arm64: dts: meson: libretech-cc: add internal DAC support
[platform/kernel/linux-starfive.git] / drivers / scsi / scsi_lib.c
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/ratelimit.h>
25 #include <asm/unaligned.h>
26
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_cmnd.h>
29 #include <scsi/scsi_dbg.h>
30 #include <scsi/scsi_device.h>
31 #include <scsi/scsi_driver.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_host.h>
34 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
35 #include <scsi/scsi_dh.h>
36
37 #include <trace/events/scsi.h>
38
39 #include "scsi_debugfs.h"
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
42
43 /*
44  * Size of integrity metadata is usually small, 1 inline sg should
45  * cover normal cases.
46  */
47 #ifdef CONFIG_ARCH_NO_SG_CHAIN
48 #define  SCSI_INLINE_PROT_SG_CNT  0
49 #define  SCSI_INLINE_SG_CNT  0
50 #else
51 #define  SCSI_INLINE_PROT_SG_CNT  1
52 #define  SCSI_INLINE_SG_CNT  2
53 #endif
54
55 static struct kmem_cache *scsi_sdb_cache;
56 static struct kmem_cache *scsi_sense_cache;
57 static struct kmem_cache *scsi_sense_isadma_cache;
58 static DEFINE_MUTEX(scsi_sense_cache_mutex);
59
60 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
61
62 static inline struct kmem_cache *
63 scsi_select_sense_cache(bool unchecked_isa_dma)
64 {
65         return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
66 }
67
68 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
69                                    unsigned char *sense_buffer)
70 {
71         kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
72                         sense_buffer);
73 }
74
75 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
76         gfp_t gfp_mask, int numa_node)
77 {
78         return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
79                                      gfp_mask, numa_node);
80 }
81
82 int scsi_init_sense_cache(struct Scsi_Host *shost)
83 {
84         struct kmem_cache *cache;
85         int ret = 0;
86
87         mutex_lock(&scsi_sense_cache_mutex);
88         cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
89         if (cache)
90                 goto exit;
91
92         if (shost->unchecked_isa_dma) {
93                 scsi_sense_isadma_cache =
94                         kmem_cache_create("scsi_sense_cache(DMA)",
95                                 SCSI_SENSE_BUFFERSIZE, 0,
96                                 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
97                 if (!scsi_sense_isadma_cache)
98                         ret = -ENOMEM;
99         } else {
100                 scsi_sense_cache =
101                         kmem_cache_create_usercopy("scsi_sense_cache",
102                                 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
103                                 0, SCSI_SENSE_BUFFERSIZE, NULL);
104                 if (!scsi_sense_cache)
105                         ret = -ENOMEM;
106         }
107  exit:
108         mutex_unlock(&scsi_sense_cache_mutex);
109         return ret;
110 }
111
112 /*
113  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
114  * not change behaviour from the previous unplug mechanism, experimentation
115  * may prove this needs changing.
116  */
117 #define SCSI_QUEUE_DELAY        3
118
119 static void
120 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
121 {
122         struct Scsi_Host *host = cmd->device->host;
123         struct scsi_device *device = cmd->device;
124         struct scsi_target *starget = scsi_target(device);
125
126         /*
127          * Set the appropriate busy bit for the device/host.
128          *
129          * If the host/device isn't busy, assume that something actually
130          * completed, and that we should be able to queue a command now.
131          *
132          * Note that the prior mid-layer assumption that any host could
133          * always queue at least one command is now broken.  The mid-layer
134          * will implement a user specifiable stall (see
135          * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
136          * if a command is requeued with no other commands outstanding
137          * either for the device or for the host.
138          */
139         switch (reason) {
140         case SCSI_MLQUEUE_HOST_BUSY:
141                 atomic_set(&host->host_blocked, host->max_host_blocked);
142                 break;
143         case SCSI_MLQUEUE_DEVICE_BUSY:
144         case SCSI_MLQUEUE_EH_RETRY:
145                 atomic_set(&device->device_blocked,
146                            device->max_device_blocked);
147                 break;
148         case SCSI_MLQUEUE_TARGET_BUSY:
149                 atomic_set(&starget->target_blocked,
150                            starget->max_target_blocked);
151                 break;
152         }
153 }
154
155 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
156 {
157         if (cmd->request->rq_flags & RQF_DONTPREP) {
158                 cmd->request->rq_flags &= ~RQF_DONTPREP;
159                 scsi_mq_uninit_cmd(cmd);
160         } else {
161                 WARN_ON_ONCE(true);
162         }
163         blk_mq_requeue_request(cmd->request, true);
164 }
165
166 /**
167  * __scsi_queue_insert - private queue insertion
168  * @cmd: The SCSI command being requeued
169  * @reason:  The reason for the requeue
170  * @unbusy: Whether the queue should be unbusied
171  *
172  * This is a private queue insertion.  The public interface
173  * scsi_queue_insert() always assumes the queue should be unbusied
174  * because it's always called before the completion.  This function is
175  * for a requeue after completion, which should only occur in this
176  * file.
177  */
178 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
179 {
180         struct scsi_device *device = cmd->device;
181
182         SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
183                 "Inserting command %p into mlqueue\n", cmd));
184
185         scsi_set_blocked(cmd, reason);
186
187         /*
188          * Decrement the counters, since these commands are no longer
189          * active on the host/device.
190          */
191         if (unbusy)
192                 scsi_device_unbusy(device, cmd);
193
194         /*
195          * Requeue this command.  It will go before all other commands
196          * that are already in the queue. Schedule requeue work under
197          * lock such that the kblockd_schedule_work() call happens
198          * before blk_cleanup_queue() finishes.
199          */
200         cmd->result = 0;
201
202         blk_mq_requeue_request(cmd->request, true);
203 }
204
205 /*
206  * Function:    scsi_queue_insert()
207  *
208  * Purpose:     Insert a command in the midlevel queue.
209  *
210  * Arguments:   cmd    - command that we are adding to queue.
211  *              reason - why we are inserting command to queue.
212  *
213  * Lock status: Assumed that lock is not held upon entry.
214  *
215  * Returns:     Nothing.
216  *
217  * Notes:       We do this for one of two cases.  Either the host is busy
218  *              and it cannot accept any more commands for the time being,
219  *              or the device returned QUEUE_FULL and can accept no more
220  *              commands.
221  * Notes:       This could be called either from an interrupt context or a
222  *              normal process context.
223  */
224 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
225 {
226         __scsi_queue_insert(cmd, reason, true);
227 }
228
229
230 /**
231  * __scsi_execute - insert request and wait for the result
232  * @sdev:       scsi device
233  * @cmd:        scsi command
234  * @data_direction: data direction
235  * @buffer:     data buffer
236  * @bufflen:    len of buffer
237  * @sense:      optional sense buffer
238  * @sshdr:      optional decoded sense header
239  * @timeout:    request timeout in seconds
240  * @retries:    number of times to retry request
241  * @flags:      flags for ->cmd_flags
242  * @rq_flags:   flags for ->rq_flags
243  * @resid:      optional residual length
244  *
245  * Returns the scsi_cmnd result field if a command was executed, or a negative
246  * Linux error code if we didn't get that far.
247  */
248 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
249                  int data_direction, void *buffer, unsigned bufflen,
250                  unsigned char *sense, struct scsi_sense_hdr *sshdr,
251                  int timeout, int retries, u64 flags, req_flags_t rq_flags,
252                  int *resid)
253 {
254         struct request *req;
255         struct scsi_request *rq;
256         int ret = DRIVER_ERROR << 24;
257
258         req = blk_get_request(sdev->request_queue,
259                         data_direction == DMA_TO_DEVICE ?
260                         REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
261         if (IS_ERR(req))
262                 return ret;
263         rq = scsi_req(req);
264
265         if (bufflen &&  blk_rq_map_kern(sdev->request_queue, req,
266                                         buffer, bufflen, GFP_NOIO))
267                 goto out;
268
269         rq->cmd_len = COMMAND_SIZE(cmd[0]);
270         memcpy(rq->cmd, cmd, rq->cmd_len);
271         rq->retries = retries;
272         req->timeout = timeout;
273         req->cmd_flags |= flags;
274         req->rq_flags |= rq_flags | RQF_QUIET;
275
276         /*
277          * head injection *required* here otherwise quiesce won't work
278          */
279         blk_execute_rq(req->q, NULL, req, 1);
280
281         /*
282          * Some devices (USB mass-storage in particular) may transfer
283          * garbage data together with a residue indicating that the data
284          * is invalid.  Prevent the garbage from being misinterpreted
285          * and prevent security leaks by zeroing out the excess data.
286          */
287         if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
288                 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
289
290         if (resid)
291                 *resid = rq->resid_len;
292         if (sense && rq->sense_len)
293                 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
294         if (sshdr)
295                 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
296         ret = rq->result;
297  out:
298         blk_put_request(req);
299
300         return ret;
301 }
302 EXPORT_SYMBOL(__scsi_execute);
303
304 /*
305  * Function:    scsi_init_cmd_errh()
306  *
307  * Purpose:     Initialize cmd fields related to error handling.
308  *
309  * Arguments:   cmd     - command that is ready to be queued.
310  *
311  * Notes:       This function has the job of initializing a number of
312  *              fields related to error handling.   Typically this will
313  *              be called once for each command, as required.
314  */
315 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
316 {
317         scsi_set_resid(cmd, 0);
318         memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
319         if (cmd->cmd_len == 0)
320                 cmd->cmd_len = scsi_command_size(cmd->cmnd);
321 }
322
323 /*
324  * Wake up the error handler if necessary. Avoid as follows that the error
325  * handler is not woken up if host in-flight requests number ==
326  * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
327  * with an RCU read lock in this function to ensure that this function in
328  * its entirety either finishes before scsi_eh_scmd_add() increases the
329  * host_failed counter or that it notices the shost state change made by
330  * scsi_eh_scmd_add().
331  */
332 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
333 {
334         unsigned long flags;
335
336         rcu_read_lock();
337         __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
338         if (unlikely(scsi_host_in_recovery(shost))) {
339                 spin_lock_irqsave(shost->host_lock, flags);
340                 if (shost->host_failed || shost->host_eh_scheduled)
341                         scsi_eh_wakeup(shost);
342                 spin_unlock_irqrestore(shost->host_lock, flags);
343         }
344         rcu_read_unlock();
345 }
346
347 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
348 {
349         struct Scsi_Host *shost = sdev->host;
350         struct scsi_target *starget = scsi_target(sdev);
351
352         scsi_dec_host_busy(shost, cmd);
353
354         if (starget->can_queue > 0)
355                 atomic_dec(&starget->target_busy);
356
357         atomic_dec(&sdev->device_busy);
358 }
359
360 static void scsi_kick_queue(struct request_queue *q)
361 {
362         blk_mq_run_hw_queues(q, false);
363 }
364
365 /*
366  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
367  * and call blk_run_queue for all the scsi_devices on the target -
368  * including current_sdev first.
369  *
370  * Called with *no* scsi locks held.
371  */
372 static void scsi_single_lun_run(struct scsi_device *current_sdev)
373 {
374         struct Scsi_Host *shost = current_sdev->host;
375         struct scsi_device *sdev, *tmp;
376         struct scsi_target *starget = scsi_target(current_sdev);
377         unsigned long flags;
378
379         spin_lock_irqsave(shost->host_lock, flags);
380         starget->starget_sdev_user = NULL;
381         spin_unlock_irqrestore(shost->host_lock, flags);
382
383         /*
384          * Call blk_run_queue for all LUNs on the target, starting with
385          * current_sdev. We race with others (to set starget_sdev_user),
386          * but in most cases, we will be first. Ideally, each LU on the
387          * target would get some limited time or requests on the target.
