9cf6a80fe29754fc93d96d41edb090db2fdd67f4
[sfrench/cifs-2.6.git] / drivers / scsi / scsi_lib.c
1 /*
2  * Copyright (C) 1999 Eric Youngdale
3  * Copyright (C) 2014 Christoph Hellwig
4  *
5  *  SCSI queueing library.
6  *      Initial versions: Eric Youngdale (eric@andante.org).
7  *                        Based upon conversations with large numbers
8  *                        of people at Linux Expo.
9  */
10
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
22 #include <linux/blk-mq.h>
23 #include <linux/ratelimit.h>
24 #include <asm/unaligned.h>
25
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_driver.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_host.h>
33 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
34 #include <scsi/scsi_dh.h>
35
36 #include <trace/events/scsi.h>
37
38 #include "scsi_debugfs.h"
39 #include "scsi_priv.h"
40 #include "scsi_logging.h"
41
42 static struct kmem_cache *scsi_sdb_cache;
43 static struct kmem_cache *scsi_sense_cache;
44 static struct kmem_cache *scsi_sense_isadma_cache;
45 static DEFINE_MUTEX(scsi_sense_cache_mutex);
46
47 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
48
49 static inline struct kmem_cache *
50 scsi_select_sense_cache(bool unchecked_isa_dma)
51 {
52         return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
53 }
54
55 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
56                                    unsigned char *sense_buffer)
57 {
58         kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
59                         sense_buffer);
60 }
61
62 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
63         gfp_t gfp_mask, int numa_node)
64 {
65         return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
66                                      gfp_mask, numa_node);
67 }
68
69 int scsi_init_sense_cache(struct Scsi_Host *shost)
70 {
71         struct kmem_cache *cache;
72         int ret = 0;
73
74         cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
75         if (cache)
76                 return 0;
77
78         mutex_lock(&scsi_sense_cache_mutex);
79         if (shost->unchecked_isa_dma) {
80                 scsi_sense_isadma_cache =
81                         kmem_cache_create("scsi_sense_cache(DMA)",
82                         SCSI_SENSE_BUFFERSIZE, 0,
83                         SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
84                 if (!scsi_sense_isadma_cache)
85                         ret = -ENOMEM;
86         } else {
87                 scsi_sense_cache =
88                         kmem_cache_create("scsi_sense_cache",
89                         SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN, NULL);
90                 if (!scsi_sense_cache)
91                         ret = -ENOMEM;
92         }
93
94         mutex_unlock(&scsi_sense_cache_mutex);
95         return ret;
96 }
97
98 /*
99  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
100  * not change behaviour from the previous unplug mechanism, experimentation
101  * may prove this needs changing.
102  */
103 #define SCSI_QUEUE_DELAY        3
104
105 static void
106 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
107 {
108         struct Scsi_Host *host = cmd->device->host;
109         struct scsi_device *device = cmd->device;
110         struct scsi_target *starget = scsi_target(device);
111
112         /*
113          * Set the appropriate busy bit for the device/host.
114          *
115          * If the host/device isn't busy, assume that something actually
116          * completed, and that we should be able to queue a command now.
117          *
118          * Note that the prior mid-layer assumption that any host could
119          * always queue at least one command is now broken.  The mid-layer
120          * will implement a user specifiable stall (see
121          * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
122          * if a command is requeued with no other commands outstanding
123          * either for the device or for the host.
124          */
125         switch (reason) {
126         case SCSI_MLQUEUE_HOST_BUSY:
127                 atomic_set(&host->host_blocked, host->max_host_blocked);
128                 break;
129         case SCSI_MLQUEUE_DEVICE_BUSY:
130         case SCSI_MLQUEUE_EH_RETRY:
131                 atomic_set(&device->device_blocked,
132                            device->max_device_blocked);
133                 break;
134         case SCSI_MLQUEUE_TARGET_BUSY:
135                 atomic_set(&starget->target_blocked,
136                            starget->max_target_blocked);
137                 break;
138         }
139 }
140
141 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
142 {
143         struct scsi_device *sdev = cmd->device;
144
145         if (cmd->request->rq_flags & RQF_DONTPREP) {
146                 cmd->request->rq_flags &= ~RQF_DONTPREP;
147                 scsi_mq_uninit_cmd(cmd);
148         } else {
149                 WARN_ON_ONCE(true);
150         }
151         blk_mq_requeue_request(cmd->request, true);
152         put_device(&sdev->sdev_gendev);
153 }
154
155 /**
156  * __scsi_queue_insert - private queue insertion
157  * @cmd: The SCSI command being requeued
158  * @reason:  The reason for the requeue
159  * @unbusy: Whether the queue should be unbusied
160  *
161  * This is a private queue insertion.  The public interface
162  * scsi_queue_insert() always assumes the queue should be unbusied
163  * because it's always called before the completion.  This function is
164  * for a requeue after completion, which should only occur in this
165  * file.
166  */
167 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
168 {
169         struct scsi_device *device = cmd->device;
170         struct request_queue *q = device->request_queue;
171         unsigned long flags;
172
173         SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
174                 "Inserting command %p into mlqueue\n", cmd));
175
176         scsi_set_blocked(cmd, reason);
177
178         /*
179          * Decrement the counters, since these commands are no longer
180          * active on the host/device.
181          */
182         if (unbusy)
183                 scsi_device_unbusy(device);
184
185         /*
186          * Requeue this command.  It will go before all other commands
187          * that are already in the queue. Schedule requeue work under
188          * lock such that the kblockd_schedule_work() call happens
189          * before blk_cleanup_queue() finishes.
190          */
191         cmd->result = 0;
192         if (q->mq_ops) {
193                 scsi_mq_requeue_cmd(cmd);
194                 return;
195         }
196         spin_lock_irqsave(q->queue_lock, flags);
197         blk_requeue_request(q, cmd->request);
198         kblockd_schedule_work(&device->requeue_work);
199         spin_unlock_irqrestore(q->queue_lock, flags);
200 }
201
202 /*
203  * Function:    scsi_queue_insert()
204  *
205  * Purpose:     Insert a command in the midlevel queue.
206  *
207  * Arguments:   cmd    - command that we are adding to queue.
208  *              reason - why we are inserting command to queue.
209  *
210  * Lock status: Assumed that lock is not held upon entry.
211  *
212  * Returns:     Nothing.
213  *
214  * Notes:       We do this for one of two cases.  Either the host is busy
215  *              and it cannot accept any more commands for the time being,
216  *              or the device returned QUEUE_FULL and can accept no more
217  *              commands.
218  * Notes:       This could be called either from an interrupt context or a
219  *              normal process context.
220  */
221 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
222 {
223         __scsi_queue_insert(cmd, reason, 1);
224 }
225
226
227 /**
228  * scsi_execute - insert request and wait for the result
229  * @sdev:       scsi device
230  * @cmd:        scsi command
231  * @data_direction: data direction
232  * @buffer:     data buffer
233  * @bufflen:    len of buffer
234  * @sense:      optional sense buffer
235  * @sshdr:      optional decoded sense header
236  * @timeout:    request timeout in seconds
237  * @retries:    number of times to retry request
238  * @flags:      flags for ->cmd_flags
239  * @rq_flags:   flags for ->rq_flags
240  * @resid:      optional residual length
241  *
242  * Returns the scsi_cmnd result field if a command was executed, or a negative
243  * Linux error code if we didn't get that far.
244  */
245 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
246                  int data_direction, void *buffer, unsigned bufflen,
247                  unsigned char *sense, struct scsi_sense_hdr *sshdr,
248                  int timeout, int retries, u64 flags, req_flags_t rq_flags,
249                  int *resid)
250 {
251         struct request *req;
252         struct scsi_request *rq;
253         int ret = DRIVER_ERROR << 24;
254
255         req = blk_get_request(sdev->request_queue,
256                         data_direction == DMA_TO_DEVICE ?
257                         REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, __GFP_RECLAIM);
258         if (IS_ERR(req))
259                 return ret;
260         rq = scsi_req(req);
261
262         if (bufflen &&  blk_rq_map_kern(sdev->request_queue, req,
263                                         buffer, bufflen, __GFP_RECLAIM))
264                 goto out;
265
266         rq->cmd_len = COMMAND_SIZE(cmd[0]);
267         memcpy(rq->cmd, cmd, rq->cmd_len);
268         rq->retries = retries;
269         req->timeout = timeout;
270         req->cmd_flags |= flags;
271         req->rq_flags |= rq_flags | RQF_QUIET | RQF_PREEMPT;
272
273         /*
274          * head injection *required* here otherwise quiesce won't work
275          */
276         blk_execute_rq(req->q, NULL, req, 1);
277
278         /*
279          * Some devices (USB mass-storage in particular) may transfer
280          * garbage data together with a residue indicating that the data
281          * is invalid.  Prevent the garbage from being misinterpreted
282          * and prevent security leaks by zeroing out the excess data.
283          */
284         if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
285                 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
286
287         if (resid)
288                 *resid = rq->resid_len;
289         if (sense && rq->sense_len)
290                 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
291         if (sshdr)
292                 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
293         ret = rq->result;
294  out:
295         blk_put_request(req);
296
297         return ret;
298 }
299 EXPORT_SYMBOL(scsi_execute);
300
301 /*
302  * Function:    scsi_init_cmd_errh()
303  *
304  * Purpose:     Initialize cmd fields related to error handling.
305  *
306  * Arguments:   cmd     - command that is ready to be queued.
