (unlikely(result)) /* a nz result may or may not be an error */
962 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
963
964 /*
965 * Next deal with any sectors which we were able to correctly
966 * handle.
967 */
968 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
969 "%u sectors total, %d bytes done.\n",
970 blk_rq_sectors(req), good_bytes));
971
972 /*
973 * Failed, zero length commands always need to drop down
974 * to retry code. Fast path should return in this block.
975 */
976 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
977 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
978 return; /* no bytes remaining */
979 }
980
981 /* Kill remainder if no retries. */
982 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
983 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
984 WARN_ONCE(true,
985 "Bytes remaining after failed, no-retry command");
986 return;
987 }
988
989 /*
990 * If there had been no error, but we have leftover bytes in the
991 * request just queue the command up again.
992 */
993 if (likely(result == 0))
994 scsi_mq_requeue_cmd(cmd, 0);
995 else
996 scsi_io_completion_action(cmd, result);
997}
998
999static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1000 struct request *rq)
1001{
1002 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1003 !op_is_write(req_op(rq)) &&
1004 sdev->host->hostt->dma_need_drain(rq);
1005}
1006
1007/**
1008 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1009 * @cmd: SCSI command data structure to initialize.
1010 *
1011 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1012 * for @cmd.
1013 *
1014 * Returns:
1015 * * BLK_STS_OK - on success
1016 * * BLK_STS_RESOURCE - if the failure is retryable
1017 * * BLK_STS_IOERR - if the failure is fatal
1018 */
1019blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1020{
1021 struct scsi_device *sdev = cmd->device;
1022 struct request *rq = scsi_cmd_to_rq(cmd);
1023 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1024 struct scatterlist *last_sg = NULL;
1025 blk_status_t ret;
1026 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1027 int count;
1028
1029 if (WARN_ON_ONCE(!nr_segs))
1030 return BLK_STS_IOERR;
1031
1032 /*
1033 * Make sure there is space for the drain. The driver must adjust
1034 * max_hw_segments to be prepared for this.
1035 */
1036 if (need_drain)
1037 nr_segs++;
1038
1039 /*
1040 * If sg table allocation fails, requeue request later.
1041 */
1042 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1043 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1044 return BLK_STS_RESOURCE;
1045
1046 /*
1047 * Next, walk the list, and fill in the addresses and sizes of
1048 * each segment.
1049 */
1050 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1051
1052 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1053 unsigned int pad_len =
1054 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1055
1056 last_sg->length += pad_len;
1057 cmd->extra_len += pad_len;
1058 }
1059
1060 if (need_drain) {
1061 sg_unmark_end(last_sg);
1062 last_sg = sg_next(last_sg);
1063 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1064 sg_mark_end(last_sg);
1065
1066 cmd->extra_len += sdev->dma_drain_len;
1067 count++;
1068 }
1069
1070 BUG_ON(count > cmd->sdb.table.nents);
1071 cmd->sdb.table.nents = count;
1072 cmd->sdb.length = blk_rq_payload_bytes(rq);
1073
1074 if (blk_integrity_rq(rq)) {
1075 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1076 int ivecs;
1077
1078 if (WARN_ON_ONCE(!prot_sdb)) {
1079 /*
1080 * This can happen if someone (e.g. multipath)
1081 * queues a command to a device on an adapter
1082 * that does not support DIX.
1083 */
1084 ret = BLK_STS_IOERR;
1085 goto out_free_sgtables;
1086 }
1087
1088 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1089
1090 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1091 prot_sdb->table.sgl,
1092 SCSI_INLINE_PROT_SG_CNT)) {
1093 ret = BLK_STS_RESOURCE;
1094 goto out_free_sgtables;
1095 }
1096
1097 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1098 prot_sdb->table.sgl);
1099 BUG_ON(count > ivecs);
1100 BUG_ON(count > queue_max_integrity_segments(rq->q));
1101
1102 cmd->prot_sdb = prot_sdb;
1103 cmd->prot_sdb->table.nents = count;
1104 }
1105
1106 return BLK_STS_OK;
1107out_free_sgtables:
1108 scsi_free_sgtables(cmd);
1109 return ret;
1110}
1111EXPORT_SYMBOL(scsi_alloc_sgtables);
1112
1113/**
1114 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1115 * @rq: Request associated with the SCSI command to be initialized.
1116 *
1117 * This function initializes the members of struct scsi_cmnd that must be
1118 * initialized before request processing starts and that won't be
1119 * reinitialized if a SCSI command is requeued.
1120 */
1121static void scsi_initialize_rq(struct request *rq)
1122{
1123 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1124
1125 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1126 cmd->cmd_len = MAX_COMMAND_SIZE;
1127 cmd->sense_len = 0;
1128 init_rcu_head(&cmd->rcu);
1129 cmd->jiffies_at_alloc = jiffies;
1130 cmd->retries = 0;
1131}
1132
1133struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1134 blk_mq_req_flags_t flags)
1135{
1136 struct request *rq;
1137
1138 rq = blk_mq_alloc_request(q, opf, flags);
1139 if (!IS_ERR(rq))
1140 scsi_initialize_rq(rq);
1141 return rq;
1142}
1143EXPORT_SYMBOL_GPL(scsi_alloc_request);
1144
1145/*
1146 * Only called when the request isn't completed by SCSI, and not freed by
1147 * SCSI
1148 */
1149static void scsi_cleanup_rq(struct request *rq)
1150{
1151 if (rq->rq_flags & RQF_DONTPREP) {
1152 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1153 rq->rq_flags &= ~RQF_DONTPREP;
1154 }
1155}
1156
1157/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1158void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1159{
1160 struct request *rq = scsi_cmd_to_rq(cmd);
1161
1162 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1163 cmd->flags |= SCMD_INITIALIZED;
1164 scsi_initialize_rq(rq);
1165 }
1166
1167 cmd->device = dev;
1168 INIT_LIST_HEAD(&cmd->eh_entry);
1169 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1170}
1171
1172static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1173 struct request *req)
1174{
1175 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1176
1177 /*
1178 * Passthrough requests may transfer data, in which case they must
1179 * a bio attached to them. Or they might contain a SCSI command
1180 * that does not transfer data, in which case they may optionally
1181 * submit a request without an attached bio.
1182 */
1183 if (req->bio) {
1184 blk_status_t ret = scsi_alloc_sgtables(cmd);
1185 if (unlikely(ret != BLK_STS_OK))
1186 return ret;
1187 } else {
1188 BUG_ON(blk_rq_bytes(req));
1189
1190 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1191 }
1192
1193 cmd->transfersize = blk_rq_bytes(req);
1194 return BLK_STS_OK;
1195}
1196
1197static blk_status_t
1198scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1199{
1200 switch (sdev->sdev_state) {
1201 case SDEV_CREATED:
1202 return BLK_STS_OK;
1203 case SDEV_OFFLINE:
1204 case SDEV_TRANSPORT_OFFLINE:
1205 /*
1206 * If the device is offline we refuse to process any
1207 * commands. The device must be brought online
1208 * before trying any recovery commands.
1209 */
1210 if (!sdev->offline_already) {
1211 sdev->offline_already = true;
1212 sdev_printk(KERN_ERR, sdev,
1213 "rejecting I/O to offline device\n");
1214 }
1215 return BLK_STS_IOERR;
1216 case SDEV_DEL:
1217 /*
1218 * If the device is fully deleted, we refuse to
1219 * process any commands as well.
1220 */
1221 sdev_printk(KERN_ERR, sdev,
1222 "rejecting I/O to dead device\n");
1223 return BLK_STS_IOERR;
1224 case SDEV_BLOCK:
1225 case SDEV_CREATED_BLOCK:
1226 return BLK_STS_RESOURCE;
1227 case SDEV_QUIESCE:
1228 /*
1229 * If the device is blocked we only accept power management
1230 * commands.
1231 */
1232 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1233 return BLK_STS_RESOURCE;
1234 return BLK_STS_OK;
1235 default:
1236 /*
1237 * For any other not fully online state we only allow
1238 * power management commands.
1239 */
1240 if (req && !(req->rq_flags & RQF_PM))
1241 return BLK_STS_OFFLINE;
1242 return BLK_STS_OK;
1243 }
1244}
1245
1246/*
1247 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1248 * and return the token else return -1.
1249 */
1250static inline int scsi_dev_queue_ready(struct request_queue *q,
1251 struct scsi_device *sdev)
1252{
1253 int token;
1254
1255 token = sbitmap_get(&sdev->budget_map);
1256 if (token < 0)
1257 return -1;
1258
1259 if (!atomic_read(&sdev->device_blocked))
1260 return token;
1261
1262 /*
1263 * Only unblock if no other commands are pending and
1264 * if device_blocked has decreased to zero
1265 */
1266 if (scsi_device_busy(sdev) > 1 ||
1267 atomic_dec_return(&sdev->device_blocked) > 0) {
1268 sbitmap_put(&sdev->budget_map, token);
1269 return -1;
1270 }
1271
1272 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1273 "unblocking device at zero depth\n"));
1274
1275 return token;
1276}
1277
1278/*
1279 * scsi_target_queue_ready: checks if there we can send commands to target
1280 * @sdev: scsi device on starget to check.
1281 */
1282static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1283 struct scsi_device *sdev)
1284{
1285 struct scsi_target *starget = scsi_target(sdev);
1286 unsigned int busy;
1287
1288 if (starget->single_lun) {
1289 spin_lock_irq(shost->host_lock);
1290 if (starget->starget_sdev_user &&
1291 starget->starget_sdev_user != sdev) {
1292 spin_unlock_irq(shost->host_lock);
1293 return 0;
1294 }
1295 starget->starget_sdev_user = sdev;
1296 spin_unlock_irq(shost->host_lock);
1297 }
1298
1299 if (starget->can_queue <= 0)
1300 return 1;
1301
1302 busy = atomic_inc_return(&starget->target_busy) - 1;
1303 if (atomic_read(&starget->target_blocked) > 0) {
1304 if (busy)
1305 goto starved;
1306
1307 /*
1308 * unblock after target_blocked iterates to zero
1309 */
1310 if (atomic_dec_return(&starget->target_blocked) > 0)
1311 goto out_dec;
1312
1313 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1314 "unblocking target at zero depth\n"));
1315 }
1316
1317 if (busy >= starget->can_queue)
1318 goto starved;
1319
1320 return 1;
1321
1322starved:
1323 spin_lock_irq(shost->host_lock);
1324 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1325 spin_unlock_irq(shost->host_lock);
1326out_dec:
1327 if (starget->can_queue > 0)
1328 atomic_dec(&starget->target_busy);
1329 return 0;
1330}
1331
1332/*
1333 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1334 * return 0. We must end up running the queue again whenever 0 is
1335 * returned, else IO can hang.
1336 */
1337static inline int scsi_host_queue_ready(struct request_queue *q,
1338 struct Scsi_Host *shost,
1339 struct scsi_device *sdev,
1340 struct scsi_cmnd *cmd)
1341{
1342 if (atomic_read(&shost->host_blocked) > 0) {
1343 if (scsi_host_busy(shost) > 0)
1344 goto starved;
1345
1346 /*
1347 * unblock after host_blocked iterates to zero
1348 */
1349 if (atomic_dec_return(&shost->host_blocked) > 0)
1350 goto out_dec;
1351
1352 SCSI_LOG_MLQUEUE(3,
1353 shost_printk(KERN_INFO, shost,
1354 "unblocking host at zero depth\n"));
1355 }
1356
1357 if (shost->host_self_blocked)
1358 goto starved;
1359
1360 /* We're OK to process the command, so we can't be starved */
1361 if (!list_empty(&sdev->starved_entry)) {
1362 spin_lock_irq(shost->host_lock);
1363 if (!list_empty(&sdev->starved_entry))
1364 list_del_init(&sdev->starved_entry);
1365 spin_unlock_irq(shost->host_lock);
1366 }
1367
1368 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1369
1370 return 1;
1371
1372starved:
1373 spin_lock_irq(shost->host_lock);
1374 if (list_empty(&sdev->starved_entry))
1375 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1376 spin_unlock_irq(shost->host_lock);
1377out_dec:
1378 scsi_dec_host_busy(shost, cmd);
1379 return 0;
1380}
1381
1382/*
1383 * Busy state exporting function for request stacking drivers.
1384 *
1385 * For efficiency, no lock is taken to check the busy state of
1386 * shost/starget/sdev, since the returned value is not guaranteed and
1387 * may be changed after request stacking drivers call the function,
1388 * regardless of taking lock or not.
1389 *
1390 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1391 * needs to return 'not busy'. Otherwise, request stacking drivers
1392 * may hold requests forever.
1393 */
1394static bool scsi_mq_lld_busy(struct request_queue *q)
1395{
1396 struct scsi_device *sdev = q->queuedata;
1397 struct Scsi_Host *shost;
1398
1399 if (blk_queue_dying(q))
1400 return false;
1401
1402 shost = sdev->host;
1403
1404 /*
1405 * Ignore host/starget busy state.
1406 * Since block layer does not have a concept of fairness across
1407 * multiple queues, congestion of host/starget needs to be handled
1408 * in SCSI layer.
1409 */
1410 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1411 return true;
1412
1413 return false;
1414}
1415
1416/*
1417 * Block layer request completion callback. May be called from interrupt
1418 * context.
1419 */
1420static void scsi_complete(struct request *rq)
1421{
1422 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1423 enum scsi_disposition disposition;
1424
1425 INIT_LIST_HEAD(&cmd->eh_entry);
1426
1427 atomic_inc(&cmd->device->iodone_cnt);
1428 if (cmd->result)
1429 atomic_inc(&cmd->device->ioerr_cnt);
1430
1431 disposition = scsi_decide_disposition(cmd);
1432 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1433 disposition = SUCCESS;
1434
1435 scsi_log_completion(cmd, disposition);
1436
1437 switch (disposition) {
1438 case SUCCESS:
1439 scsi_finish_command(cmd);
1440 break;
1441 case NEEDS_RETRY:
1442 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1443 break;
1444 case ADD_TO_MLQUEUE:
1445 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1446 break;
1447 default:
1448 scsi_eh_scmd_add(cmd);
1449 break;
1450 }
1451}
1452
1453/**
1454 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1455 * @cmd: command block we are dispatching.
1456 *
1457 * Return: nonzero return request was rejected and device's queue needs to be
1458 * plugged.
1459 */
1460static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1461{
1462 struct Scsi_Host *host = cmd->device->host;
1463 int rtn = 0;
1464
1465 atomic_inc(&cmd->device->iorequest_cnt);
1466
1467 /* check if the device is still usable */
1468 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1469 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1470 * returns an immediate error upwards, and signals
1471 * that the device is no longer present */
1472 cmd->result = DID_NO_CONNECT << 16;
1473 goto done;
1474 }
1475
1476 /* Check to see if the scsi lld made this device blocked. */
1477 if (unlikely(scsi_device_blocked(cmd->device))) {
1478 /*
1479 * in blocked state, the command is just put back on
1480 * the device queue. The suspend state has already
1481 * blocked the queue so future requests should not
1482 * occur until the device transitions out of the
1483 * suspend state.
1484 */
1485 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1486 "queuecommand : device blocked\n"));
1487 atomic_dec(&cmd->device->iorequest_cnt);
1488 return SCSI_MLQUEUE_DEVICE_BUSY;
1489 }
1490
1491 /* Store the LUN value in cmnd, if needed. */
1492 if (cmd->device->lun_in_cdb)
1493 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1494 (cmd->device->lun << 5 & 0xe0);
1495
1496 scsi_log_send(cmd);
1497
1498 /*
1499 * Before we queue this command, check if the command
1500 * length exceeds what the host adapter can handle.
1501 */
1502 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1503 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1504 "queuecommand : command too long. "
1505 "cdb_size=%d host->max_cmd_len=%d\n",
1506 cmd->cmd_len, cmd->device->host->max_cmd_len));
1507 cmd->result = (DID_ABORT << 16);
1508 goto done;
1509 }
1510
1511 if (unlikely(host->shost_state == SHOST_DEL)) {
1512 cmd->result = (DID_NO_CONNECT << 16);
1513 goto done;
1514
1515 }
1516
1517 trace_scsi_dispatch_cmd_start(cmd);
1518 rtn = host->hostt->queuecommand(host, cmd);
1519 if (rtn) {
1520 atomic_dec(&cmd->device->iorequest_cnt);
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. */
1537static 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
1543static 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 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1549 struct scatterlist *sg;
1550
1551 scsi_init_command(sdev, cmd);
1552
1553 cmd->eh_eflags = 0;
1554 cmd->prot_type = 0;
1555 cmd->prot_flags = 0;
1556 cmd->submitter = 0;
1557 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1558 cmd->underflow = 0;
1559 cmd->transfersize = 0;
1560 cmd->host_scribble = NULL;
1561 cmd->result = 0;
1562 cmd->extra_len = 0;
1563 cmd->state = 0;
1564 if (in_flight)
1565 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1566
1567 /*
1568 * Only clear the driver-private command data if the LLD does not supply
1569 * a function to initialize that data.
1570 */
1571 if (!shost->hostt->init_cmd_priv)
1572 memset(cmd + 1, 0, shost->hostt->cmd_size);
1573
1574 cmd->prot_op = SCSI_PROT_NORMAL;
1575 if (blk_rq_bytes(req))
1576 cmd->sc_data_direction = rq_dma_dir(req);
1577 else
1578 cmd->sc_data_direction = DMA_NONE;
1579
1580 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1581 cmd->sdb.table.sgl = sg;
1582
1583 if (scsi_host_get_prot(shost)) {
1584 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1585
1586 cmd->prot_sdb->table.sgl =
1587 (struct scatterlist *)(cmd->prot_sdb + 1);
1588 }
1589
1590 /*
1591 * Special handling for passthrough commands, which don't go to the ULP
1592 * at all:
1593 */
1594 if (blk_rq_is_passthrough(req))
1595 return scsi_setup_scsi_cmnd(sdev, req);
1596
1597 if (sdev->handler && sdev->handler->prep_fn) {
1598 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1599
1600 if (ret != BLK_STS_OK)
1601 return ret;
1602 }
1603
1604 /* Usually overridden by the ULP */
1605 cmd->allowed = 0;
1606 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1607 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1608}
1609
1610static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1611{
1612 struct request *req = scsi_cmd_to_rq(cmd);
1613
1614 switch (cmd->submitter) {
1615 case SUBMITTED_BY_BLOCK_LAYER:
1616 break;
1617 case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1618 return scsi_eh_done(cmd);
1619 case SUBMITTED_BY_SCSI_RESET_IOCTL:
1620 return;
1621 }
1622
1623 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1624 return;
1625 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1626 return;
1627 trace_scsi_dispatch_cmd_done(cmd);
1628
1629 if (complete_directly)
1630 blk_mq_complete_request_direct(req, scsi_complete);
1631 else
1632 blk_mq_complete_request(req);
1633}
1634
1635void scsi_done(struct scsi_cmnd *cmd)
1636{
1637 scsi_done_internal(cmd, false);
1638}
1639EXPORT_SYMBOL(scsi_done);
1640
1641void scsi_done_direct(struct scsi_cmnd *cmd)
1642{
1643 scsi_done_internal(cmd, true);
1644}
1645EXPORT_SYMBOL(scsi_done_direct);
1646
1647static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1648{
1649 struct scsi_device *sdev = q->queuedata;
1650
1651 sbitmap_put(&sdev->budget_map, budget_token);
1652}
1653
1654/*
1655 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1656 * not change behaviour from the previous unplug mechanism, experimentation
1657 * may prove this needs changing.
1658 */
1659#define SCSI_QUEUE_DELAY 3
1660
1661static int scsi_mq_get_budget(struct request_queue *q)
1662{
1663 struct scsi_device *sdev = q->queuedata;
1664 int token = scsi_dev_queue_ready(q, sdev);
1665
1666 if (token >= 0)
1667 return token;
1668
1669 atomic_inc(&sdev->restarts);
1670
1671 /*
1672 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1673 * .restarts must be incremented before .device_busy is read because the
1674 * code in scsi_run_queue_async() depends on the order of these operations.
1675 */
1676 smp_mb__after_atomic();
1677
1678 /*
1679 * If all in-flight requests originated from this LUN are completed
1680 * before reading .device_busy, sdev->device_busy will be observed as
1681 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1682 * soon. Otherwise, completion of one of these requests will observe
1683 * the .restarts flag, and the request queue will be run for handling
1684 * this request, see scsi_end_request().
1685 */
1686 if (unlikely(scsi_device_busy(sdev) == 0 &&
1687 !scsi_device_blocked(sdev)))
1688 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1689 return -1;
1690}
1691
1692static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1693{
1694 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1695
1696 cmd->budget_token = token;
1697}
1698
1699static int scsi_mq_get_rq_budget_token(struct request *req)
1700{
1701 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1702
1703 return cmd->budget_token;
1704}
1705
1706static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1707 const struct blk_mq_queue_data *bd)
1708{
1709 struct request *req = bd->rq;
1710 struct request_queue *q = req->q;
1711 struct scsi_device *sdev = q->queuedata;
1712 struct Scsi_Host *shost = sdev->host;
1713 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1714 blk_status_t ret;
1715 int reason;
1716
1717 WARN_ON_ONCE(cmd->budget_token < 0);
1718
1719 /*
1720 * If the device is not in running state we will reject some or all
1721 * commands.