388          */
389         scsi_kick_queue(current_sdev->request_queue);
390
391         spin_lock_irqsave(shost->host_lock, flags);
392         if (starget->starget_sdev_user)
393                 goto out;
394         list_for_each_entry_safe(sdev, tmp, &starget->devices,
395                         same_target_siblings) {
396                 if (sdev == current_sdev)
397                         continue;
398                 if (scsi_device_get(sdev))
399                         continue;
400
401                 spin_unlock_irqrestore(shost->host_lock, flags);
402                 scsi_kick_queue(sdev->request_queue);
403                 spin_lock_irqsave(shost->host_lock, flags);
404         
405                 scsi_device_put(sdev);
406         }
407  out:
408         spin_unlock_irqrestore(shost->host_lock, flags);
409 }
410
411 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
412 {
413         if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
414                 return true;
415         if (atomic_read(&sdev->device_blocked) > 0)
416                 return true;
417         return false;
418 }
419
420 static inline bool scsi_target_is_busy(struct scsi_target *starget)
421 {
422         if (starget->can_queue > 0) {
423                 if (atomic_read(&starget->target_busy) >= starget->can_queue)
424                         return true;
425                 if (atomic_read(&starget->target_blocked) > 0)
426                         return true;
427         }
428         return false;
429 }
430
431 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
432 {
433         if (atomic_read(&shost->host_blocked) > 0)
434                 return true;
435         if (shost->host_self_blocked)
436                 return true;
437         return false;
438 }
439
440 static void scsi_starved_list_run(struct Scsi_Host *shost)
441 {
442         LIST_HEAD(starved_list);
443         struct scsi_device *sdev;
444         unsigned long flags;
445
446         spin_lock_irqsave(shost->host_lock, flags);
447         list_splice_init(&shost->starved_list, &starved_list);
448
449         while (!list_empty(&starved_list)) {
450                 struct request_queue *slq;
451
452                 /*
453                  * As long as shost is accepting commands and we have
454                  * starved queues, call blk_run_queue. scsi_request_fn
455                  * drops the queue_lock and can add us back to the
456                  * starved_list.
457                  *
458                  * host_lock protects the starved_list and starved_entry.
459                  * scsi_request_fn must get the host_lock before checking
460                  * or modifying starved_list or starved_entry.
461                  */
462                 if (scsi_host_is_busy(shost))
463                         break;
464
465                 sdev = list_entry(starved_list.next,
466                                   struct scsi_device, starved_entry);
467                 list_del_init(&sdev->starved_entry);
468                 if (scsi_target_is_busy(scsi_target(sdev))) {
469                         list_move_tail(&sdev->starved_entry,
470                                        &shost->starved_list);
471                         continue;
472                 }
473
474                 /*
475                  * Once we drop the host lock, a racing scsi_remove_device()
476                  * call may remove the sdev from the starved list and destroy
477                  * it and the queue.  Mitigate by taking a reference to the
478                  * queue and never touching the sdev again after we drop the
479                  * host lock.  Note: if __scsi_remove_device() invokes
480                  * blk_cleanup_queue() before the queue is run from this
481                  * function then blk_run_queue() will return immediately since
482                  * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
483                  */
484                 slq = sdev->request_queue;
485                 if (!blk_get_queue(slq))
486                         continue;
487                 spin_unlock_irqrestore(shost->host_lock, flags);
488
489                 scsi_kick_queue(slq);
490                 blk_put_queue(slq);
491
492                 spin_lock_irqsave(shost->host_lock, flags);
493         }
494         /* put any unprocessed entries back */
495         list_splice(&starved_list, &shost->starved_list);
496         spin_unlock_irqrestore(shost->host_lock, flags);
497 }
498
499 /*
500  * Function:   scsi_run_queue()
501  *
502  * Purpose:    Select a proper request queue to serve next
503  *
504  * Arguments:  q       - last request's queue
505  *
506  * Returns:     Nothing
507  *
508  * Notes:      The previous command was completely finished, start
509  *             a new one if possible.
510  */
511 static void scsi_run_queue(struct request_queue *q)
512 {
513         struct scsi_device *sdev = q->queuedata;
514
515         if (scsi_target(sdev)->single_lun)
516                 scsi_single_lun_run(sdev);
517         if (!list_empty(&sdev->host->starved_list))
518                 scsi_starved_list_run(sdev->host);
519
520         blk_mq_run_hw_queues(q, false);
521 }
522
523 void scsi_requeue_run_queue(struct work_struct *work)
524 {
525         struct scsi_device *sdev;
526         struct request_queue *q;
527
528         sdev = container_of(work, struct scsi_device, requeue_work);
529         q = sdev->request_queue;
530         scsi_run_queue(q);
531 }
532
533 void scsi_run_host_queues(struct Scsi_Host *shost)
534 {
535         struct scsi_device *sdev;
536
537         shost_for_each_device(sdev, shost)
538                 scsi_run_queue(sdev->request_queue);
539 }
540
541 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
542 {
543         if (!blk_rq_is_passthrough(cmd->request)) {
544                 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
545
546                 if (drv->uninit_command)
547                         drv->uninit_command(cmd);
548         }
549 }
550
551 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
552 {
553         if (cmd->sdb.table.nents)
554                 sg_free_table_chained(&cmd->sdb.table,
555                                 SCSI_INLINE_SG_CNT);
556         if (scsi_prot_sg_count(cmd))
557                 sg_free_table_chained(&cmd->prot_sdb->table,
558                                 SCSI_INLINE_PROT_SG_CNT);
559 }
560
561 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
562 {
563         scsi_mq_free_sgtables(cmd);
564         scsi_uninit_cmd(cmd);
565 }
566
567 /* Returns false when no more bytes to process, true if there are more */
568 static bool scsi_end_request(struct request *req, blk_status_t error,
569                 unsigned int bytes)
570 {
571         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
572         struct scsi_device *sdev = cmd->device;
573         struct request_queue *q = sdev->request_queue;
574
575         if (blk_update_request(req, error, bytes))
576                 return true;
577
578         if (blk_queue_add_random(q))
579                 add_disk_randomness(req->rq_disk);
580
581         if (!blk_rq_is_scsi(req)) {
582                 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
583                 cmd->flags &= ~SCMD_INITIALIZED;
584         }
585
586         /*
587          * Calling rcu_barrier() is not necessary here because the
588          * SCSI error handler guarantees that the function called by
589          * call_rcu() has been called before scsi_end_request() is
590          * called.
591          */
592         destroy_rcu_head(&cmd->rcu);
593
594         /*
595          * In the MQ case the command gets freed by __blk_mq_end_request,
596          * so we have to do all cleanup that depends on it earlier.
597          *
598          * We also can't kick the queues from irq context, so we
599          * will have to defer it to a workqueue.
600          */
601         scsi_mq_uninit_cmd(cmd);
602
603         /*
604          * queue is still alive, so grab the ref for preventing it
605          * from being cleaned up during running queue.
606          */
607         percpu_ref_get(&q->q_usage_counter);
608
609         __blk_mq_end_request(req, error);
610
611         if (scsi_target(sdev)->single_lun ||
612             !list_empty(&sdev->host->starved_list))
613                 kblockd_schedule_work(&sdev->requeue_work);
614         else
615                 blk_mq_run_hw_queues(q, true);
616
617         percpu_ref_put(&q->q_usage_counter);
618         return false;
619 }
620
621 /**
622  * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
623  * @cmd:        SCSI command
624  * @result:     scsi error code
625  *
626  * Translate a SCSI result code into a blk_status_t value. May reset the host
627  * byte of @cmd->result.
628  */
629 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
630 {
631         switch (host_byte(result)) {
632         case DID_OK:
633                 /*
634                  * Also check the other bytes than the status byte in result
635                  * to handle the case when a SCSI LLD sets result to
636                  * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
637                  */
638                 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
639                         return BLK_STS_OK;
640                 return BLK_STS_IOERR;
641         case DID_TRANSPORT_FAILFAST:
642                 return BLK_STS_TRANSPORT;
643         case DID_TARGET_FAILURE:
644                 set_host_byte(cmd, DID_OK);
645                 return BLK_STS_TARGET;
646         case DID_NEXUS_FAILURE:
647                 set_host_byte(cmd, DID_OK);
648                 return BLK_STS_NEXUS;
649         case DID_ALLOC_FAILURE:
650                 set_host_byte(cmd, DID_OK);
651                 return BLK_STS_NOSPC;
652         case DID_MEDIUM_ERROR:
653                 set_host_byte(cmd, DID_OK);
654                 return BLK_STS_MEDIUM;
655         default:
656                 return BLK_STS_IOERR;
657         }
658 }
659
660 /* Helper for scsi_io_completion() when "reprep" action required. */
661 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
662                                       struct request_queue *q)
663 {
664         /* A new command will be prepared and issued. */
665         scsi_mq_requeue_cmd(cmd);
666 }
667
668 /* Helper for scsi_io_completion() when special action required. */
669 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
670 {
671         struct request_queue *q = cmd->device->request_queue;
672         struct request *req = cmd->request;
673         int level = 0;
674         enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
675               ACTION_DELAYED_RETRY} action;
676         unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
677         struct scsi_sense_hdr sshdr;
678         bool sense_valid;
679         bool sense_current = true;      /* false implies "deferred sense" */
680         blk_status_t blk_stat;
681
682         sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
683         if (sense_valid)
684                 sense_current = !scsi_sense_is_deferred(&sshdr);
685
686         blk_stat = scsi_result_to_blk_status(cmd, result);
687
688         if (host_byte(result) == DID_RESET) {
689                 /* Third party bus reset or reset for error recovery
690                  * reasons.  Just retry the command and see what
691                  * happens.
692                  */
693                 action = ACTION_RETRY;
694         } else if (sense_valid && sense_current) {
695                 switch (sshdr.sense_key) {
696                 case UNIT_ATTENTION:
697                         if (cmd->device->removable) {
698                                 /* Detected disc change.  Set a bit
699                                  * and quietly refuse further access.
700                                  */
701                                 cmd->device->changed = 1;
702                                 action = ACTION_FAIL;
703                         } else {
704                                 /* Must have been a power glitch, or a
705                                  * bus reset.  Could not have been a
706                                  * media change, so we just retry the
707                                  * command and see what happens.
708                                  */
709                                 action = ACTION_RETRY;
710                         }
711                         break;
712                 case ILLEGAL_REQUEST:
713                         /* If we had an ILLEGAL REQUEST returned, then
714                          * we may have performed an unsupported
715                          * command.  The only thing this should be
716                          * would be a ten byte read where only a six
717                          * byte read was supported.  Also, on a system
718                          * where READ CAPACITY failed, we may have
719                          * read past the end of the disk.
720                          */
721                         if ((cmd->device->use_10_for_rw &&
722                             sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
723                             (cmd->cmnd[0] == READ_10 ||
724                              cmd->cmnd[0] == WRITE_10)) {
725                                 /* This will issue a new 6-byte command. */
726                                 cmd->device->use_10_for_rw = 0;
727                                 action = ACTION_REPREP;
728                         } else if (sshdr.asc == 0x10) /* DIX */ {
729                                 action = ACTION_FAIL;
730                                 blk_stat = BLK_STS_PROTECTION;
731                         /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
732                         } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
733                                 action = ACTION_FAIL;
734                                 blk_stat = BLK_STS_TARGET;
735                         } else
736                                 action = ACTION_FAIL;
737                         break;
738                 case ABORTED_COMMAND:
739                         action = ACTION_FAIL;
740                         if (sshdr.asc == 0x10) /* DIF */
741                                 blk_stat = BLK_STS_PROTECTION;
742                         break;
743                 case NOT_READY:
744                         /* If the device is in the process of becoming
745                          * ready, or has a temporary blockage, retry.
746                          */
747                         if (sshdr.asc == 0x04) {
748                                 switch (sshdr.ascq) {
749                                 case 0x01: /* becoming ready */
750                                 case 0x04: /* format in progress */
751                                 case 0x05: /* rebuild in progress */
752                                 case 0x06: /* recalculation in progress */
753                                 case 0x07: /* operation in progress */
754                                 case 0x08: /* Long write in progress */
755                                 case 0x09: /* self test in progress */
756                                 case 0x14: /* space allocation in progress */
757                                 case 0x1a: /* start stop unit in progress */
758                                 case 0x1b: /* sanitize in progress */
759                                 case 0x1d: /* configuration in progress */
760                                 case 0x24: /* depopulation in progress */
761                                         action = ACTION_DELAYED_RETRY;
762                                         break;
763                                 default:
764                                         action = ACTION_FAIL;
765                                         break;
766                                 }
767                         } else
768                                 action = ACTION_FAIL;
769                         break;
770                 case VOLUME_OVERFLOW:
771                         /* See SSC3rXX or current. */
772                         action = ACTION_FAIL;
773                         break;
774                 default:
775                         action = ACTION_FAIL;
776                         break;
777                 }
778         } else
779                 action = ACTION_FAIL;
780
781         if (action != ACTION_FAIL &&
782             time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
783                 action = ACTION_FAIL;
784
785         switch (action) {
786         case ACTION_FAIL:
787                 /* Give up and fail the remainder of the request */
788                 if (!(req->rq_flags & RQF_QUIET)) {
789                         static DEFINE_RATELIMIT_STATE(_rs,
790                                         DEFAULT_RATELIMIT_INTERVAL,
791                                         DEFAULT_RATELIMIT_BURST);
792
793                         if (unlikely(scsi_logging_level))
794                                 level =
795                                      SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
796                                                     SCSI_LOG_MLCOMPLETE_BITS);
797
798                         /*
799                          * if logging is enabled the failure will be printed
800                          * in scsi_log_completion(), so avoid duplicate messages
801                          */
802                         if (!level && __ratelimit(&_rs)) {
803                                 scsi_print_result(cmd, NULL, FAILED);
804                                 if (driver_byte(result) == DRIVER_SENSE)
805                                         scsi_print_sense(cmd);
806                                 scsi_print_command(cmd);
807                         }
808                 }
809                 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
810                         return;
811                 /*FALLTHRU*/
812         case ACTION_REPREP:
813                 scsi_io_completion_reprep(cmd, q);
814                 break;
815         case ACTION_RETRY:
816                 /* Retry the same command immediately */
817                 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
818                 break;
819         case ACTION_DELAYED_RETRY:
820                 /* Retry the same command after a delay */
821                 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
822                 break;
823         }
824 }
825
826 /*
827  * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
828  * new result that may suppress further error checking. Also modifies
829  * *blk_statp in some cases.