307  *
308  * Notes:       This function has the job of initializing a number of
309  *              fields related to error handling.   Typically this will
310  *              be called once for each command, as required.
311  */
312 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
313 {
314         cmd->serial_number = 0;
315         scsi_set_resid(cmd, 0);
316         memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
317         if (cmd->cmd_len == 0)
318                 cmd->cmd_len = scsi_command_size(cmd->cmnd);
319 }
320
321 void scsi_device_unbusy(struct scsi_device *sdev)
322 {
323         struct Scsi_Host *shost = sdev->host;
324         struct scsi_target *starget = scsi_target(sdev);
325         unsigned long flags;
326
327         atomic_dec(&shost->host_busy);
328         if (starget->can_queue > 0)
329                 atomic_dec(&starget->target_busy);
330
331         if (unlikely(scsi_host_in_recovery(shost) &&
332                      (shost->host_failed || shost->host_eh_scheduled))) {
333                 spin_lock_irqsave(shost->host_lock, flags);
334                 scsi_eh_wakeup(shost);
335                 spin_unlock_irqrestore(shost->host_lock, flags);
336         }
337
338         atomic_dec(&sdev->device_busy);
339 }
340
341 static void scsi_kick_queue(struct request_queue *q)
342 {
343         if (q->mq_ops)
344                 blk_mq_start_hw_queues(q);
345         else
346                 blk_run_queue(q);
347 }
348
349 /*
350  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
351  * and call blk_run_queue for all the scsi_devices on the target -
352  * including current_sdev first.
353  *
354  * Called with *no* scsi locks held.
355  */
356 static void scsi_single_lun_run(struct scsi_device *current_sdev)
357 {
358         struct Scsi_Host *shost = current_sdev->host;
359         struct scsi_device *sdev, *tmp;
360         struct scsi_target *starget = scsi_target(current_sdev);
361         unsigned long flags;
362
363         spin_lock_irqsave(shost->host_lock, flags);
364         starget->starget_sdev_user = NULL;
365         spin_unlock_irqrestore(shost->host_lock, flags);
366
367         /*
368          * Call blk_run_queue for all LUNs on the target, starting with
369          * current_sdev. We race with others (to set starget_sdev_user),
370          * but in most cases, we will be first. Ideally, each LU on the
371          * target would get some limited time or requests on the target.
372          */
373         scsi_kick_queue(current_sdev->request_queue);
374
375         spin_lock_irqsave(shost->host_lock, flags);
376         if (starget->starget_sdev_user)
377                 goto out;
378         list_for_each_entry_safe(sdev, tmp, &starget->devices,
379                         same_target_siblings) {
380                 if (sdev == current_sdev)
381                         continue;
382                 if (scsi_device_get(sdev))
383                         continue;
384
385                 spin_unlock_irqrestore(shost->host_lock, flags);
386                 scsi_kick_queue(sdev->request_queue);
387                 spin_lock_irqsave(shost->host_lock, flags);
388         
389                 scsi_device_put(sdev);
390         }
391  out:
392         spin_unlock_irqrestore(shost->host_lock, flags);
393 }
394
395 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
396 {
397         if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
398                 return true;
399         if (atomic_read(&sdev->device_blocked) > 0)
400                 return true;
401         return false;
402 }
403
404 static inline bool scsi_target_is_busy(struct scsi_target *starget)
405 {
406         if (starget->can_queue > 0) {
407                 if (atomic_read(&starget->target_busy) >= starget->can_queue)
408                         return true;
409                 if (atomic_read(&starget->target_blocked) > 0)
410                         return true;
411         }
412         return false;
413 }
414
415 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
416 {
417         if (shost->can_queue > 0 &&
418             atomic_read(&shost->host_busy) >= shost->can_queue)
419                 return true;
420         if (atomic_read(&shost->host_blocked) > 0)
421                 return true;
422         if (shost->host_self_blocked)
423                 return true;
424         return false;
425 }
426
427 static void scsi_starved_list_run(struct Scsi_Host *shost)
428 {
429         LIST_HEAD(starved_list);
430         struct scsi_device *sdev;
431         unsigned long flags;
432
433         spin_lock_irqsave(shost->host_lock, flags);
434         list_splice_init(&shost->starved_list, &starved_list);
435
436         while (!list_empty(&starved_list)) {
437                 struct request_queue *slq;
438
439                 /*
440                  * As long as shost is accepting commands and we have
441                  * starved queues, call blk_run_queue. scsi_request_fn
442                  * drops the queue_lock and can add us back to the
443                  * starved_list.
444                  *
445                  * host_lock protects the starved_list and starved_entry.
446                  * scsi_request_fn must get the host_lock before checking
447                  * or modifying starved_list or starved_entry.
448                  */
449                 if (scsi_host_is_busy(shost))
450                         break;
451
452                 sdev = list_entry(starved_list.next,
453                                   struct scsi_device, starved_entry);
454                 list_del_init(&sdev->starved_entry);
455                 if (scsi_target_is_busy(scsi_target(sdev))) {
456                         list_move_tail(&sdev->starved_entry,
457                                        &shost->starved_list);
458                         continue;
459                 }
460
461                 /*
462                  * Once we drop the host lock, a racing scsi_remove_device()
463                  * call may remove the sdev from the starved list and destroy
464                  * it and the queue.  Mitigate by taking a reference to the
465                  * queue and never touching the sdev again after we drop the
466                  * host lock.  Note: if __scsi_remove_device() invokes
467                  * blk_cleanup_queue() before the queue is run from this
468                  * function then blk_run_queue() will return immediately since
469                  * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
470                  */
471                 slq = sdev->request_queue;
472                 if (!blk_get_queue(slq))
473                         continue;
474                 spin_unlock_irqrestore(shost->host_lock, flags);
475
476                 scsi_kick_queue(slq);
477                 blk_put_queue(slq);
478
479                 spin_lock_irqsave(shost->host_lock, flags);
480         }
481         /* put any unprocessed entries back */
482         list_splice(&starved_list, &shost->starved_list);
483         spin_unlock_irqrestore(shost->host_lock, flags);
484 }
485
486 /*
487  * Function:   scsi_run_queue()
488  *
489  * Purpose:    Select a proper request queue to serve next
490  *
491  * Arguments:  q       - last request's queue
492  *
493  * Returns:     Nothing
494  *
495  * Notes:      The previous command was completely finished, start
496  *             a new one if possible.
497  */
498 static void scsi_run_queue(struct request_queue *q)
499 {
500         struct scsi_device *sdev = q->queuedata;
501
502         if (scsi_target(sdev)->single_lun)
503                 scsi_single_lun_run(sdev);
504         if (!list_empty(&sdev->host->starved_list))
505                 scsi_starved_list_run(sdev->host);
506
507         if (q->mq_ops)
508                 blk_mq_run_hw_queues(q, false);
509         else
510                 blk_run_queue(q);
511 }
512
513 void scsi_requeue_run_queue(struct work_struct *work)
514 {
515         struct scsi_device *sdev;
516         struct request_queue *q;
517
518         sdev = container_of(work, struct scsi_device, requeue_work);
519         q = sdev->request_queue;
520         scsi_run_queue(q);
521 }
522
523 /*
524  * Function:    scsi_requeue_command()
525  *
526  * Purpose:     Handle post-processing of completed commands.
527  *
528  * Arguments:   q       - queue to operate on
529  *              cmd     - command that may need to be requeued.
530  *
531  * Returns:     Nothing
532  *
533  * Notes:       After command completion, there may be blocks left
534  *              over which weren't finished by the previous command
535  *              this can be for a number of reasons - the main one is
536  *              I/O errors in the middle of the request, in which case
537  *              we need to request the blocks that come after the bad
538  *              sector.
539  * Notes:       Upon return, cmd is a stale pointer.
540  */
541 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
542 {
543         struct scsi_device *sdev = cmd->device;
544         struct request *req = cmd->request;
545         unsigned long flags;
546
547         spin_lock_irqsave(q->queue_lock, flags);
548         blk_unprep_request(req);
549         req->special = NULL;
550         scsi_put_command(cmd);
551         blk_requeue_request(q, req);
552         spin_unlock_irqrestore(q->queue_lock, flags);
553
554         scsi_run_queue(q);
555
556         put_device(&sdev->sdev_gendev);
557 }
558
559 void scsi_run_host_queues(struct Scsi_Host *shost)
560 {
561         struct scsi_device *sdev;
562
563         shost_for_each_device(sdev, shost)
564                 scsi_run_queue(sdev->request_queue);
565 }
566
567 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
568 {
569         if (!blk_rq_is_passthrough(cmd->request)) {
570                 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
571
572                 if (drv->uninit_command)
573                         drv->uninit_command(cmd);
574         }
575 }
576
577 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
578 {
579         struct scsi_data_buffer *sdb;
580
581         if (cmd->sdb.table.nents)
582                 sg_free_table_chained(&cmd->sdb.table, true);
583         if (cmd->request->next_rq) {
584                 sdb = cmd->request->next_rq->special;
585                 if (sdb)
586                         sg_free_table_chained(&sdb->table, true);
587         }
588         if (scsi_prot_sg_count(cmd))
589                 sg_free_table_chained(&cmd->prot_sdb->table, true);
590 }
591
592 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
593 {
594         scsi_mq_free_sgtables(cmd);
595         scsi_uninit_cmd(cmd);
596         scsi_del_cmd_from_list(cmd);
597 }
598
599 /*
600  * Function:    scsi_release_buffers()
601  *
602  * Purpose:     Free resources allocate for a scsi_command.