1722 */
1723 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1724 ret = scsi_device_state_check(sdev, req);
1725 if (ret != BLK_STS_OK)
1726 goto out_put_budget;
1727 }
1728
1729 ret = BLK_STS_RESOURCE;
1730 if (!scsi_target_queue_ready(shost, sdev))
1731 goto out_put_budget;
1732 if (unlikely(scsi_host_in_recovery(shost))) {
1733 if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1734 ret = BLK_STS_OFFLINE;
1735 goto out_dec_target_busy;
1736 }
1737 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1738 goto out_dec_target_busy;
1739
1740 if (!(req->rq_flags & RQF_DONTPREP)) {
1741 ret = scsi_prepare_cmd(req);
1742 if (ret != BLK_STS_OK)
1743 goto out_dec_host_busy;
1744 req->rq_flags |= RQF_DONTPREP;
1745 } else {
1746 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1747 }
1748
1749 cmd->flags &= SCMD_PRESERVED_FLAGS;
1750 if (sdev->simple_tags)
1751 cmd->flags |= SCMD_TAGGED;
1752 if (bd->last)
1753 cmd->flags |= SCMD_LAST;
1754
1755 scsi_set_resid(cmd, 0);
1756 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1757 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1758
1759 blk_mq_start_request(req);
1760 reason = scsi_dispatch_cmd(cmd);
1761 if (reason) {
1762 scsi_set_blocked(cmd, reason);
1763 ret = BLK_STS_RESOURCE;
1764 goto out_dec_host_busy;
1765 }
1766
1767 return BLK_STS_OK;
1768
1769out_dec_host_busy:
1770 scsi_dec_host_busy(shost, cmd);
1771out_dec_target_busy:
1772 if (scsi_target(sdev)->can_queue > 0)
1773 atomic_dec(&scsi_target(sdev)->target_busy);
1774out_put_budget:
1775 scsi_mq_put_budget(q, cmd->budget_token);
1776 cmd->budget_token = -1;
1777 switch (ret) {
1778 case BLK_STS_OK:
1779 break;
1780 case BLK_STS_RESOURCE:
1781 case BLK_STS_ZONE_RESOURCE:
1782 if (scsi_device_blocked(sdev))
1783 ret = BLK_STS_DEV_RESOURCE;
1784 break;
1785 case BLK_STS_AGAIN:
1786 cmd->result = DID_BUS_BUSY << 16;
1787 if (req->rq_flags & RQF_DONTPREP)
1788 scsi_mq_uninit_cmd(cmd);
1789 break;
1790 default:
1791 if (unlikely(!scsi_device_online(sdev)))
1792 cmd->result = DID_NO_CONNECT << 16;
1793 else
1794 cmd->result = DID_ERROR << 16;
1795 /*
1796 * Make sure to release all allocated resources when
1797 * we hit an error, as we will never see this command
1798 * again.
1799 */
1800 if (req->rq_flags & RQF_DONTPREP)
1801 scsi_mq_uninit_cmd(cmd);
1802 scsi_run_queue_async(sdev);
1803 break;
1804 }
1805 return ret;
1806}
1807
1808static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1809 unsigned int hctx_idx, unsigned int numa_node)
1810{
1811 struct Scsi_Host *shost = set->driver_data;
1812 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1813 struct scatterlist *sg;
1814 int ret = 0;
1815
1816 cmd->sense_buffer =
1817 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1818 if (!cmd->sense_buffer)
1819 return -ENOMEM;
1820
1821 if (scsi_host_get_prot(shost)) {
1822 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1823 shost->hostt->cmd_size;
1824 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1825 }
1826
1827 if (shost->hostt->init_cmd_priv) {
1828 ret = shost->hostt->init_cmd_priv(shost, cmd);
1829 if (ret < 0)
1830 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1831 }
1832
1833 return ret;
1834}
1835
1836static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1837 unsigned int hctx_idx)
1838{
1839 struct Scsi_Host *shost = set->driver_data;
1840 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1841
1842 if (shost->hostt->exit_cmd_priv)
1843 shost->hostt->exit_cmd_priv(shost, cmd);
1844 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1845}
1846
1847
1848static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1849{
1850 struct Scsi_Host *shost = hctx->driver_data;
1851
1852 if (shost->hostt->mq_poll)
1853 return shost->hostt->mq_poll(shost, hctx->queue_num);
1854
1855 return 0;
1856}
1857
1858static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1859 unsigned int hctx_idx)
1860{
1861 struct Scsi_Host *shost = data;
1862
1863 hctx->driver_data = shost;
1864 return 0;
1865}
1866
1867static void scsi_map_queues(struct blk_mq_tag_set *set)
1868{
1869 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1870
1871 if (shost->hostt->map_queues)
1872 return shost->hostt->map_queues(shost);
1873 blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1874}
1875
1876void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1877{
1878 struct device *dev = shost->dma_dev;
1879
1880 /*
1881 * this limit is imposed by hardware restrictions
1882 */
1883 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1884 SG_MAX_SEGMENTS));
1885
1886 if (scsi_host_prot_dma(shost)) {
1887 shost->sg_prot_tablesize =
1888 min_not_zero(shost->sg_prot_tablesize,
1889 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1890 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1891 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1892 }
1893
1894 blk_queue_max_hw_sectors(q, shost->max_sectors);
1895 blk_queue_segment_boundary(q, shost->dma_boundary);
1896 dma_set_seg_boundary(dev, shost->dma_boundary);
1897
1898 blk_queue_max_segment_size(q, shost->max_segment_size);
1899 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1900 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1901
1902 /*
1903 * Set a reasonable default alignment: The larger of 32-byte (dword),
1904 * which is a common minimum for HBAs, and the minimum DMA alignment,
1905 * which is set by the platform.
1906 *
1907 * Devices that require a bigger alignment can increase it later.
1908 */
1909 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1910}
1911EXPORT_SYMBOL_GPL(__scsi_init_queue);
1912
1913static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1914 .get_budget = scsi_mq_get_budget,
1915 .put_budget = scsi_mq_put_budget,
1916 .queue_rq = scsi_queue_rq,
1917 .complete = scsi_complete,
1918 .timeout = scsi_timeout,
1919#ifdef CONFIG_BLK_DEBUG_FS
1920 .show_rq = scsi_show_rq,
1921#endif
1922 .init_request = scsi_mq_init_request,
1923 .exit_request = scsi_mq_exit_request,
1924 .cleanup_rq = scsi_cleanup_rq,
1925 .busy = scsi_mq_lld_busy,
1926 .map_queues = scsi_map_queues,
1927 .init_hctx = scsi_init_hctx,
1928 .poll = scsi_mq_poll,
1929 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1930 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1931};
1932
1933
1934static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1935{
1936 struct Scsi_Host *shost = hctx->driver_data;
1937
1938 shost->hostt->commit_rqs(shost, hctx->queue_num);
1939}
1940
1941static const struct blk_mq_ops scsi_mq_ops = {
1942 .get_budget = scsi_mq_get_budget,
1943 .put_budget = scsi_mq_put_budget,
1944 .queue_rq = scsi_queue_rq,
1945 .commit_rqs = scsi_commit_rqs,
1946 .complete = scsi_complete,
1947 .timeout = scsi_timeout,
1948#ifdef CONFIG_BLK_DEBUG_FS
1949 .show_rq = scsi_show_rq,
1950#endif
1951 .init_request = scsi_mq_init_request,
1952 .exit_request = scsi_mq_exit_request,
1953 .cleanup_rq = scsi_cleanup_rq,
1954 .busy = scsi_mq_lld_busy,
1955 .map_queues = scsi_map_queues,
1956 .init_hctx = scsi_init_hctx,
1957 .poll = scsi_mq_poll,
1958 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1959 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1960};
1961
1962int scsi_mq_setup_tags(struct Scsi_Host *shost)
1963{
1964 unsigned int cmd_size, sgl_size;
1965 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1966
1967 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1968 scsi_mq_inline_sgl_size(shost));
1969 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1970 if (scsi_host_get_prot(shost))
1971 cmd_size += sizeof(struct scsi_data_buffer) +
1972 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1973
1974 memset(tag_set, 0, sizeof(*tag_set));
1975 if (shost->hostt->commit_rqs)
1976 tag_set->ops = &scsi_mq_ops;
1977 else
1978 tag_set->ops = &scsi_mq_ops_no_commit;
1979 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1980 tag_set->nr_maps = shost->nr_maps ? : 1;
1981 tag_set->queue_depth = shost->can_queue;
1982 tag_set->cmd_size = cmd_size;
1983 tag_set->numa_node = dev_to_node(shost->dma_dev);
1984 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1985 tag_set->flags |=
1986 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1987 if (shost->queuecommand_may_block)
1988 tag_set->flags |= BLK_MQ_F_BLOCKING;
1989 tag_set->driver_data = shost;
1990 if (shost->host_tagset)
1991 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1992
1993 return blk_mq_alloc_tag_set(tag_set);
1994}
1995
1996void scsi_mq_free_tags(struct kref *kref)
1997{
1998 struct Scsi_Host *shost = container_of(kref, typeof(*shost),
1999 tagset_refcnt);
2000
2001 blk_mq_free_tag_set(&shost->tag_set);
2002 complete(&shost->tagset_freed);
2003}
2004
2005/**
2006 * scsi_device_from_queue - return sdev associated with a request_queue
2007 * @q: The request queue to return the sdev from
2008 *
2009 * Return the sdev associated with a request queue or NULL if the
2010 * request_queue does not reference a SCSI device.
2011 */
2012struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2013{
2014 struct scsi_device *sdev = NULL;
2015
2016 if (q->mq_ops == &scsi_mq_ops_no_commit ||
2017 q->mq_ops == &scsi_mq_ops)
2018 sdev = q->queuedata;
2019 if (!sdev || !get_device(&sdev->sdev_gendev))
2020 sdev = NULL;
2021
2022 return sdev;
2023}
2024/*
2025 * pktcdvd should have been integrated into the SCSI layers, but for historical
2026 * reasons like the old IDE driver it isn't. This export allows it to safely
2027 * probe if a given device is a SCSI one and only attach to that.
2028 */
2029#ifdef CONFIG_CDROM_PKTCDVD_MODULE
2030EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2031#endif
2032
2033/**
2034 * scsi_block_requests - Utility function used by low-level drivers to prevent
2035 * further commands from being queued to the device.
2036 * @shost: host in question
2037 *
2038 * There is no timer nor any other means by which the requests get unblocked
2039 * other than the low-level driver calling scsi_unblock_requests().
2040 */
2041void scsi_block_requests(struct Scsi_Host *shost)
2042{
2043 shost->host_self_blocked = 1;
2044}
2045EXPORT_SYMBOL(scsi_block_requests);
2046
2047/**
2048 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2049 * further commands to be queued to the device.
2050 * @shost: host in question
2051 *
2052 * There is no timer nor any other means by which the requests get unblocked
2053 * other than the low-level driver calling scsi_unblock_requests(). This is done
2054 * as an API function so that changes to the internals of the scsi mid-layer
2055 * won't require wholesale changes to drivers that use this feature.
2056 */
2057void scsi_unblock_requests(struct Scsi_Host *shost)
2058{
2059 shost->host_self_blocked = 0;
2060 scsi_run_host_queues(shost);
2061}
2062EXPORT_SYMBOL(scsi_unblock_requests);
2063
2064void scsi_exit_queue(void)
2065{
2066 kmem_cache_destroy(scsi_sense_cache);
2067}
2068
2069/**
2070 * scsi_mode_select - issue a mode select
2071 * @sdev: SCSI device to be queried
2072 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2073 * @sp: Save page bit (0 == don't save, 1 == save)
2074 * @buffer: request buffer (may not be smaller than eight bytes)
2075 * @len: length of request buffer.
2076 * @timeout: command timeout
2077 * @retries: number of retries before failing
2078 * @data: returns a structure abstracting the mode header data
2079 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2080 * must be SCSI_SENSE_BUFFERSIZE big.
2081 *
2082 * Returns zero if successful; negative error number or scsi
2083 * status on error
2084 *
2085 */
2086int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2087 unsigned char *buffer, int len, int timeout, int retries,
2088 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2089{
2090 unsigned char cmd[10];
2091 unsigned char *real_buffer;
2092 const struct scsi_exec_args exec_args = {
2093 .sshdr = sshdr,
2094 };
2095 int ret;
2096
2097 memset(cmd, 0, sizeof(cmd));
2098 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2099
2100 /*
2101 * Use MODE SELECT(10) if the device asked for it or if the mode page
2102 * and the mode select header cannot fit within the maximumm 255 bytes
2103 * of the MODE SELECT(6) command.
2104 */
2105 if (sdev->use_10_for_ms ||
2106 len + 4 > 255 ||
2107 data->block_descriptor_length > 255) {
2108 if (len > 65535 - 8)
2109 return -EINVAL;
2110 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2111 if (!real_buffer)
2112 return -ENOMEM;
2113 memcpy(real_buffer + 8, buffer, len);
2114 len += 8;
2115 real_buffer[0] = 0;
2116 real_buffer[1] = 0;
2117 real_buffer[2] = data->medium_type;
2118 real_buffer[3] = data->device_specific;
2119 real_buffer[4] = data->longlba ? 0x01 : 0;
2120 real_buffer[5] = 0;
2121 put_unaligned_be16(data->block_descriptor_length,
2122 &real_buffer[6]);
2123
2124 cmd[0] = MODE_SELECT_10;
2125 put_unaligned_be16(len, &cmd[7]);
2126 } else {
2127 if (data->longlba)
2128 return -EINVAL;
2129
2130 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2131 if (!real_buffer)
2132 return -ENOMEM;
2133 memcpy(real_buffer + 4, buffer, len);
2134 len += 4;
2135 real_buffer[0] = 0;
2136 real_buffer[1] = data->medium_type;
2137 real_buffer[2] = data->device_specific;
2138 real_buffer[3] = data->block_descriptor_length;
2139
2140 cmd[0] = MODE_SELECT;
2141 cmd[4] = len;
2142 }
2143
2144 ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2145 timeout, retries, &exec_args);
2146 kfree(real_buffer);
2147 return ret;
2148}
2149EXPORT_SYMBOL_GPL(scsi_mode_select);
2150
2151/**
2152 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2153 * @sdev: SCSI device to be queried
2154 * @dbd: set to prevent mode sense from returning block descriptors
2155 * @modepage: mode page being requested
2156 * @subpage: sub-page of the mode page being requested
2157 * @buffer: request buffer (may not be smaller than eight bytes)
2158 * @len: length of request buffer.
2159 * @timeout: command timeout
2160 * @retries: number of retries before failing
2161 * @data: returns a structure abstracting the mode header data
2162 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2163 * must be SCSI_SENSE_BUFFERSIZE big.
2164 *
2165 * Returns zero if successful, or a negative error number on failure
2166 */
2167int
2168scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,
2169 unsigned char *buffer, int len, int timeout, int retries,
2170 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2171{
2172 unsigned char cmd[12];
2173 int use_10_for_ms;
2174 int header_length;
2175 int result, retry_count = retries;
2176 struct scsi_sense_hdr my_sshdr;
2177 const struct scsi_exec_args exec_args = {
2178 /* caller might not be interested in sense, but we need it */
2179 .sshdr = sshdr ? : &my_sshdr,
2180 };
2181
2182 memset(data, 0, sizeof(*data));
2183 memset(&cmd[0], 0, 12);
2184
2185 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2186 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2187 cmd[2] = modepage;
2188 cmd[3] = subpage;
2189
2190 sshdr = exec_args.sshdr;
2191
2192 retry:
2193 use_10_for_ms = sdev->use_10_for_ms || len > 255;
2194
2195 if (use_10_for_ms) {
2196 if (len < 8 || len > 65535)
2197 return -EINVAL;
2198
2199 cmd[0] = MODE_SENSE_10;
2200 put_unaligned_be16(len, &cmd[7]);
2201 header_length = 8;
2202 } else {
2203 if (len < 4)
2204 return -EINVAL;
2205
2206 cmd[0] = MODE_SENSE;
2207 cmd[4] = len;
2208 header_length = 4;
2209 }
2210
2211 memset(buffer, 0, len);
2212
2213 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2214 timeout, retries, &exec_args);
2215 if (result < 0)
2216 return result;
2217
2218 /* This code looks awful: what it's doing is making sure an
2219 * ILLEGAL REQUEST sense return identifies the actual command
2220 * byte as the problem. MODE_SENSE commands can return
2221 * ILLEGAL REQUEST if the code page isn't supported */
2222
2223 if (!scsi_status_is_good(result)) {
2224 if (scsi_sense_valid(sshdr)) {
2225 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2226 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2227 /*
2228 * Invalid command operation code: retry using
2229 * MODE SENSE(6) if this was a MODE SENSE(10)
2230 * request, except if the request mode page is
2231 * too large for MODE SENSE single byte
2232 * allocation length field.
2233 */
2234 if (use_10_for_ms) {
2235 if (len > 255)
2236 return -EIO;
2237 sdev->use_10_for_ms = 0;
2238 goto retry;
2239 }
2240 }
2241 if (scsi_status_is_check_condition(result) &&
2242 sshdr->sense_key == UNIT_ATTENTION &&
2243 retry_count) {
2244 retry_count--;
2245 goto retry;
2246 }
2247 }
2248 return -EIO;
2249 }
2250 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2251 (modepage == 6 || modepage == 8))) {
2252 /* Initio breakage? */
2253 header_length = 0;
2254 data->length = 13;
2255 data->medium_type = 0;
2256 data->device_specific = 0;
2257 data->longlba = 0;
2258 data->block_descriptor_length = 0;
2259 } else if (use_10_for_ms) {
2260 data->length = get_unaligned_be16(&buffer[0]) + 2;
2261 data->medium_type = buffer[2];
2262 data->device_specific = buffer[3];
2263 data->longlba = buffer[4] & 0x01;
2264 data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2265 } else {
2266 data->length = buffer[0] + 1;
2267 data->medium_type = buffer[1];
2268 data->device_specific = buffer[2];
2269 data->block_descriptor_length = buffer[3];
2270 }
2271 data->header_length = header_length;
2272
2273 return 0;
2274}
2275EXPORT_SYMBOL(scsi_mode_sense);
2276
2277/**
2278 * scsi_test_unit_ready - test if unit is ready
2279 * @sdev: scsi device to change the state of.
2280 * @timeout: command timeout
2281 * @retries: number of retries before failing
2282 * @sshdr: outpout pointer for decoded sense information.
2283 *
2284 * Returns zero if unsuccessful or an error if TUR failed. For
2285 * removable media, UNIT_ATTENTION sets ->changed flag.
2286 **/
2287int
2288scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2289 struct scsi_sense_hdr *sshdr)
2290{
2291 char cmd[] = {
2292 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2293 };
2294 const struct scsi_exec_args exec_args = {
2295 .sshdr = sshdr,
2296 };
2297 int result;
2298
2299 /* try to eat the UNIT_ATTENTION if there are enough retries */
2300 do {
2301 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2302 timeout, 1, &exec_args);
2303 if (sdev->removable && result > 0 && scsi_sense_valid(sshdr) &&
2304 sshdr->sense_key == UNIT_ATTENTION)
2305 sdev->changed = 1;
2306 } while (result > 0 && scsi_sense_valid(sshdr) &&
2307 sshdr->sense_key == UNIT_ATTENTION && --retries);
2308
2309 return result;
2310}
2311EXPORT_SYMBOL(scsi_test_unit_ready);
2312
2313/**
2314 * scsi_device_set_state - Take the given device through the device state model.
2315 * @sdev: scsi device to change the state of.
2316 * @state: state to change to.
2317 *
2318 * Returns zero if successful or an error if the requested
2319 * transition is illegal.