830  */
831 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
832                                         blk_status_t *blk_statp)
833 {
834         bool sense_valid;
835         bool sense_current = true;      /* false implies "deferred sense" */
836         struct request *req = cmd->request;
837         struct scsi_sense_hdr sshdr;
838
839         sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
840         if (sense_valid)
841                 sense_current = !scsi_sense_is_deferred(&sshdr);
842
843         if (blk_rq_is_passthrough(req)) {
844                 if (sense_valid) {
845                         /*
846                          * SG_IO wants current and deferred errors
847                          */
848                         scsi_req(req)->sense_len =
849                                 min(8 + cmd->sense_buffer[7],
850                                     SCSI_SENSE_BUFFERSIZE);
851                 }
852                 if (sense_current)
853                         *blk_statp = scsi_result_to_blk_status(cmd, result);
854         } else if (blk_rq_bytes(req) == 0 && sense_current) {
855                 /*
856                  * Flush commands do not transfers any data, and thus cannot use
857                  * good_bytes != blk_rq_bytes(req) as the signal for an error.
858                  * This sets *blk_statp explicitly for the problem case.
859                  */
860                 *blk_statp = scsi_result_to_blk_status(cmd, result);
861         }
862         /*
863          * Recovered errors need reporting, but they're always treated as
864          * success, so fiddle the result code here.  For passthrough requests
865          * we already took a copy of the original into sreq->result which
866          * is what gets returned to the user
867          */
868         if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
869                 bool do_print = true;
870                 /*
871                  * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
872                  * skip print since caller wants ATA registers. Only occurs
873                  * on SCSI ATA PASS_THROUGH commands when CK_COND=1
874                  */
875                 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
876                         do_print = false;
877                 else if (req->rq_flags & RQF_QUIET)
878                         do_print = false;
879                 if (do_print)
880                         scsi_print_sense(cmd);
881                 result = 0;
882                 /* for passthrough, *blk_statp may be set */
883                 *blk_statp = BLK_STS_OK;
884         }
885         /*
886          * Another corner case: the SCSI status byte is non-zero but 'good'.
887          * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
888          * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
889          * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
890          * intermediate statuses (both obsolete in SAM-4) as good.
891          */
892         if (status_byte(result) && scsi_status_is_good(result)) {
893                 result = 0;
894                 *blk_statp = BLK_STS_OK;
895         }
896         return result;
897 }
898
899 /*
900  * Function:    scsi_io_completion()
901  *
902  * Purpose:     Completion processing for block device I/O requests.
903  *
904  * Arguments:   cmd   - command that is finished.
905  *
906  * Lock status: Assumed that no lock is held upon entry.
907  *
908  * Returns:     Nothing
909  *
910  * Notes:       We will finish off the specified number of sectors.  If we
911  *              are done, the command block will be released and the queue
912  *              function will be goosed.  If we are not done then we have to
913  *              figure out what to do next:
914  *
915  *              a) We can call scsi_requeue_command().  The request
916  *                 will be unprepared and put back on the queue.  Then
917  *                 a new command will be created for it.  This should
918  *                 be used if we made forward progress, or if we want
919  *                 to switch from READ(10) to READ(6) for example.
920  *
921  *              b) We can call __scsi_queue_insert().  The request will
922  *                 be put back on the queue and retried using the same
923  *                 command as before, possibly after a delay.
924  *
925  *              c) We can call scsi_end_request() with blk_stat other than
926  *                 BLK_STS_OK, to fail the remainder of the request.
927  */
928 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
929 {
930         int result = cmd->result;
931         struct request_queue *q = cmd->device->request_queue;
932         struct request *req = cmd->request;
933         blk_status_t blk_stat = BLK_STS_OK;
934
935         if (unlikely(result))   /* a nz result may or may not be an error */
936                 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
937
938         if (unlikely(blk_rq_is_passthrough(req))) {
939                 /*
940                  * scsi_result_to_blk_status may have reset the host_byte
941                  */
942                 scsi_req(req)->result = cmd->result;
943         }
944
945         /*
946          * Next deal with any sectors which we were able to correctly
947          * handle.
948          */
949         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
950                 "%u sectors total, %d bytes done.\n",
951                 blk_rq_sectors(req), good_bytes));
952
953         /*
954          * Next deal with any sectors which we were able to correctly
955          * handle. Failed, zero length commands always need to drop down
956          * to retry code. Fast path should return in this block.
957          */
958         if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
959                 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
960                         return; /* no bytes remaining */
961         }
962
963         /* Kill remainder if no retries. */
964         if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
965                 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
966                         WARN_ONCE(true,
967                             "Bytes remaining after failed, no-retry command");
968                 return;
969         }
970
971         /*
972          * If there had been no error, but we have leftover bytes in the
973          * requeues just queue the command up again.
974          */
975         if (likely(result == 0))
976                 scsi_io_completion_reprep(cmd, q);
977         else
978                 scsi_io_completion_action(cmd, result);
979 }
980
981 static blk_status_t scsi_init_sgtable(struct request *req,
982                 struct scsi_data_buffer *sdb)
983 {
984         int count;
985
986         /*
987          * If sg table allocation fails, requeue request later.
988          */
989         if (unlikely(sg_alloc_table_chained(&sdb->table,
990                         blk_rq_nr_phys_segments(req), sdb->table.sgl,
991                         SCSI_INLINE_SG_CNT)))
992                 return BLK_STS_RESOURCE;
993
994         /* 
995          * Next, walk the list, and fill in the addresses and sizes of
996          * each segment.
997          */
998         count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
999         BUG_ON(count > sdb->table.nents);
1000         sdb->table.nents = count;
1001         sdb->length = blk_rq_payload_bytes(req);
1002         return BLK_STS_OK;
1003 }
1004
1005 /*
1006  * Function:    scsi_init_io()
1007  *
1008  * Purpose:     SCSI I/O initialize function.
1009  *
1010  * Arguments:   cmd   - Command descriptor we wish to initialize
1011  *
1012  * Returns:     BLK_STS_OK on success
1013  *              BLK_STS_RESOURCE if the failure is retryable
1014  *              BLK_STS_IOERR if the failure is fatal
1015  */
1016 blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
1017 {
1018         struct request *rq = cmd->request;
1019         blk_status_t ret;
1020
1021         if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1022                 return BLK_STS_IOERR;
1023
1024         ret = scsi_init_sgtable(rq, &cmd->sdb);
1025         if (ret)
1026                 return ret;
1027
1028         if (blk_integrity_rq(rq)) {
1029                 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1030                 int ivecs, count;
1031
1032                 if (WARN_ON_ONCE(!prot_sdb)) {
1033                         /*
1034                          * This can happen if someone (e.g. multipath)
1035                          * queues a command to a device on an adapter
1036                          * that does not support DIX.
1037                          */
1038                         ret = BLK_STS_IOERR;
1039                         goto out_free_sgtables;
1040                 }
1041
1042                 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1043
1044                 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1045                                 prot_sdb->table.sgl,
1046                                 SCSI_INLINE_PROT_SG_CNT)) {
1047                         ret = BLK_STS_RESOURCE;
1048                         goto out_free_sgtables;
1049                 }
1050
1051                 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1052                                                 prot_sdb->table.sgl);
1053                 BUG_ON(count > ivecs);
1054                 BUG_ON(count > queue_max_integrity_segments(rq->q));
1055
1056                 cmd->prot_sdb = prot_sdb;
1057                 cmd->prot_sdb->table.nents = count;
1058         }
1059
1060         return BLK_STS_OK;
1061 out_free_sgtables:
1062         scsi_mq_free_sgtables(cmd);
1063         return ret;
1064 }
1065 EXPORT_SYMBOL(scsi_init_io);
1066
1067 /**
1068  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1069  * @rq: Request associated with the SCSI command to be initialized.
1070  *
1071  * This function initializes the members of struct scsi_cmnd that must be
1072  * initialized before request processing starts and that won't be
1073  * reinitialized if a SCSI command is requeued.
1074  *
1075  * Called from inside blk_get_request() for pass-through requests and from
1076  * inside scsi_init_command() for filesystem requests.
1077  */
1078 static void scsi_initialize_rq(struct request *rq)
1079 {
1080         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1081
1082         scsi_req_init(&cmd->req);
1083         init_rcu_head(&cmd->rcu);
1084         cmd->jiffies_at_alloc = jiffies;
1085         cmd->retries = 0;
1086 }
1087
1088 /*
1089  * Only called when the request isn't completed by SCSI, and not freed by
1090  * SCSI
1091  */
1092 static void scsi_cleanup_rq(struct request *rq)
1093 {
1094         if (rq->rq_flags & RQF_DONTPREP) {
1095                 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1096                 rq->rq_flags &= ~RQF_DONTPREP;
1097         }
1098 }
1099
1100 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1101 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1102 {
1103         void *buf = cmd->sense_buffer;
1104         void *prot = cmd->prot_sdb;
1105         struct request *rq = blk_mq_rq_from_pdu(cmd);
1106         unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1107         unsigned long jiffies_at_alloc;
1108         int retries, to_clear;
1109         bool in_flight;
1110
1111         if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1112                 flags |= SCMD_INITIALIZED;
1113                 scsi_initialize_rq(rq);
1114         }
1115
1116         jiffies_at_alloc = cmd->jiffies_at_alloc;
1117         retries = cmd->retries;
1118         in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1119         /*
1120          * Zero out the cmd, except for the embedded scsi_request. Only clear
1121          * the driver-private command data if the LLD does not supply a
1122          * function to initialize that data.
1123          */
1124         to_clear = sizeof(*cmd) - sizeof(cmd->req);
1125         if (!dev->host->hostt->init_cmd_priv)
1126                 to_clear += dev->host->hostt->cmd_size;
1127         memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1128
1129         cmd->device = dev;
1130         cmd->sense_buffer = buf;
1131         cmd->prot_sdb = prot;
1132         cmd->flags = flags;
1133         INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1134         cmd->jiffies_at_alloc = jiffies_at_alloc;
1135         cmd->retries = retries;
1136         if (in_flight)
1137                 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1138
1139 }
1140
1141 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1142                 struct request *req)
1143 {
1144         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1145
1146         /*
1147          * Passthrough requests may transfer data, in which case they must
1148          * a bio attached to them.  Or they might contain a SCSI command
1149          * that does not transfer data, in which case they may optionally
1150          * submit a request without an attached bio.
1151          */
1152         if (req->bio) {
1153                 blk_status_t ret = scsi_init_io(cmd);
1154                 if (unlikely(ret != BLK_STS_OK))
1155                         return ret;
1156         } else {
1157                 BUG_ON(blk_rq_bytes(req));
1158
1159                 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1160         }
1161
1162         cmd->cmd_len = scsi_req(req)->cmd_len;
1163         cmd->cmnd = scsi_req(req)->cmd;
1164         cmd->transfersize = blk_rq_bytes(req);
1165         cmd->allowed = scsi_req(req)->retries;
1166         return BLK_STS_OK;
1167 }
1168
1169 /*
1170  * Setup a normal block command.  These are simple request from filesystems
1171  * that still need to be translated to SCSI CDBs from the ULD.