603  *
604  * Arguments:   cmd     - command that we are bailing.
605  *
606  * Lock status: Assumed that no lock is held upon entry.
607  *
608  * Returns:     Nothing
609  *
610  * Notes:       In the event that an upper level driver rejects a
611  *              command, we must release resources allocated during
612  *              the __init_io() function.  Primarily this would involve
613  *              the scatter-gather table.
614  */
615 static void scsi_release_buffers(struct scsi_cmnd *cmd)
616 {
617         if (cmd->sdb.table.nents)
618                 sg_free_table_chained(&cmd->sdb.table, false);
619
620         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
621
622         if (scsi_prot_sg_count(cmd))
623                 sg_free_table_chained(&cmd->prot_sdb->table, false);
624 }
625
626 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
627 {
628         struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
629
630         sg_free_table_chained(&bidi_sdb->table, false);
631         kmem_cache_free(scsi_sdb_cache, bidi_sdb);
632         cmd->request->next_rq->special = NULL;
633 }
634
635 static bool scsi_end_request(struct request *req, blk_status_t error,
636                 unsigned int bytes, unsigned int bidi_bytes)
637 {
638         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
639         struct scsi_device *sdev = cmd->device;
640         struct request_queue *q = sdev->request_queue;
641
642         if (blk_update_request(req, error, bytes))
643                 return true;
644
645         /* Bidi request must be completed as a whole */
646         if (unlikely(bidi_bytes) &&
647             blk_update_request(req->next_rq, error, bidi_bytes))
648                 return true;
649
650         if (blk_queue_add_random(q))
651                 add_disk_randomness(req->rq_disk);
652
653         if (!blk_rq_is_scsi(req)) {
654                 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
655                 cmd->flags &= ~SCMD_INITIALIZED;
656         }
657
658         if (req->mq_ctx) {
659                 /*
660                  * In the MQ case the command gets freed by __blk_mq_end_request,
661                  * so we have to do all cleanup that depends on it earlier.
662                  *
663                  * We also can't kick the queues from irq context, so we
664                  * will have to defer it to a workqueue.
665                  */
666                 scsi_mq_uninit_cmd(cmd);
667
668                 __blk_mq_end_request(req, error);
669
670                 if (scsi_target(sdev)->single_lun ||
671                     !list_empty(&sdev->host->starved_list))
672                         kblockd_schedule_work(&sdev->requeue_work);
673                 else
674                         blk_mq_run_hw_queues(q, true);
675         } else {
676                 unsigned long flags;
677
678                 if (bidi_bytes)
679                         scsi_release_bidi_buffers(cmd);
680                 scsi_release_buffers(cmd);
681                 scsi_put_command(cmd);
682
683                 spin_lock_irqsave(q->queue_lock, flags);
684                 blk_finish_request(req, error);
685                 spin_unlock_irqrestore(q->queue_lock, flags);
686
687                 scsi_run_queue(q);
688         }
689
690         put_device(&sdev->sdev_gendev);
691         return false;
692 }
693
694 /**
695  * __scsi_error_from_host_byte - translate SCSI error code into errno
696  * @cmd:        SCSI command (unused)
697  * @result:     scsi error code
698  *
699  * Translate SCSI error code into block errors.
700  */
701 static blk_status_t __scsi_error_from_host_byte(struct scsi_cmnd *cmd,
702                 int result)
703 {
704         switch (host_byte(result)) {
705         case DID_TRANSPORT_FAILFAST:
706                 return BLK_STS_TRANSPORT;
707         case DID_TARGET_FAILURE:
708                 set_host_byte(cmd, DID_OK);
709                 return BLK_STS_TARGET;
710         case DID_NEXUS_FAILURE:
711                 return BLK_STS_NEXUS;
712         case DID_ALLOC_FAILURE:
713                 set_host_byte(cmd, DID_OK);
714                 return BLK_STS_NOSPC;
715         case DID_MEDIUM_ERROR:
716                 set_host_byte(cmd, DID_OK);
717                 return BLK_STS_MEDIUM;
718         default:
719                 return BLK_STS_IOERR;
720         }
721 }
722
723 /*
724  * Function:    scsi_io_completion()
725  *
726  * Purpose:     Completion processing for block device I/O requests.
727  *
728  * Arguments:   cmd   - command that is finished.
729  *
730  * Lock status: Assumed that no lock is held upon entry.
731  *
732  * Returns:     Nothing
733  *
734  * Notes:       We will finish off the specified number of sectors.  If we
735  *              are done, the command block will be released and the queue
736  *              function will be goosed.  If we are not done then we have to
737  *              figure out what to do next:
738  *
739  *              a) We can call scsi_requeue_command().  The request
740  *                 will be unprepared and put back on the queue.  Then
741  *                 a new command will be created for it.  This should
742  *                 be used if we made forward progress, or if we want
743  *                 to switch from READ(10) to READ(6) for example.
744  *
745  *              b) We can call __scsi_queue_insert().  The request will
746  *                 be put back on the queue and retried using the same
747  *                 command as before, possibly after a delay.
748  *
749  *              c) We can call scsi_end_request() with -EIO to fail
750  *                 the remainder of the request.
751  */
752 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
753 {
754         int result = cmd->result;
755         struct request_queue *q = cmd->device->request_queue;
756         struct request *req = cmd->request;
757         blk_status_t error = BLK_STS_OK;
758         struct scsi_sense_hdr sshdr;
759         bool sense_valid = false;
760         int sense_deferred = 0, level = 0;
761         enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
762               ACTION_DELAYED_RETRY} action;
763         unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
764
765         if (result) {
766                 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
767                 if (sense_valid)
768                         sense_deferred = scsi_sense_is_deferred(&sshdr);
769         }
770
771         if (blk_rq_is_passthrough(req)) {
772                 if (result) {
773                         if (sense_valid) {
774                                 /*
775                                  * SG_IO wants current and deferred errors
776                                  */
777                                 scsi_req(req)->sense_len =
778                                         min(8 + cmd->sense_buffer[7],
779                                             SCSI_SENSE_BUFFERSIZE);
780                         }
781                         if (!sense_deferred)
782                                 error = __scsi_error_from_host_byte(cmd, result);
783                 }
784                 /*
785                  * __scsi_error_from_host_byte may have reset the host_byte
786                  */
787                 scsi_req(req)->result = cmd->result;
788                 scsi_req(req)->resid_len = scsi_get_resid(cmd);
789
790                 if (scsi_bidi_cmnd(cmd)) {
791                         /*
792                          * Bidi commands Must be complete as a whole,
793                          * both sides at once.
794                          */
795                         scsi_req(req->next_rq)->resid_len = scsi_in(cmd)->resid;
796                         if (scsi_end_request(req, BLK_STS_OK, blk_rq_bytes(req),
797                                         blk_rq_bytes(req->next_rq)))
798                                 BUG();
799                         return;
800                 }
801         } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
802                 /*
803                  * Flush commands do not transfers any data, and thus cannot use
804                  * good_bytes != blk_rq_bytes(req) as the signal for an error.
805                  * This sets the error explicitly for the problem case.
806                  */
807                 error = __scsi_error_from_host_byte(cmd, result);
808         }
809
810         /* no bidi support for !blk_rq_is_passthrough yet */
811         BUG_ON(blk_bidi_rq(req));
812
813         /*
814          * Next deal with any sectors which we were able to correctly
815          * handle.
816          */
817         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
818                 "%u sectors total, %d bytes done.\n",
819                 blk_rq_sectors(req), good_bytes));
820
821         /*
822          * Recovered errors need reporting, but they're always treated as
823          * success, so fiddle the result code here.  For passthrough requests
824          * we already took a copy of the original into sreq->result which
825          * is what gets returned to the user
826          */
827         if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
828                 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
829                  * print since caller wants ATA registers. Only occurs on
830                  * SCSI ATA PASS_THROUGH commands when CK_COND=1
831                  */
832                 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
833                         ;
834                 else if (!(req->rq_flags & RQF_QUIET))
835                         scsi_print_sense(cmd);
836                 result = 0;
837                 /* for passthrough error may be set */
838                 error = BLK_STS_OK;
839         }
840
841         /*
842          * special case: failed zero length commands always need to
843          * drop down into the retry code. Otherwise, if we finished
844          * all bytes in the request we are done now.
845          */
846         if (!(blk_rq_bytes(req) == 0 && error) &&
847             !scsi_end_request(req, error, good_bytes, 0))
848                 return;
849
850         /*
851          * Kill remainder if no retrys.
852          */
853         if (error && scsi_noretry_cmd(cmd)) {
854                 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
855                         BUG();
856                 return;
857         }
858
859         /*
860          * If there had been no error, but we have leftover bytes in the
861          * requeues just queue the command up again.
862          */
863         if (result == 0)
864                 goto requeue;
865
866         error = __scsi_error_from_host_byte(cmd, result);
867
868         if (host_byte(result) == DID_RESET) {
869                 /* Third party bus reset or reset for error recovery
870                  * reasons.  Just retry the command and see what
871                  * happens.
872                  */
873                 action = ACTION_RETRY;
874         } else if (sense_valid && !sense_deferred) {
875                 switch (sshdr.sense_key) {
876                 case UNIT_ATTENTION:
877                         if (cmd->device->removable) {
878                                 /* Detected disc change.  Set a bit
879                                  * and quietly refuse further access.
880                                  */
881                                 cmd->device->changed = 1;
882                                 action = ACTION_FAIL;
883                         } else {
884                                 /* Must have been a power glitch, or a
885                                  * bus reset.  Could not have been a
886                                  * media change, so we just retry the
887                                  * command and see what happens.