2320 */
2321int
2322scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2323{
2324 enum scsi_device_state oldstate = sdev->sdev_state;
2325
2326 if (state == oldstate)
2327 return 0;
2328
2329 switch (state) {
2330 case SDEV_CREATED:
2331 switch (oldstate) {
2332 case SDEV_CREATED_BLOCK:
2333 break;
2334 default:
2335 goto illegal;
2336 }
2337 break;
2338
2339 case SDEV_RUNNING:
2340 switch (oldstate) {
2341 case SDEV_CREATED:
2342 case SDEV_OFFLINE:
2343 case SDEV_TRANSPORT_OFFLINE:
2344 case SDEV_QUIESCE:
2345 case SDEV_BLOCK:
2346 break;
2347 default:
2348 goto illegal;
2349 }
2350 break;
2351
2352 case SDEV_QUIESCE:
2353 switch (oldstate) {
2354 case SDEV_RUNNING:
2355 case SDEV_OFFLINE:
2356 case SDEV_TRANSPORT_OFFLINE:
2357 break;
2358 default:
2359 goto illegal;
2360 }
2361 break;
2362
2363 case SDEV_OFFLINE:
2364 case SDEV_TRANSPORT_OFFLINE:
2365 switch (oldstate) {
2366 case SDEV_CREATED:
2367 case SDEV_RUNNING:
2368 case SDEV_QUIESCE:
2369 case SDEV_BLOCK:
2370 break;
2371 default:
2372 goto illegal;
2373 }
2374 break;
2375
2376 case SDEV_BLOCK:
2377 switch (oldstate) {
2378 case SDEV_RUNNING:
2379 case SDEV_CREATED_BLOCK:
2380 case SDEV_QUIESCE:
2381 case SDEV_OFFLINE:
2382 break;
2383 default:
2384 goto illegal;
2385 }
2386 break;
2387
2388 case SDEV_CREATED_BLOCK:
2389 switch (oldstate) {
2390 case SDEV_CREATED:
2391 break;
2392 default:
2393 goto illegal;
2394 }
2395 break;
2396
2397 case SDEV_CANCEL:
2398 switch (oldstate) {
2399 case SDEV_CREATED:
2400 case SDEV_RUNNING:
2401 case SDEV_QUIESCE:
2402 case SDEV_OFFLINE:
2403 case SDEV_TRANSPORT_OFFLINE:
2404 break;
2405 default:
2406 goto illegal;
2407 }
2408 break;
2409
2410 case SDEV_DEL:
2411 switch (oldstate) {
2412 case SDEV_CREATED:
2413 case SDEV_RUNNING:
2414 case SDEV_OFFLINE:
2415 case SDEV_TRANSPORT_OFFLINE:
2416 case SDEV_CANCEL:
2417 case SDEV_BLOCK:
2418 case SDEV_CREATED_BLOCK:
2419 break;
2420 default:
2421 goto illegal;
2422 }
2423 break;
2424
2425 }
2426 sdev->offline_already = false;
2427 sdev->sdev_state = state;
2428 return 0;
2429
2430 illegal:
2431 SCSI_LOG_ERROR_RECOVERY(1,
2432 sdev_printk(KERN_ERR, sdev,
2433 "Illegal state transition %s->%s",
2434 scsi_device_state_name(oldstate),
2435 scsi_device_state_name(state))
2436 );
2437 return -EINVAL;
2438}
2439EXPORT_SYMBOL(scsi_device_set_state);
2440
2441/**
2442 * scsi_evt_emit - emit a single SCSI device uevent
2443 * @sdev: associated SCSI device
2444 * @evt: event to emit
2445 *
2446 * Send a single uevent (scsi_event) to the associated scsi_device.
2447 */
2448static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2449{
2450 int idx = 0;
2451 char *envp[3];
2452
2453 switch (evt->evt_type) {
2454 case SDEV_EVT_MEDIA_CHANGE:
2455 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2456 break;
2457 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2458 scsi_rescan_device(sdev);
2459 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2460 break;
2461 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2462 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2463 break;
2464 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2465 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2466 break;
2467 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2468 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2469 break;
2470 case SDEV_EVT_LUN_CHANGE_REPORTED:
2471 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2472 break;
2473 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2474 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2475 break;
2476 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2477 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2478 break;
2479 default:
2480 /* do nothing */
2481 break;
2482 }
2483
2484 envp[idx++] = NULL;
2485
2486 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2487}
2488
2489/**
2490 * scsi_evt_thread - send a uevent for each scsi event
2491 * @work: work struct for scsi_device
2492 *
2493 * Dispatch queued events to their associated scsi_device kobjects
2494 * as uevents.
2495 */
2496void scsi_evt_thread(struct work_struct *work)
2497{
2498 struct scsi_device *sdev;
2499 enum scsi_device_event evt_type;
2500 LIST_HEAD(event_list);
2501
2502 sdev = container_of(work, struct scsi_device, event_work);
2503
2504 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2505 if (test_and_clear_bit(evt_type, sdev->pending_events))
2506 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2507
2508 while (1) {
2509 struct scsi_event *evt;
2510 struct list_head *this, *tmp;
2511 unsigned long flags;
2512
2513 spin_lock_irqsave(&sdev->list_lock, flags);
2514 list_splice_init(&sdev->event_list, &event_list);
2515 spin_unlock_irqrestore(&sdev->list_lock, flags);
2516
2517 if (list_empty(&event_list))
2518 break;
2519
2520 list_for_each_safe(this, tmp, &event_list) {
2521 evt = list_entry(this, struct scsi_event, node);
2522 list_del(&evt->node);
2523 scsi_evt_emit(sdev, evt);
2524 kfree(evt);
2525 }
2526 }
2527}
2528
2529/**
2530 * sdev_evt_send - send asserted event to uevent thread
2531 * @sdev: scsi_device event occurred on
2532 * @evt: event to send
2533 *
2534 * Assert scsi device event asynchronously.
2535 */
2536void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2537{
2538 unsigned long flags;
2539
2540#if 0
2541 /* FIXME: currently this check eliminates all media change events
2542 * for polled devices. Need to update to discriminate between AN
2543 * and polled events */
2544 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2545 kfree(evt);
2546 return;
2547 }
2548#endif
2549
2550 spin_lock_irqsave(&sdev->list_lock, flags);
2551 list_add_tail(&evt->node, &sdev->event_list);
2552 schedule_work(&sdev->event_work);
2553 spin_unlock_irqrestore(&sdev->list_lock, flags);
2554}
2555EXPORT_SYMBOL_GPL(sdev_evt_send);
2556
2557/**
2558 * sdev_evt_alloc - allocate a new scsi event
2559 * @evt_type: type of event to allocate
2560 * @gfpflags: GFP flags for allocation
2561 *
2562 * Allocates and returns a new scsi_event.
2563 */
2564struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2565 gfp_t gfpflags)
2566{
2567 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2568 if (!evt)
2569 return NULL;
2570
2571 evt->evt_type = evt_type;
2572 INIT_LIST_HEAD(&evt->node);
2573
2574 /* evt_type-specific initialization, if any */
2575 switch (evt_type) {
2576 case SDEV_EVT_MEDIA_CHANGE:
2577 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2578 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2579 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2580 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2581 case SDEV_EVT_LUN_CHANGE_REPORTED:
2582 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2583 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2584 default:
2585 /* do nothing */
2586 break;
2587 }
2588
2589 return evt;
2590}
2591EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2592
2593/**
2594 * sdev_evt_send_simple - send asserted event to uevent thread
2595 * @sdev: scsi_device event occurred on
2596 * @evt_type: type of event to send
2597 * @gfpflags: GFP flags for allocation
2598 *
2599 * Assert scsi device event asynchronously, given an event type.
2600 */
2601void sdev_evt_send_simple(struct scsi_device *sdev,
2602 enum scsi_device_event evt_type, gfp_t gfpflags)
2603{
2604 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2605 if (!evt) {
2606 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2607 evt_type);
2608 return;
2609 }
2610
2611 sdev_evt_send(sdev, evt);
2612}
2613EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2614
2615/**
2616 * scsi_device_quiesce - Block all commands except power management.
2617 * @sdev: scsi device to quiesce.
2618 *
2619 * This works by trying to transition to the SDEV_QUIESCE state
2620 * (which must be a legal transition). When the device is in this
2621 * state, only power management requests will be accepted, all others will
2622 * be deferred.
2623 *
2624 * Must be called with user context, may sleep.
2625 *
2626 * Returns zero if unsuccessful or an error if not.
2627 */
2628int
2629scsi_device_quiesce(struct scsi_device *sdev)
2630{
2631 struct request_queue *q = sdev->request_queue;
2632 int err;
2633
2634 /*
2635 * It is allowed to call scsi_device_quiesce() multiple times from
2636 * the same context but concurrent scsi_device_quiesce() calls are
2637 * not allowed.
2638 */
2639 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2640
2641 if (sdev->quiesced_by == current)
2642 return 0;
2643
2644 blk_set_pm_only(q);
2645
2646 blk_mq_freeze_queue(q);
2647 /*
2648 * Ensure that the effect of blk_set_pm_only() will be visible
2649 * for percpu_ref_tryget() callers that occur after the queue
2650 * unfreeze even if the queue was already frozen before this function
2651 * was called. See also https://lwn.net/Articles/573497/.
2652 */
2653 synchronize_rcu();
2654 blk_mq_unfreeze_queue(q);
2655
2656 mutex_lock(&sdev->state_mutex);
2657 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2658 if (err == 0)
2659 sdev->quiesced_by = current;
2660 else
2661 blk_clear_pm_only(q);
2662 mutex_unlock(&sdev->state_mutex);
2663
2664 return err;
2665}
2666EXPORT_SYMBOL(scsi_device_quiesce);
2667
2668/**
2669 * scsi_device_resume - Restart user issued commands to a quiesced device.
2670 * @sdev: scsi device to resume.
2671 *
2672 * Moves the device from quiesced back to running and restarts the
2673 * queues.
2674 *
2675 * Must be called with user context, may sleep.
2676 */
2677void scsi_device_resume(struct scsi_device *sdev)
2678{
2679 /* check if the device state was mutated prior to resume, and if
2680 * so assume the state is being managed elsewhere (for example
2681 * device deleted during suspend)
2682 */
2683 mutex_lock(&sdev->state_mutex);
2684 if (sdev->sdev_state == SDEV_QUIESCE)
2685 scsi_device_set_state(sdev, SDEV_RUNNING);
2686 if (sdev->quiesced_by) {
2687 sdev->quiesced_by = NULL;
2688 blk_clear_pm_only(sdev->request_queue);
2689 }
2690 mutex_unlock(&sdev->state_mutex);
2691}
2692EXPORT_SYMBOL(scsi_device_resume);
2693
2694static void
2695device_quiesce_fn(struct scsi_device *sdev, void *data)
2696{
2697 scsi_device_quiesce(sdev);
2698}
2699
2700void
2701scsi_target_quiesce(struct scsi_target *starget)
2702{
2703 starget_for_each_device(starget, NULL, device_quiesce_fn);
2704}
2705EXPORT_SYMBOL(scsi_target_quiesce);
2706
2707static void
2708device_resume_fn(struct scsi_device *sdev, void *data)
2709{
2710 scsi_device_resume(sdev);
2711}
2712
2713void
2714scsi_target_resume(struct scsi_target *starget)
2715{
2716 starget_for_each_device(starget, NULL, device_resume_fn);
2717}
2718EXPORT_SYMBOL(scsi_target_resume);
2719
2720static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2721{
2722 if (scsi_device_set_state(sdev, SDEV_BLOCK))
2723 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2724
2725 return 0;
2726}
2727
2728void scsi_start_queue(struct scsi_device *sdev)
2729{
2730 if (cmpxchg(&sdev->queue_stopped, 1, 0))
2731 blk_mq_unquiesce_queue(sdev->request_queue);
2732}
2733
2734static void scsi_stop_queue(struct scsi_device *sdev)
2735{
2736 /*
2737 * The atomic variable of ->queue_stopped covers that
2738 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2739 *
2740 * The caller needs to wait until quiesce is done.
2741 */
2742 if (!cmpxchg(&sdev->queue_stopped, 0, 1))
2743 blk_mq_quiesce_queue_nowait(sdev->request_queue);
2744}
2745
2746/**
2747 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2748 * @sdev: device to block
2749 *
2750 * Pause SCSI command processing on the specified device. Does not sleep.
2751 *
2752 * Returns zero if successful or a negative error code upon failure.
2753 *
2754 * Notes:
2755 * This routine transitions the device to the SDEV_BLOCK state (which must be
2756 * a legal transition). When the device is in this state, command processing
2757 * is paused until the device leaves the SDEV_BLOCK state. See also
2758 * scsi_internal_device_unblock_nowait().
2759 */
2760int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2761{
2762 int ret = __scsi_internal_device_block_nowait(sdev);
2763
2764 /*
2765 * The device has transitioned to SDEV_BLOCK. Stop the
2766 * block layer from calling the midlayer with this device's
2767 * request queue.
2768 */
2769 if (!ret)
2770 scsi_stop_queue(sdev);
2771 return ret;
2772}
2773EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2774
2775/**
2776 * scsi_device_block - try to transition to the SDEV_BLOCK state
2777 * @sdev: device to block
2778 * @data: dummy argument, ignored
2779 *
2780 * Pause SCSI command processing on the specified device. Callers must wait
2781 * until all ongoing scsi_queue_rq() calls have finished after this function
2782 * returns.
2783 *
2784 * Note:
2785 * This routine transitions the device to the SDEV_BLOCK state (which must be
2786 * a legal transition). When the device is in this state, command processing
2787 * is paused until the device leaves the SDEV_BLOCK state. See also
2788 * scsi_internal_device_unblock().
2789 */
2790static void scsi_device_block(struct scsi_device *sdev, void *data)
2791{
2792 int err;
2793 enum scsi_device_state state;
2794
2795 mutex_lock(&sdev->state_mutex);
2796 err = __scsi_internal_device_block_nowait(sdev);
2797 state = sdev->sdev_state;
2798 if (err == 0)
2799 /*
2800 * scsi_stop_queue() must be called with the state_mutex
2801 * held. Otherwise a simultaneous scsi_start_queue() call
2802 * might unquiesce the queue before we quiesce it.
2803 */
2804 scsi_stop_queue(sdev);
2805
2806 mutex_unlock(&sdev->state_mutex);
2807
2808 WARN_ONCE(err, "%s: failed to block %s in state %d\n",
2809 __func__, dev_name(&sdev->sdev_gendev), state);
2810}
2811
2812/**
2813 * scsi_internal_device_unblock_nowait - 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. Does not
2818 * sleep.
2819 *
2820 * Returns zero if successful or a negative error code upon failure.
2821 *
2822 * Notes:
2823 * This routine transitions the device to the SDEV_RUNNING state or to one of
2824 * the offline states (which must be a legal transition) allowing the midlayer
2825 * to goose the queue for this device.
2826 */
2827int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2828 enum scsi_device_state new_state)
2829{
2830 switch (new_state) {
2831 case SDEV_RUNNING:
2832 case SDEV_TRANSPORT_OFFLINE:
2833 break;
2834 default:
2835 return -EINVAL;
2836 }
2837
2838 /*
2839 * Try to transition the scsi device to SDEV_RUNNING or one of the
2840 * offlined states and goose the device queue if successful.
2841 */
2842 switch (sdev->sdev_state) {
2843 case SDEV_BLOCK:
2844 case SDEV_TRANSPORT_OFFLINE:
2845 sdev->sdev_state = new_state;
2846 break;
2847 case SDEV_CREATED_BLOCK:
2848 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2849 new_state == SDEV_OFFLINE)
2850 sdev->sdev_state = new_state;
2851 else
2852 sdev->sdev_state = SDEV_CREATED;
2853 break;
2854 case SDEV_CANCEL:
2855 case SDEV_OFFLINE:
2856 break;
2857 default:
2858 return -EINVAL;
2859 }
2860 scsi_start_queue(sdev);
2861
2862 return 0;
2863}
2864EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2865
2866/**
2867 * scsi_internal_device_unblock - resume a device after a block request
2868 * @sdev: device to resume
2869 * @new_state: state to set the device to after unblocking
2870 *
2871 * Restart the device queue for a previously suspended SCSI device. May sleep.
2872 *
2873 * Returns zero if successful or a negative error code upon failure.
2874 *
2875 * Notes:
2876 * This routine transitions the device to the SDEV_RUNNING state or to one of
2877 * the offline states (which must be a legal transition) allowing the midlayer
2878 * to goose the queue for this device.
2879 */
2880static int scsi_internal_device_unblock(struct scsi_device *sdev,
2881 enum scsi_device_state new_state)
2882{
2883 int ret;
2884
2885 mutex_lock(&sdev->state_mutex);
2886 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2887 mutex_unlock(&sdev->state_mutex);
2888
2889 return ret;
2890}
2891
2892static int
2893target_block(struct device *dev, void *data)
2894{
2895 if (scsi_is_target_device(dev))
2896 starget_for_each_device(to_scsi_target(dev), NULL,
2897 scsi_device_block);
2898 return 0;
2899}
2900
2901/**
2902 * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
2903 * @dev: a parent device of one or more scsi_target devices
2904 * @shost: the Scsi_Host to which this device belongs
2905 *
2906 * Iterate over all children of @dev, which should be scsi_target devices,
2907 * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
2908 * ongoing scsi_queue_rq() calls to finish. May sleep.
2909 *
2910 * Note:
2911 * @dev must not itself be a scsi_target device.
2912 */
2913void
2914scsi_block_targets(struct Scsi_Host *shost, struct device *dev)
2915{
2916 WARN_ON_ONCE(scsi_is_target_device(dev));
2917 device_for_each_child(dev, NULL, target_block);
2918 blk_mq_wait_quiesce_done(&shost->tag_set);
2919}
2920EXPORT_SYMBOL_GPL(scsi_block_targets);
2921
2922static void
2923device_unblock(struct scsi_device *sdev, void *data)
2924{
2925 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2926}
2927
2928static int
2929target_unblock(struct device *dev, void *data)
2930{
2931 if (scsi_is_target_device(dev))
2932 starget_for_each_device(to_scsi_target(dev), data,
2933 device_unblock);
2934 return 0;
2935}
2936
2937void
2938scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2939{
2940 if (scsi_is_target_device(dev))
2941 starget_for_each_device(to_scsi_target(dev), &new_state,
2942 device_unblock);
2943 else
2944 device_for_each_child(dev, &new_state, target_unblock);
2945}
2946EXPORT_SYMBOL_GPL(scsi_target_unblock);
2947
2948/**
2949 * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
2950 * @shost: device to block
2951 *
2952 * Pause SCSI command processing for all logical units associated with the SCSI
2953 * host and wait until pending scsi_queue_rq() calls have finished.
2954 *
2955 * Returns zero if successful or a negative error code upon failure.
2956 */
2957int
2958scsi_host_block(struct Scsi_Host *shost)
2959{
2960 struct scsi_device *sdev;
2961 int ret;
2962
2963 /*
2964 * Call scsi_internal_device_block_nowait so we can avoid
2965 * calling synchronize_rcu() for each LUN.
2966 */
2967 shost_for_each_device(sdev, shost) {
2968 mutex_lock(&sdev->state_mutex);
2969 ret = scsi_internal_device_block_nowait(sdev);
2970 mutex_unlock(&sdev->state_mutex);
2971 if (ret) {
2972 scsi_device_put(sdev);
2973 return ret;
2974 }
2975 }
2976
2977 /* Wait for ongoing scsi_queue_rq() calls to finish. */
2978 blk_mq_wait_quiesce_done(&shost->tag_set);
2979
2980 return 0;
2981}
2982EXPORT_SYMBOL_GPL(scsi_host_block);
2983
2984int
2985scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2986{
2987 struct scsi_device *sdev;
2988 int ret = 0;
2989
2990 shost_for_each_device(sdev, shost) {
2991 ret = scsi_internal_device_unblock(sdev, new_state);
2992 if (ret) {
2993 scsi_device_put(sdev);
2994 break;
2995 }
2996 }
2997 return ret;
2998}
2999EXPORT_SYMBOL_GPL(scsi_host_unblock);
3000
3001/**
3002 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3003 * @sgl: scatter-gather list
3004 * @sg_count: number of segments in sg
3005 * @offset: offset in bytes into sg, on return offset into the mapped area
3006 * @len: bytes to map, on return number of bytes mapped
3007 *
3008 * Returns virtual address of the start of the mapped page
3009 */
3010void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3011 size_t *offset, size_t *len)
3012{
3013 int i;
3014 size_t sg_len = 0, len_complete = 0;
3015 struct scatterlist *sg;
3016 struct page *page;
3017
3018 WARN_ON(!irqs_disabled());
3019
3020 for_each_sg(sgl, sg, sg_count, i) {
3021 len_complete = sg_len; /* Complete sg-entries */
3022 sg_len += sg->length;
3023 if (sg_len > *offset)
3024 break;
3025 }
3026
3027 if (unlikely(i == sg_count)) {
3028 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3029 "elements %d\n",
3030 __func__, sg_len, *offset, sg_count);
3031 WARN_ON(1);
3032 return NULL;
3033 }
3034
3035 /* Offset starting from the beginning of first page in this sg-entry */
3036 *offset = *offset - len_complete + sg->offset;
3037
3038 /* Assumption: contiguous pages can be accessed as "page + i" */
3039 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3040 *offset &= ~PAGE_MASK;
3041
3042 /* Bytes in this sg-entry from *offset to the end of the page */
3043 sg_len = PAGE_SIZE - *offset;
3044 if (*len > sg_len)
3045 *len = sg_len;
3046
3047 return kmap_atomic(page);
3048}
3049EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3050
3051/**
3052 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3053 * @virt: virtual address to be unmapped
3054 */
3055void scsi_kunmap_atomic_sg(void *virt)
3056{
3057 kunmap_atomic(virt);
3058}
3059EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3060
3061void sdev_disable_disk_events(struct scsi_device *sdev)
3062{
3063 atomic_inc(&sdev->disk_events_disable_depth);
3064}
3065EXPORT_SYMBOL(sdev_disable_disk_events);
3066
3067void sdev_enable_disk_events(struct scsi_device *sdev)
3068{
3069 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3070 return;
3071 atomic_dec(&sdev->disk_events_disable_depth);
3072}
3073EXPORT_SYMBOL(sdev_enable_disk_events);
3074
3075static unsigned char designator_prio(const unsigned char *d)
3076{
3077 if (d[1] & 0x30)
3078 /* not associated with LUN */
3079 return 0;
3080
3081 if (d[3] == 0)
3082 /* invalid length */
3083 return 0;
3084
3085 /*
3086 * Order of preference for lun descriptor:
3087 * - SCSI name string
3088 * - NAA IEEE Registered Extended
3089 * - EUI-64 based 16-byte
3090 * - EUI-64 based 12-byte
3091 * - NAA IEEE Registered
3092 * - NAA IEEE Extended
3093 * - EUI-64 based 8-byte
3094 * - SCSI name string (truncated)
3095 * - T10 Vendor ID
3096 * as longer descriptors reduce the likelyhood
3097 * of identification clashes.