1172  */
1173 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1174                 struct request *req)
1175 {
1176         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1177
1178         if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1179                 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1180                 if (ret != BLK_STS_OK)
1181                         return ret;
1182         }
1183
1184         cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1185         memset(cmd->cmnd, 0, BLK_MAX_CDB);
1186         return scsi_cmd_to_driver(cmd)->init_command(cmd);
1187 }
1188
1189 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1190                 struct request *req)
1191 {
1192         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1193
1194         if (!blk_rq_bytes(req))
1195                 cmd->sc_data_direction = DMA_NONE;
1196         else if (rq_data_dir(req) == WRITE)
1197                 cmd->sc_data_direction = DMA_TO_DEVICE;
1198         else
1199                 cmd->sc_data_direction = DMA_FROM_DEVICE;
1200
1201         if (blk_rq_is_scsi(req))
1202                 return scsi_setup_scsi_cmnd(sdev, req);
1203         else
1204                 return scsi_setup_fs_cmnd(sdev, req);
1205 }
1206
1207 static blk_status_t
1208 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1209 {
1210         switch (sdev->sdev_state) {
1211         case SDEV_OFFLINE:
1212         case SDEV_TRANSPORT_OFFLINE:
1213                 /*
1214                  * If the device is offline we refuse to process any
1215                  * commands.  The device must be brought online
1216                  * before trying any recovery commands.
1217                  */
1218                 if (!sdev->offline_already) {
1219                         sdev->offline_already = true;
1220                         sdev_printk(KERN_ERR, sdev,
1221                                     "rejecting I/O to offline device\n");
1222                 }
1223                 return BLK_STS_IOERR;
1224         case SDEV_DEL:
1225                 /*
1226                  * If the device is fully deleted, we refuse to
1227                  * process any commands as well.
1228                  */
1229                 sdev_printk(KERN_ERR, sdev,
1230                             "rejecting I/O to dead device\n");
1231                 return BLK_STS_IOERR;
1232         case SDEV_BLOCK:
1233         case SDEV_CREATED_BLOCK:
1234                 return BLK_STS_RESOURCE;
1235         case SDEV_QUIESCE:
1236                 /*
1237                  * If the devices is blocked we defer normal commands.
1238                  */
1239                 if (req && !(req->rq_flags & RQF_PREEMPT))
1240                         return BLK_STS_RESOURCE;
1241                 return BLK_STS_OK;
1242         default:
1243                 /*
1244                  * For any other not fully online state we only allow
1245                  * special commands.  In particular any user initiated
1246                  * command is not allowed.
1247                  */
1248                 if (req && !(req->rq_flags & RQF_PREEMPT))
1249                         return BLK_STS_IOERR;
1250                 return BLK_STS_OK;
1251         }
1252 }
1253
1254 /*
1255  * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1256  * return 0.
1257  *
1258  * Called with the queue_lock held.
1259  */
1260 static inline int scsi_dev_queue_ready(struct request_queue *q,
1261                                   struct scsi_device *sdev)
1262 {
1263         unsigned int busy;
1264
1265         busy = atomic_inc_return(&sdev->device_busy) - 1;
1266         if (atomic_read(&sdev->device_blocked)) {
1267                 if (busy)
1268                         goto out_dec;
1269
1270                 /*
1271                  * unblock after device_blocked iterates to zero
1272                  */
1273                 if (atomic_dec_return(&sdev->device_blocked) > 0)
1274                         goto out_dec;
1275                 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1276                                    "unblocking device at zero depth\n"));
1277         }
1278
1279         if (busy >= sdev->queue_depth)
1280                 goto out_dec;
1281
1282         return 1;
1283 out_dec:
1284         atomic_dec(&sdev->device_busy);
1285         return 0;
1286 }
1287
1288 /*
1289  * scsi_target_queue_ready: checks if there we can send commands to target
1290  * @sdev: scsi device on starget to check.
1291  */
1292 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1293                                            struct scsi_device *sdev)
1294 {
1295         struct scsi_target *starget = scsi_target(sdev);
1296         unsigned int busy;
1297
1298         if (starget->single_lun) {
1299                 spin_lock_irq(shost->host_lock);
1300                 if (starget->starget_sdev_user &&
1301                     starget->starget_sdev_user != sdev) {
1302                         spin_unlock_irq(shost->host_lock);
1303                         return 0;
1304                 }
1305                 starget->starget_sdev_user = sdev;
1306                 spin_unlock_irq(shost->host_lock);
1307         }
1308
1309         if (starget->can_queue <= 0)
1310                 return 1;
1311
1312         busy = atomic_inc_return(&starget->target_busy) - 1;
1313         if (atomic_read(&starget->target_blocked) > 0) {
1314                 if (busy)
1315                         goto starved;
1316
1317                 /*
1318                  * unblock after target_blocked iterates to zero
1319                  */
1320                 if (atomic_dec_return(&starget->target_blocked) > 0)
1321                         goto out_dec;
1322
1323                 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1324                                  "unblocking target at zero depth\n"));
1325         }
1326
1327         if (busy >= starget->can_queue)
1328                 goto starved;
1329
1330         return 1;
1331
1332 starved:
1333         spin_lock_irq(shost->host_lock);
1334         list_move_tail(&sdev->starved_entry, &shost->starved_list);
1335         spin_unlock_irq(shost->host_lock);
1336 out_dec:
1337         if (starget->can_queue > 0)
1338                 atomic_dec(&starget->target_busy);
1339         return 0;
1340 }
1341
1342 /*
1343  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1344  * return 0. We must end up running the queue again whenever 0 is
1345  * returned, else IO can hang.
1346  */
1347 static inline int scsi_host_queue_ready(struct request_queue *q,
1348                                    struct Scsi_Host *shost,
1349                                    struct scsi_device *sdev,
1350                                    struct scsi_cmnd *cmd)
1351 {
1352         if (scsi_host_in_recovery(shost))
1353                 return 0;
1354
1355         if (atomic_read(&shost->host_blocked) > 0) {
1356                 if (scsi_host_busy(shost) > 0)
1357                         goto starved;
1358
1359                 /*
1360                  * unblock after host_blocked iterates to zero
1361                  */
1362                 if (atomic_dec_return(&shost->host_blocked) > 0)
1363                         goto out_dec;
1364
1365                 SCSI_LOG_MLQUEUE(3,
1366                         shost_printk(KERN_INFO, shost,
1367                                      "unblocking host at zero depth\n"));
1368         }
1369
1370         if (shost->host_self_blocked)
1371                 goto starved;
1372
1373         /* We're OK to process the command, so we can't be starved */
1374         if (!list_empty(&sdev->starved_entry)) {
1375                 spin_lock_irq(shost->host_lock);
1376                 if (!list_empty(&sdev->starved_entry))
1377                         list_del_init(&sdev->starved_entry);
1378                 spin_unlock_irq(shost->host_lock);
1379         }
1380
1381         __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1382
1383         return 1;
1384
1385 starved:
1386         spin_lock_irq(shost->host_lock);
1387         if (list_empty(&sdev->starved_entry))
1388                 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1389         spin_unlock_irq(shost->host_lock);
1390 out_dec:
1391         scsi_dec_host_busy(shost, cmd);
1392         return 0;
1393 }
1394
1395 /*
1396  * Busy state exporting function for request stacking drivers.
1397  *
1398  * For efficiency, no lock is taken to check the busy state of
1399  * shost/starget/sdev, since the returned value is not guaranteed and
1400  * may be changed after request stacking drivers call the function,
1401  * regardless of taking lock or not.
1402  *
1403  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1404  * needs to return 'not busy'. Otherwise, request stacking drivers
1405  * may hold requests forever.
1406  */
1407 static bool scsi_mq_lld_busy(struct request_queue *q)
1408 {
1409         struct scsi_device *sdev = q->queuedata;
1410         struct Scsi_Host *shost;
1411
1412         if (blk_queue_dying(q))
1413                 return false;
1414
1415         shost = sdev->host;
1416
1417         /*
1418          * Ignore host/starget busy state.
1419          * Since block layer does not have a concept of fairness across
1420          * multiple queues, congestion of host/starget needs to be handled
1421          * in SCSI layer.
1422          */
1423         if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1424                 return true;
1425
1426         return false;
1427 }
1428
1429 static void scsi_softirq_done(struct request *rq)
1430 {
1431         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1432         unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1433         int disposition;
1434
1435         INIT_LIST_HEAD(&cmd->eh_entry);
1436
1437         atomic_inc(&cmd->device->iodone_cnt);
1438         if (cmd->result)
1439                 atomic_inc(&cmd->device->ioerr_cnt);
1440
1441         disposition = scsi_decide_disposition(cmd);
1442         if (disposition != SUCCESS &&
1443             time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1444                 scmd_printk(KERN_ERR, cmd,
1445                             "timing out command, waited %lus\n",
1446                             wait_for/HZ);
1447                 disposition = SUCCESS;
1448         }
1449
1450         scsi_log_completion(cmd, disposition);
1451
1452         switch (disposition) {
1453                 case SUCCESS:
1454                         scsi_finish_command(cmd);
1455                         break;
1456                 case NEEDS_RETRY:
1457                         scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1458                         break;
1459                 case ADD_TO_MLQUEUE:
1460                         scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1461                         break;
1462                 default:
1463                         scsi_eh_scmd_add(cmd);
1464                         break;
1465         }
1466 }
1467
1468 /**
1469  * scsi_dispatch_command - Dispatch a command to the low-level driver.
1470  * @cmd: command block we are dispatching.
1471  *
1472  * Return: nonzero return request was rejected and device's queue needs to be
1473  * plugged.
1474  */
1475 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1476 {
1477         struct Scsi_Host *host = cmd->device->host;
1478         int rtn = 0;
1479
1480         atomic_inc(&cmd->device->iorequest_cnt);
1481
1482         /* check if the device is still usable */
1483         if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1484                 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1485                  * returns an immediate error upwards, and signals
1486                  * that the device is no longer present */
1487                 cmd->result = DID_NO_CONNECT << 16;
1488                 goto done;
1489         }
1490
1491         /* Check to see if the scsi lld made this device blocked. */
1492         if (unlikely(scsi_device_blocked(cmd->device))) {
1493                 /*
1494                  * in blocked state, the command is just put back on
1495                  * the device queue.  The suspend state has already
1496                  * blocked the queue so future requests should not
1497                  * occur until the device transitions out of the
1498                  * suspend state.
1499                  */
1500                 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1501                         "queuecommand : device blocked\n"));
1502                 return SCSI_MLQUEUE_DEVICE_BUSY;
1503         }
1504
1505         /* Store the LUN value in cmnd, if needed. */
1506         if (cmd->device->lun_in_cdb)
1507                 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1508                                (cmd->device->lun << 5 & 0xe0);
1509
1510         scsi_log_send(cmd);
1511
1512         /*
1513          * Before we queue this command, check if the command
1514          * length exceeds what the host adapter can handle.
1515          */
1516         if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1517                 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1518                                "queuecommand : command too long. "
1519                                "cdb_size=%d host->max_cmd_len=%d\n",
1520                                cmd->cmd_len, cmd->device->host->max_cmd_len));
1521                 cmd->result = (DID_ABORT << 16);
1522                 goto done;
1523         }
1524
1525         if (unlikely(host->shost_state == SHOST_DEL)) {
1526                 cmd->result = (DID_NO_CONNECT << 16);
1527                 goto done;
1528
1529         }
1530
1531         trace_scsi_dispatch_cmd_start(cmd);
1532         rtn = host->hostt->queuecommand(host, cmd);
1533         if (rtn) {
1534                 trace_scsi_dispatch_cmd_error(cmd, rtn);
1535                 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1536                     rtn != SCSI_MLQUEUE_TARGET_BUSY)
1537                         rtn = SCSI_MLQUEUE_HOST_BUSY;
1538
1539                 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1540                         "queuecommand : request rejected\n"));
1541         }
1542
1543         return rtn;
1544  done:
1545         cmd->scsi_done(cmd);
1546         return 0;
1547 }
1548
1549 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1550 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1551 {
1552         return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1553                 sizeof(struct scatterlist);
1554 }
1555
1556 static blk_status_t scsi_mq_prep_fn(struct request *req)
1557 {
1558         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1559         struct scsi_device *sdev = req->q->queuedata;
1560         struct Scsi_Host *shost = sdev->host;
1561         struct scatterlist *sg;
1562
1563         scsi_init_command(sdev, cmd);
1564
1565         cmd->request = req;
1566         cmd->tag = req->tag;
1567         cmd->prot_op = SCSI_PROT_NORMAL;
1568
1569         sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1570         cmd->sdb.table.sgl = sg;
1571
1572         if (scsi_host_get_prot(shost)) {
1573                 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1574
1575                 cmd->prot_sdb->table.sgl =
1576                         (struct scatterlist *)(cmd->prot_sdb + 1);
1577         }
1578
1579         blk_mq_start_request(req);
1580
1581         return scsi_setup_cmnd(sdev, req);
1582 }
1583
1584 static void scsi_mq_done(struct scsi_cmnd *cmd)
1585 {
1586         if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1587                 return;
1588         trace_scsi_dispatch_cmd_done(cmd);
1589
1590         /*
1591          * If the block layer didn't complete the request due to a timeout
1592          * injection, scsi must clear its internal completed state so that the
1593          * timeout handler will see it needs to escalate its own error
1594          * recovery.