888                                  */
889                                 action = ACTION_RETRY;
890                         }
891                         break;
892                 case ILLEGAL_REQUEST:
893                         /* If we had an ILLEGAL REQUEST returned, then
894                          * we may have performed an unsupported
895                          * command.  The only thing this should be
896                          * would be a ten byte read where only a six
897                          * byte read was supported.  Also, on a system
898                          * where READ CAPACITY failed, we may have
899                          * read past the end of the disk.
900                          */
901                         if ((cmd->device->use_10_for_rw &&
902                             sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
903                             (cmd->cmnd[0] == READ_10 ||
904                              cmd->cmnd[0] == WRITE_10)) {
905                                 /* This will issue a new 6-byte command. */
906                                 cmd->device->use_10_for_rw = 0;
907                                 action = ACTION_REPREP;
908                         } else if (sshdr.asc == 0x10) /* DIX */ {
909                                 action = ACTION_FAIL;
910                                 error = BLK_STS_PROTECTION;
911                         /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
912                         } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
913                                 action = ACTION_FAIL;
914                                 error = BLK_STS_TARGET;
915                         } else
916                                 action = ACTION_FAIL;
917                         break;
918                 case ABORTED_COMMAND:
919                         action = ACTION_FAIL;
920                         if (sshdr.asc == 0x10) /* DIF */
921                                 error = BLK_STS_PROTECTION;
922                         break;
923                 case NOT_READY:
924                         /* If the device is in the process of becoming
925                          * ready, or has a temporary blockage, retry.
926                          */
927                         if (sshdr.asc == 0x04) {
928                                 switch (sshdr.ascq) {
929                                 case 0x01: /* becoming ready */
930                                 case 0x04: /* format in progress */
931                                 case 0x05: /* rebuild in progress */
932                                 case 0x06: /* recalculation in progress */
933                                 case 0x07: /* operation in progress */
934                                 case 0x08: /* Long write in progress */
935                                 case 0x09: /* self test in progress */
936                                 case 0x14: /* space allocation in progress */
937                                         action = ACTION_DELAYED_RETRY;
938                                         break;
939                                 default:
940                                         action = ACTION_FAIL;
941                                         break;
942                                 }
943                         } else
944                                 action = ACTION_FAIL;
945                         break;
946                 case VOLUME_OVERFLOW:
947                         /* See SSC3rXX or current. */
948                         action = ACTION_FAIL;
949                         break;
950                 default:
951                         action = ACTION_FAIL;
952                         break;
953                 }
954         } else
955                 action = ACTION_FAIL;
956
957         if (action != ACTION_FAIL &&
958             time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
959                 action = ACTION_FAIL;
960
961         switch (action) {
962         case ACTION_FAIL:
963                 /* Give up and fail the remainder of the request */
964                 if (!(req->rq_flags & RQF_QUIET)) {
965                         static DEFINE_RATELIMIT_STATE(_rs,
966                                         DEFAULT_RATELIMIT_INTERVAL,
967                                         DEFAULT_RATELIMIT_BURST);
968
969                         if (unlikely(scsi_logging_level))
970                                 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
971                                                        SCSI_LOG_MLCOMPLETE_BITS);
972
973                         /*
974                          * if logging is enabled the failure will be printed
975                          * in scsi_log_completion(), so avoid duplicate messages
976                          */
977                         if (!level && __ratelimit(&_rs)) {
978                                 scsi_print_result(cmd, NULL, FAILED);
979                                 if (driver_byte(result) & DRIVER_SENSE)
980                                         scsi_print_sense(cmd);
981                                 scsi_print_command(cmd);
982                         }
983                 }
984                 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
985                         return;
986                 /*FALLTHRU*/
987         case ACTION_REPREP:
988         requeue:
989                 /* Unprep the request and put it back at the head of the queue.
990                  * A new command will be prepared and issued.
991                  */
992                 if (q->mq_ops) {
993                         scsi_mq_requeue_cmd(cmd);
994                 } else {
995                         scsi_release_buffers(cmd);
996                         scsi_requeue_command(q, cmd);
997                 }
998                 break;
999         case ACTION_RETRY:
1000                 /* Retry the same command immediately */
1001                 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1002                 break;
1003         case ACTION_DELAYED_RETRY:
1004                 /* Retry the same command after a delay */
1005                 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1006                 break;
1007         }
1008 }
1009
1010 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1011 {
1012         int count;
1013
1014         /*
1015          * If sg table allocation fails, requeue request later.
1016          */
1017         if (unlikely(sg_alloc_table_chained(&sdb->table,
1018                         blk_rq_nr_phys_segments(req), sdb->table.sgl)))
1019                 return BLKPREP_DEFER;
1020
1021         /* 
1022          * Next, walk the list, and fill in the addresses and sizes of
1023          * each segment.
1024          */
1025         count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1026         BUG_ON(count > sdb->table.nents);
1027         sdb->table.nents = count;
1028         sdb->length = blk_rq_payload_bytes(req);
1029         return BLKPREP_OK;
1030 }
1031
1032 /*
1033  * Function:    scsi_init_io()
1034  *
1035  * Purpose:     SCSI I/O initialize function.
1036  *
1037  * Arguments:   cmd   - Command descriptor we wish to initialize
1038  *
1039  * Returns:     0 on success
1040  *              BLKPREP_DEFER if the failure is retryable
1041  *              BLKPREP_KILL if the failure is fatal
1042  */
1043 int scsi_init_io(struct scsi_cmnd *cmd)
1044 {
1045         struct scsi_device *sdev = cmd->device;
1046         struct request *rq = cmd->request;
1047         bool is_mq = (rq->mq_ctx != NULL);
1048         int error = BLKPREP_KILL;
1049
1050         if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1051                 goto err_exit;
1052
1053         error = scsi_init_sgtable(rq, &cmd->sdb);
1054         if (error)
1055                 goto err_exit;
1056
1057         if (blk_bidi_rq(rq)) {
1058                 if (!rq->q->mq_ops) {
1059                         struct scsi_data_buffer *bidi_sdb =
1060                                 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1061                         if (!bidi_sdb) {
1062                                 error = BLKPREP_DEFER;
1063                                 goto err_exit;
1064                         }
1065
1066                         rq->next_rq->special = bidi_sdb;
1067                 }
1068
1069                 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1070                 if (error)
1071                         goto err_exit;
1072         }
1073
1074         if (blk_integrity_rq(rq)) {
1075                 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1076                 int ivecs, count;
1077
1078                 if (prot_sdb == NULL) {
1079                         /*
1080                          * This can happen if someone (e.g. multipath)
1081                          * queues a command to a device on an adapter
1082                          * that does not support DIX.
1083                          */
1084                         WARN_ON_ONCE(1);
1085                         error = BLKPREP_KILL;
1086                         goto err_exit;
1087                 }
1088
1089                 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1090
1091                 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1092                                 prot_sdb->table.sgl)) {
1093                         error = BLKPREP_DEFER;
1094                         goto err_exit;
1095                 }
1096
1097                 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1098                                                 prot_sdb->table.sgl);
1099                 BUG_ON(unlikely(count > ivecs));
1100                 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1101
1102                 cmd->prot_sdb = prot_sdb;
1103                 cmd->prot_sdb->table.nents = count;
1104         }
1105
1106         return BLKPREP_OK;
1107 err_exit:
1108         if (is_mq) {
1109                 scsi_mq_free_sgtables(cmd);
1110         } else {
1111                 scsi_release_buffers(cmd);
1112                 cmd->request->special = NULL;
1113                 scsi_put_command(cmd);
1114                 put_device(&sdev->sdev_gendev);
1115         }
1116         return error;
1117 }
1118 EXPORT_SYMBOL(scsi_init_io);
1119
1120 /**
1121  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1122  * @rq: Request associated with the SCSI command to be initialized.
1123  *
1124  * This function initializes the members of struct scsi_cmnd that must be
1125  * initialized before request processing starts and that won't be
1126  * reinitialized if a SCSI command is requeued.
1127  *
1128  * Called from inside blk_get_request() for pass-through requests and from
1129  * inside scsi_init_command() for filesystem requests.