3098 */
3099
3100 switch (d[1] & 0xf) {
3101 case 8:
3102 /* SCSI name string, variable-length UTF-8 */
3103 return 9;
3104 case 3:
3105 switch (d[4] >> 4) {
3106 case 6:
3107 /* NAA registered extended */
3108 return 8;
3109 case 5:
3110 /* NAA registered */
3111 return 5;
3112 case 4:
3113 /* NAA extended */
3114 return 4;
3115 case 3:
3116 /* NAA locally assigned */
3117 return 1;
3118 default:
3119 break;
3120 }
3121 break;
3122 case 2:
3123 switch (d[3]) {
3124 case 16:
3125 /* EUI64-based, 16 byte */
3126 return 7;
3127 case 12:
3128 /* EUI64-based, 12 byte */
3129 return 6;
3130 case 8:
3131 /* EUI64-based, 8 byte */
3132 return 3;
3133 default:
3134 break;
3135 }
3136 break;
3137 case 1:
3138 /* T10 vendor ID */
3139 return 1;
3140 default:
3141 break;
3142 }
3143
3144 return 0;
3145}
3146
3147/**
3148 * scsi_vpd_lun_id - return a unique device identification
3149 * @sdev: SCSI device
3150 * @id: buffer for the identification
3151 * @id_len: length of the buffer
3152 *
3153 * Copies a unique device identification into @id based
3154 * on the information in the VPD page 0x83 of the device.
3155 * The string will be formatted as a SCSI name string.
3156 *
3157 * Returns the length of the identification or error on failure.
3158 * If the identifier is longer than the supplied buffer the actual
3159 * identifier length is returned and the buffer is not zero-padded.
3160 */
3161int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3162{
3163 u8 cur_id_prio = 0;
3164 u8 cur_id_size = 0;
3165 const unsigned char *d, *cur_id_str;
3166 const struct scsi_vpd *vpd_pg83;
3167 int id_size = -EINVAL;
3168
3169 rcu_read_lock();
3170 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3171 if (!vpd_pg83) {
3172 rcu_read_unlock();
3173 return -ENXIO;
3174 }
3175
3176 /* The id string must be at least 20 bytes + terminating NULL byte */
3177 if (id_len < 21) {
3178 rcu_read_unlock();
3179 return -EINVAL;
3180 }
3181
3182 memset(id, 0, id_len);
3183 for (d = vpd_pg83->data + 4;
3184 d < vpd_pg83->data + vpd_pg83->len;
3185 d += d[3] + 4) {
3186 u8 prio = designator_prio(d);
3187
3188 if (prio == 0 || cur_id_prio > prio)
3189 continue;
3190
3191 switch (d[1] & 0xf) {
3192 case 0x1:
3193 /* T10 Vendor ID */
3194 if (cur_id_size > d[3])
3195 break;
3196 cur_id_prio = prio;
3197 cur_id_size = d[3];
3198 if (cur_id_size + 4 > id_len)
3199 cur_id_size = id_len - 4;
3200 cur_id_str = d + 4;
3201 id_size = snprintf(id, id_len, "t10.%*pE",
3202 cur_id_size, cur_id_str);
3203 break;
3204 case 0x2:
3205 /* EUI-64 */
3206 cur_id_prio = prio;
3207 cur_id_size = d[3];
3208 cur_id_str = d + 4;
3209 switch (cur_id_size) {
3210 case 8:
3211 id_size = snprintf(id, id_len,
3212 "eui.%8phN",
3213 cur_id_str);
3214 break;
3215 case 12:
3216 id_size = snprintf(id, id_len,
3217 "eui.%12phN",
3218 cur_id_str);
3219 break;
3220 case 16:
3221 id_size = snprintf(id, id_len,
3222 "eui.%16phN",
3223 cur_id_str);
3224 break;
3225 default:
3226 break;
3227 }
3228 break;
3229 case 0x3:
3230 /* NAA */
3231 cur_id_prio = prio;
3232 cur_id_size = d[3];
3233 cur_id_str = d + 4;
3234 switch (cur_id_size) {
3235 case 8:
3236 id_size = snprintf(id, id_len,
3237 "naa.%8phN",
3238 cur_id_str);
3239 break;
3240 case 16:
3241 id_size = snprintf(id, id_len,
3242 "naa.%16phN",
3243 cur_id_str);
3244 break;
3245 default:
3246 break;
3247 }
3248 break;
3249 case 0x8:
3250 /* SCSI name string */
3251 if (cur_id_size > d[3])
3252 break;
3253 /* Prefer others for truncated descriptor */
3254 if (d[3] > id_len) {
3255 prio = 2;
3256 if (cur_id_prio > prio)
3257 break;
3258 }
3259 cur_id_prio = prio;
3260 cur_id_size = id_size = d[3];
3261 cur_id_str = d + 4;
3262 if (cur_id_size >= id_len)
3263 cur_id_size = id_len - 1;
3264 memcpy(id, cur_id_str, cur_id_size);
3265 break;
3266 default:
3267 break;
3268 }
3269 }
3270 rcu_read_unlock();
3271
3272 return id_size;
3273}
3274EXPORT_SYMBOL(scsi_vpd_lun_id);
3275
3276/*
3277 * scsi_vpd_tpg_id - return a target port group identifier
3278 * @sdev: SCSI device
3279 *
3280 * Returns the Target Port Group identifier from the information
3281 * froom VPD page 0x83 of the device.
3282 *
3283 * Returns the identifier or error on failure.
3284 */
3285int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3286{
3287 const unsigned char *d;
3288 const struct scsi_vpd *vpd_pg83;
3289 int group_id = -EAGAIN, rel_port = -1;
3290
3291 rcu_read_lock();
3292 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3293 if (!vpd_pg83) {
3294 rcu_read_unlock();
3295 return -ENXIO;
3296 }
3297
3298 d = vpd_pg83->data + 4;
3299 while (d < vpd_pg83->data + vpd_pg83->len) {
3300 switch (d[1] & 0xf) {
3301 case 0x4:
3302 /* Relative target port */
3303 rel_port = get_unaligned_be16(&d[6]);
3304 break;
3305 case 0x5:
3306 /* Target port group */
3307 group_id = get_unaligned_be16(&d[6]);
3308 break;
3309 default:
3310 break;
3311 }
3312 d += d[3] + 4;
3313 }
3314 rcu_read_unlock();
3315
3316 if (group_id >= 0 && rel_id && rel_port != -1)
3317 *rel_id = rel_port;
3318
3319 return group_id;
3320}
3321EXPORT_SYMBOL(scsi_vpd_tpg_id);
3322
3323/**
3324 * scsi_build_sense - build sense data for a command
3325 * @scmd: scsi command for which the sense should be formatted
3326 * @desc: Sense format (non-zero == descriptor format,
3327 * 0 == fixed format)
3328 * @key: Sense key
3329 * @asc: Additional sense code
3330 * @ascq: Additional sense code qualifier
3331 *
3332 **/
3333void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3334{
3335 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3336 scmd->result = SAM_STAT_CHECK_CONDITION;
3337}
3338EXPORT_SYMBOL_GPL(scsi_build_sense);
1/*
2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
4 *
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
9 */
10
11#include <linux/bio.h>
12#include <linux/bitops.h>
13#include <linux/blkdev.h>
14#include <linux/completion.h>
15#include <linux/kernel.h>
16#include <linux/export.h>
17#include <linux/mempool.h>
18#include <linux/slab.h>
19#include <linux/init.h>
20#include <linux/pci.h>
21#include <linux/delay.h>
22#include <linux/hardirq.h>
23#include <linux/scatterlist.h>
24#include <linux/blk-mq.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_dh.h>
36
37#include <trace/events/scsi.h>
38
39#include "scsi_priv.h"
40#include "scsi_logging.h"
41
42
43#define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
44#define SG_MEMPOOL_SIZE 2
45
46struct scsi_host_sg_pool {
47 size_t size;
48 char *name;
49 struct kmem_cache *slab;
50 mempool_t *pool;
51};
52
53#define SP(x) { .size = x, "sgpool-" __stringify(x) }
54#if (SCSI_MAX_SG_SEGMENTS < 32)
55#error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
56#endif
57static struct scsi_host_sg_pool scsi_sg_pools[] = {
58 SP(8),
59 SP(16),
60#if (SCSI_MAX_SG_SEGMENTS > 32)
61 SP(32),
62#if (SCSI_MAX_SG_SEGMENTS > 64)
63 SP(64),
64#if (SCSI_MAX_SG_SEGMENTS > 128)
65 SP(128),
66#if (SCSI_MAX_SG_SEGMENTS > 256)
67#error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
68#endif
69#endif
70#endif
71#endif
72 SP(SCSI_MAX_SG_SEGMENTS)
73};
74#undef SP
75
76struct kmem_cache *scsi_sdb_cache;
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
85static void
86scsi_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
121static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
122{
123 struct scsi_device *sdev = cmd->device;
124 struct request_queue *q = cmd->request->q;
125
126 blk_mq_requeue_request(cmd->request);
127 blk_mq_kick_requeue_list(q);
128 put_device(&sdev->sdev_gendev);
129}
130
131/**
132 * __scsi_queue_insert - private queue insertion
133 * @cmd: The SCSI command being requeued
134 * @reason: The reason for the requeue
135 * @unbusy: Whether the queue should be unbusied
136 *
137 * This is a private queue insertion. The public interface
138 * scsi_queue_insert() always assumes the queue should be unbusied
139 * because it's always called before the completion. This function is
140 * for a requeue after completion, which should only occur in this
141 * file.
142 */
143static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
144{
145 struct scsi_device *device = cmd->device;
146 struct request_queue *q = device->request_queue;
147 unsigned long flags;
148
149 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
150 "Inserting command %p into mlqueue\n", cmd));
151
152 scsi_set_blocked(cmd, reason);
153
154 /*
155 * Decrement the counters, since these commands are no longer
156 * active on the host/device.
157 */
158 if (unbusy)
159 scsi_device_unbusy(device);
160
161 /*
162 * Requeue this command. It will go before all other commands
163 * that are already in the queue. Schedule requeue work under
164 * lock such that the kblockd_schedule_work() call happens
165 * before blk_cleanup_queue() finishes.
166 */
167 cmd->result = 0;
168 if (q->mq_ops) {
169 scsi_mq_requeue_cmd(cmd);
170 return;
171 }
172 spin_lock_irqsave(q->queue_lock, flags);
173 blk_requeue_request(q, cmd->request);
174 kblockd_schedule_work(&device->requeue_work);
175 spin_unlock_irqrestore(q->queue_lock, flags);
176}
177
178/*
179 * Function: scsi_queue_insert()
180 *
181 * Purpose: Insert a command in the midlevel queue.
182 *
183 * Arguments: cmd - command that we are adding to queue.
184 * reason - why we are inserting command to queue.
185 *
186 * Lock status: Assumed that lock is not held upon entry.
187 *
188 * Returns: Nothing.
189 *
190 * Notes: We do this for one of two cases. Either the host is busy
191 * and it cannot accept any more commands for the time being,
192 * or the device returned QUEUE_FULL and can accept no more
193 * commands.
194 * Notes: This could be called either from an interrupt context or a
195 * normal process context.
196 */
197void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
198{
199 __scsi_queue_insert(cmd, reason, 1);
200}
201/**
202 * scsi_execute - insert request and wait for the result
203 * @sdev: scsi device
204 * @cmd: scsi command
205 * @data_direction: data direction
206 * @buffer: data buffer
207 * @bufflen: len of buffer
208 * @sense: optional sense buffer
209 * @timeout: request timeout in seconds
210 * @retries: number of times to retry request
211 * @flags: or into request flags;
212 * @resid: optional residual length
213 *
214 * returns the req->errors value which is the scsi_cmnd result
215 * field.
216 */
217int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
218 int data_direction, void *buffer, unsigned bufflen,
219 unsigned char *sense, int timeout, int retries, u64 flags,
220 int *resid)
221{
222 struct request *req;
223 int write = (data_direction == DMA_TO_DEVICE);
224 int ret = DRIVER_ERROR << 24;
225
226 req = blk_get_request(sdev->request_queue, write, __GFP_RECLAIM);
227 if (IS_ERR(req))
228 return ret;
229 blk_rq_set_block_pc(req);
230
231 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
232 buffer, bufflen, __GFP_RECLAIM))
233 goto out;
234
235 req->cmd_len = COMMAND_SIZE(cmd[0]);
236 memcpy(req->cmd, cmd, req->cmd_len);
237 req->sense = sense;
238 req->sense_len = 0;
239 req->retries = retries;
240 req->timeout = timeout;
241 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
242
243 /*
244 * head injection *required* here otherwise quiesce won't work
245 */
246 blk_execute_rq(req->q, NULL, req, 1);
247
248 /*
249 * Some devices (USB mass-storage in particular) may transfer
250 * garbage data together with a residue indicating that the data
251 * is invalid. Prevent the garbage from being misinterpreted
252 * and prevent security leaks by zeroing out the excess data.
253 */
254 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
255 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
256
257 if (resid)
258 *resid = req->resid_len;
259 ret = req->errors;
260 out:
261 blk_put_request(req);
262
263 return ret;
264}
265EXPORT_SYMBOL(scsi_execute);
266
267int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
268 int data_direction, void *buffer, unsigned bufflen,
269 struct scsi_sense_hdr *sshdr, int timeout, int retries,
270 int *resid, u64 flags)
271{
272 char *sense = NULL;
273 int result;
274
275 if (sshdr) {
276 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
277 if (!sense)
278 return DRIVER_ERROR << 24;
279 }
280 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
281 sense, timeout, retries, flags, resid);
282 if (sshdr)
283 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
284
285 kfree(sense);
286 return result;
287}
288EXPORT_SYMBOL(scsi_execute_req_flags);
289
290/*
291 * Function: scsi_init_cmd_errh()
292 *
293 * Purpose: Initialize cmd fields related to error handling.
294 *
295 * Arguments: cmd - command that is ready to be queued.
296 *
297 * Notes: This function has the job of initializing a number of
298 * fields related to error handling. Typically this will
299 * be called once for each command, as required.
300 */
301static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
302{
303 cmd->serial_number = 0;
304 scsi_set_resid(cmd, 0);
305 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
306 if (cmd->cmd_len == 0)
307 cmd->cmd_len = scsi_command_size(cmd->cmnd);
308}
309
310void scsi_device_unbusy(struct scsi_device *sdev)
311{
312 struct Scsi_Host *shost = sdev->host;
313 struct scsi_target *starget = scsi_target(sdev);
314 unsigned long flags;
315
316 atomic_dec(&shost->host_busy);
317 if (starget->can_queue > 0)
318 atomic_dec(&starget->target_busy);
319
320 if (unlikely(scsi_host_in_recovery(shost) &&
321 (shost->host_failed || shost->host_eh_scheduled))) {
322 spin_lock_irqsave(shost->host_lock, flags);
323 scsi_eh_wakeup(shost);
324 spin_unlock_irqrestore(shost->host_lock, flags);
325 }
326
327 atomic_dec(&sdev->device_busy);
328}
329
330static void scsi_kick_queue(struct request_queue *q)
331{
332 if (q->mq_ops)
333 blk_mq_start_hw_queues(q);
334 else
335 blk_run_queue(q);
336}
337
338/*
339 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
340 * and call blk_run_queue for all the scsi_devices on the target -
341 * including current_sdev first.
342 *
343 * Called with *no* scsi locks held.
344 */
345static void scsi_single_lun_run(struct scsi_device *current_sdev)
346{
347 struct Scsi_Host *shost = current_sdev->host;
348 struct scsi_device *sdev, *tmp;
349 struct scsi_target *starget = scsi_target(current_sdev);
350 unsigned long flags;
351
352 spin_lock_irqsave(shost->host_lock, flags);
353 starget->starget_sdev_user = NULL;
354 spin_unlock_irqrestore(shost->host_lock, flags);
355
356 /*
357 * Call blk_run_queue for all LUNs on the target, starting with
358 * current_sdev. We race with others (to set starget_sdev_user),
359 * but in most cases, we will be first. Ideally, each LU on the
360 * target would get some limited time or requests on the target.
361 */
362 scsi_kick_queue(current_sdev->request_queue);
363
364 spin_lock_irqsave(shost->host_lock, flags);
365 if (starget->starget_sdev_user)
366 goto out;
367 list_for_each_entry_safe(sdev, tmp, &starget->devices,
368 same_target_siblings) {
369 if (sdev == current_sdev)
370 continue;
371 if (scsi_device_get(sdev))
372 continue;
373
374 spin_unlock_irqrestore(shost->host_lock, flags);
375 scsi_kick_queue(sdev->request_queue);
376 spin_lock_irqsave(shost->host_lock, flags);
377
378 scsi_device_put(sdev);
379 }
380 out:
381 spin_unlock_irqrestore(shost->host_lock, flags);
382}
383
384static inline bool scsi_device_is_busy(struct scsi_device *sdev)
385{
386 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
387 return true;
388 if (atomic_read(&sdev->device_blocked) > 0)
389 return true;
390 return false;
391}
392
393static inline bool scsi_target_is_busy(struct scsi_target *starget)
394{
395 if (starget->can_queue > 0) {
396 if (atomic_read(&starget->target_busy) >= starget->can_queue)
397 return true;
398 if (atomic_read(&starget->target_blocked) > 0)
399 return true;
400 }
401 return false;
402}
403
404static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
405{
406 if (shost->can_queue > 0 &&
407 atomic_read(&shost->host_busy) >= shost->can_queue)
408 return true;
409 if (atomic_read(&shost->host_blocked) > 0)
410 return true;
411 if (shost->host_self_blocked)
412 return true;
413 return false;
414}
415
416static void scsi_starved_list_run(struct Scsi_Host *shost)
417{
418 LIST_HEAD(starved_list);
419 struct scsi_device *sdev;
420 unsigned long flags;
421
422 spin_lock_irqsave(shost->host_lock, flags);
423 list_splice_init(&shost->starved_list, &starved_list);
424
425 while (!list_empty(&starved_list)) {
426 struct request_queue *slq;
427
428 /*
429 * As long as shost is accepting commands and we have
430 * starved queues, call blk_run_queue. scsi_request_fn
431 * drops the queue_lock and can add us back to the
432 * starved_list.
433 *
434 * host_lock protects the starved_list and starved_entry.
435 * scsi_request_fn must get the host_lock before checking
436 * or modifying starved_list or starved_entry.
437 */
438 if (scsi_host_is_busy(shost))
439 break;
440
441 sdev = list_entry(starved_list.next,
442 struct scsi_device, starved_entry);
443 list_del_init(&sdev->starved_entry);
444 if (scsi_target_is_busy(scsi_target(sdev))) {
445 list_move_tail(&sdev->starved_entry,
446 &shost->starved_list);
447 continue;
448 }
449
450 /*
451 * Once we drop the host lock, a racing scsi_remove_device()
452 * call may remove the sdev from the starved list and destroy
453 * it and the queue. Mitigate by taking a reference to the
454 * queue and never touching the sdev again after we drop the
455 * host lock. Note: if __scsi_remove_device() invokes
456 * blk_cleanup_queue() before the queue is run from this
457 * function then blk_run_queue() will return immediately since
458 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
459 */
460 slq = sdev->request_queue;
461 if (!blk_get_queue(slq))
462 continue;
463 spin_unlock_irqrestore(shost->host_lock, flags);
464
465 scsi_kick_queue(slq);
466 blk_put_queue(slq);
467
468 spin_lock_irqsave(shost->host_lock, flags);
469 }
470 /* put any unprocessed entries back */
471 list_splice(&starved_list, &shost->starved_list);
472 spin_unlock_irqrestore(shost->host_lock, flags);
473}
474
475/*
476 * Function: scsi_run_queue()
477 *
478 * Purpose: Select a proper request queue to serve next
479 *
480 * Arguments: q - last request's queue
481 *
482 * Returns: Nothing
483 *
484 * Notes: The previous command was completely finished, start
485 * a new one if possible.
486 */
487static void scsi_run_queue(struct request_queue *q)
488{
489 struct scsi_device *sdev = q->queuedata;
490
491 if (scsi_target(sdev)->single_lun)
492 scsi_single_lun_run(sdev);
493 if (!list_empty(&sdev->host->starved_list))
494 scsi_starved_list_run(sdev->host);
495
496 if (q->mq_ops)
497 blk_mq_start_stopped_hw_queues(q, false);
498 else
499 blk_run_queue(q);
500}
501
502void scsi_requeue_run_queue(struct work_struct *work)
503{
504 struct scsi_device *sdev;
505 struct request_queue *q;
506
507 sdev = container_of(work, struct scsi_device, requeue_work);
508 q = sdev->request_queue;
509 scsi_run_queue(q);
510}
511
512/*
513 * Function: scsi_requeue_command()
514 *
515 * Purpose: Handle post-processing of completed commands.
516 *
517 * Arguments: q - queue to operate on
518 * cmd - command that may need to be requeued.
519 *
520 * Returns: Nothing
521 *
522 * Notes: After command completion, there may be blocks left
523 * over which weren't finished by the previous command
524 * this can be for a number of reasons - the main one is
525 * I/O errors in the middle of the request, in which case
526 * we need to request the blocks that come after the bad
527 * sector.