1595          */
1596         if (unlikely(!blk_mq_complete_request(cmd->request)))
1597                 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1598 }
1599
1600 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1601 {
1602         struct request_queue *q = hctx->queue;
1603         struct scsi_device *sdev = q->queuedata;
1604
1605         atomic_dec(&sdev->device_busy);
1606 }
1607
1608 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1609 {
1610         struct request_queue *q = hctx->queue;
1611         struct scsi_device *sdev = q->queuedata;
1612
1613         if (scsi_dev_queue_ready(q, sdev))
1614                 return true;
1615
1616         if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1617                 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1618         return false;
1619 }
1620
1621 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1622                          const struct blk_mq_queue_data *bd)
1623 {
1624         struct request *req = bd->rq;
1625         struct request_queue *q = req->q;
1626         struct scsi_device *sdev = q->queuedata;
1627         struct Scsi_Host *shost = sdev->host;
1628         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1629         blk_status_t ret;
1630         int reason;
1631
1632         /*
1633          * If the device is not in running state we will reject some or all
1634          * commands.
1635          */
1636         if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1637                 ret = scsi_prep_state_check(sdev, req);
1638                 if (ret != BLK_STS_OK)
1639                         goto out_put_budget;
1640         }
1641
1642         ret = BLK_STS_RESOURCE;
1643         if (!scsi_target_queue_ready(shost, sdev))
1644                 goto out_put_budget;
1645         if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1646                 goto out_dec_target_busy;
1647
1648         if (!(req->rq_flags & RQF_DONTPREP)) {
1649                 ret = scsi_mq_prep_fn(req);
1650                 if (ret != BLK_STS_OK)
1651                         goto out_dec_host_busy;
1652                 req->rq_flags |= RQF_DONTPREP;
1653         } else {
1654                 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1655                 blk_mq_start_request(req);
1656         }
1657
1658         cmd->flags &= SCMD_PRESERVED_FLAGS;
1659         if (sdev->simple_tags)
1660                 cmd->flags |= SCMD_TAGGED;
1661         if (bd->last)
1662                 cmd->flags |= SCMD_LAST;
1663
1664         scsi_init_cmd_errh(cmd);
1665         cmd->scsi_done = scsi_mq_done;
1666
1667         reason = scsi_dispatch_cmd(cmd);
1668         if (reason) {
1669                 scsi_set_blocked(cmd, reason);
1670                 ret = BLK_STS_RESOURCE;
1671                 goto out_dec_host_busy;
1672         }
1673
1674         return BLK_STS_OK;
1675
1676 out_dec_host_busy:
1677         scsi_dec_host_busy(shost, cmd);
1678 out_dec_target_busy:
1679         if (scsi_target(sdev)->can_queue > 0)
1680                 atomic_dec(&scsi_target(sdev)->target_busy);
1681 out_put_budget:
1682         scsi_mq_put_budget(hctx);
1683         switch (ret) {
1684         case BLK_STS_OK:
1685                 break;
1686         case BLK_STS_RESOURCE:
1687                 if (atomic_read(&sdev->device_busy) ||
1688                     scsi_device_blocked(sdev))
1689                         ret = BLK_STS_DEV_RESOURCE;
1690                 break;
1691         default:
1692                 if (unlikely(!scsi_device_online(sdev)))
1693                         scsi_req(req)->result = DID_NO_CONNECT << 16;
1694                 else
1695                         scsi_req(req)->result = DID_ERROR << 16;
1696                 /*
1697                  * Make sure to release all allocated resources when
1698                  * we hit an error, as we will never see this command
1699                  * again.
1700                  */
1701                 if (req->rq_flags & RQF_DONTPREP)
1702                         scsi_mq_uninit_cmd(cmd);
1703                 break;
1704         }
1705         return ret;
1706 }
1707
1708 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1709                 bool reserved)
1710 {
1711         if (reserved)
1712                 return BLK_EH_RESET_TIMER;
1713         return scsi_times_out(req);
1714 }
1715
1716 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1717                                 unsigned int hctx_idx, unsigned int numa_node)
1718 {
1719         struct Scsi_Host *shost = set->driver_data;
1720         const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1721         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1722         struct scatterlist *sg;
1723         int ret = 0;
1724
1725         if (unchecked_isa_dma)
1726                 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1727         cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1728                                                     GFP_KERNEL, numa_node);
1729         if (!cmd->sense_buffer)
1730                 return -ENOMEM;
1731         cmd->req.sense = cmd->sense_buffer;
1732
1733         if (scsi_host_get_prot(shost)) {
1734                 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1735                         shost->hostt->cmd_size;
1736                 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1737         }
1738
1739         if (shost->hostt->init_cmd_priv) {
1740                 ret = shost->hostt->init_cmd_priv(shost, cmd);
1741                 if (ret < 0)
1742                         scsi_free_sense_buffer(unchecked_isa_dma,
1743                                                cmd->sense_buffer);
1744         }
1745
1746         return ret;
1747 }
1748
1749 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1750                                  unsigned int hctx_idx)
1751 {
1752         struct Scsi_Host *shost = set->driver_data;
1753         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1754
1755         if (shost->hostt->exit_cmd_priv)
1756                 shost->hostt->exit_cmd_priv(shost, cmd);
1757         scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1758                                cmd->sense_buffer);
1759 }
1760
1761 static int scsi_map_queues(struct blk_mq_tag_set *set)
1762 {
1763         struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1764
1765         if (shost->hostt->map_queues)
1766                 return shost->hostt->map_queues(shost);
1767         return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1768 }
1769
1770 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1771 {
1772         struct device *dev = shost->dma_dev;
1773
1774         /*
1775          * this limit is imposed by hardware restrictions
1776          */
1777         blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1778                                         SG_MAX_SEGMENTS));
1779
1780         if (scsi_host_prot_dma(shost)) {
1781                 shost->sg_prot_tablesize =
1782                         min_not_zero(shost->sg_prot_tablesize,
1783                                      (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1784                 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1785                 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1786         }
1787
1788         if (dev->dma_mask) {
1789                 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1790                                 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1791         }
1792         blk_queue_max_hw_sectors(q, shost->max_sectors);
1793         if (shost->unchecked_isa_dma)
1794                 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1795         blk_queue_segment_boundary(q, shost->dma_boundary);
1796         dma_set_seg_boundary(dev, shost->dma_boundary);
1797
1798         blk_queue_max_segment_size(q, shost->max_segment_size);
1799         blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1800         dma_set_max_seg_size(dev, queue_max_segment_size(q));
1801
1802         /*
1803          * Set a reasonable default alignment:  The larger of 32-byte (dword),
1804          * which is a common minimum for HBAs, and the minimum DMA alignment,
1805          * which is set by the platform.
1806          *
1807          * Devices that require a bigger alignment can increase it later.
1808          */
1809         blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1810 }
1811 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1812
1813 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1814         .get_budget     = scsi_mq_get_budget,
1815         .put_budget     = scsi_mq_put_budget,
1816         .queue_rq       = scsi_queue_rq,
1817         .complete       = scsi_softirq_done,
1818         .timeout        = scsi_timeout,
1819 #ifdef CONFIG_BLK_DEBUG_FS
1820         .show_rq        = scsi_show_rq,
1821 #endif
1822         .init_request   = scsi_mq_init_request,
1823         .exit_request   = scsi_mq_exit_request,
1824         .initialize_rq_fn = scsi_initialize_rq,
1825         .cleanup_rq     = scsi_cleanup_rq,
1826         .busy           = scsi_mq_lld_busy,
1827         .map_queues     = scsi_map_queues,
1828 };
1829
1830
1831 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1832 {
1833         struct request_queue *q = hctx->queue;
1834         struct scsi_device *sdev = q->queuedata;
1835         struct Scsi_Host *shost = sdev->host;
1836
1837         shost->hostt->commit_rqs(shost, hctx->queue_num);
1838 }
1839
1840 static const struct blk_mq_ops scsi_mq_ops = {
1841         .get_budget     = scsi_mq_get_budget,
1842         .put_budget     = scsi_mq_put_budget,
1843         .queue_rq       = scsi_queue_rq,
1844         .commit_rqs     = scsi_commit_rqs,
1845         .complete       = scsi_softirq_done,
1846         .timeout        = scsi_timeout,
1847 #ifdef CONFIG_BLK_DEBUG_FS
1848         .show_rq        = scsi_show_rq,
1849 #endif
1850         .init_request   = scsi_mq_init_request,
1851         .exit_request   = scsi_mq_exit_request,
1852         .initialize_rq_fn = scsi_initialize_rq,
1853         .cleanup_rq     = scsi_cleanup_rq,
1854         .busy           = scsi_mq_lld_busy,
1855         .map_queues     = scsi_map_queues,
1856 };
1857
1858 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1859 {
1860         sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1861         if (IS_ERR(sdev->request_queue))
1862                 return NULL;
1863
1864         sdev->request_queue->queuedata = sdev;
1865         __scsi_init_queue(sdev->host, sdev->request_queue);
1866         blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1867         return sdev->request_queue;
1868 }
1869
1870 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1871 {
1872         unsigned int cmd_size, sgl_size;
1873
1874         sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1875                                 scsi_mq_inline_sgl_size(shost));
1876         cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1877         if (scsi_host_get_prot(shost))
1878                 cmd_size += sizeof(struct scsi_data_buffer) +
1879                         sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1880
1881         memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1882         if (shost->hostt->commit_rqs)
1883                 shost->tag_set.ops = &scsi_mq_ops;
1884         else
1885                 shost->tag_set.ops = &scsi_mq_ops_no_commit;
1886         shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1887         shost->tag_set.queue_depth = shost->can_queue;
1888         shost->tag_set.cmd_size = cmd_size;
1889         shost->tag_set.numa_node = NUMA_NO_NODE;
1890         shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1891         shost->tag_set.flags |=
1892                 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1893         shost->tag_set.driver_data = shost;
1894
1895         return blk_mq_alloc_tag_set(&shost->tag_set);
1896 }
1897
1898 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1899 {
1900         blk_mq_free_tag_set(&shost->tag_set);
1901 }
1902
1903 /**
1904  * scsi_device_from_queue - return sdev associated with a request_queue
1905  * @q: The request queue to return the sdev from
1906  *
1907  * Return the sdev associated with a request queue or NULL if the
1908  * request_queue does not reference a SCSI device.
1909  */
1910 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1911 {
1912         struct scsi_device *sdev = NULL;
1913
1914         if (q->mq_ops == &scsi_mq_ops_no_commit ||
1915             q->mq_ops == &scsi_mq_ops)
1916                 sdev = q->queuedata;
1917         if (!sdev || !get_device(&sdev->sdev_gendev))
1918                 sdev = NULL;
1919
1920         return sdev;
1921 }
1922 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1923
1924 /*
1925  * Function:    scsi_block_requests()
1926  *
1927  * Purpose:     Utility function used by low-level drivers to prevent further
1928  *              commands from being queued to the device.
1929  *
1930  * Arguments:   shost       - Host in question
1931  *
1932  * Returns:     Nothing
1933  *
1934  * Lock status: No locks are assumed held.
1935  *
1936  * Notes:       There is no timer nor any other means by which the requests
1937  *              get unblocked other than the low-level driver calling
1938  *              scsi_unblock_requests().
1939  */
1940 void scsi_block_requests(struct Scsi_Host *shost)
1941 {
1942         shost->host_self_blocked = 1;
1943 }
1944 EXPORT_SYMBOL(scsi_block_requests);
1945
1946 /*
1947  * Function:    scsi_unblock_requests()
1948  *
1949  * Purpose:     Utility function used by low-level drivers to allow further
1950  *              commands from being queued to the device.
1951  *
1952  * Arguments:   shost       - Host in question
1953  *
1954  * Returns:     Nothing
1955  *
1956  * Lock status: No locks are assumed held.
1957  *
1958  * Notes:       There is no timer nor any other means by which the requests
1959  *              get unblocked other than the low-level driver calling
1960  *              scsi_unblock_requests().
1961  *
1962  *              This is done as an API function so that changes to the
1963  *              internals of the scsi mid-layer won't require wholesale
1964  *              changes to drivers that use this feature.