1130  */
1131 void scsi_initialize_rq(struct request *rq)
1132 {
1133         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1134
1135         scsi_req_init(&cmd->req);
1136         cmd->jiffies_at_alloc = jiffies;
1137         cmd->retries = 0;
1138 }
1139 EXPORT_SYMBOL(scsi_initialize_rq);
1140
1141 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1142 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1143 {
1144         struct scsi_device *sdev = cmd->device;
1145         struct Scsi_Host *shost = sdev->host;
1146         unsigned long flags;
1147
1148         if (shost->use_cmd_list) {
1149                 spin_lock_irqsave(&sdev->list_lock, flags);
1150                 list_add_tail(&cmd->list, &sdev->cmd_list);
1151                 spin_unlock_irqrestore(&sdev->list_lock, flags);
1152         }
1153 }
1154
1155 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1156 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1157 {
1158         struct scsi_device *sdev = cmd->device;
1159         struct Scsi_Host *shost = sdev->host;
1160         unsigned long flags;
1161
1162         if (shost->use_cmd_list) {
1163                 spin_lock_irqsave(&sdev->list_lock, flags);
1164                 BUG_ON(list_empty(&cmd->list));
1165                 list_del_init(&cmd->list);
1166                 spin_unlock_irqrestore(&sdev->list_lock, flags);
1167         }
1168 }
1169
1170 /* Called after a request has been started. */
1171 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1172 {
1173         void *buf = cmd->sense_buffer;
1174         void *prot = cmd->prot_sdb;
1175         struct request *rq = blk_mq_rq_from_pdu(cmd);
1176         unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1177         unsigned long jiffies_at_alloc;
1178         int retries;
1179
1180         if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1181                 flags |= SCMD_INITIALIZED;
1182                 scsi_initialize_rq(rq);
1183         }
1184
1185         jiffies_at_alloc = cmd->jiffies_at_alloc;
1186         retries = cmd->retries;
1187         /* zero out the cmd, except for the embedded scsi_request */
1188         memset((char *)cmd + sizeof(cmd->req), 0,
1189                 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1190
1191         cmd->device = dev;
1192         cmd->sense_buffer = buf;
1193         cmd->prot_sdb = prot;
1194         cmd->flags = flags;
1195         INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1196         cmd->jiffies_at_alloc = jiffies_at_alloc;
1197         cmd->retries = retries;
1198
1199         scsi_add_cmd_to_list(cmd);
1200 }
1201
1202 static int scsi_setup_scsi_cmnd(struct scsi_device *sdev, struct request *req)
1203 {
1204         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1205
1206         /*
1207          * Passthrough requests may transfer data, in which case they must
1208          * a bio attached to them.  Or they might contain a SCSI command
1209          * that does not transfer data, in which case they may optionally
1210          * submit a request without an attached bio.
1211          */
1212         if (req->bio) {
1213                 int ret = scsi_init_io(cmd);
1214                 if (unlikely(ret))
1215                         return ret;
1216         } else {
1217                 BUG_ON(blk_rq_bytes(req));
1218
1219                 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1220         }
1221
1222         cmd->cmd_len = scsi_req(req)->cmd_len;
1223         cmd->cmnd = scsi_req(req)->cmd;
1224         cmd->transfersize = blk_rq_bytes(req);
1225         cmd->allowed = scsi_req(req)->retries;
1226         return BLKPREP_OK;
1227 }
1228
1229 /*
1230  * Setup a normal block command.  These are simple request from filesystems
1231  * that still need to be translated to SCSI CDBs from the ULD.
1232  */
1233 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1234 {
1235         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1236
1237         if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1238                 int ret = sdev->handler->prep_fn(sdev, req);
1239                 if (ret != BLKPREP_OK)
1240                         return ret;
1241         }
1242
1243         cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1244         memset(cmd->cmnd, 0, BLK_MAX_CDB);
1245         return scsi_cmd_to_driver(cmd)->init_command(cmd);
1246 }
1247
1248 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1249 {
1250         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1251
1252         if (!blk_rq_bytes(req))
1253                 cmd->sc_data_direction = DMA_NONE;
1254         else if (rq_data_dir(req) == WRITE)
1255                 cmd->sc_data_direction = DMA_TO_DEVICE;
1256         else
1257                 cmd->sc_data_direction = DMA_FROM_DEVICE;
1258
1259         if (blk_rq_is_scsi(req))
1260                 return scsi_setup_scsi_cmnd(sdev, req);
1261         else
1262                 return scsi_setup_fs_cmnd(sdev, req);
1263 }
1264
1265 static int
1266 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1267 {
1268         int ret = BLKPREP_OK;
1269
1270         /*
1271          * If the device is not in running state we will reject some
1272          * or all commands.
1273          */
1274         if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1275                 switch (sdev->sdev_state) {
1276                 case SDEV_OFFLINE:
1277                 case SDEV_TRANSPORT_OFFLINE:
1278                         /*
1279                          * If the device is offline we refuse to process any
1280                          * commands.  The device must be brought online
1281                          * before trying any recovery commands.
1282                          */
1283                         sdev_printk(KERN_ERR, sdev,
1284                                     "rejecting I/O to offline device\n");
1285                         ret = BLKPREP_KILL;
1286                         break;
1287                 case SDEV_DEL:
1288                         /*
1289                          * If the device is fully deleted, we refuse to
1290                          * process any commands as well.
1291                          */
1292                         sdev_printk(KERN_ERR, sdev,
1293                                     "rejecting I/O to dead device\n");
1294                         ret = BLKPREP_KILL;
1295                         break;
1296                 case SDEV_BLOCK:
1297                 case SDEV_CREATED_BLOCK:
1298                         ret = BLKPREP_DEFER;
1299                         break;
1300                 case SDEV_QUIESCE:
1301                         /*
1302                          * If the devices is blocked we defer normal commands.
1303                          */
1304                         if (!(req->rq_flags & RQF_PREEMPT))
1305                                 ret = BLKPREP_DEFER;
1306                         break;
1307                 default:
1308                         /*
1309                          * For any other not fully online state we only allow
1310                          * special commands.  In particular any user initiated
1311                          * command is not allowed.
1312                          */
1313                         if (!(req->rq_flags & RQF_PREEMPT))
1314                                 ret = BLKPREP_KILL;
1315                         break;
1316                 }
1317         }
1318         return ret;
1319 }
1320
1321 static int
1322 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1323 {
1324         struct scsi_device *sdev = q->queuedata;
1325
1326         switch (ret) {
1327         case BLKPREP_KILL:
1328         case BLKPREP_INVALID:
1329                 scsi_req(req)->result = DID_NO_CONNECT << 16;
1330                 /* release the command and kill it */
1331                 if (req->special) {
1332                         struct scsi_cmnd *cmd = req->special;
1333                         scsi_release_buffers(cmd);
1334                         scsi_put_command(cmd);
1335                         put_device(&sdev->sdev_gendev);
1336                         req->special = NULL;
1337                 }
1338                 break;
1339         case BLKPREP_DEFER:
1340                 /*
1341                  * If we defer, the blk_peek_request() returns NULL, but the
1342                  * queue must be restarted, so we schedule a callback to happen
1343                  * shortly.
1344                  */
1345                 if (atomic_read(&sdev->device_busy) == 0)
1346                         blk_delay_queue(q, SCSI_QUEUE_DELAY);
1347                 break;
1348         default:
1349                 req->rq_flags |= RQF_DONTPREP;
1350         }
1351
1352         return ret;
1353 }
1354
1355 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1356 {
1357         struct scsi_device *sdev = q->queuedata;
1358         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1359         int ret;
1360
1361         ret = scsi_prep_state_check(sdev, req);
1362         if (ret != BLKPREP_OK)
1363                 goto out;
1364
1365         if (!req->special) {
1366                 /* Bail if we can't get a reference to the device */
1367                 if (unlikely(!get_device(&sdev->sdev_gendev))) {
1368                         ret = BLKPREP_DEFER;
1369                         goto out;
1370                 }
1371
1372                 scsi_init_command(sdev, cmd);
1373                 req->special = cmd;
1374         }
1375
1376         cmd->tag = req->tag;
1377         cmd->request = req;
1378         cmd->prot_op = SCSI_PROT_NORMAL;
1379
1380         ret = scsi_setup_cmnd(sdev, req);
1381 out:
1382         if (ret != BLKPREP_OK)
1383                 cmd->flags &= ~SCMD_INITIALIZED;
1384         return scsi_prep_return(q, req, ret);
1385 }
1386
1387 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1388 {
1389         scsi_uninit_cmd(blk_mq_rq_to_pdu(req));
1390 }
1391
1392 /*
1393  * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1394  * return 0.
1395  *
1396  * Called with the queue_lock held.
1397  */
1398 static inline int scsi_dev_queue_ready(struct request_queue *q,
1399                                   struct scsi_device *sdev)
1400 {
1401         unsigned int busy;
1402
1403         busy = atomic_inc_return(&sdev->device_busy) - 1;
1404         if (atomic_read(&sdev->device_blocked)) {
1405                 if (busy)
1406                         goto out_dec;
1407
1408                 /*
1409                  * unblock after device_blocked iterates to zero
1410                  */
1411                 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1412                         /*
1413                          * For the MQ case we take care of this in the caller.
1414                          */
1415                         if (!q->mq_ops)
1416                                 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1417                         goto out_dec;
1418                 }
1419                 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1420                                    "unblocking device at zero depth\n"));
1421         }
1422
1423         if (busy >= sdev->queue_depth)
1424                 goto out_dec;
1425
1426         return 1;
1427 out_dec:
1428         atomic_dec(&sdev->device_busy);
1429         return 0;
1430 }
1431
1432 /*
1433  * scsi_target_queue_ready: checks if there we can send commands to target
1434  * @sdev: scsi device on starget to check.
1435  */
1436 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1437                                            struct scsi_device *sdev)
1438 {
1439         struct scsi_target *starget = scsi_target(sdev);
1440         unsigned int busy;
1441
1442         if (starget->single_lun) {
1443                 spin_lock_irq(shost->host_lock);
1444                 if (starget->starget_sdev_user &&
1445                     starget->starget_sdev_user != sdev) {
1446                         spin_unlock_irq(shost->host_lock);
1447                         return 0;
1448                 }
1449                 starget->starget_sdev_user = sdev;
1450                 spin_unlock_irq(shost->host_lock);
1451         }
1452
1453         if (starget->can_queue <= 0)
1454                 return 1;
1455
1456         busy = atomic_inc_return(&starget->target_busy) - 1;
1457         if (atomic_read(&starget->target_blocked) > 0) {
1458                 if (busy)
1459                         goto starved;
1460
1461                 /*
1462                  * unblock after target_blocked iterates to zero
1463                  */
1464                 if (atomic_dec_return(&starget->target_blocked) > 0)
1465                         goto out_dec;
1466
1467                 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1468                                  "unblocking target at zero depth\n"));
1469         }
1470
1471         if (busy >= starget->can_queue)
1472                 goto starved;
1473
1474         return 1;
1475
1476 starved:
1477         spin_lock_irq(shost->host_lock);
1478         list_move_tail(&sdev->starved_entry, &shost->starved_list);
1479         spin_unlock_irq(shost->host_lock);
1480 out_dec:
1481         if (starget->can_queue > 0)
1482                 atomic_dec(&starget->target_busy);
1483         return 0;
1484 }
1485
1486 /*
1487  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1488  * return 0. We must end up running the queue again whenever 0 is
1489  * returned, else IO can hang.