528 * Notes: Upon return, cmd is a stale pointer.
529 */
530static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
531{
532 struct scsi_device *sdev = cmd->device;
533 struct request *req = cmd->request;
534 unsigned long flags;
535
536 spin_lock_irqsave(q->queue_lock, flags);
537 blk_unprep_request(req);
538 req->special = NULL;
539 scsi_put_command(cmd);
540 blk_requeue_request(q, req);
541 spin_unlock_irqrestore(q->queue_lock, flags);
542
543 scsi_run_queue(q);
544
545 put_device(&sdev->sdev_gendev);
546}
547
548void scsi_run_host_queues(struct Scsi_Host *shost)
549{
550 struct scsi_device *sdev;
551
552 shost_for_each_device(sdev, shost)
553 scsi_run_queue(sdev->request_queue);
554}
555
556static inline unsigned int scsi_sgtable_index(unsigned short nents)
557{
558 unsigned int index;
559
560 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
561
562 if (nents <= 8)
563 index = 0;
564 else
565 index = get_count_order(nents) - 3;
566
567 return index;
568}
569
570static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
571{
572 struct scsi_host_sg_pool *sgp;
573
574 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
575 mempool_free(sgl, sgp->pool);
576}
577
578static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
579{
580 struct scsi_host_sg_pool *sgp;
581
582 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
583 return mempool_alloc(sgp->pool, gfp_mask);
584}
585
586static void scsi_free_sgtable(struct scsi_data_buffer *sdb, bool mq)
587{
588 if (mq && sdb->table.orig_nents <= SCSI_MAX_SG_SEGMENTS)
589 return;
590 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, mq, scsi_sg_free);
591}
592
593static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents, bool mq)
594{
595 struct scatterlist *first_chunk = NULL;
596 int ret;
597
598 BUG_ON(!nents);
599
600 if (mq) {
601 if (nents <= SCSI_MAX_SG_SEGMENTS) {
602 sdb->table.nents = sdb->table.orig_nents = nents;
603 sg_init_table(sdb->table.sgl, nents);
604 return 0;
605 }
606 first_chunk = sdb->table.sgl;
607 }
608
609 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
610 first_chunk, GFP_ATOMIC, scsi_sg_alloc);
611 if (unlikely(ret))
612 scsi_free_sgtable(sdb, mq);
613 return ret;
614}
615
616static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
617{
618 if (cmd->request->cmd_type == REQ_TYPE_FS) {
619 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
620
621 if (drv->uninit_command)
622 drv->uninit_command(cmd);
623 }
624}
625
626static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
627{
628 if (cmd->sdb.table.nents)
629 scsi_free_sgtable(&cmd->sdb, true);
630 if (cmd->request->next_rq && cmd->request->next_rq->special)
631 scsi_free_sgtable(cmd->request->next_rq->special, true);
632 if (scsi_prot_sg_count(cmd))
633 scsi_free_sgtable(cmd->prot_sdb, true);
634}
635
636static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
637{
638 struct scsi_device *sdev = cmd->device;
639 struct Scsi_Host *shost = sdev->host;
640 unsigned long flags;
641
642 scsi_mq_free_sgtables(cmd);
643 scsi_uninit_cmd(cmd);
644
645 if (shost->use_cmd_list) {
646 BUG_ON(list_empty(&cmd->list));
647 spin_lock_irqsave(&sdev->list_lock, flags);
648 list_del_init(&cmd->list);
649 spin_unlock_irqrestore(&sdev->list_lock, flags);
650 }
651}
652
653/*
654 * Function: scsi_release_buffers()
655 *
656 * Purpose: Free resources allocate for a scsi_command.
657 *
658 * Arguments: cmd - command that we are bailing.
659 *
660 * Lock status: Assumed that no lock is held upon entry.
661 *
662 * Returns: Nothing
663 *
664 * Notes: In the event that an upper level driver rejects a
665 * command, we must release resources allocated during
666 * the __init_io() function. Primarily this would involve
667 * the scatter-gather table.
668 */
669static void scsi_release_buffers(struct scsi_cmnd *cmd)
670{
671 if (cmd->sdb.table.nents)
672 scsi_free_sgtable(&cmd->sdb, false);
673
674 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
675
676 if (scsi_prot_sg_count(cmd))
677 scsi_free_sgtable(cmd->prot_sdb, false);
678}
679
680static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
681{
682 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
683
684 scsi_free_sgtable(bidi_sdb, false);
685 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
686 cmd->request->next_rq->special = NULL;
687}
688
689static bool scsi_end_request(struct request *req, int error,
690 unsigned int bytes, unsigned int bidi_bytes)
691{
692 struct scsi_cmnd *cmd = req->special;
693 struct scsi_device *sdev = cmd->device;
694 struct request_queue *q = sdev->request_queue;
695
696 if (blk_update_request(req, error, bytes))
697 return true;
698
699 /* Bidi request must be completed as a whole */
700 if (unlikely(bidi_bytes) &&
701 blk_update_request(req->next_rq, error, bidi_bytes))
702 return true;
703
704 if (blk_queue_add_random(q))
705 add_disk_randomness(req->rq_disk);
706
707 if (req->mq_ctx) {
708 /*
709 * In the MQ case the command gets freed by __blk_mq_end_request,
710 * so we have to do all cleanup that depends on it earlier.
711 *
712 * We also can't kick the queues from irq context, so we
713 * will have to defer it to a workqueue.
714 */
715 scsi_mq_uninit_cmd(cmd);
716
717 __blk_mq_end_request(req, error);
718
719 if (scsi_target(sdev)->single_lun ||
720 !list_empty(&sdev->host->starved_list))
721 kblockd_schedule_work(&sdev->requeue_work);
722 else
723 blk_mq_start_stopped_hw_queues(q, true);
724 } else {
725 unsigned long flags;
726
727 if (bidi_bytes)
728 scsi_release_bidi_buffers(cmd);
729
730 spin_lock_irqsave(q->queue_lock, flags);
731 blk_finish_request(req, error);
732 spin_unlock_irqrestore(q->queue_lock, flags);
733
734 scsi_release_buffers(cmd);
735
736 scsi_put_command(cmd);
737 scsi_run_queue(q);
738 }
739
740 put_device(&sdev->sdev_gendev);
741 return false;
742}
743
744/**
745 * __scsi_error_from_host_byte - translate SCSI error code into errno
746 * @cmd: SCSI command (unused)
747 * @result: scsi error code
748 *
749 * Translate SCSI error code into standard UNIX errno.
750 * Return values:
751 * -ENOLINK temporary transport failure
752 * -EREMOTEIO permanent target failure, do not retry
753 * -EBADE permanent nexus failure, retry on other path
754 * -ENOSPC No write space available
755 * -ENODATA Medium error
756 * -EIO unspecified I/O error
757 */
758static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
759{
760 int error = 0;
761
762 switch(host_byte(result)) {
763 case DID_TRANSPORT_FAILFAST:
764 error = -ENOLINK;
765 break;
766 case DID_TARGET_FAILURE:
767 set_host_byte(cmd, DID_OK);
768 error = -EREMOTEIO;
769 break;
770 case DID_NEXUS_FAILURE:
771 set_host_byte(cmd, DID_OK);
772 error = -EBADE;
773 break;
774 case DID_ALLOC_FAILURE:
775 set_host_byte(cmd, DID_OK);
776 error = -ENOSPC;
777 break;
778 case DID_MEDIUM_ERROR:
779 set_host_byte(cmd, DID_OK);
780 error = -ENODATA;
781 break;
782 default:
783 error = -EIO;
784 break;
785 }
786
787 return error;
788}
789
790/*
791 * Function: scsi_io_completion()
792 *
793 * Purpose: Completion processing for block device I/O requests.
794 *
795 * Arguments: cmd - command that is finished.
796 *
797 * Lock status: Assumed that no lock is held upon entry.
798 *
799 * Returns: Nothing
800 *
801 * Notes: We will finish off the specified number of sectors. If we
802 * are done, the command block will be released and the queue
803 * function will be goosed. If we are not done then we have to
804 * figure out what to do next:
805 *
806 * a) We can call scsi_requeue_command(). The request
807 * will be unprepared and put back on the queue. Then
808 * a new command will be created for it. This should
809 * be used if we made forward progress, or if we want
810 * to switch from READ(10) to READ(6) for example.
811 *
812 * b) We can call __scsi_queue_insert(). The request will
813 * be put back on the queue and retried using the same
814 * command as before, possibly after a delay.
815 *
816 * c) We can call scsi_end_request() with -EIO to fail
817 * the remainder of the request.
818 */
819void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
820{
821 int result = cmd->result;
822 struct request_queue *q = cmd->device->request_queue;
823 struct request *req = cmd->request;
824 int error = 0;
825 struct scsi_sense_hdr sshdr;
826 bool sense_valid = false;
827 int sense_deferred = 0, level = 0;
828 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
829 ACTION_DELAYED_RETRY} action;
830 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
831
832 if (result) {
833 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
834 if (sense_valid)
835 sense_deferred = scsi_sense_is_deferred(&sshdr);
836 }
837
838 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
839 if (result) {
840 if (sense_valid && req->sense) {
841 /*
842 * SG_IO wants current and deferred errors
843 */
844 int len = 8 + cmd->sense_buffer[7];
845
846 if (len > SCSI_SENSE_BUFFERSIZE)
847 len = SCSI_SENSE_BUFFERSIZE;
848 memcpy(req->sense, cmd->sense_buffer, len);
849 req->sense_len = len;
850 }
851 if (!sense_deferred)
852 error = __scsi_error_from_host_byte(cmd, result);
853 }
854 /*
855 * __scsi_error_from_host_byte may have reset the host_byte
856 */
857 req->errors = cmd->result;
858
859 req->resid_len = scsi_get_resid(cmd);
860
861 if (scsi_bidi_cmnd(cmd)) {
862 /*
863 * Bidi commands Must be complete as a whole,
864 * both sides at once.
865 */
866 req->next_rq->resid_len = scsi_in(cmd)->resid;
867 if (scsi_end_request(req, 0, blk_rq_bytes(req),
868 blk_rq_bytes(req->next_rq)))
869 BUG();
870 return;
871 }
872 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
873 /*
874 * Certain non BLOCK_PC requests are commands that don't
875 * actually transfer anything (FLUSH), so cannot use
876 * good_bytes != blk_rq_bytes(req) as the signal for an error.
877 * This sets the error explicitly for the problem case.
878 */
879 error = __scsi_error_from_host_byte(cmd, result);
880 }
881
882 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
883 BUG_ON(blk_bidi_rq(req));
884
885 /*
886 * Next deal with any sectors which we were able to correctly
887 * handle.
888 */
889 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
890 "%u sectors total, %d bytes done.\n",
891 blk_rq_sectors(req), good_bytes));
892
893 /*
894 * Recovered errors need reporting, but they're always treated
895 * as success, so fiddle the result code here. For BLOCK_PC
896 * we already took a copy of the original into rq->errors which
897 * is what gets returned to the user
898 */
899 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
900 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
901 * print since caller wants ATA registers. Only occurs on
902 * SCSI ATA PASS_THROUGH commands when CK_COND=1
903 */
904 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
905 ;
906 else if (!(req->cmd_flags & REQ_QUIET))
907 scsi_print_sense(cmd);
908 result = 0;
909 /* BLOCK_PC may have set error */
910 error = 0;
911 }
912
913 /*
914 * If we finished all bytes in the request we are done now.
915 */
916 if (!scsi_end_request(req, error, good_bytes, 0))
917 return;
918
919 /*
920 * Kill remainder if no retrys.
921 */
922 if (error && scsi_noretry_cmd(cmd)) {
923 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
924 BUG();
925 return;
926 }
927
928 /*
929 * If there had been no error, but we have leftover bytes in the
930 * requeues just queue the command up again.
931 */
932 if (result == 0)
933 goto requeue;
934
935 error = __scsi_error_from_host_byte(cmd, result);
936
937 if (host_byte(result) == DID_RESET) {
938 /* Third party bus reset or reset for error recovery
939 * reasons. Just retry the command and see what
940 * happens.
941 */
942 action = ACTION_RETRY;
943 } else if (sense_valid && !sense_deferred) {
944 switch (sshdr.sense_key) {
945 case UNIT_ATTENTION:
946 if (cmd->device->removable) {
947 /* Detected disc change. Set a bit
948 * and quietly refuse further access.
949 */
950 cmd->device->changed = 1;
951 action = ACTION_FAIL;
952 } else {
953 /* Must have been a power glitch, or a
954 * bus reset. Could not have been a
955 * media change, so we just retry the
956 * command and see what happens.
957 */
958 action = ACTION_RETRY;
959 }
960 break;
961 case ILLEGAL_REQUEST:
962 /* If we had an ILLEGAL REQUEST returned, then
963 * we may have performed an unsupported
964 * command. The only thing this should be
965 * would be a ten byte read where only a six
966 * byte read was supported. Also, on a system
967 * where READ CAPACITY failed, we may have
968 * read past the end of the disk.
969 */
970 if ((cmd->device->use_10_for_rw &&
971 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
972 (cmd->cmnd[0] == READ_10 ||
973 cmd->cmnd[0] == WRITE_10)) {
974 /* This will issue a new 6-byte command. */
975 cmd->device->use_10_for_rw = 0;
976 action = ACTION_REPREP;
977 } else if (sshdr.asc == 0x10) /* DIX */ {
978 action = ACTION_FAIL;
979 error = -EILSEQ;
980 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
981 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
982 action = ACTION_FAIL;
983 error = -EREMOTEIO;
984 } else
985 action = ACTION_FAIL;
986 break;
987 case ABORTED_COMMAND:
988 action = ACTION_FAIL;
989 if (sshdr.asc == 0x10) /* DIF */
990 error = -EILSEQ;
991 break;
992 case NOT_READY:
993 /* If the device is in the process of becoming
994 * ready, or has a temporary blockage, retry.
995 */
996 if (sshdr.asc == 0x04) {
997 switch (sshdr.ascq) {
998 case 0x01: /* becoming ready */
999 case 0x04: /* format in progress */
1000 case 0x05: /* rebuild in progress */
1001 case 0x06: /* recalculation in progress */
1002 case 0x07: /* operation in progress */
1003 case 0x08: /* Long write in progress */
1004 case 0x09: /* self test in progress */
1005 case 0x14: /* space allocation in progress */
1006 action = ACTION_DELAYED_RETRY;
1007 break;
1008 default:
1009 action = ACTION_FAIL;
1010 break;
1011 }
1012 } else
1013 action = ACTION_FAIL;
1014 break;
1015 case VOLUME_OVERFLOW:
1016 /* See SSC3rXX or current. */
1017 action = ACTION_FAIL;
1018 break;
1019 default:
1020 action = ACTION_FAIL;
1021 break;
1022 }
1023 } else
1024 action = ACTION_FAIL;
1025
1026 if (action != ACTION_FAIL &&
1027 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
1028 action = ACTION_FAIL;
1029
1030 switch (action) {
1031 case ACTION_FAIL:
1032 /* Give up and fail the remainder of the request */
1033 if (!(req->cmd_flags & REQ_QUIET)) {
1034 static DEFINE_RATELIMIT_STATE(_rs,
1035 DEFAULT_RATELIMIT_INTERVAL,
1036 DEFAULT_RATELIMIT_BURST);
1037
1038 if (unlikely(scsi_logging_level))
1039 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
1040 SCSI_LOG_MLCOMPLETE_BITS);
1041
1042 /*
1043 * if logging is enabled the failure will be printed
1044 * in scsi_log_completion(), so avoid duplicate messages
1045 */
1046 if (!level && __ratelimit(&_rs)) {
1047 scsi_print_result(cmd, NULL, FAILED);
1048 if (driver_byte(result) & DRIVER_SENSE)
1049 scsi_print_sense(cmd);
1050 scsi_print_command(cmd);
1051 }
1052 }
1053 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
1054 return;
1055 /*FALLTHRU*/
1056 case ACTION_REPREP:
1057 requeue:
1058 /* Unprep the request and put it back at the head of the queue.
1059 * A new command will be prepared and issued.
1060 */
1061 if (q->mq_ops) {
1062 cmd->request->cmd_flags &= ~REQ_DONTPREP;
1063 scsi_mq_uninit_cmd(cmd);
1064 scsi_mq_requeue_cmd(cmd);
1065 } else {
1066 scsi_release_buffers(cmd);
1067 scsi_requeue_command(q, cmd);
1068 }
1069 break;
1070 case ACTION_RETRY:
1071 /* Retry the same command immediately */
1072 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1073 break;
1074 case ACTION_DELAYED_RETRY:
1075 /* Retry the same command after a delay */
1076 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1077 break;
1078 }
1079}
1080
1081static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1082{
1083 int count;
1084
1085 /*
1086 * If sg table allocation fails, requeue request later.
1087 */
1088 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1089 req->mq_ctx != NULL)))
1090 return BLKPREP_DEFER;
1091
1092 /*
1093 * Next, walk the list, and fill in the addresses and sizes of
1094 * each segment.
1095 */
1096 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1097 BUG_ON(count > sdb->table.nents);
1098 sdb->table.nents = count;
1099 sdb->length = blk_rq_bytes(req);
1100 return BLKPREP_OK;
1101}
1102
1103/*
1104 * Function: scsi_init_io()
1105 *
1106 * Purpose: SCSI I/O initialize function.
1107 *
1108 * Arguments: cmd - Command descriptor we wish to initialize
1109 *
1110 * Returns: 0 on success
1111 * BLKPREP_DEFER if the failure is retryable
1112 * BLKPREP_KILL if the failure is fatal
1113 */
1114int scsi_init_io(struct scsi_cmnd *cmd)
1115{
1116 struct scsi_device *sdev = cmd->device;
1117 struct request *rq = cmd->request;
1118 bool is_mq = (rq->mq_ctx != NULL);
1119 int error;
1120
1121 BUG_ON(!rq->nr_phys_segments);
1122
1123 error = scsi_init_sgtable(rq, &cmd->sdb);
1124 if (error)
1125 goto err_exit;
1126
1127 if (blk_bidi_rq(rq)) {
1128 if (!rq->q->mq_ops) {
1129 struct scsi_data_buffer *bidi_sdb =
1130 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1131 if (!bidi_sdb) {
1132 error = BLKPREP_DEFER;
1133 goto err_exit;
1134 }
1135
1136 rq->next_rq->special = bidi_sdb;
1137 }
1138
1139 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1140 if (error)
1141 goto err_exit;
1142 }
1143
1144 if (blk_integrity_rq(rq)) {
1145 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1146 int ivecs, count;
1147
1148 if (prot_sdb == NULL) {
1149 /*
1150 * This can happen if someone (e.g. multipath)
1151 * queues a command to a device on an adapter
1152 * that does not support DIX.
1153 */
1154 WARN_ON_ONCE(1);
1155 error = BLKPREP_KILL;
1156 goto err_exit;
1157 }
1158
1159 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1160
1161 if (scsi_alloc_sgtable(prot_sdb, ivecs, is_mq)) {
1162 error = BLKPREP_DEFER;
1163 goto err_exit;
1164 }
1165
1166 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1167 prot_sdb->table.sgl);
1168 BUG_ON(unlikely(count > ivecs));
1169 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1170
1171 cmd->prot_sdb = prot_sdb;
1172 cmd->prot_sdb->table.nents = count;
1173 }
1174
1175 return BLKPREP_OK;
1176err_exit:
1177 if (is_mq) {
1178 scsi_mq_free_sgtables(cmd);
1179 } else {
1180 scsi_release_buffers(cmd);
1181 cmd->request->special = NULL;
1182 scsi_put_command(cmd);
1183 put_device(&sdev->sdev_gendev);
1184 }
1185 return error;
1186}
1187EXPORT_SYMBOL(scsi_init_io);
1188
1189static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1190 struct request *req)
1191{
1192 struct scsi_cmnd *cmd;
1193
1194 if (!req->special) {
1195 /* Bail if we can't get a reference to the device */
1196 if (!get_device(&sdev->sdev_gendev))
1197 return NULL;
1198
1199 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1200 if (unlikely(!cmd)) {
1201 put_device(&sdev->sdev_gendev);
1202 return NULL;
1203 }
1204 req->special = cmd;
1205 } else {
1206 cmd = req->special;
1207 }
1208
1209 /* pull a tag out of the request if we have one */
1210 cmd->tag = req->tag;
1211 cmd->request = req;
1212
1213 cmd->cmnd = req->cmd;
1214 cmd->prot_op = SCSI_PROT_NORMAL;
1215
1216 return cmd;
1217}
1218
1219static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1220{
1221 struct scsi_cmnd *cmd = req->special;
1222
1223 /*
1224 * BLOCK_PC requests may transfer data, in which case they must
1225 * a bio attached to them. Or they might contain a SCSI command
1226 * that does not transfer data, in which case they may optionally
1227 * submit a request without an attached bio.
1228 */
1229 if (req->bio) {
1230 int ret = scsi_init_io(cmd);
1231 if (unlikely(ret))
1232 return ret;
1233 } else {
1234 BUG_ON(blk_rq_bytes(req));
1235
1236 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1237 }
1238
1239 cmd->cmd_len = req->cmd_len;
1240 cmd->transfersize = blk_rq_bytes(req);
1241 cmd->allowed = req->retries;
1242 return BLKPREP_OK;
1243}
1244
1245/*
1246 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1247 * that still need to be translated to SCSI CDBs from the ULD.