1965  */
1966 void scsi_unblock_requests(struct Scsi_Host *shost)
1967 {
1968         shost->host_self_blocked = 0;
1969         scsi_run_host_queues(shost);
1970 }
1971 EXPORT_SYMBOL(scsi_unblock_requests);
1972
1973 int __init scsi_init_queue(void)
1974 {
1975         scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1976                                            sizeof(struct scsi_data_buffer),
1977                                            0, 0, NULL);
1978         if (!scsi_sdb_cache) {
1979                 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1980                 return -ENOMEM;
1981         }
1982
1983         return 0;
1984 }
1985
1986 void scsi_exit_queue(void)
1987 {
1988         kmem_cache_destroy(scsi_sense_cache);
1989         kmem_cache_destroy(scsi_sense_isadma_cache);
1990         kmem_cache_destroy(scsi_sdb_cache);
1991 }
1992
1993 /**
1994  *      scsi_mode_select - issue a mode select
1995  *      @sdev:  SCSI device to be queried
1996  *      @pf:    Page format bit (1 == standard, 0 == vendor specific)
1997  *      @sp:    Save page bit (0 == don't save, 1 == save)
1998  *      @modepage: mode page being requested
1999  *      @buffer: request buffer (may not be smaller than eight bytes)
2000  *      @len:   length of request buffer.
2001  *      @timeout: command timeout
2002  *      @retries: number of retries before failing
2003  *      @data: returns a structure abstracting the mode header data
2004  *      @sshdr: place to put sense data (or NULL if no sense to be collected).
2005  *              must be SCSI_SENSE_BUFFERSIZE big.
2006  *
2007  *      Returns zero if successful; negative error number or scsi
2008  *      status on error
2009  *
2010  */
2011 int
2012 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2013                  unsigned char *buffer, int len, int timeout, int retries,
2014                  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2015 {
2016         unsigned char cmd[10];
2017         unsigned char *real_buffer;
2018         int ret;
2019
2020         memset(cmd, 0, sizeof(cmd));
2021         cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2022
2023         if (sdev->use_10_for_ms) {
2024                 if (len > 65535)
2025                         return -EINVAL;
2026                 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2027                 if (!real_buffer)
2028                         return -ENOMEM;
2029                 memcpy(real_buffer + 8, buffer, len);
2030                 len += 8;
2031                 real_buffer[0] = 0;
2032                 real_buffer[1] = 0;
2033                 real_buffer[2] = data->medium_type;
2034                 real_buffer[3] = data->device_specific;
2035                 real_buffer[4] = data->longlba ? 0x01 : 0;
2036                 real_buffer[5] = 0;
2037                 real_buffer[6] = data->block_descriptor_length >> 8;
2038                 real_buffer[7] = data->block_descriptor_length;
2039
2040                 cmd[0] = MODE_SELECT_10;
2041                 cmd[7] = len >> 8;
2042                 cmd[8] = len;
2043         } else {
2044                 if (len > 255 || data->block_descriptor_length > 255 ||
2045                     data->longlba)
2046                         return -EINVAL;
2047
2048                 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2049                 if (!real_buffer)
2050                         return -ENOMEM;
2051                 memcpy(real_buffer + 4, buffer, len);
2052                 len += 4;
2053                 real_buffer[0] = 0;
2054                 real_buffer[1] = data->medium_type;
2055                 real_buffer[2] = data->device_specific;
2056                 real_buffer[3] = data->block_descriptor_length;
2057                 
2058
2059                 cmd[0] = MODE_SELECT;
2060                 cmd[4] = len;
2061         }
2062
2063         ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2064                                sshdr, timeout, retries, NULL);
2065         kfree(real_buffer);
2066         return ret;
2067 }
2068 EXPORT_SYMBOL_GPL(scsi_mode_select);
2069
2070 /**
2071  *      scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2072  *      @sdev:  SCSI device to be queried
2073  *      @dbd:   set if mode sense will allow block descriptors to be returned
2074  *      @modepage: mode page being requested
2075  *      @buffer: request buffer (may not be smaller than eight bytes)
2076  *      @len:   length of request buffer.
2077  *      @timeout: command timeout
2078  *      @retries: number of retries before failing
2079  *      @data: returns a structure abstracting the mode header data
2080  *      @sshdr: place to put sense data (or NULL if no sense to be collected).
2081  *              must be SCSI_SENSE_BUFFERSIZE big.
2082  *
2083  *      Returns zero if unsuccessful, or the header offset (either 4
2084  *      or 8 depending on whether a six or ten byte command was
2085  *      issued) if successful.
2086  */
2087 int
2088 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2089                   unsigned char *buffer, int len, int timeout, int retries,
2090                   struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2091 {
2092         unsigned char cmd[12];
2093         int use_10_for_ms;
2094         int header_length;
2095         int result, retry_count = retries;
2096         struct scsi_sense_hdr my_sshdr;
2097
2098         memset(data, 0, sizeof(*data));
2099         memset(&cmd[0], 0, 12);
2100
2101         dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2102         cmd[1] = dbd & 0x18;    /* allows DBD and LLBA bits */
2103         cmd[2] = modepage;
2104
2105         /* caller might not be interested in sense, but we need it */
2106         if (!sshdr)
2107                 sshdr = &my_sshdr;
2108
2109  retry:
2110         use_10_for_ms = sdev->use_10_for_ms;
2111
2112         if (use_10_for_ms) {
2113                 if (len < 8)
2114                         len = 8;
2115
2116                 cmd[0] = MODE_SENSE_10;
2117                 cmd[8] = len;
2118                 header_length = 8;
2119         } else {
2120                 if (len < 4)
2121                         len = 4;
2122
2123                 cmd[0] = MODE_SENSE;
2124                 cmd[4] = len;
2125                 header_length = 4;
2126         }
2127
2128         memset(buffer, 0, len);
2129
2130         result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2131                                   sshdr, timeout, retries, NULL);
2132
2133         /* This code looks awful: what it's doing is making sure an
2134          * ILLEGAL REQUEST sense return identifies the actual command
2135          * byte as the problem.  MODE_SENSE commands can return
2136          * ILLEGAL REQUEST if the code page isn't supported */
2137
2138         if (use_10_for_ms && !scsi_status_is_good(result) &&
2139             driver_byte(result) == DRIVER_SENSE) {
2140                 if (scsi_sense_valid(sshdr)) {
2141                         if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2142                             (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2143                                 /* 
2144                                  * Invalid command operation code
2145                                  */
2146                                 sdev->use_10_for_ms = 0;
2147                                 goto retry;
2148                         }
2149                 }
2150         }
2151
2152         if(scsi_status_is_good(result)) {
2153                 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2154                              (modepage == 6 || modepage == 8))) {
2155                         /* Initio breakage? */
2156                         header_length = 0;
2157                         data->length = 13;
2158                         data->medium_type = 0;
2159                         data->device_specific = 0;
2160                         data->longlba = 0;
2161                         data->block_descriptor_length = 0;
2162                 } else if(use_10_for_ms) {
2163                         data->length = buffer[0]*256 + buffer[1] + 2;
2164                         data->medium_type = buffer[2];
2165                         data->device_specific = buffer[3];
2166                         data->longlba = buffer[4] & 0x01;
2167                         data->block_descriptor_length = buffer[6]*256
2168                                 + buffer[7];
2169                 } else {
2170                         data->length = buffer[0] + 1;
2171                         data->medium_type = buffer[1];
2172                         data->device_specific = buffer[2];
2173                         data->block_descriptor_length = buffer[3];
2174                 }
2175                 data->header_length = header_length;
2176         } else if ((status_byte(result) == CHECK_CONDITION) &&
2177                    scsi_sense_valid(sshdr) &&
2178                    sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2179                 retry_count--;
2180                 goto retry;
2181         }
2182
2183         return result;
2184 }
2185 EXPORT_SYMBOL(scsi_mode_sense);
2186
2187 /**
2188  *      scsi_test_unit_ready - test if unit is ready
2189  *      @sdev:  scsi device to change the state of.
2190  *      @timeout: command timeout
2191  *      @retries: number of retries before failing
2192  *      @sshdr: outpout pointer for decoded sense information.
2193  *
2194  *      Returns zero if unsuccessful or an error if TUR failed.  For
2195  *      removable media, UNIT_ATTENTION sets ->changed flag.
2196  **/
2197 int
2198 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2199                      struct scsi_sense_hdr *sshdr)
2200 {
2201         char cmd[] = {
2202                 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2203         };
2204         int result;
2205
2206         /* try to eat the UNIT_ATTENTION if there are enough retries */
2207         do {
2208                 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2209                                           timeout, 1, NULL);
2210                 if (sdev->removable && scsi_sense_valid(sshdr) &&
2211                     sshdr->sense_key == UNIT_ATTENTION)
2212                         sdev->changed = 1;
2213         } while (scsi_sense_valid(sshdr) &&
2214                  sshdr->sense_key == UNIT_ATTENTION && --retries);
2215
2216         return result;
2217 }
2218 EXPORT_SYMBOL(scsi_test_unit_ready);
2219
2220 /**
2221  *      scsi_device_set_state - Take the given device through the device state model.
2222  *      @sdev:  scsi device to change the state of.
2223  *      @state: state to change to.
2224  *
2225  *      Returns zero if successful or an error if the requested
2226  *      transition is illegal.
2227  */
2228 int
2229 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2230 {
2231         enum scsi_device_state oldstate = sdev->sdev_state;
2232
2233         if (state == oldstate)
2234                 return 0;
2235
2236         switch (state) {
2237         case SDEV_CREATED:
2238                 switch (oldstate) {
2239                 case SDEV_CREATED_BLOCK:
2240                         break;
2241                 default:
2242                         goto illegal;
2243                 }
2244                 break;
2245                         
2246         case SDEV_RUNNING:
2247                 switch (oldstate) {
2248                 case SDEV_CREATED:
2249                 case SDEV_OFFLINE:
2250                 case SDEV_TRANSPORT_OFFLINE:
2251                 case SDEV_QUIESCE:
2252                 case SDEV_BLOCK:
2253                         break;
2254                 default:
2255                         goto illegal;
2256                 }
2257                 break;
2258
2259         case SDEV_QUIESCE:
2260                 switch (oldstate) {
2261                 case SDEV_RUNNING:
2262                 case SDEV_OFFLINE:
2263                 case SDEV_TRANSPORT_OFFLINE:
2264                         break;
2265                 default:
2266                         goto illegal;
2267                 }
2268                 break;
2269
2270         case SDEV_OFFLINE:
2271         case SDEV_TRANSPORT_OFFLINE:
2272                 switch (oldstate) {
2273                 case SDEV_CREATED:
2274                 case SDEV_RUNNING:
2275                 case SDEV_QUIESCE:
2276                 case SDEV_BLOCK:
2277                         break;
2278                 default:
2279                         goto illegal;
2280                 }
2281                 break;
2282
2283         case SDEV_BLOCK:
2284                 switch (oldstate) {
2285                 case SDEV_RUNNING:
2286                 case SDEV_CREATED_BLOCK:
2287                 case SDEV_OFFLINE:
2288                         break;
2289                 default:
2290                         goto illegal;
2291                 }
2292                 break;
2293
2294         case SDEV_CREATED_BLOCK:
2295                 switch (oldstate) {
2296                 case SDEV_CREATED:
2297                         break;
2298                 default:
2299                         goto illegal;
2300                 }
2301                 break;
2302
2303         case SDEV_CANCEL:
2304                 switch (oldstate) {
2305                 case SDEV_CREATED:
2306                 case SDEV_RUNNING:
2307                 case SDEV_QUIESCE:
2308                 case SDEV_OFFLINE:
2309                 case SDEV_TRANSPORT_OFFLINE:
2310                         break;
2311                 default:
2312                         goto illegal;
2313                 }
2314                 break;
2315
2316         case SDEV_DEL:
2317                 switch (oldstate) {
2318                 case SDEV_CREATED:
2319                 case SDEV_RUNNING:
2320                 case SDEV_OFFLINE:
2321                 case SDEV_TRANSPORT_OFFLINE:
2322                 case SDEV_CANCEL:
2323                 case SDEV_BLOCK:
2324                 case SDEV_CREATED_BLOCK:
2325                         break;
2326                 default:
2327                         goto illegal;
2328                 }
2329                 break;
2330
2331         }
2332         sdev->offline_already = false;
2333         sdev->sdev_state = state;
2334         return 0;
2335
2336  illegal:
2337         SCSI_LOG_ERROR_RECOVERY(1,
2338                                 sdev_printk(KERN_ERR, sdev,
2339                                             "Illegal state transition %s->%s",
2340                                             scsi_device_state_name(oldstate),
2341                                             scsi_device_state_name(state))
2342                                 );
2343         return -EINVAL;
2344 }
2345 EXPORT_SYMBOL(scsi_device_set_state);
2346
2347 /**
2348  *      sdev_evt_emit - emit a single SCSI device uevent
2349  *      @sdev: associated SCSI device
2350  *      @evt: event to emit
2351  *
2352  *      Send a single uevent (scsi_event) to the associated scsi_device.