1490  */
1491 static inline int scsi_host_queue_ready(struct request_queue *q,
1492                                    struct Scsi_Host *shost,
1493                                    struct scsi_device *sdev)
1494 {
1495         unsigned int busy;
1496
1497         if (scsi_host_in_recovery(shost))
1498                 return 0;
1499
1500         busy = atomic_inc_return(&shost->host_busy) - 1;
1501         if (atomic_read(&shost->host_blocked) > 0) {
1502                 if (busy)
1503                         goto starved;
1504
1505                 /*
1506                  * unblock after host_blocked iterates to zero
1507                  */
1508                 if (atomic_dec_return(&shost->host_blocked) > 0)
1509                         goto out_dec;
1510
1511                 SCSI_LOG_MLQUEUE(3,
1512                         shost_printk(KERN_INFO, shost,
1513                                      "unblocking host at zero depth\n"));
1514         }
1515
1516         if (shost->can_queue > 0 && busy >= shost->can_queue)
1517                 goto starved;
1518         if (shost->host_self_blocked)
1519                 goto starved;
1520
1521         /* We're OK to process the command, so we can't be starved */
1522         if (!list_empty(&sdev->starved_entry)) {
1523                 spin_lock_irq(shost->host_lock);
1524                 if (!list_empty(&sdev->starved_entry))
1525                         list_del_init(&sdev->starved_entry);
1526                 spin_unlock_irq(shost->host_lock);
1527         }
1528
1529         return 1;
1530
1531 starved:
1532         spin_lock_irq(shost->host_lock);
1533         if (list_empty(&sdev->starved_entry))
1534                 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1535         spin_unlock_irq(shost->host_lock);
1536 out_dec:
1537         atomic_dec(&shost->host_busy);
1538         return 0;
1539 }
1540
1541 /*
1542  * Busy state exporting function for request stacking drivers.
1543  *
1544  * For efficiency, no lock is taken to check the busy state of
1545  * shost/starget/sdev, since the returned value is not guaranteed and
1546  * may be changed after request stacking drivers call the function,
1547  * regardless of taking lock or not.
1548  *
1549  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1550  * needs to return 'not busy'. Otherwise, request stacking drivers
1551  * may hold requests forever.
1552  */
1553 static int scsi_lld_busy(struct request_queue *q)
1554 {
1555         struct scsi_device *sdev = q->queuedata;
1556         struct Scsi_Host *shost;
1557
1558         if (blk_queue_dying(q))
1559                 return 0;
1560
1561         shost = sdev->host;
1562
1563         /*
1564          * Ignore host/starget busy state.
1565          * Since block layer does not have a concept of fairness across
1566          * multiple queues, congestion of host/starget needs to be handled
1567          * in SCSI layer.
1568          */
1569         if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1570                 return 1;
1571
1572         return 0;
1573 }
1574
1575 /*
1576  * Kill a request for a dead device
1577  */
1578 static void scsi_kill_request(struct request *req, struct request_queue *q)
1579 {
1580         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1581         struct scsi_device *sdev;
1582         struct scsi_target *starget;
1583         struct Scsi_Host *shost;
1584
1585         blk_start_request(req);
1586
1587         scmd_printk(KERN_INFO, cmd, "killing request\n");
1588
1589         sdev = cmd->device;
1590         starget = scsi_target(sdev);
1591         shost = sdev->host;
1592         scsi_init_cmd_errh(cmd);
1593         cmd->result = DID_NO_CONNECT << 16;
1594         atomic_inc(&cmd->device->iorequest_cnt);
1595
1596         /*
1597          * SCSI request completion path will do scsi_device_unbusy(),
1598          * bump busy counts.  To bump the counters, we need to dance
1599          * with the locks as normal issue path does.
1600          */
1601         atomic_inc(&sdev->device_busy);
1602         atomic_inc(&shost->host_busy);
1603         if (starget->can_queue > 0)
1604                 atomic_inc(&starget->target_busy);
1605
1606         blk_complete_request(req);
1607 }
1608
1609 static void scsi_softirq_done(struct request *rq)
1610 {
1611         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1612         unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1613         int disposition;
1614
1615         INIT_LIST_HEAD(&cmd->eh_entry);
1616
1617         atomic_inc(&cmd->device->iodone_cnt);
1618         if (cmd->result)
1619                 atomic_inc(&cmd->device->ioerr_cnt);
1620
1621         disposition = scsi_decide_disposition(cmd);
1622         if (disposition != SUCCESS &&
1623             time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1624                 sdev_printk(KERN_ERR, cmd->device,
1625                             "timing out command, waited %lus\n",
1626                             wait_for/HZ);
1627                 disposition = SUCCESS;
1628         }
1629
1630         scsi_log_completion(cmd, disposition);
1631
1632         switch (disposition) {
1633                 case SUCCESS:
1634                         scsi_finish_command(cmd);
1635                         break;
1636                 case NEEDS_RETRY:
1637                         scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1638                         break;
1639                 case ADD_TO_MLQUEUE:
1640                         scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1641                         break;
1642                 default:
1643                         scsi_eh_scmd_add(cmd);
1644                         break;
1645         }
1646 }
1647
1648 /**
1649  * scsi_dispatch_command - Dispatch a command to the low-level driver.
1650  * @cmd: command block we are dispatching.
1651  *
1652  * Return: nonzero return request was rejected and device's queue needs to be
1653  * plugged.
1654  */
1655 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1656 {
1657         struct Scsi_Host *host = cmd->device->host;
1658         int rtn = 0;
1659
1660         atomic_inc(&cmd->device->iorequest_cnt);
1661
1662         /* check if the device is still usable */
1663         if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1664                 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1665                  * returns an immediate error upwards, and signals
1666                  * that the device is no longer present */
1667                 cmd->result = DID_NO_CONNECT << 16;
1668                 goto done;
1669         }
1670
1671         /* Check to see if the scsi lld made this device blocked. */
1672         if (unlikely(scsi_device_blocked(cmd->device))) {
1673                 /*
1674                  * in blocked state, the command is just put back on
1675                  * the device queue.  The suspend state has already
1676                  * blocked the queue so future requests should not
1677                  * occur until the device transitions out of the
1678                  * suspend state.
1679                  */
1680                 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1681                         "queuecommand : device blocked\n"));
1682                 return SCSI_MLQUEUE_DEVICE_BUSY;
1683         }
1684
1685         /* Store the LUN value in cmnd, if needed. */
1686         if (cmd->device->lun_in_cdb)
1687                 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1688                                (cmd->device->lun << 5 & 0xe0);
1689
1690         scsi_log_send(cmd);
1691
1692         /*
1693          * Before we queue this command, check if the command
1694          * length exceeds what the host adapter can handle.
1695          */
1696         if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1697                 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1698                                "queuecommand : command too long. "
1699                                "cdb_size=%d host->max_cmd_len=%d\n",
1700                                cmd->cmd_len, cmd->device->host->max_cmd_len));
1701                 cmd->result = (DID_ABORT << 16);
1702                 goto done;
1703         }
1704
1705         if (unlikely(host->shost_state == SHOST_DEL)) {
1706                 cmd->result = (DID_NO_CONNECT << 16);
1707                 goto done;
1708
1709         }
1710
1711         trace_scsi_dispatch_cmd_start(cmd);
1712         rtn = host->hostt->queuecommand(host, cmd);
1713         if (rtn) {
1714                 trace_scsi_dispatch_cmd_error(cmd, rtn);
1715                 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1716                     rtn != SCSI_MLQUEUE_TARGET_BUSY)
1717                         rtn = SCSI_MLQUEUE_HOST_BUSY;
1718
1719                 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1720                         "queuecommand : request rejected\n"));
1721         }
1722
1723         return rtn;
1724  done:
1725         cmd->scsi_done(cmd);
1726         return 0;
1727 }
1728
1729 /**
1730  * scsi_done - Invoke completion on finished SCSI command.
1731  * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1732  * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1733  *
1734  * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1735  * which regains ownership of the SCSI command (de facto) from a LLDD, and
1736  * calls blk_complete_request() for further processing.
1737  *
1738  * This function is interrupt context safe.
1739  */
1740 static void scsi_done(struct scsi_cmnd *cmd)
1741 {
1742         trace_scsi_dispatch_cmd_done(cmd);
1743         blk_complete_request(cmd->request);
1744 }
1745
1746 /*
1747  * Function:    scsi_request_fn()
1748  *
1749  * Purpose:     Main strategy routine for SCSI.
1750  *
1751  * Arguments:   q       - Pointer to actual queue.
1752  *
1753  * Returns:     Nothing
1754  *
1755  * Lock status: IO request lock assumed to be held when called.