1248 */
1249static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1250{
1251 struct scsi_cmnd *cmd = req->special;
1252
1253 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1254 int ret = sdev->handler->prep_fn(sdev, req);
1255 if (ret != BLKPREP_OK)
1256 return ret;
1257 }
1258
1259 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1260 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1261}
1262
1263static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1264{
1265 struct scsi_cmnd *cmd = req->special;
1266
1267 if (!blk_rq_bytes(req))
1268 cmd->sc_data_direction = DMA_NONE;
1269 else if (rq_data_dir(req) == WRITE)
1270 cmd->sc_data_direction = DMA_TO_DEVICE;
1271 else
1272 cmd->sc_data_direction = DMA_FROM_DEVICE;
1273
1274 switch (req->cmd_type) {
1275 case REQ_TYPE_FS:
1276 return scsi_setup_fs_cmnd(sdev, req);
1277 case REQ_TYPE_BLOCK_PC:
1278 return scsi_setup_blk_pc_cmnd(sdev, req);
1279 default:
1280 return BLKPREP_KILL;
1281 }
1282}
1283
1284static int
1285scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1286{
1287 int ret = BLKPREP_OK;
1288
1289 /*
1290 * If the device is not in running state we will reject some
1291 * or all commands.
1292 */
1293 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1294 switch (sdev->sdev_state) {
1295 case SDEV_OFFLINE:
1296 case SDEV_TRANSPORT_OFFLINE:
1297 /*
1298 * If the device is offline we refuse to process any
1299 * commands. The device must be brought online
1300 * before trying any recovery commands.
1301 */
1302 sdev_printk(KERN_ERR, sdev,
1303 "rejecting I/O to offline device\n");
1304 ret = BLKPREP_KILL;
1305 break;
1306 case SDEV_DEL:
1307 /*
1308 * If the device is fully deleted, we refuse to
1309 * process any commands as well.
1310 */
1311 sdev_printk(KERN_ERR, sdev,
1312 "rejecting I/O to dead device\n");
1313 ret = BLKPREP_KILL;
1314 break;
1315 case SDEV_BLOCK:
1316 case SDEV_CREATED_BLOCK:
1317 ret = BLKPREP_DEFER;
1318 break;
1319 case SDEV_QUIESCE:
1320 /*
1321 * If the devices is blocked we defer normal commands.
1322 */
1323 if (!(req->cmd_flags & REQ_PREEMPT))
1324 ret = BLKPREP_DEFER;
1325 break;
1326 default:
1327 /*
1328 * For any other not fully online state we only allow
1329 * special commands. In particular any user initiated
1330 * command is not allowed.
1331 */
1332 if (!(req->cmd_flags & REQ_PREEMPT))
1333 ret = BLKPREP_KILL;
1334 break;
1335 }
1336 }
1337 return ret;
1338}
1339
1340static int
1341scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1342{
1343 struct scsi_device *sdev = q->queuedata;
1344
1345 switch (ret) {
1346 case BLKPREP_KILL:
1347 case BLKPREP_INVALID:
1348 req->errors = DID_NO_CONNECT << 16;
1349 /* release the command and kill it */
1350 if (req->special) {
1351 struct scsi_cmnd *cmd = req->special;
1352 scsi_release_buffers(cmd);
1353 scsi_put_command(cmd);
1354 put_device(&sdev->sdev_gendev);
1355 req->special = NULL;
1356 }
1357 break;
1358 case BLKPREP_DEFER:
1359 /*
1360 * If we defer, the blk_peek_request() returns NULL, but the
1361 * queue must be restarted, so we schedule a callback to happen
1362 * shortly.
1363 */
1364 if (atomic_read(&sdev->device_busy) == 0)
1365 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1366 break;
1367 default:
1368 req->cmd_flags |= REQ_DONTPREP;
1369 }
1370
1371 return ret;
1372}
1373
1374static int scsi_prep_fn(struct request_queue *q, struct request *req)
1375{
1376 struct scsi_device *sdev = q->queuedata;
1377 struct scsi_cmnd *cmd;
1378 int ret;
1379
1380 ret = scsi_prep_state_check(sdev, req);
1381 if (ret != BLKPREP_OK)
1382 goto out;
1383
1384 cmd = scsi_get_cmd_from_req(sdev, req);
1385 if (unlikely(!cmd)) {
1386 ret = BLKPREP_DEFER;
1387 goto out;
1388 }
1389
1390 ret = scsi_setup_cmnd(sdev, req);
1391out:
1392 return scsi_prep_return(q, req, ret);
1393}
1394
1395static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1396{
1397 scsi_uninit_cmd(req->special);
1398}
1399
1400/*
1401 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1402 * return 0.
1403 *
1404 * Called with the queue_lock held.
1405 */
1406static inline int scsi_dev_queue_ready(struct request_queue *q,
1407 struct scsi_device *sdev)
1408{
1409 unsigned int busy;
1410
1411 busy = atomic_inc_return(&sdev->device_busy) - 1;
1412 if (atomic_read(&sdev->device_blocked)) {
1413 if (busy)
1414 goto out_dec;
1415
1416 /*
1417 * unblock after device_blocked iterates to zero
1418 */
1419 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1420 /*
1421 * For the MQ case we take care of this in the caller.
1422 */
1423 if (!q->mq_ops)
1424 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1425 goto out_dec;
1426 }
1427 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1428 "unblocking device at zero depth\n"));
1429 }
1430
1431 if (busy >= sdev->queue_depth)
1432 goto out_dec;
1433
1434 return 1;
1435out_dec:
1436 atomic_dec(&sdev->device_busy);
1437 return 0;
1438}
1439
1440/*
1441 * scsi_target_queue_ready: checks if there we can send commands to target
1442 * @sdev: scsi device on starget to check.
1443 */
1444static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1445 struct scsi_device *sdev)
1446{
1447 struct scsi_target *starget = scsi_target(sdev);
1448 unsigned int busy;
1449
1450 if (starget->single_lun) {
1451 spin_lock_irq(shost->host_lock);
1452 if (starget->starget_sdev_user &&
1453 starget->starget_sdev_user != sdev) {
1454 spin_unlock_irq(shost->host_lock);
1455 return 0;
1456 }
1457 starget->starget_sdev_user = sdev;
1458 spin_unlock_irq(shost->host_lock);
1459 }
1460
1461 if (starget->can_queue <= 0)
1462 return 1;
1463
1464 busy = atomic_inc_return(&starget->target_busy) - 1;
1465 if (atomic_read(&starget->target_blocked) > 0) {
1466 if (busy)
1467 goto starved;
1468
1469 /*
1470 * unblock after target_blocked iterates to zero
1471 */
1472 if (atomic_dec_return(&starget->target_blocked) > 0)
1473 goto out_dec;
1474
1475 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1476 "unblocking target at zero depth\n"));
1477 }
1478
1479 if (busy >= starget->can_queue)
1480 goto starved;
1481
1482 return 1;
1483
1484starved:
1485 spin_lock_irq(shost->host_lock);
1486 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1487 spin_unlock_irq(shost->host_lock);
1488out_dec:
1489 if (starget->can_queue > 0)
1490 atomic_dec(&starget->target_busy);
1491 return 0;
1492}
1493
1494/*
1495 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1496 * return 0. We must end up running the queue again whenever 0 is
1497 * returned, else IO can hang.
1498 */
1499static inline int scsi_host_queue_ready(struct request_queue *q,
1500 struct Scsi_Host *shost,
1501 struct scsi_device *sdev)
1502{
1503 unsigned int busy;
1504
1505 if (scsi_host_in_recovery(shost))
1506 return 0;
1507
1508 busy = atomic_inc_return(&shost->host_busy) - 1;
1509 if (atomic_read(&shost->host_blocked) > 0) {
1510 if (busy)
1511 goto starved;
1512
1513 /*
1514 * unblock after host_blocked iterates to zero
1515 */
1516 if (atomic_dec_return(&shost->host_blocked) > 0)
1517 goto out_dec;
1518
1519 SCSI_LOG_MLQUEUE(3,
1520 shost_printk(KERN_INFO, shost,
1521 "unblocking host at zero depth\n"));
1522 }
1523
1524 if (shost->can_queue > 0 && busy >= shost->can_queue)
1525 goto starved;
1526 if (shost->host_self_blocked)
1527 goto starved;
1528
1529 /* We're OK to process the command, so we can't be starved */
1530 if (!list_empty(&sdev->starved_entry)) {
1531 spin_lock_irq(shost->host_lock);
1532 if (!list_empty(&sdev->starved_entry))
1533 list_del_init(&sdev->starved_entry);
1534 spin_unlock_irq(shost->host_lock);
1535 }
1536
1537 return 1;
1538
1539starved:
1540 spin_lock_irq(shost->host_lock);
1541 if (list_empty(&sdev->starved_entry))
1542 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1543 spin_unlock_irq(shost->host_lock);
1544out_dec:
1545 atomic_dec(&shost->host_busy);
1546 return 0;
1547}
1548
1549/*
1550 * Busy state exporting function for request stacking drivers.
1551 *
1552 * For efficiency, no lock is taken to check the busy state of
1553 * shost/starget/sdev, since the returned value is not guaranteed and
1554 * may be changed after request stacking drivers call the function,
1555 * regardless of taking lock or not.
1556 *
1557 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1558 * needs to return 'not busy'. Otherwise, request stacking drivers
1559 * may hold requests forever.
1560 */
1561static int scsi_lld_busy(struct request_queue *q)
1562{
1563 struct scsi_device *sdev = q->queuedata;
1564 struct Scsi_Host *shost;
1565
1566 if (blk_queue_dying(q))
1567 return 0;
1568
1569 shost = sdev->host;
1570
1571 /*
1572 * Ignore host/starget busy state.
1573 * Since block layer does not have a concept of fairness across
1574 * multiple queues, congestion of host/starget needs to be handled
1575 * in SCSI layer.
1576 */
1577 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1578 return 1;
1579
1580 return 0;
1581}
1582
1583/*
1584 * Kill a request for a dead device
1585 */
1586static void scsi_kill_request(struct request *req, struct request_queue *q)
1587{
1588 struct scsi_cmnd *cmd = req->special;
1589 struct scsi_device *sdev;
1590 struct scsi_target *starget;
1591 struct Scsi_Host *shost;
1592
1593 blk_start_request(req);
1594
1595 scmd_printk(KERN_INFO, cmd, "killing request\n");
1596
1597 sdev = cmd->device;
1598 starget = scsi_target(sdev);
1599 shost = sdev->host;
1600 scsi_init_cmd_errh(cmd);
1601 cmd->result = DID_NO_CONNECT << 16;
1602 atomic_inc(&cmd->device->iorequest_cnt);
1603
1604 /*
1605 * SCSI request completion path will do scsi_device_unbusy(),
1606 * bump busy counts. To bump the counters, we need to dance
1607 * with the locks as normal issue path does.
1608 */
1609 atomic_inc(&sdev->device_busy);
1610 atomic_inc(&shost->host_busy);
1611 if (starget->can_queue > 0)
1612 atomic_inc(&starget->target_busy);
1613
1614 blk_complete_request(req);
1615}
1616
1617static void scsi_softirq_done(struct request *rq)
1618{
1619 struct scsi_cmnd *cmd = rq->special;
1620 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1621 int disposition;
1622
1623 INIT_LIST_HEAD(&cmd->eh_entry);
1624
1625 atomic_inc(&cmd->device->iodone_cnt);
1626 if (cmd->result)
1627 atomic_inc(&cmd->device->ioerr_cnt);
1628
1629 disposition = scsi_decide_disposition(cmd);
1630 if (disposition != SUCCESS &&
1631 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1632 sdev_printk(KERN_ERR, cmd->device,
1633 "timing out command, waited %lus\n",
1634 wait_for/HZ);
1635 disposition = SUCCESS;
1636 }
1637
1638 scsi_log_completion(cmd, disposition);
1639
1640 switch (disposition) {
1641 case SUCCESS:
1642 scsi_finish_command(cmd);
1643 break;
1644 case NEEDS_RETRY:
1645 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1646 break;
1647 case ADD_TO_MLQUEUE:
1648 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1649 break;
1650 default:
1651 if (!scsi_eh_scmd_add(cmd, 0))
1652 scsi_finish_command(cmd);
1653 }
1654}
1655
1656/**
1657 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1658 * @cmd: command block we are dispatching.
1659 *
1660 * Return: nonzero return request was rejected and device's queue needs to be
1661 * plugged.
1662 */
1663static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1664{
1665 struct Scsi_Host *host = cmd->device->host;
1666 int rtn = 0;
1667
1668 atomic_inc(&cmd->device->iorequest_cnt);
1669
1670 /* check if the device is still usable */
1671 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1672 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1673 * returns an immediate error upwards, and signals
1674 * that the device is no longer present */
1675 cmd->result = DID_NO_CONNECT << 16;
1676 goto done;
1677 }
1678
1679 /* Check to see if the scsi lld made this device blocked. */
1680 if (unlikely(scsi_device_blocked(cmd->device))) {
1681 /*
1682 * in blocked state, the command is just put back on
1683 * the device queue. The suspend state has already
1684 * blocked the queue so future requests should not
1685 * occur until the device transitions out of the
1686 * suspend state.
1687 */
1688 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1689 "queuecommand : device blocked\n"));
1690 return SCSI_MLQUEUE_DEVICE_BUSY;
1691 }
1692
1693 /* Store the LUN value in cmnd, if needed. */
1694 if (cmd->device->lun_in_cdb)
1695 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1696 (cmd->device->lun << 5 & 0xe0);
1697
1698 scsi_log_send(cmd);
1699
1700 /*
1701 * Before we queue this command, check if the command
1702 * length exceeds what the host adapter can handle.
1703 */
1704 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1705 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1706 "queuecommand : command too long. "
1707 "cdb_size=%d host->max_cmd_len=%d\n",
1708 cmd->cmd_len, cmd->device->host->max_cmd_len));
1709 cmd->result = (DID_ABORT << 16);
1710 goto done;
1711 }
1712
1713 if (unlikely(host->shost_state == SHOST_DEL)) {
1714 cmd->result = (DID_NO_CONNECT << 16);
1715 goto done;
1716
1717 }
1718
1719 trace_scsi_dispatch_cmd_start(cmd);
1720 rtn = host->hostt->queuecommand(host, cmd);
1721 if (rtn) {
1722 trace_scsi_dispatch_cmd_error(cmd, rtn);
1723 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1724 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1725 rtn = SCSI_MLQUEUE_HOST_BUSY;
1726
1727 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1728 "queuecommand : request rejected\n"));
1729 }
1730
1731 return rtn;
1732 done:
1733 cmd->scsi_done(cmd);
1734 return 0;
1735}
1736
1737/**
1738 * scsi_done - Invoke completion on finished SCSI command.
1739 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1740 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1741 *
1742 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1743 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1744 * calls blk_complete_request() for further processing.
1745 *
1746 * This function is interrupt context safe.
1747 */
1748static void scsi_done(struct scsi_cmnd *cmd)
1749{
1750 trace_scsi_dispatch_cmd_done(cmd);
1751 blk_complete_request(cmd->request);
1752}
1753
1754/*
1755 * Function: scsi_request_fn()
1756 *
1757 * Purpose: Main strategy routine for SCSI.
1758 *
1759 * Arguments: q - Pointer to actual queue.
1760 *
1761 * Returns: Nothing
1762 *
1763 * Lock status: IO request lock assumed to be held when called.
1764 */
1765static void scsi_request_fn(struct request_queue *q)
1766 __releases(q->queue_lock)
1767 __acquires(q->queue_lock)
1768{
1769 struct scsi_device *sdev = q->queuedata;
1770 struct Scsi_Host *shost;
1771 struct scsi_cmnd *cmd;
1772 struct request *req;
1773
1774 /*
1775 * To start with, we keep looping until the queue is empty, or until
1776 * the host is no longer able to accept any more requests.
1777 */
1778 shost = sdev->host;
1779 for (;;) {
1780 int rtn;
1781 /*
1782 * get next queueable request. We do this early to make sure
1783 * that the request is fully prepared even if we cannot
1784 * accept it.
1785 */
1786 req = blk_peek_request(q);
1787 if (!req)
1788 break;
1789
1790 if (unlikely(!scsi_device_online(sdev))) {
1791 sdev_printk(KERN_ERR, sdev,
1792 "rejecting I/O to offline device\n");
1793 scsi_kill_request(req, q);
1794 continue;
1795 }
1796
1797 if (!scsi_dev_queue_ready(q, sdev))
1798 break;
1799
1800 /*
1801 * Remove the request from the request list.
1802 */
1803 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1804 blk_start_request(req);
1805
1806 spin_unlock_irq(q->queue_lock);
1807 cmd = req->special;
1808 if (unlikely(cmd == NULL)) {
1809 printk(KERN_CRIT "impossible request in %s.\n"
1810 "please mail a stack trace to "
1811 "linux-scsi@vger.kernel.org\n",
1812 __func__);
1813 blk_dump_rq_flags(req, "foo");
1814 BUG();
1815 }
1816
1817 /*
1818 * We hit this when the driver is using a host wide
1819 * tag map. For device level tag maps the queue_depth check
1820 * in the device ready fn would prevent us from trying
1821 * to allocate a tag. Since the map is a shared host resource
1822 * we add the dev to the starved list so it eventually gets
1823 * a run when a tag is freed.
1824 */
1825 if (blk_queue_tagged(q) && !(req->cmd_flags & REQ_QUEUED)) {
1826 spin_lock_irq(shost->host_lock);
1827 if (list_empty(&sdev->starved_entry))
1828 list_add_tail(&sdev->starved_entry,
1829 &shost->starved_list);
1830 spin_unlock_irq(shost->host_lock);
1831 goto not_ready;
1832 }
1833
1834 if (!scsi_target_queue_ready(shost, sdev))
1835 goto not_ready;
1836
1837 if (!scsi_host_queue_ready(q, shost, sdev))
1838 goto host_not_ready;
1839
1840 if (sdev->simple_tags)
1841 cmd->flags |= SCMD_TAGGED;
1842 else
1843 cmd->flags &= ~SCMD_TAGGED;
1844
1845 /*
1846 * Finally, initialize any error handling parameters, and set up
1847 * the timers for timeouts.
1848 */
1849 scsi_init_cmd_errh(cmd);
1850
1851 /*
1852 * Dispatch the command to the low-level driver.
1853 */
1854 cmd->scsi_done = scsi_done;
1855 rtn = scsi_dispatch_cmd(cmd);
1856 if (rtn) {
1857 scsi_queue_insert(cmd, rtn);
1858 spin_lock_irq(q->queue_lock);
1859 goto out_delay;
1860 }
1861 spin_lock_irq(q->queue_lock);
1862 }
1863
1864 return;
1865
1866 host_not_ready:
1867 if (scsi_target(sdev)->can_queue > 0)
1868 atomic_dec(&scsi_target(sdev)->target_busy);
1869 not_ready:
1870 /*
1871 * lock q, handle tag, requeue req, and decrement device_busy. We
1872 * must return with queue_lock held.
1873 *
1874 * Decrementing device_busy without checking it is OK, as all such
1875 * cases (host limits or settings) should run the queue at some
1876 * later time.