2353  */
2354 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2355 {
2356         int idx = 0;
2357         char *envp[3];
2358
2359         switch (evt->evt_type) {
2360         case SDEV_EVT_MEDIA_CHANGE:
2361                 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2362                 break;
2363         case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2364                 scsi_rescan_device(&sdev->sdev_gendev);
2365                 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2366                 break;
2367         case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2368                 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2369                 break;
2370         case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2371                envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2372                 break;
2373         case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2374                 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2375                 break;
2376         case SDEV_EVT_LUN_CHANGE_REPORTED:
2377                 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2378                 break;
2379         case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2380                 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2381                 break;
2382         case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2383                 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2384                 break;
2385         default:
2386                 /* do nothing */
2387                 break;
2388         }
2389
2390         envp[idx++] = NULL;
2391
2392         kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2393 }
2394
2395 /**
2396  *      sdev_evt_thread - send a uevent for each scsi event
2397  *      @work: work struct for scsi_device
2398  *
2399  *      Dispatch queued events to their associated scsi_device kobjects
2400  *      as uevents.
2401  */
2402 void scsi_evt_thread(struct work_struct *work)
2403 {
2404         struct scsi_device *sdev;
2405         enum scsi_device_event evt_type;
2406         LIST_HEAD(event_list);
2407
2408         sdev = container_of(work, struct scsi_device, event_work);
2409
2410         for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2411                 if (test_and_clear_bit(evt_type, sdev->pending_events))
2412                         sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2413
2414         while (1) {
2415                 struct scsi_event *evt;
2416                 struct list_head *this, *tmp;
2417                 unsigned long flags;
2418
2419                 spin_lock_irqsave(&sdev->list_lock, flags);
2420                 list_splice_init(&sdev->event_list, &event_list);
2421                 spin_unlock_irqrestore(&sdev->list_lock, flags);
2422
2423                 if (list_empty(&event_list))
2424                         break;
2425
2426                 list_for_each_safe(this, tmp, &event_list) {
2427                         evt = list_entry(this, struct scsi_event, node);
2428                         list_del(&evt->node);
2429                         scsi_evt_emit(sdev, evt);
2430                         kfree(evt);
2431                 }
2432         }
2433 }
2434
2435 /**
2436  *      sdev_evt_send - send asserted event to uevent thread
2437  *      @sdev: scsi_device event occurred on
2438  *      @evt: event to send
2439  *
2440  *      Assert scsi device event asynchronously.
2441  */
2442 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2443 {
2444         unsigned long flags;
2445
2446 #if 0
2447         /* FIXME: currently this check eliminates all media change events
2448          * for polled devices.  Need to update to discriminate between AN
2449          * and polled events */
2450         if (!test_bit(evt->evt_type, sdev->supported_events)) {
2451                 kfree(evt);
2452                 return;
2453         }
2454 #endif
2455
2456         spin_lock_irqsave(&sdev->list_lock, flags);
2457         list_add_tail(&evt->node, &sdev->event_list);
2458         schedule_work(&sdev->event_work);
2459         spin_unlock_irqrestore(&sdev->list_lock, flags);
2460 }
2461 EXPORT_SYMBOL_GPL(sdev_evt_send);
2462
2463 /**
2464  *      sdev_evt_alloc - allocate a new scsi event
2465  *      @evt_type: type of event to allocate
2466  *      @gfpflags: GFP flags for allocation
2467  *
2468  *      Allocates and returns a new scsi_event.
2469  */
2470 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2471                                   gfp_t gfpflags)
2472 {
2473         struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2474         if (!evt)
2475                 return NULL;
2476
2477         evt->evt_type = evt_type;
2478         INIT_LIST_HEAD(&evt->node);
2479
2480         /* evt_type-specific initialization, if any */
2481         switch (evt_type) {
2482         case SDEV_EVT_MEDIA_CHANGE:
2483         case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2484         case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2485         case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2486         case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2487         case SDEV_EVT_LUN_CHANGE_REPORTED:
2488         case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2489         case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2490         default:
2491                 /* do nothing */
2492                 break;
2493         }
2494
2495         return evt;
2496 }
2497 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2498
2499 /**
2500  *      sdev_evt_send_simple - send asserted event to uevent thread
2501  *      @sdev: scsi_device event occurred on
2502  *      @evt_type: type of event to send
2503  *      @gfpflags: GFP flags for allocation
2504  *
2505  *      Assert scsi device event asynchronously, given an event type.
2506  */
2507 void sdev_evt_send_simple(struct scsi_device *sdev,
2508                           enum scsi_device_event evt_type, gfp_t gfpflags)
2509 {
2510         struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2511         if (!evt) {
2512                 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2513                             evt_type);
2514                 return;
2515         }
2516
2517         sdev_evt_send(sdev, evt);
2518 }
2519 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2520
2521 /**
2522  *      scsi_device_quiesce - Block user issued commands.
2523  *      @sdev:  scsi device to quiesce.
2524  *
2525  *      This works by trying to transition to the SDEV_QUIESCE state
2526  *      (which must be a legal transition).  When the device is in this
2527  *      state, only special requests will be accepted, all others will
2528  *      be deferred.  Since special requests may also be requeued requests,
2529  *      a successful return doesn't guarantee the device will be 
2530  *      totally quiescent.
2531  *
2532  *      Must be called with user context, may sleep.
2533  *
2534  *      Returns zero if unsuccessful or an error if not.
2535  */
2536 int
2537 scsi_device_quiesce(struct scsi_device *sdev)
2538 {
2539         struct request_queue *q = sdev->request_queue;
2540         int err;
2541
2542         /*
2543          * It is allowed to call scsi_device_quiesce() multiple times from
2544          * the same context but concurrent scsi_device_quiesce() calls are
2545          * not allowed.
2546          */
2547         WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2548
2549         if (sdev->quiesced_by == current)
2550                 return 0;
2551
2552         blk_set_pm_only(q);
2553
2554         blk_mq_freeze_queue(q);
2555         /*
2556          * Ensure that the effect of blk_set_pm_only() will be visible
2557          * for percpu_ref_tryget() callers that occur after the queue
2558          * unfreeze even if the queue was already frozen before this function
2559          * was called. See also https://lwn.net/Articles/573497/.
2560          */
2561         synchronize_rcu();
2562         blk_mq_unfreeze_queue(q);
2563
2564         mutex_lock(&sdev->state_mutex);
2565         err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2566         if (err == 0)
2567                 sdev->quiesced_by = current;
2568         else
2569                 blk_clear_pm_only(q);
2570         mutex_unlock(&sdev->state_mutex);
2571
2572         return err;
2573 }
2574 EXPORT_SYMBOL(scsi_device_quiesce);
2575
2576 /**
2577  *      scsi_device_resume - Restart user issued commands to a quiesced device.
2578  *      @sdev:  scsi device to resume.
2579  *
2580  *      Moves the device from quiesced back to running and restarts the
2581  *      queues.
2582  *
2583  *      Must be called with user context, may sleep.
2584  */
2585 void scsi_device_resume(struct scsi_device *sdev)
2586 {
2587         /* check if the device state was mutated prior to resume, and if
2588          * so assume the state is being managed elsewhere (for example
2589          * device deleted during suspend)
2590          */
2591         mutex_lock(&sdev->state_mutex);
2592         if (sdev->quiesced_by) {
2593                 sdev->quiesced_by = NULL;
2594                 blk_clear_pm_only(sdev->request_queue);
2595         }
2596         if (sdev->sdev_state == SDEV_QUIESCE)
2597                 scsi_device_set_state(sdev, SDEV_RUNNING);
2598         mutex_unlock(&sdev->state_mutex);
2599 }
2600 EXPORT_SYMBOL(scsi_device_resume);
2601
2602 static void
2603 device_quiesce_fn(struct scsi_device *sdev, void *data)
2604 {
2605         scsi_device_quiesce(sdev);
2606 }
2607
2608 void
2609 scsi_target_quiesce(struct scsi_target *starget)
2610 {
2611         starget_for_each_device(starget, NULL, device_quiesce_fn);
2612 }
2613 EXPORT_SYMBOL(scsi_target_quiesce);
2614
2615 static void
2616 device_resume_fn(struct scsi_device *sdev, void *data)
2617 {
2618         scsi_device_resume(sdev);
2619 }
2620
2621 void
2622 scsi_target_resume(struct scsi_target *starget)
2623 {
2624         starget_for_each_device(starget, NULL, device_resume_fn);
2625 }
2626 EXPORT_SYMBOL(scsi_target_resume);
2627
2628 /**
2629  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2630  * @sdev: device to block
2631  *
2632  * Pause SCSI command processing on the specified device. Does not sleep.
2633  *
2634  * Returns zero if successful or a negative error code upon failure.
2635  *
2636  * Notes:
2637  * This routine transitions the device to the SDEV_BLOCK state (which must be
2638  * a legal transition). When the device is in this state, command processing
2639  * is paused until the device leaves the SDEV_BLOCK state. See also
2640  * scsi_internal_device_unblock_nowait().
2641  */
2642 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2643 {
2644         struct request_queue *q = sdev->request_queue;
2645         int err = 0;
2646
2647         err = scsi_device_set_state(sdev, SDEV_BLOCK);
2648         if (err) {
2649                 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2650
2651                 if (err)
2652                         return err;
2653         }
2654
2655         /* 
2656          * The device has transitioned to SDEV_BLOCK.  Stop the
2657          * block layer from calling the midlayer with this device's
2658          * request queue. 
2659          */
2660         blk_mq_quiesce_queue_nowait(q);
2661         return 0;
2662 }
2663 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2664
2665 /**
2666  * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2667  * @sdev: device to block
2668  *
2669  * Pause SCSI command processing on the specified device and wait until all
2670  * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2671  *
2672  * Returns zero if successful or a negative error code upon failure.
2673  *
2674  * Note:
2675  * This routine transitions the device to the SDEV_BLOCK state (which must be
2676  * a legal transition). When the device is in this state, command processing
2677  * is paused until the device leaves the SDEV_BLOCK state. See also
2678  * scsi_internal_device_unblock().
2679  */
2680 static int scsi_internal_device_block(struct scsi_device *sdev)
2681 {
2682         struct request_queue *q = sdev->request_queue;
2683         int err;
2684
2685         mutex_lock(&sdev->state_mutex);
2686         err = scsi_internal_device_block_nowait(sdev);
2687         if (err == 0)
2688                 blk_mq_quiesce_queue(q);
2689         mutex_unlock(&sdev->state_mutex);
2690
2691         return err;
2692 }
2693  
2694 void scsi_start_queue(struct scsi_device *sdev)
2695 {
2696         struct request_queue *q = sdev->request_queue;
2697
2698         blk_mq_unquiesce_queue(q);
2699 }
2700
2701 /**
2702  * scsi_internal_device_unblock_nowait - resume a device after a block request
2703  * @sdev:       device to resume
2704  * @new_state:  state to set the device to after unblocking
2705  *
2706  * Restart the device queue for a previously suspended SCSI device. Does not
2707  * sleep.
2708  *
2709  * Returns zero if successful or a negative error code upon failure.
2710  *
2711  * Notes:
2712  * This routine transitions the device to the SDEV_RUNNING state or to one of
2713  * the offline states (which must be a legal transition) allowing the midlayer
2714  * to goose the queue for this device.
2715  */
2716 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2717                                         enum scsi_device_state new_state)
2718 {
2719         switch (new_state) {
2720         case SDEV_RUNNING:
2721         case SDEV_TRANSPORT_OFFLINE:
2722                 break;
2723         default:
2724                 return -EINVAL;
2725         }
2726
2727         /*
2728          * Try to transition the scsi device to SDEV_RUNNING or one of the
2729          * offlined states and goose the device queue if successful.
2730          */
2731         switch (sdev->sdev_state) {
2732         case SDEV_BLOCK:
2733         case SDEV_TRANSPORT_OFFLINE:
2734                 sdev->sdev_state = new_state;
2735                 break;
2736         case SDEV_CREATED_BLOCK:
2737                 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2738                     new_state == SDEV_OFFLINE)
2739                         sdev->sdev_state = new_state;
2740                 else
2741                         sdev->sdev_state = SDEV_CREATED;
2742                 break;
2743         case SDEV_CANCEL:
2744         case SDEV_OFFLINE:
2745                 break;
2746         default:
2747                 return -EINVAL;
2748         }
2749         scsi_start_queue(sdev);
2750
2751         return 0;
2752 }
2753 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2754
2755 /**
2756  * scsi_internal_device_unblock - resume a device after a block request
2757  * @sdev:       device to resume
2758  * @new_state:  state to set the device to after unblocking
2759  *
2760  * Restart the device queue for a previously suspended SCSI device. May sleep.