1756  */
1757 static void scsi_request_fn(struct request_queue *q)
1758         __releases(q->queue_lock)
1759         __acquires(q->queue_lock)
1760 {
1761         struct scsi_device *sdev = q->queuedata;
1762         struct Scsi_Host *shost;
1763         struct scsi_cmnd *cmd;
1764         struct request *req;
1765
1766         /*
1767          * To start with, we keep looping until the queue is empty, or until
1768          * the host is no longer able to accept any more requests.
1769          */
1770         shost = sdev->host;
1771         for (;;) {
1772                 int rtn;
1773                 /*
1774                  * get next queueable request.  We do this early to make sure
1775                  * that the request is fully prepared even if we cannot
1776                  * accept it.
1777                  */
1778                 req = blk_peek_request(q);
1779                 if (!req)
1780                         break;
1781
1782                 if (unlikely(!scsi_device_online(sdev))) {
1783                         sdev_printk(KERN_ERR, sdev,
1784                                     "rejecting I/O to offline device\n");
1785                         scsi_kill_request(req, q);
1786                         continue;
1787                 }
1788
1789                 if (!scsi_dev_queue_ready(q, sdev))
1790                         break;
1791
1792                 /*
1793                  * Remove the request from the request list.
1794                  */
1795                 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1796                         blk_start_request(req);
1797
1798                 spin_unlock_irq(q->queue_lock);
1799                 cmd = blk_mq_rq_to_pdu(req);
1800                 if (cmd != req->special) {
1801                         printk(KERN_CRIT "impossible request in %s.\n"
1802                                          "please mail a stack trace to "
1803                                          "linux-scsi@vger.kernel.org\n",
1804                                          __func__);
1805                         blk_dump_rq_flags(req, "foo");
1806                         BUG();
1807                 }
1808
1809                 /*
1810                  * We hit this when the driver is using a host wide
1811                  * tag map. For device level tag maps the queue_depth check
1812                  * in the device ready fn would prevent us from trying
1813                  * to allocate a tag. Since the map is a shared host resource
1814                  * we add the dev to the starved list so it eventually gets
1815                  * a run when a tag is freed.
1816                  */
1817                 if (blk_queue_tagged(q) && !(req->rq_flags & RQF_QUEUED)) {
1818                         spin_lock_irq(shost->host_lock);
1819                         if (list_empty(&sdev->starved_entry))
1820                                 list_add_tail(&sdev->starved_entry,
1821                                               &shost->starved_list);
1822                         spin_unlock_irq(shost->host_lock);
1823                         goto not_ready;
1824                 }
1825
1826                 if (!scsi_target_queue_ready(shost, sdev))
1827                         goto not_ready;
1828
1829                 if (!scsi_host_queue_ready(q, shost, sdev))
1830                         goto host_not_ready;
1831         
1832                 if (sdev->simple_tags)
1833                         cmd->flags |= SCMD_TAGGED;
1834                 else
1835                         cmd->flags &= ~SCMD_TAGGED;
1836
1837                 /*
1838                  * Finally, initialize any error handling parameters, and set up
1839                  * the timers for timeouts.
1840                  */
1841                 scsi_init_cmd_errh(cmd);
1842
1843                 /*
1844                  * Dispatch the command to the low-level driver.
1845                  */
1846                 cmd->scsi_done = scsi_done;
1847                 rtn = scsi_dispatch_cmd(cmd);
1848                 if (rtn) {
1849                         scsi_queue_insert(cmd, rtn);
1850                         spin_lock_irq(q->queue_lock);
1851                         goto out_delay;
1852                 }
1853                 spin_lock_irq(q->queue_lock);
1854         }
1855
1856         return;
1857
1858  host_not_ready:
1859         if (scsi_target(sdev)->can_queue > 0)
1860                 atomic_dec(&scsi_target(sdev)->target_busy);
1861  not_ready:
1862         /*
1863          * lock q, handle tag, requeue req, and decrement device_busy. We
1864          * must return with queue_lock held.
1865          *
1866          * Decrementing device_busy without checking it is OK, as all such
1867          * cases (host limits or settings) should run the queue at some
1868          * later time.
1869          */
1870         spin_lock_irq(q->queue_lock);
1871         blk_requeue_request(q, req);
1872         atomic_dec(&sdev->device_busy);
1873 out_delay:
1874         if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1875                 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1876 }
1877
1878 static inline blk_status_t prep_to_mq(int ret)
1879 {
1880         switch (ret) {
1881         case BLKPREP_OK:
1882                 return BLK_STS_OK;
1883         case BLKPREP_DEFER:
1884                 return BLK_STS_RESOURCE;
1885         default:
1886                 return BLK_STS_IOERR;
1887         }
1888 }
1889
1890 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1891 static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost)
1892 {
1893         return min_t(unsigned int, shost->sg_tablesize, SG_CHUNK_SIZE) *
1894                 sizeof(struct scatterlist);
1895 }
1896
1897 static int scsi_mq_prep_fn(struct request *req)
1898 {
1899         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1900         struct scsi_device *sdev = req->q->queuedata;
1901         struct Scsi_Host *shost = sdev->host;
1902         struct scatterlist *sg;
1903         int ret;
1904
1905         scsi_init_command(sdev, cmd);
1906
1907         req->special = cmd;
1908
1909         cmd->request = req;
1910
1911         cmd->tag = req->tag;
1912         cmd->prot_op = SCSI_PROT_NORMAL;
1913
1914         sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1915         cmd->sdb.table.sgl = sg;
1916
1917         if (scsi_host_get_prot(shost)) {
1918                 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1919
1920                 cmd->prot_sdb->table.sgl =
1921                         (struct scatterlist *)(cmd->prot_sdb + 1);
1922         }
1923
1924         if (blk_bidi_rq(req)) {
1925                 struct request *next_rq = req->next_rq;
1926                 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1927
1928                 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1929                 bidi_sdb->table.sgl =
1930                         (struct scatterlist *)(bidi_sdb + 1);
1931
1932                 next_rq->special = bidi_sdb;
1933         }
1934
1935         blk_mq_start_request(req);
1936
1937         ret = scsi_setup_cmnd(sdev, req);
1938         if (ret != BLK_STS_OK)
1939                 cmd->flags &= ~SCMD_INITIALIZED;
1940         return ret;
1941 }
1942
1943 static void scsi_mq_done(struct scsi_cmnd *cmd)
1944 {
1945         trace_scsi_dispatch_cmd_done(cmd);
1946         blk_mq_complete_request(cmd->request);
1947 }
1948
1949 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1950                          const struct blk_mq_queue_data *bd)
1951 {
1952         struct request *req = bd->rq;
1953         struct request_queue *q = req->q;
1954         struct scsi_device *sdev = q->queuedata;
1955         struct Scsi_Host *shost = sdev->host;
1956         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1957         blk_status_t ret;
1958         int reason;
1959
1960         ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1961         if (ret != BLK_STS_OK)
1962                 goto out;
1963
1964         ret = BLK_STS_RESOURCE;
1965         if (!get_device(&sdev->sdev_gendev))
1966                 goto out;
1967
1968         if (!scsi_dev_queue_ready(q, sdev))
1969                 goto out_put_device;
1970         if (!scsi_target_queue_ready(shost, sdev))
1971                 goto out_dec_device_busy;
1972         if (!scsi_host_queue_ready(q, shost, sdev))
1973                 goto out_dec_target_busy;
1974
1975         if (!(req->rq_flags & RQF_DONTPREP)) {
1976                 ret = prep_to_mq(scsi_mq_prep_fn(req));
1977                 if (ret != BLK_STS_OK)
1978                         goto out_dec_host_busy;
1979                 req->rq_flags |= RQF_DONTPREP;
1980         } else {
1981                 blk_mq_start_request(req);
1982         }
1983
1984         if (sdev->simple_tags)
1985                 cmd->flags |= SCMD_TAGGED;
1986         else
1987                 cmd->flags &= ~SCMD_TAGGED;
1988
1989         scsi_init_cmd_errh(cmd);
1990         cmd->scsi_done = scsi_mq_done;
1991
1992         reason = scsi_dispatch_cmd(cmd);
1993         if (reason) {
1994                 scsi_set_blocked(cmd, reason);
1995                 ret = BLK_STS_RESOURCE;
1996                 goto out_dec_host_busy;
1997         }
1998
1999         return BLK_STS_OK;
2000
2001 out_dec_host_busy:
2002         atomic_dec(&shost->host_busy);
2003 out_dec_target_busy:
2004         if (scsi_target(sdev)->can_queue > 0)
2005                 atomic_dec(&scsi_target(sdev)->target_busy);
2006 out_dec_device_busy:
2007         atomic_dec(&sdev->device_busy);
2008 out_put_device:
2009         put_device(&sdev->sdev_gendev);
2010 out:
2011         switch (ret) {
2012         case BLK_STS_OK:
2013                 break;
2014         case BLK_STS_RESOURCE:
2015                 if (atomic_read(&sdev->device_busy) == 0 &&
2016                     !scsi_device_blocked(sdev))
2017                         blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
2018                 break;
2019         default:
2020                 /*
2021                  * Make sure to release all allocated ressources when
2022                  * we hit an error, as we will never see this command
2023                  * again.