1877 */
1878 spin_lock_irq(q->queue_lock);
1879 blk_requeue_request(q, req);
1880 atomic_dec(&sdev->device_busy);
1881out_delay:
1882 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1883 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1884}
1885
1886static inline int prep_to_mq(int ret)
1887{
1888 switch (ret) {
1889 case BLKPREP_OK:
1890 return 0;
1891 case BLKPREP_DEFER:
1892 return BLK_MQ_RQ_QUEUE_BUSY;
1893 default:
1894 return BLK_MQ_RQ_QUEUE_ERROR;
1895 }
1896}
1897
1898static int scsi_mq_prep_fn(struct request *req)
1899{
1900 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1901 struct scsi_device *sdev = req->q->queuedata;
1902 struct Scsi_Host *shost = sdev->host;
1903 unsigned char *sense_buf = cmd->sense_buffer;
1904 struct scatterlist *sg;
1905
1906 memset(cmd, 0, sizeof(struct scsi_cmnd));
1907
1908 req->special = cmd;
1909
1910 cmd->request = req;
1911 cmd->device = sdev;
1912 cmd->sense_buffer = sense_buf;
1913
1914 cmd->tag = req->tag;
1915
1916 cmd->cmnd = req->cmd;
1917 cmd->prot_op = SCSI_PROT_NORMAL;
1918
1919 INIT_LIST_HEAD(&cmd->list);
1920 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1921 cmd->jiffies_at_alloc = jiffies;
1922
1923 if (shost->use_cmd_list) {
1924 spin_lock_irq(&sdev->list_lock);
1925 list_add_tail(&cmd->list, &sdev->cmd_list);
1926 spin_unlock_irq(&sdev->list_lock);
1927 }
1928
1929 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1930 cmd->sdb.table.sgl = sg;
1931
1932 if (scsi_host_get_prot(shost)) {
1933 cmd->prot_sdb = (void *)sg +
1934 min_t(unsigned int,
1935 shost->sg_tablesize, SCSI_MAX_SG_SEGMENTS) *
1936 sizeof(struct scatterlist);
1937 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1938
1939 cmd->prot_sdb->table.sgl =
1940 (struct scatterlist *)(cmd->prot_sdb + 1);
1941 }
1942
1943 if (blk_bidi_rq(req)) {
1944 struct request *next_rq = req->next_rq;
1945 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1946
1947 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1948 bidi_sdb->table.sgl =
1949 (struct scatterlist *)(bidi_sdb + 1);
1950
1951 next_rq->special = bidi_sdb;
1952 }
1953
1954 blk_mq_start_request(req);
1955
1956 return scsi_setup_cmnd(sdev, req);
1957}
1958
1959static void scsi_mq_done(struct scsi_cmnd *cmd)
1960{
1961 trace_scsi_dispatch_cmd_done(cmd);
1962 blk_mq_complete_request(cmd->request, cmd->request->errors);
1963}
1964
1965static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1966 const struct blk_mq_queue_data *bd)
1967{
1968 struct request *req = bd->rq;
1969 struct request_queue *q = req->q;
1970 struct scsi_device *sdev = q->queuedata;
1971 struct Scsi_Host *shost = sdev->host;
1972 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1973 int ret;
1974 int reason;
1975
1976 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1977 if (ret)
1978 goto out;
1979
1980 ret = BLK_MQ_RQ_QUEUE_BUSY;
1981 if (!get_device(&sdev->sdev_gendev))
1982 goto out;
1983
1984 if (!scsi_dev_queue_ready(q, sdev))
1985 goto out_put_device;
1986 if (!scsi_target_queue_ready(shost, sdev))
1987 goto out_dec_device_busy;
1988 if (!scsi_host_queue_ready(q, shost, sdev))
1989 goto out_dec_target_busy;
1990
1991
1992 if (!(req->cmd_flags & REQ_DONTPREP)) {
1993 ret = prep_to_mq(scsi_mq_prep_fn(req));
1994 if (ret)
1995 goto out_dec_host_busy;
1996 req->cmd_flags |= REQ_DONTPREP;
1997 } else {
1998 blk_mq_start_request(req);
1999 }
2000
2001 if (sdev->simple_tags)
2002 cmd->flags |= SCMD_TAGGED;
2003 else
2004 cmd->flags &= ~SCMD_TAGGED;
2005
2006 scsi_init_cmd_errh(cmd);
2007 cmd->scsi_done = scsi_mq_done;
2008
2009 reason = scsi_dispatch_cmd(cmd);
2010 if (reason) {
2011 scsi_set_blocked(cmd, reason);
2012 ret = BLK_MQ_RQ_QUEUE_BUSY;
2013 goto out_dec_host_busy;
2014 }
2015
2016 return BLK_MQ_RQ_QUEUE_OK;
2017
2018out_dec_host_busy:
2019 atomic_dec(&shost->host_busy);
2020out_dec_target_busy:
2021 if (scsi_target(sdev)->can_queue > 0)
2022 atomic_dec(&scsi_target(sdev)->target_busy);
2023out_dec_device_busy:
2024 atomic_dec(&sdev->device_busy);
2025out_put_device:
2026 put_device(&sdev->sdev_gendev);
2027out:
2028 switch (ret) {
2029 case BLK_MQ_RQ_QUEUE_BUSY:
2030 blk_mq_stop_hw_queue(hctx);
2031 if (atomic_read(&sdev->device_busy) == 0 &&
2032 !scsi_device_blocked(sdev))
2033 blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
2034 break;
2035 case BLK_MQ_RQ_QUEUE_ERROR:
2036 /*
2037 * Make sure to release all allocated ressources when
2038 * we hit an error, as we will never see this command
2039 * again.
2040 */
2041 if (req->cmd_flags & REQ_DONTPREP)
2042 scsi_mq_uninit_cmd(cmd);
2043 break;
2044 default:
2045 break;
2046 }
2047 return ret;
2048}
2049
2050static enum blk_eh_timer_return scsi_timeout(struct request *req,
2051 bool reserved)
2052{
2053 if (reserved)
2054 return BLK_EH_RESET_TIMER;
2055 return scsi_times_out(req);
2056}
2057
2058static int scsi_init_request(void *data, struct request *rq,
2059 unsigned int hctx_idx, unsigned int request_idx,
2060 unsigned int numa_node)
2061{
2062 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2063
2064 cmd->sense_buffer = kzalloc_node(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL,
2065 numa_node);
2066 if (!cmd->sense_buffer)
2067 return -ENOMEM;
2068 return 0;
2069}
2070
2071static void scsi_exit_request(void *data, struct request *rq,
2072 unsigned int hctx_idx, unsigned int request_idx)
2073{
2074 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2075
2076 kfree(cmd->sense_buffer);
2077}
2078
2079static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
2080{
2081 struct device *host_dev;
2082 u64 bounce_limit = 0xffffffff;
2083
2084 if (shost->unchecked_isa_dma)
2085 return BLK_BOUNCE_ISA;
2086 /*
2087 * Platforms with virtual-DMA translation
2088 * hardware have no practical limit.
2089 */
2090 if (!PCI_DMA_BUS_IS_PHYS)
2091 return BLK_BOUNCE_ANY;
2092
2093 host_dev = scsi_get_device(shost);
2094 if (host_dev && host_dev->dma_mask)
2095 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2096
2097 return bounce_limit;
2098}
2099
2100static void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2101{
2102 struct device *dev = shost->dma_dev;
2103
2104 /*
2105 * this limit is imposed by hardware restrictions
2106 */
2107 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2108 SCSI_MAX_SG_CHAIN_SEGMENTS));
2109
2110 if (scsi_host_prot_dma(shost)) {
2111 shost->sg_prot_tablesize =
2112 min_not_zero(shost->sg_prot_tablesize,
2113 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2114 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2115 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2116 }
2117
2118 blk_queue_max_hw_sectors(q, shost->max_sectors);
2119 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2120 blk_queue_segment_boundary(q, shost->dma_boundary);
2121 dma_set_seg_boundary(dev, shost->dma_boundary);
2122
2123 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2124
2125 if (!shost->use_clustering)
2126 q->limits.cluster = 0;
2127
2128 /*
2129 * set a reasonable default alignment on word boundaries: the
2130 * host and device may alter it using
2131 * blk_queue_update_dma_alignment() later.
2132 */
2133 blk_queue_dma_alignment(q, 0x03);
2134}
2135
2136struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
2137 request_fn_proc *request_fn)
2138{
2139 struct request_queue *q;
2140
2141 q = blk_init_queue(request_fn, NULL);
2142 if (!q)
2143 return NULL;
2144 __scsi_init_queue(shost, q);
2145 return q;
2146}
2147EXPORT_SYMBOL(__scsi_alloc_queue);
2148
2149struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2150{
2151 struct request_queue *q;
2152
2153 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
2154 if (!q)
2155 return NULL;
2156
2157 blk_queue_prep_rq(q, scsi_prep_fn);
2158 blk_queue_unprep_rq(q, scsi_unprep_fn);
2159 blk_queue_softirq_done(q, scsi_softirq_done);
2160 blk_queue_rq_timed_out(q, scsi_times_out);
2161 blk_queue_lld_busy(q, scsi_lld_busy);
2162 return q;
2163}
2164
2165static struct blk_mq_ops scsi_mq_ops = {
2166 .map_queue = blk_mq_map_queue,
2167 .queue_rq = scsi_queue_rq,
2168 .complete = scsi_softirq_done,
2169 .timeout = scsi_timeout,
2170 .init_request = scsi_init_request,
2171 .exit_request = scsi_exit_request,
2172};
2173
2174struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2175{
2176 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2177 if (IS_ERR(sdev->request_queue))
2178 return NULL;
2179
2180 sdev->request_queue->queuedata = sdev;
2181 __scsi_init_queue(sdev->host, sdev->request_queue);
2182 return sdev->request_queue;
2183}
2184
2185int scsi_mq_setup_tags(struct Scsi_Host *shost)
2186{
2187 unsigned int cmd_size, sgl_size, tbl_size;
2188
2189 tbl_size = shost->sg_tablesize;
2190 if (tbl_size > SCSI_MAX_SG_SEGMENTS)
2191 tbl_size = SCSI_MAX_SG_SEGMENTS;
2192 sgl_size = tbl_size * sizeof(struct scatterlist);
2193 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2194 if (scsi_host_get_prot(shost))
2195 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2196
2197 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2198 shost->tag_set.ops = &scsi_mq_ops;
2199 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2200 shost->tag_set.queue_depth = shost->can_queue;
2201 shost->tag_set.cmd_size = cmd_size;
2202 shost->tag_set.numa_node = NUMA_NO_NODE;
2203 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2204 shost->tag_set.flags |=
2205 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2206 shost->tag_set.driver_data = shost;
2207
2208 return blk_mq_alloc_tag_set(&shost->tag_set);
2209}
2210
2211void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2212{
2213 blk_mq_free_tag_set(&shost->tag_set);
2214}
2215
2216/*
2217 * Function: scsi_block_requests()
2218 *
2219 * Purpose: Utility function used by low-level drivers to prevent further
2220 * commands from being queued to the device.
2221 *
2222 * Arguments: shost - Host in question
2223 *
2224 * Returns: Nothing
2225 *
2226 * Lock status: No locks are assumed held.
2227 *
2228 * Notes: There is no timer nor any other means by which the requests
2229 * get unblocked other than the low-level driver calling
2230 * scsi_unblock_requests().
2231 */
2232void scsi_block_requests(struct Scsi_Host *shost)
2233{
2234 shost->host_self_blocked = 1;
2235}
2236EXPORT_SYMBOL(scsi_block_requests);
2237
2238/*
2239 * Function: scsi_unblock_requests()
2240 *
2241 * Purpose: Utility function used by low-level drivers to allow further
2242 * commands from being queued to the device.
2243 *
2244 * Arguments: shost - Host in question
2245 *
2246 * Returns: Nothing
2247 *
2248 * Lock status: No locks are assumed held.
2249 *
2250 * Notes: There is no timer nor any other means by which the requests
2251 * get unblocked other than the low-level driver calling
2252 * scsi_unblock_requests().
2253 *
2254 * This is done as an API function so that changes to the
2255 * internals of the scsi mid-layer won't require wholesale
2256 * changes to drivers that use this feature.
2257 */
2258void scsi_unblock_requests(struct Scsi_Host *shost)
2259{
2260 shost->host_self_blocked = 0;
2261 scsi_run_host_queues(shost);
2262}
2263EXPORT_SYMBOL(scsi_unblock_requests);
2264
2265int __init scsi_init_queue(void)
2266{
2267 int i;
2268
2269 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2270 sizeof(struct scsi_data_buffer),
2271 0, 0, NULL);
2272 if (!scsi_sdb_cache) {
2273 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2274 return -ENOMEM;
2275 }
2276
2277 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2278 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2279 int size = sgp->size * sizeof(struct scatterlist);
2280
2281 sgp->slab = kmem_cache_create(sgp->name, size, 0,
2282 SLAB_HWCACHE_ALIGN, NULL);
2283 if (!sgp->slab) {
2284 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
2285 sgp->name);
2286 goto cleanup_sdb;
2287 }
2288
2289 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
2290 sgp->slab);
2291 if (!sgp->pool) {
2292 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
2293 sgp->name);
2294 goto cleanup_sdb;
2295 }
2296 }
2297
2298 return 0;
2299
2300cleanup_sdb:
2301 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2302 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2303 if (sgp->pool)
2304 mempool_destroy(sgp->pool);
2305 if (sgp->slab)
2306 kmem_cache_destroy(sgp->slab);
2307 }
2308 kmem_cache_destroy(scsi_sdb_cache);
2309
2310 return -ENOMEM;
2311}
2312
2313void scsi_exit_queue(void)
2314{
2315 int i;
2316
2317 kmem_cache_destroy(scsi_sdb_cache);
2318
2319 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2320 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2321 mempool_destroy(sgp->pool);
2322 kmem_cache_destroy(sgp->slab);
2323 }
2324}
2325
2326/**
2327 * scsi_mode_select - issue a mode select
2328 * @sdev: SCSI device to be queried
2329 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2330 * @sp: Save page bit (0 == don't save, 1 == save)
2331 * @modepage: mode page being requested
2332 * @buffer: request buffer (may not be smaller than eight bytes)
2333 * @len: length of request buffer.
2334 * @timeout: command timeout
2335 * @retries: number of retries before failing
2336 * @data: returns a structure abstracting the mode header data
2337 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2338 * must be SCSI_SENSE_BUFFERSIZE big.
2339 *
2340 * Returns zero if successful; negative error number or scsi
2341 * status on error
2342 *
2343 */
2344int
2345scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2346 unsigned char *buffer, int len, int timeout, int retries,
2347 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2348{
2349 unsigned char cmd[10];
2350 unsigned char *real_buffer;
2351 int ret;
2352
2353 memset(cmd, 0, sizeof(cmd));
2354 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2355
2356 if (sdev->use_10_for_ms) {
2357 if (len > 65535)
2358 return -EINVAL;
2359 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2360 if (!real_buffer)
2361 return -ENOMEM;
2362 memcpy(real_buffer + 8, buffer, len);
2363 len += 8;
2364 real_buffer[0] = 0;
2365 real_buffer[1] = 0;
2366 real_buffer[2] = data->medium_type;
2367 real_buffer[3] = data->device_specific;
2368 real_buffer[4] = data->longlba ? 0x01 : 0;
2369 real_buffer[5] = 0;
2370 real_buffer[6] = data->block_descriptor_length >> 8;
2371 real_buffer[7] = data->block_descriptor_length;
2372
2373 cmd[0] = MODE_SELECT_10;
2374 cmd[7] = len >> 8;
2375 cmd[8] = len;
2376 } else {
2377 if (len > 255 || data->block_descriptor_length > 255 ||
2378 data->longlba)
2379 return -EINVAL;
2380
2381 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2382 if (!real_buffer)
2383 return -ENOMEM;
2384 memcpy(real_buffer + 4, buffer, len);
2385 len += 4;
2386 real_buffer[0] = 0;
2387 real_buffer[1] = data->medium_type;
2388 real_buffer[2] = data->device_specific;
2389 real_buffer[3] = data->block_descriptor_length;
2390
2391
2392 cmd[0] = MODE_SELECT;
2393 cmd[4] = len;
2394 }
2395
2396 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2397 sshdr, timeout, retries, NULL);
2398 kfree(real_buffer);
2399 return ret;
2400}
2401EXPORT_SYMBOL_GPL(scsi_mode_select);
2402
2403/**
2404 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2405 * @sdev: SCSI device to be queried
2406 * @dbd: set if mode sense will allow block descriptors to be returned
2407 * @modepage: mode page being requested
2408 * @buffer: request buffer (may not be smaller than eight bytes)
2409 * @len: length of request buffer.
2410 * @timeout: command timeout
2411 * @retries: number of retries before failing
2412 * @data: returns a structure abstracting the mode header data
2413 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2414 * must be SCSI_SENSE_BUFFERSIZE big.
2415 *
2416 * Returns zero if unsuccessful, or the header offset (either 4
2417 * or 8 depending on whether a six or ten byte command was
2418 * issued) if successful.
2419 */
2420int
2421scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2422 unsigned char *buffer, int len, int timeout, int retries,
2423 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2424{
2425 unsigned char cmd[12];
2426 int use_10_for_ms;
2427 int header_length;
2428 int result, retry_count = retries;
2429 struct scsi_sense_hdr my_sshdr;
2430
2431 memset(data, 0, sizeof(*data));
2432 memset(&cmd[0], 0, 12);
2433 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2434 cmd[2] = modepage;
2435
2436 /* caller might not be interested in sense, but we need it */
2437 if (!sshdr)
2438 sshdr = &my_sshdr;
2439
2440 retry:
2441 use_10_for_ms = sdev->use_10_for_ms;
2442
2443 if (use_10_for_ms) {
2444 if (len < 8)
2445 len = 8;
2446
2447 cmd[0] = MODE_SENSE_10;
2448 cmd[8] = len;
2449 header_length = 8;
2450 } else {
2451 if (len < 4)
2452 len = 4;
2453
2454 cmd[0] = MODE_SENSE;
2455 cmd[4] = len;
2456 header_length = 4;
2457 }
2458
2459 memset(buffer, 0, len);
2460
2461 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2462 sshdr, timeout, retries, NULL);
2463
2464 /* This code looks awful: what it's doing is making sure an
2465 * ILLEGAL REQUEST sense return identifies the actual command
2466 * byte as the problem. MODE_SENSE commands can return
2467 * ILLEGAL REQUEST if the code page isn't supported */
2468
2469 if (use_10_for_ms && !scsi_status_is_good(result) &&
2470 (driver_byte(result) & DRIVER_SENSE)) {
2471 if (scsi_sense_valid(sshdr)) {
2472 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2473 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2474 /*
2475 * Invalid command operation code
2476 */
2477 sdev->use_10_for_ms = 0;
2478 goto retry;
2479 }
2480 }
2481 }
2482
2483 if(scsi_status_is_good(result)) {
2484 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2485 (modepage == 6 || modepage == 8))) {
2486 /* Initio breakage? */
2487 header_length = 0;
2488 data->length = 13;
2489 data->medium_type = 0;
2490 data->device_specific = 0;
2491 data->longlba = 0;
2492 data->block_descriptor_length = 0;
2493 } else if(use_10_for_ms) {
2494 data->length = buffer[0]*256 + buffer[1] + 2;
2495 data->medium_type = buffer[2];
2496 data->device_specific = buffer[3];
2497 data->longlba = buffer[4] & 0x01;
2498 data->block_descriptor_length = buffer[6]*256
2499 + buffer[7];
2500 } else {
2501 data->length = buffer[0] + 1;
2502 data->medium_type = buffer[1];
2503 data->device_specific = buffer[2];
2504 data->block_descriptor_length = buffer[3];
2505 }
2506 data->header_length = header_length;
2507 } else if ((status_byte(result) == CHECK_CONDITION) &&
2508 scsi_sense_valid(sshdr) &&
2509 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2510 retry_count--;
2511 goto retry;
2512 }
2513
2514 return result;
2515}
2516EXPORT_SYMBOL(scsi_mode_sense);
2517
2518/**
2519 * scsi_test_unit_ready - test if unit is ready
2520 * @sdev: scsi device to change the state of.
2521 * @timeout: command timeout
2522 * @retries: number of retries before failing
2523 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2524 * returning sense. Make sure that this is cleared before passing
2525 * in.
2526 *
2527 * Returns zero if unsuccessful or an error if TUR failed. For
2528 * removable media, UNIT_ATTENTION sets ->changed flag.
2529 **/
2530int
2531scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2532 struct scsi_sense_hdr *sshdr_external)
2533{
2534 char cmd[] = {
2535 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2536 };
2537 struct scsi_sense_hdr *sshdr;
2538 int result;
2539
2540 if (!sshdr_external)
2541 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2542 else
2543 sshdr = sshdr_external;
2544
2545 /* try to eat the UNIT_ATTENTION if there are enough retries */
2546 do {
2547 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2548 timeout, retries, NULL);
2549 if (sdev->removable && scsi_sense_valid(sshdr) &&
2550 sshdr->sense_key == UNIT_ATTENTION)
2551 sdev->changed = 1;
2552 } while (scsi_sense_valid(sshdr) &&
2553 sshdr->sense_key == UNIT_ATTENTION && --retries);
2554
2555 if (!sshdr_external)
2556 kfree(sshdr);
2557 return result;
2558}
2559EXPORT_SYMBOL(scsi_test_unit_ready);
2560
2561/**
2562 * scsi_device_set_state - Take the given device through the device state model.
2563 * @sdev: scsi device to change the state of.
2564 * @state: state to change to.
2565 *
2566 * Returns zero if unsuccessful or an error if the requested
2567 * transition is illegal.