2761  *
2762  * Returns zero if successful or a negative error code upon failure.
2763  *
2764  * Notes:
2765  * This routine transitions the device to the SDEV_RUNNING state or to one of
2766  * the offline states (which must be a legal transition) allowing the midlayer
2767  * to goose the queue for this device.
2768  */
2769 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2770                                         enum scsi_device_state new_state)
2771 {
2772         int ret;
2773
2774         mutex_lock(&sdev->state_mutex);
2775         ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2776         mutex_unlock(&sdev->state_mutex);
2777
2778         return ret;
2779 }
2780
2781 static void
2782 device_block(struct scsi_device *sdev, void *data)
2783 {
2784         int ret;
2785
2786         ret = scsi_internal_device_block(sdev);
2787
2788         WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2789                   dev_name(&sdev->sdev_gendev), ret);
2790 }
2791
2792 static int
2793 target_block(struct device *dev, void *data)
2794 {
2795         if (scsi_is_target_device(dev))
2796                 starget_for_each_device(to_scsi_target(dev), NULL,
2797                                         device_block);
2798         return 0;
2799 }
2800
2801 void
2802 scsi_target_block(struct device *dev)
2803 {
2804         if (scsi_is_target_device(dev))
2805                 starget_for_each_device(to_scsi_target(dev), NULL,
2806                                         device_block);
2807         else
2808                 device_for_each_child(dev, NULL, target_block);
2809 }
2810 EXPORT_SYMBOL_GPL(scsi_target_block);
2811
2812 static void
2813 device_unblock(struct scsi_device *sdev, void *data)
2814 {
2815         scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2816 }
2817
2818 static int
2819 target_unblock(struct device *dev, void *data)
2820 {
2821         if (scsi_is_target_device(dev))
2822                 starget_for_each_device(to_scsi_target(dev), data,
2823                                         device_unblock);
2824         return 0;
2825 }
2826
2827 void
2828 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2829 {
2830         if (scsi_is_target_device(dev))
2831                 starget_for_each_device(to_scsi_target(dev), &new_state,
2832                                         device_unblock);
2833         else
2834                 device_for_each_child(dev, &new_state, target_unblock);
2835 }
2836 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2837
2838 int
2839 scsi_host_block(struct Scsi_Host *shost)
2840 {
2841         struct scsi_device *sdev;
2842         int ret = 0;
2843
2844         shost_for_each_device(sdev, shost) {
2845                 ret = scsi_internal_device_block(sdev);
2846                 if (ret)
2847                         break;
2848         }
2849         return ret;
2850 }
2851 EXPORT_SYMBOL_GPL(scsi_host_block);
2852
2853 int
2854 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2855 {
2856         struct scsi_device *sdev;
2857         int ret = 0;
2858
2859         shost_for_each_device(sdev, shost) {
2860                 ret = scsi_internal_device_unblock(sdev, new_state);
2861                 if (ret)
2862                         break;
2863         }
2864         return ret;
2865 }
2866 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2867
2868 /**
2869  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2870  * @sgl:        scatter-gather list
2871  * @sg_count:   number of segments in sg
2872  * @offset:     offset in bytes into sg, on return offset into the mapped area
2873  * @len:        bytes to map, on return number of bytes mapped
2874  *
2875  * Returns virtual address of the start of the mapped page
2876  */
2877 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2878                           size_t *offset, size_t *len)
2879 {
2880         int i;
2881         size_t sg_len = 0, len_complete = 0;
2882         struct scatterlist *sg;
2883         struct page *page;
2884
2885         WARN_ON(!irqs_disabled());
2886
2887         for_each_sg(sgl, sg, sg_count, i) {
2888                 len_complete = sg_len; /* Complete sg-entries */
2889                 sg_len += sg->length;
2890                 if (sg_len > *offset)
2891                         break;
2892         }
2893
2894         if (unlikely(i == sg_count)) {
2895                 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2896                         "elements %d\n",
2897                        __func__, sg_len, *offset, sg_count);
2898                 WARN_ON(1);
2899                 return NULL;
2900         }
2901
2902         /* Offset starting from the beginning of first page in this sg-entry */
2903         *offset = *offset - len_complete + sg->offset;
2904
2905         /* Assumption: contiguous pages can be accessed as "page + i" */
2906         page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2907         *offset &= ~PAGE_MASK;
2908
2909         /* Bytes in this sg-entry from *offset to the end of the page */
2910         sg_len = PAGE_SIZE - *offset;
2911         if (*len > sg_len)
2912                 *len = sg_len;
2913
2914         return kmap_atomic(page);
2915 }
2916 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2917
2918 /**
2919  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2920  * @virt:       virtual address to be unmapped
2921  */
2922 void scsi_kunmap_atomic_sg(void *virt)
2923 {
2924         kunmap_atomic(virt);
2925 }
2926 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2927
2928 void sdev_disable_disk_events(struct scsi_device *sdev)
2929 {
2930         atomic_inc(&sdev->disk_events_disable_depth);
2931 }
2932 EXPORT_SYMBOL(sdev_disable_disk_events);
2933
2934 void sdev_enable_disk_events(struct scsi_device *sdev)
2935 {
2936         if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2937                 return;
2938         atomic_dec(&sdev->disk_events_disable_depth);
2939 }
2940 EXPORT_SYMBOL(sdev_enable_disk_events);
2941
2942 /**
2943  * scsi_vpd_lun_id - return a unique device identification
2944  * @sdev: SCSI device
2945  * @id:   buffer for the identification
2946  * @id_len:  length of the buffer
2947  *
2948  * Copies a unique device identification into @id based
2949  * on the information in the VPD page 0x83 of the device.
2950  * The string will be formatted as a SCSI name string.
2951  *
2952  * Returns the length of the identification or error on failure.
2953  * If the identifier is longer than the supplied buffer the actual
2954  * identifier length is returned and the buffer is not zero-padded.
2955  */
2956 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2957 {
2958         u8 cur_id_type = 0xff;
2959         u8 cur_id_size = 0;
2960         const unsigned char *d, *cur_id_str;
2961         const struct scsi_vpd *vpd_pg83;
2962         int id_size = -EINVAL;
2963
2964         rcu_read_lock();
2965         vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2966         if (!vpd_pg83) {
2967                 rcu_read_unlock();
2968                 return -ENXIO;
2969         }
2970
2971         /*
2972          * Look for the correct descriptor.
2973          * Order of preference for lun descriptor:
2974          * - SCSI name string
2975          * - NAA IEEE Registered Extended
2976          * - EUI-64 based 16-byte
2977          * - EUI-64 based 12-byte
2978          * - NAA IEEE Registered
2979          * - NAA IEEE Extended
2980          * - T10 Vendor ID
2981          * as longer descriptors reduce the likelyhood
2982          * of identification clashes.
2983          */
2984
2985         /* The id string must be at least 20 bytes + terminating NULL byte */
2986         if (id_len < 21) {
2987                 rcu_read_unlock();
2988                 return -EINVAL;
2989         }
2990
2991         memset(id, 0, id_len);
2992         d = vpd_pg83->data + 4;
2993         while (d < vpd_pg83->data + vpd_pg83->len) {
2994                 /* Skip designators not referring to the LUN */
2995                 if ((d[1] & 0x30) != 0x00)
2996                         goto next_desig;
2997
2998                 switch (d[1] & 0xf) {
2999                 case 0x1:
3000                         /* T10 Vendor ID */
3001                         if (cur_id_size > d[3])
3002                                 break;
3003                         /* Prefer anything */
3004                         if (cur_id_type > 0x01 && cur_id_type != 0xff)
3005                                 break;
3006                         cur_id_size = d[3];
3007                         if (cur_id_size + 4 > id_len)
3008                                 cur_id_size = id_len - 4;
3009                         cur_id_str = d + 4;
3010                         cur_id_type = d[1] & 0xf;
3011                         id_size = snprintf(id, id_len, "t10.%*pE",
3012                                            cur_id_size, cur_id_str);
3013                         break;
3014                 case 0x2:
3015                         /* EUI-64 */
3016                         if (cur_id_size > d[3])
3017                                 break;
3018                         /* Prefer NAA IEEE Registered Extended */
3019                         if (cur_id_type == 0x3 &&
3020                             cur_id_size == d[3])
3021                                 break;
3022                         cur_id_size = d[3];
3023                         cur_id_str = d + 4;
3024                         cur_id_type = d[1] & 0xf;
3025                         switch (cur_id_size) {
3026                         case 8:
3027                                 id_size = snprintf(id, id_len,
3028                                                    "eui.%8phN",
3029                                                    cur_id_str);
3030                                 break;
3031                         case 12:
3032                                 id_size = snprintf(id, id_len,
3033                                                    "eui.%12phN",
3034                                                    cur_id_str);
3035                                 break;
3036                         case 16:
3037                                 id_size = snprintf(id, id_len,
3038                                                    "eui.%16phN",
3039                                                    cur_id_str);
3040                                 break;
3041                         default:
3042                                 cur_id_size = 0;
3043                                 break;
3044                         }
3045                         break;
3046                 case 0x3:
3047                         /* NAA */
3048                         if (cur_id_size > d[3])
3049                                 break;
3050                         cur_id_size = d[3];
3051                         cur_id_str = d + 4;
3052                         cur_id_type = d[1] & 0xf;
3053                         switch (cur_id_size) {
3054                         case 8:
3055                                 id_size = snprintf(id, id_len,
3056                                                    "naa.%8phN",
3057                                                    cur_id_str);
3058                                 break;
3059                         case 16:
3060                                 id_size = snprintf(id, id_len,
3061                                                    "naa.%16phN",
3062                                                    cur_id_str);
3063                                 break;
3064                         default:
3065                                 cur_id_size = 0;
3066                                 break;
3067                         }
3068                         break;
3069                 case 0x8:
3070                         /* SCSI name string */
3071                         if (cur_id_size + 4 > d[3])
3072                                 break;
3073                         /* Prefer others for truncated descriptor */
3074                         if (cur_id_size && d[3] > id_len)
3075                                 break;
3076                         cur_id_size = id_size = d[3];
3077                         cur_id_str = d + 4;
3078                         cur_id_type = d[1] & 0xf;
3079                         if (cur_id_size >= id_len)
3080                                 cur_id_size = id_len - 1;
3081                         memcpy(id, cur_id_str, cur_id_size);
3082                         /* Decrease priority for truncated descriptor */
3083                         if (cur_id_size != id_size)
3084                                 cur_id_size = 6;
3085                         break;
3086                 default:
3087                         break;
3088                 }
3089 next_desig:
3090                 d += d[3] + 4;
3091         }
3092         rcu_read_unlock();
3093
3094         return id_size;
3095 }
3096 EXPORT_SYMBOL(scsi_vpd_lun_id);
3097
3098 /*
3099  * scsi_vpd_tpg_id - return a target port group identifier
3100  * @sdev: SCSI device
3101  *
3102  * Returns the Target Port Group identifier from the information
3103  * froom VPD page 0x83 of the device.
3104  *
3105  * Returns the identifier or error on failure.
3106  */
3107 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3108 {
3109         const unsigned char *d;
3110         const struct scsi_vpd *vpd_pg83;
3111         int group_id = -EAGAIN, rel_port = -1;
3112
3113         rcu_read_lock();
3114         vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3115         if (!vpd_pg83) {
3116                 rcu_read_unlock();
3117                 return -ENXIO;
3118         }
3119
3120         d = vpd_pg83->data + 4;
3121         while (d < vpd_pg83->data + vpd_pg83->len) {
3122                 switch (d[1] & 0xf) {
3123                 case 0x4:
3124                         /* Relative target port */
3125                         rel_port = get_unaligned_be16(&d[6]);
3126                         break;
3127                 case 0x5:
3128                         /* Target port group */
3129                         group_id = get_unaligned_be16(&d[6]);
3130                         break;
3131                 default:
3132                         break;
3133                 }
3134                 d += d[3] + 4;
3135         }
3136         rcu_read_unlock();
3137
3138         if (group_id >= 0 && rel_id && rel_port != -1)
3139                 *rel_id = rel_port;
3140
3141         return group_id;
3142 }
3143 EXPORT_SYMBOL(scsi_vpd_tpg_id);