2568 */
2569int
2570scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2571{
2572 enum scsi_device_state oldstate = sdev->sdev_state;
2573
2574 if (state == oldstate)
2575 return 0;
2576
2577 switch (state) {
2578 case SDEV_CREATED:
2579 switch (oldstate) {
2580 case SDEV_CREATED_BLOCK:
2581 break;
2582 default:
2583 goto illegal;
2584 }
2585 break;
2586
2587 case SDEV_RUNNING:
2588 switch (oldstate) {
2589 case SDEV_CREATED:
2590 case SDEV_OFFLINE:
2591 case SDEV_TRANSPORT_OFFLINE:
2592 case SDEV_QUIESCE:
2593 case SDEV_BLOCK:
2594 break;
2595 default:
2596 goto illegal;
2597 }
2598 break;
2599
2600 case SDEV_QUIESCE:
2601 switch (oldstate) {
2602 case SDEV_RUNNING:
2603 case SDEV_OFFLINE:
2604 case SDEV_TRANSPORT_OFFLINE:
2605 break;
2606 default:
2607 goto illegal;
2608 }
2609 break;
2610
2611 case SDEV_OFFLINE:
2612 case SDEV_TRANSPORT_OFFLINE:
2613 switch (oldstate) {
2614 case SDEV_CREATED:
2615 case SDEV_RUNNING:
2616 case SDEV_QUIESCE:
2617 case SDEV_BLOCK:
2618 break;
2619 default:
2620 goto illegal;
2621 }
2622 break;
2623
2624 case SDEV_BLOCK:
2625 switch (oldstate) {
2626 case SDEV_RUNNING:
2627 case SDEV_CREATED_BLOCK:
2628 break;
2629 default:
2630 goto illegal;
2631 }
2632 break;
2633
2634 case SDEV_CREATED_BLOCK:
2635 switch (oldstate) {
2636 case SDEV_CREATED:
2637 break;
2638 default:
2639 goto illegal;
2640 }
2641 break;
2642
2643 case SDEV_CANCEL:
2644 switch (oldstate) {
2645 case SDEV_CREATED:
2646 case SDEV_RUNNING:
2647 case SDEV_QUIESCE:
2648 case SDEV_OFFLINE:
2649 case SDEV_TRANSPORT_OFFLINE:
2650 case SDEV_BLOCK:
2651 break;
2652 default:
2653 goto illegal;
2654 }
2655 break;
2656
2657 case SDEV_DEL:
2658 switch (oldstate) {
2659 case SDEV_CREATED:
2660 case SDEV_RUNNING:
2661 case SDEV_OFFLINE:
2662 case SDEV_TRANSPORT_OFFLINE:
2663 case SDEV_CANCEL:
2664 case SDEV_CREATED_BLOCK:
2665 break;
2666 default:
2667 goto illegal;
2668 }
2669 break;
2670
2671 }
2672 sdev->sdev_state = state;
2673 return 0;
2674
2675 illegal:
2676 SCSI_LOG_ERROR_RECOVERY(1,
2677 sdev_printk(KERN_ERR, sdev,
2678 "Illegal state transition %s->%s",
2679 scsi_device_state_name(oldstate),
2680 scsi_device_state_name(state))
2681 );
2682 return -EINVAL;
2683}
2684EXPORT_SYMBOL(scsi_device_set_state);
2685
2686/**
2687 * sdev_evt_emit - emit a single SCSI device uevent
2688 * @sdev: associated SCSI device
2689 * @evt: event to emit
2690 *
2691 * Send a single uevent (scsi_event) to the associated scsi_device.
2692 */
2693static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2694{
2695 int idx = 0;
2696 char *envp[3];
2697
2698 switch (evt->evt_type) {
2699 case SDEV_EVT_MEDIA_CHANGE:
2700 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2701 break;
2702 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2703 scsi_rescan_device(&sdev->sdev_gendev);
2704 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2705 break;
2706 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2707 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2708 break;
2709 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2710 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2711 break;
2712 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2713 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2714 break;
2715 case SDEV_EVT_LUN_CHANGE_REPORTED:
2716 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2717 break;
2718 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2719 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2720 break;
2721 default:
2722 /* do nothing */
2723 break;
2724 }
2725
2726 envp[idx++] = NULL;
2727
2728 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2729}
2730
2731/**
2732 * sdev_evt_thread - send a uevent for each scsi event
2733 * @work: work struct for scsi_device
2734 *
2735 * Dispatch queued events to their associated scsi_device kobjects
2736 * as uevents.
2737 */
2738void scsi_evt_thread(struct work_struct *work)
2739{
2740 struct scsi_device *sdev;
2741 enum scsi_device_event evt_type;
2742 LIST_HEAD(event_list);
2743
2744 sdev = container_of(work, struct scsi_device, event_work);
2745
2746 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2747 if (test_and_clear_bit(evt_type, sdev->pending_events))
2748 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2749
2750 while (1) {
2751 struct scsi_event *evt;
2752 struct list_head *this, *tmp;
2753 unsigned long flags;
2754
2755 spin_lock_irqsave(&sdev->list_lock, flags);
2756 list_splice_init(&sdev->event_list, &event_list);
2757 spin_unlock_irqrestore(&sdev->list_lock, flags);
2758
2759 if (list_empty(&event_list))
2760 break;
2761
2762 list_for_each_safe(this, tmp, &event_list) {
2763 evt = list_entry(this, struct scsi_event, node);
2764 list_del(&evt->node);
2765 scsi_evt_emit(sdev, evt);
2766 kfree(evt);
2767 }
2768 }
2769}
2770
2771/**
2772 * sdev_evt_send - send asserted event to uevent thread
2773 * @sdev: scsi_device event occurred on
2774 * @evt: event to send
2775 *
2776 * Assert scsi device event asynchronously.
2777 */
2778void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2779{
2780 unsigned long flags;
2781
2782#if 0
2783 /* FIXME: currently this check eliminates all media change events
2784 * for polled devices. Need to update to discriminate between AN
2785 * and polled events */
2786 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2787 kfree(evt);
2788 return;
2789 }
2790#endif
2791
2792 spin_lock_irqsave(&sdev->list_lock, flags);
2793 list_add_tail(&evt->node, &sdev->event_list);
2794 schedule_work(&sdev->event_work);
2795 spin_unlock_irqrestore(&sdev->list_lock, flags);
2796}
2797EXPORT_SYMBOL_GPL(sdev_evt_send);
2798
2799/**
2800 * sdev_evt_alloc - allocate a new scsi event
2801 * @evt_type: type of event to allocate
2802 * @gfpflags: GFP flags for allocation
2803 *
2804 * Allocates and returns a new scsi_event.
2805 */
2806struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2807 gfp_t gfpflags)
2808{
2809 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2810 if (!evt)
2811 return NULL;
2812
2813 evt->evt_type = evt_type;
2814 INIT_LIST_HEAD(&evt->node);
2815
2816 /* evt_type-specific initialization, if any */
2817 switch (evt_type) {
2818 case SDEV_EVT_MEDIA_CHANGE:
2819 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2820 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2821 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2822 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2823 case SDEV_EVT_LUN_CHANGE_REPORTED:
2824 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2825 default:
2826 /* do nothing */
2827 break;
2828 }
2829
2830 return evt;
2831}
2832EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2833
2834/**
2835 * sdev_evt_send_simple - send asserted event to uevent thread
2836 * @sdev: scsi_device event occurred on
2837 * @evt_type: type of event to send
2838 * @gfpflags: GFP flags for allocation
2839 *
2840 * Assert scsi device event asynchronously, given an event type.
2841 */
2842void sdev_evt_send_simple(struct scsi_device *sdev,
2843 enum scsi_device_event evt_type, gfp_t gfpflags)
2844{
2845 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2846 if (!evt) {
2847 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2848 evt_type);
2849 return;
2850 }
2851
2852 sdev_evt_send(sdev, evt);
2853}
2854EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2855
2856/**
2857 * scsi_device_quiesce - Block user issued commands.
2858 * @sdev: scsi device to quiesce.
2859 *
2860 * This works by trying to transition to the SDEV_QUIESCE state
2861 * (which must be a legal transition). When the device is in this
2862 * state, only special requests will be accepted, all others will
2863 * be deferred. Since special requests may also be requeued requests,
2864 * a successful return doesn't guarantee the device will be
2865 * totally quiescent.
2866 *
2867 * Must be called with user context, may sleep.
2868 *
2869 * Returns zero if unsuccessful or an error if not.
2870 */
2871int
2872scsi_device_quiesce(struct scsi_device *sdev)
2873{
2874 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2875 if (err)
2876 return err;
2877
2878 scsi_run_queue(sdev->request_queue);
2879 while (atomic_read(&sdev->device_busy)) {
2880 msleep_interruptible(200);
2881 scsi_run_queue(sdev->request_queue);
2882 }
2883 return 0;
2884}
2885EXPORT_SYMBOL(scsi_device_quiesce);
2886
2887/**
2888 * scsi_device_resume - Restart user issued commands to a quiesced device.
2889 * @sdev: scsi device to resume.
2890 *
2891 * Moves the device from quiesced back to running and restarts the
2892 * queues.
2893 *
2894 * Must be called with user context, may sleep.
2895 */
2896void scsi_device_resume(struct scsi_device *sdev)
2897{
2898 /* check if the device state was mutated prior to resume, and if
2899 * so assume the state is being managed elsewhere (for example
2900 * device deleted during suspend)
2901 */
2902 if (sdev->sdev_state != SDEV_QUIESCE ||
2903 scsi_device_set_state(sdev, SDEV_RUNNING))
2904 return;
2905 scsi_run_queue(sdev->request_queue);
2906}
2907EXPORT_SYMBOL(scsi_device_resume);
2908
2909static void
2910device_quiesce_fn(struct scsi_device *sdev, void *data)
2911{
2912 scsi_device_quiesce(sdev);
2913}
2914
2915void
2916scsi_target_quiesce(struct scsi_target *starget)
2917{
2918 starget_for_each_device(starget, NULL, device_quiesce_fn);
2919}
2920EXPORT_SYMBOL(scsi_target_quiesce);
2921
2922static void
2923device_resume_fn(struct scsi_device *sdev, void *data)
2924{
2925 scsi_device_resume(sdev);
2926}
2927
2928void
2929scsi_target_resume(struct scsi_target *starget)
2930{
2931 starget_for_each_device(starget, NULL, device_resume_fn);
2932}
2933EXPORT_SYMBOL(scsi_target_resume);
2934
2935/**
2936 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2937 * @sdev: device to block
2938 *
2939 * Block request made by scsi lld's to temporarily stop all
2940 * scsi commands on the specified device. Called from interrupt
2941 * or normal process context.
2942 *
2943 * Returns zero if successful or error if not
2944 *
2945 * Notes:
2946 * This routine transitions the device to the SDEV_BLOCK state
2947 * (which must be a legal transition). When the device is in this
2948 * state, all commands are deferred until the scsi lld reenables
2949 * the device with scsi_device_unblock or device_block_tmo fires.
2950 */
2951int
2952scsi_internal_device_block(struct scsi_device *sdev)
2953{
2954 struct request_queue *q = sdev->request_queue;
2955 unsigned long flags;
2956 int err = 0;
2957
2958 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2959 if (err) {
2960 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2961
2962 if (err)
2963 return err;
2964 }
2965
2966 /*
2967 * The device has transitioned to SDEV_BLOCK. Stop the
2968 * block layer from calling the midlayer with this device's
2969 * request queue.
2970 */
2971 if (q->mq_ops) {
2972 blk_mq_stop_hw_queues(q);
2973 } else {
2974 spin_lock_irqsave(q->queue_lock, flags);
2975 blk_stop_queue(q);
2976 spin_unlock_irqrestore(q->queue_lock, flags);
2977 }
2978
2979 return 0;
2980}
2981EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2982
2983/**
2984 * scsi_internal_device_unblock - resume a device after a block request
2985 * @sdev: device to resume
2986 * @new_state: state to set devices to after unblocking
2987 *
2988 * Called by scsi lld's or the midlayer to restart the device queue
2989 * for the previously suspended scsi device. Called from interrupt or
2990 * normal process context.
2991 *
2992 * Returns zero if successful or error if not.
2993 *
2994 * Notes:
2995 * This routine transitions the device to the SDEV_RUNNING state
2996 * or to one of the offline states (which must be a legal transition)
2997 * allowing the midlayer to goose the queue for this device.
2998 */
2999int
3000scsi_internal_device_unblock(struct scsi_device *sdev,
3001 enum scsi_device_state new_state)
3002{
3003 struct request_queue *q = sdev->request_queue;
3004 unsigned long flags;
3005
3006 /*
3007 * Try to transition the scsi device to SDEV_RUNNING or one of the
3008 * offlined states and goose the device queue if successful.
3009 */
3010 if ((sdev->sdev_state == SDEV_BLOCK) ||
3011 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
3012 sdev->sdev_state = new_state;
3013 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
3014 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3015 new_state == SDEV_OFFLINE)
3016 sdev->sdev_state = new_state;
3017 else
3018 sdev->sdev_state = SDEV_CREATED;
3019 } else if (sdev->sdev_state != SDEV_CANCEL &&
3020 sdev->sdev_state != SDEV_OFFLINE)
3021 return -EINVAL;
3022
3023 if (q->mq_ops) {
3024 blk_mq_start_stopped_hw_queues(q, false);
3025 } else {
3026 spin_lock_irqsave(q->queue_lock, flags);
3027 blk_start_queue(q);
3028 spin_unlock_irqrestore(q->queue_lock, flags);
3029 }
3030
3031 return 0;
3032}
3033EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
3034
3035static void
3036device_block(struct scsi_device *sdev, void *data)
3037{
3038 scsi_internal_device_block(sdev);
3039}
3040
3041static int
3042target_block(struct device *dev, void *data)
3043{
3044 if (scsi_is_target_device(dev))
3045 starget_for_each_device(to_scsi_target(dev), NULL,
3046 device_block);
3047 return 0;
3048}
3049
3050void
3051scsi_target_block(struct device *dev)
3052{
3053 if (scsi_is_target_device(dev))
3054 starget_for_each_device(to_scsi_target(dev), NULL,
3055 device_block);
3056 else
3057 device_for_each_child(dev, NULL, target_block);
3058}
3059EXPORT_SYMBOL_GPL(scsi_target_block);
3060
3061static void
3062device_unblock(struct scsi_device *sdev, void *data)
3063{
3064 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3065}
3066
3067static int
3068target_unblock(struct device *dev, void *data)
3069{
3070 if (scsi_is_target_device(dev))
3071 starget_for_each_device(to_scsi_target(dev), data,
3072 device_unblock);
3073 return 0;
3074}
3075
3076void
3077scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3078{
3079 if (scsi_is_target_device(dev))
3080 starget_for_each_device(to_scsi_target(dev), &new_state,
3081 device_unblock);
3082 else
3083 device_for_each_child(dev, &new_state, target_unblock);
3084}
3085EXPORT_SYMBOL_GPL(scsi_target_unblock);
3086
3087/**
3088 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3089 * @sgl: scatter-gather list
3090 * @sg_count: number of segments in sg
3091 * @offset: offset in bytes into sg, on return offset into the mapped area
3092 * @len: bytes to map, on return number of bytes mapped
3093 *
3094 * Returns virtual address of the start of the mapped page
3095 */
3096void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3097 size_t *offset, size_t *len)
3098{
3099 int i;
3100 size_t sg_len = 0, len_complete = 0;
3101 struct scatterlist *sg;
3102 struct page *page;
3103
3104 WARN_ON(!irqs_disabled());
3105
3106 for_each_sg(sgl, sg, sg_count, i) {
3107 len_complete = sg_len; /* Complete sg-entries */
3108 sg_len += sg->length;
3109 if (sg_len > *offset)
3110 break;
3111 }
3112
3113 if (unlikely(i == sg_count)) {
3114 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3115 "elements %d\n",
3116 __func__, sg_len, *offset, sg_count);
3117 WARN_ON(1);
3118 return NULL;
3119 }
3120
3121 /* Offset starting from the beginning of first page in this sg-entry */
3122 *offset = *offset - len_complete + sg->offset;
3123
3124 /* Assumption: contiguous pages can be accessed as "page + i" */
3125 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3126 *offset &= ~PAGE_MASK;
3127
3128 /* Bytes in this sg-entry from *offset to the end of the page */
3129 sg_len = PAGE_SIZE - *offset;
3130 if (*len > sg_len)
3131 *len = sg_len;
3132
3133 return kmap_atomic(page);
3134}
3135EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3136
3137/**
3138 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3139 * @virt: virtual address to be unmapped
3140 */
3141void scsi_kunmap_atomic_sg(void *virt)
3142{
3143 kunmap_atomic(virt);
3144}
3145EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3146
3147void sdev_disable_disk_events(struct scsi_device *sdev)
3148{
3149 atomic_inc(&sdev->disk_events_disable_depth);
3150}
3151EXPORT_SYMBOL(sdev_disable_disk_events);
3152
3153void sdev_enable_disk_events(struct scsi_device *sdev)
3154{
3155 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3156 return;
3157 atomic_dec(&sdev->disk_events_disable_depth);
3158}
3159EXPORT_SYMBOL(sdev_enable_disk_events);
3160
3161/**
3162 * scsi_vpd_lun_id - return a unique device identification
3163 * @sdev: SCSI device
3164 * @id: buffer for the identification
3165 * @id_len: length of the buffer
3166 *
3167 * Copies a unique device identification into @id based
3168 * on the information in the VPD page 0x83 of the device.
3169 * The string will be formatted as a SCSI name string.
3170 *
3171 * Returns the length of the identification or error on failure.
3172 * If the identifier is longer than the supplied buffer the actual
3173 * identifier length is returned and the buffer is not zero-padded.
3174 */
3175int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3176{
3177 u8 cur_id_type = 0xff;
3178 u8 cur_id_size = 0;
3179 unsigned char *d, *cur_id_str;
3180 unsigned char __rcu *vpd_pg83;
3181 int id_size = -EINVAL;
3182
3183 rcu_read_lock();
3184 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3185 if (!vpd_pg83) {
3186 rcu_read_unlock();
3187 return -ENXIO;
3188 }
3189
3190 /*
3191 * Look for the correct descriptor.
3192 * Order of preference for lun descriptor:
3193 * - SCSI name string
3194 * - NAA IEEE Registered Extended
3195 * - EUI-64 based 16-byte
3196 * - EUI-64 based 12-byte
3197 * - NAA IEEE Registered
3198 * - NAA IEEE Extended
3199 * as longer descriptors reduce the likelyhood
3200 * of identification clashes.
3201 */
3202
3203 /* The id string must be at least 20 bytes + terminating NULL byte */
3204 if (id_len < 21) {
3205 rcu_read_unlock();
3206 return -EINVAL;
3207 }
3208
3209 memset(id, 0, id_len);
3210 d = vpd_pg83 + 4;
3211 while (d < vpd_pg83 + sdev->vpd_pg83_len) {
3212 /* Skip designators not referring to the LUN */
3213 if ((d[1] & 0x30) != 0x00)
3214 goto next_desig;
3215
3216 switch (d[1] & 0xf) {
3217 case 0x2:
3218 /* EUI-64 */
3219 if (cur_id_size > d[3])
3220 break;
3221 /* Prefer NAA IEEE Registered Extended */
3222 if (cur_id_type == 0x3 &&
3223 cur_id_size == d[3])
3224 break;
3225 cur_id_size = d[3];
3226 cur_id_str = d + 4;
3227 cur_id_type = d[1] & 0xf;
3228 switch (cur_id_size) {
3229 case 8:
3230 id_size = snprintf(id, id_len,
3231 "eui.%8phN",
3232 cur_id_str);
3233 break;
3234 case 12:
3235 id_size = snprintf(id, id_len,
3236 "eui.%12phN",
3237 cur_id_str);
3238 break;
3239 case 16:
3240 id_size = snprintf(id, id_len,
3241 "eui.%16phN",
3242 cur_id_str);
3243 break;
3244 default:
3245 cur_id_size = 0;
3246 break;
3247 }
3248 break;
3249 case 0x3:
3250 /* NAA */
3251 if (cur_id_size > d[3])
3252 break;
3253 cur_id_size = d[3];
3254 cur_id_str = d + 4;
3255 cur_id_type = d[1] & 0xf;
3256 switch (cur_id_size) {
3257 case 8:
3258 id_size = snprintf(id, id_len,
3259 "naa.%8phN",
3260 cur_id_str);
3261 break;
3262 case 16:
3263 id_size = snprintf(id, id_len,
3264 "naa.%16phN",
3265 cur_id_str);
3266 break;
3267 default:
3268 cur_id_size = 0;
3269 break;
3270 }
3271 break;
3272 case 0x8:
3273 /* SCSI name string */
3274 if (cur_id_size + 4 > d[3])
3275 break;
3276 /* Prefer others for truncated descriptor */
3277 if (cur_id_size && d[3] > id_len)
3278 break;
3279 cur_id_size = id_size = d[3];
3280 cur_id_str = d + 4;
3281 cur_id_type = d[1] & 0xf;
3282 if (cur_id_size >= id_len)
3283 cur_id_size = id_len - 1;
3284 memcpy(id, cur_id_str, cur_id_size);
3285 /* Decrease priority for truncated descriptor */
3286 if (cur_id_size != id_size)
3287 cur_id_size = 6;
3288 break;
3289 default:
3290 break;
3291 }
3292next_desig:
3293 d += d[3] + 4;
3294 }
3295 rcu_read_unlock();
3296
3297 return id_size;
3298}
3299EXPORT_SYMBOL(scsi_vpd_lun_id);
3300
3301/*
3302 * scsi_vpd_tpg_id - return a target port group identifier
3303 * @sdev: SCSI device
3304 *
3305 * Returns the Target Port Group identifier from the information
3306 * froom VPD page 0x83 of the device.
3307 *
3308 * Returns the identifier or error on failure.
3309 */
3310int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3311{
3312 unsigned char *d;
3313 unsigned char __rcu *vpd_pg83;
3314 int group_id = -EAGAIN, rel_port = -1;
3315
3316 rcu_read_lock();
3317 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3318 if (!vpd_pg83) {
3319 rcu_read_unlock();
3320 return -ENXIO;
3321 }
3322
3323 d = sdev->vpd_pg83 + 4;
3324 while (d < sdev->vpd_pg83 + sdev->vpd_pg83_len) {
3325 switch (d[1] & 0xf) {
3326 case 0x4:
3327 /* Relative target port */
3328 rel_port = get_unaligned_be16(&d[6]);
3329 break;
3330 case 0x5:
3331 /* Target port group */
3332 group_id = get_unaligned_be16(&d[6]);
3333 break;
3334 default:
3335 break;
3336 }
3337 d += d[3] + 4;
3338 }
3339 rcu_read_unlock();
3340
3341 if (group_id >= 0 && rel_id && rel_port != -1)
3342 *rel_id = rel_port;
3343
3344 return group_id;
3345}
3346EXPORT_SYMBOL(scsi_vpd_tpg_id);