2024                  */
2025                 if (req->rq_flags & RQF_DONTPREP)
2026                         scsi_mq_uninit_cmd(cmd);
2027                 break;
2028         }
2029         return ret;
2030 }
2031
2032 static enum blk_eh_timer_return scsi_timeout(struct request *req,
2033                 bool reserved)
2034 {
2035         if (reserved)
2036                 return BLK_EH_RESET_TIMER;
2037         return scsi_times_out(req);
2038 }
2039
2040 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
2041                                 unsigned int hctx_idx, unsigned int numa_node)
2042 {
2043         struct Scsi_Host *shost = set->driver_data;
2044         const bool unchecked_isa_dma = shost->unchecked_isa_dma;
2045         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2046         struct scatterlist *sg;
2047
2048         if (unchecked_isa_dma)
2049                 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
2050         cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
2051                                                     GFP_KERNEL, numa_node);
2052         if (!cmd->sense_buffer)
2053                 return -ENOMEM;
2054         cmd->req.sense = cmd->sense_buffer;
2055
2056         if (scsi_host_get_prot(shost)) {
2057                 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
2058                         shost->hostt->cmd_size;
2059                 cmd->prot_sdb = (void *)sg + scsi_mq_sgl_size(shost);
2060         }
2061
2062         return 0;
2063 }
2064
2065 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
2066                                  unsigned int hctx_idx)
2067 {
2068         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2069
2070         scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
2071                                cmd->sense_buffer);
2072 }
2073
2074 static int scsi_map_queues(struct blk_mq_tag_set *set)
2075 {
2076         struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
2077
2078         if (shost->hostt->map_queues)
2079                 return shost->hostt->map_queues(shost);
2080         return blk_mq_map_queues(set);
2081 }
2082
2083 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
2084 {
2085         struct device *host_dev;
2086         u64 bounce_limit = 0xffffffff;
2087
2088         if (shost->unchecked_isa_dma)
2089                 return BLK_BOUNCE_ISA;
2090         /*
2091          * Platforms with virtual-DMA translation
2092          * hardware have no practical limit.
2093          */
2094         if (!PCI_DMA_BUS_IS_PHYS)
2095                 return BLK_BOUNCE_ANY;
2096
2097         host_dev = scsi_get_device(shost);
2098         if (host_dev && host_dev->dma_mask)
2099                 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2100
2101         return bounce_limit;
2102 }
2103
2104 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2105 {
2106         struct device *dev = shost->dma_dev;
2107
2108         queue_flag_set_unlocked(QUEUE_FLAG_SCSI_PASSTHROUGH, q);
2109
2110         /*
2111          * this limit is imposed by hardware restrictions
2112          */
2113         blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2114                                         SG_MAX_SEGMENTS));
2115
2116         if (scsi_host_prot_dma(shost)) {
2117                 shost->sg_prot_tablesize =
2118                         min_not_zero(shost->sg_prot_tablesize,
2119                                      (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2120                 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2121                 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2122         }
2123
2124         blk_queue_max_hw_sectors(q, shost->max_sectors);
2125         blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2126         blk_queue_segment_boundary(q, shost->dma_boundary);
2127         dma_set_seg_boundary(dev, shost->dma_boundary);
2128
2129         blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2130
2131         if (!shost->use_clustering)
2132                 q->limits.cluster = 0;
2133
2134         /*
2135          * set a reasonable default alignment on word boundaries: the
2136          * host and device may alter it using
2137          * blk_queue_update_dma_alignment() later.
2138          */
2139         blk_queue_dma_alignment(q, 0x03);
2140 }
2141 EXPORT_SYMBOL_GPL(__scsi_init_queue);
2142
2143 static int scsi_old_init_rq(struct request_queue *q, struct request *rq,
2144                             gfp_t gfp)
2145 {
2146         struct Scsi_Host *shost = q->rq_alloc_data;
2147         const bool unchecked_isa_dma = shost->unchecked_isa_dma;
2148         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2149
2150         memset(cmd, 0, sizeof(*cmd));
2151
2152         if (unchecked_isa_dma)
2153                 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
2154         cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma, gfp,
2155                                                     NUMA_NO_NODE);
2156         if (!cmd->sense_buffer)
2157                 goto fail;
2158         cmd->req.sense = cmd->sense_buffer;
2159
2160         if (scsi_host_get_prot(shost) >= SHOST_DIX_TYPE0_PROTECTION) {
2161                 cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp);
2162                 if (!cmd->prot_sdb)
2163                         goto fail_free_sense;
2164         }
2165
2166         return 0;
2167
2168 fail_free_sense:
2169         scsi_free_sense_buffer(unchecked_isa_dma, cmd->sense_buffer);
2170 fail:
2171         return -ENOMEM;
2172 }
2173
2174 static void scsi_old_exit_rq(struct request_queue *q, struct request *rq)
2175 {
2176         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2177
2178         if (cmd->prot_sdb)
2179                 kmem_cache_free(scsi_sdb_cache, cmd->prot_sdb);
2180         scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
2181                                cmd->sense_buffer);
2182 }
2183
2184 struct request_queue *scsi_old_alloc_queue(struct scsi_device *sdev)
2185 {
2186         struct Scsi_Host *shost = sdev->host;
2187         struct request_queue *q;
2188
2189         q = blk_alloc_queue_node(GFP_KERNEL, NUMA_NO_NODE);
2190         if (!q)
2191                 return NULL;
2192         q->cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
2193         q->rq_alloc_data = shost;
2194         q->request_fn = scsi_request_fn;
2195         q->init_rq_fn = scsi_old_init_rq;
2196         q->exit_rq_fn = scsi_old_exit_rq;
2197         q->initialize_rq_fn = scsi_initialize_rq;
2198
2199         if (blk_init_allocated_queue(q) < 0) {
2200                 blk_cleanup_queue(q);
2201                 return NULL;
2202         }
2203
2204         __scsi_init_queue(shost, q);
2205         blk_queue_prep_rq(q, scsi_prep_fn);
2206         blk_queue_unprep_rq(q, scsi_unprep_fn);
2207         blk_queue_softirq_done(q, scsi_softirq_done);
2208         blk_queue_rq_timed_out(q, scsi_times_out);
2209         blk_queue_lld_busy(q, scsi_lld_busy);
2210         return q;
2211 }
2212
2213 static const struct blk_mq_ops scsi_mq_ops = {
2214         .queue_rq       = scsi_queue_rq,
2215         .complete       = scsi_softirq_done,
2216         .timeout        = scsi_timeout,
2217 #ifdef CONFIG_BLK_DEBUG_FS
2218         .show_rq        = scsi_show_rq,
2219 #endif
2220         .init_request   = scsi_mq_init_request,
2221         .exit_request   = scsi_mq_exit_request,
2222         .initialize_rq_fn = scsi_initialize_rq,
2223         .map_queues     = scsi_map_queues,
2224 };
2225
2226 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2227 {
2228         sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2229         if (IS_ERR(sdev->request_queue))
2230                 return NULL;
2231
2232         sdev->request_queue->queuedata = sdev;
2233         __scsi_init_queue(sdev->host, sdev->request_queue);
2234         return sdev->request_queue;
2235 }
2236
2237 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2238 {
2239         unsigned int cmd_size, sgl_size;
2240
2241         sgl_size = scsi_mq_sgl_size(shost);
2242         cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2243         if (scsi_host_get_prot(shost))
2244                 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2245
2246         memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2247         shost->tag_set.ops = &scsi_mq_ops;
2248         shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2249         shost->tag_set.queue_depth = shost->can_queue;
2250         shost->tag_set.cmd_size = cmd_size;
2251         shost->tag_set.numa_node = NUMA_NO_NODE;
2252         shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2253         shost->tag_set.flags |=
2254                 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2255         shost->tag_set.driver_data = shost;
2256
2257         return blk_mq_alloc_tag_set(&shost->tag_set);
2258 }
2259
2260 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2261 {
2262         blk_mq_free_tag_set(&shost->tag_set);
2263 }
2264
2265 /**
2266  * scsi_device_from_queue - return sdev associated with a request_queue
2267  * @q: The request queue to return the sdev from
2268  *
2269  * Return the sdev associated with a request queue or NULL if the
2270  * request_queue does not reference a SCSI device.
2271  */
2272 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2273 {
2274         struct scsi_device *sdev = NULL;
2275
2276         if (q->mq_ops) {
2277                 if (q->mq_ops == &scsi_mq_ops)
2278                         sdev = q->queuedata;
2279         } else if (q->request_fn == scsi_request_fn)
2280                 sdev = q->queuedata;
2281         if (!sdev || !get_device(&sdev->sdev_gendev))
2282                 sdev = NULL;
2283
2284         return sdev;
2285 }
2286 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2287
2288 /*
2289  * Function:    scsi_block_requests()
2290  *
2291  * Purpose:     Utility function used by low-level drivers to prevent further
2292  *              commands from being queued to the device.
2293  *
2294  * Arguments:   shost       - Host in question
2295  *
2296  * Returns:     Nothing
2297  *
2298  * Lock status: No locks are assumed held.
2299  *
2300  * Notes:       There is no timer nor any other means by which the requests
2301  *              get unblocked other than the low-level driver calling
2302  *              scsi_unblock_requests().
2303  */
2304 void scsi_block_requests(struct Scsi_Host *shost)
2305 {
2306         shost->host_self_blocked = 1;
2307 }
2308 EXPORT_SYMBOL(scsi_block_requests);
2309
2310 /*
2311  * Function:    scsi_unblock_requests()
2312  *
2313  * Purpose:     Utility function used by low-level drivers to allow further
2314  *              commands from being queued to the device.
2315  *
2316  * Arguments:   shost       - Host in question
2317  *
2318  * Returns:     Nothing
2319  *
2320  * Lock status: No locks are assumed held.
2321  *
2322  * Notes:       There is no timer nor any other means by which the requests
2323  *              get unblocked other than the low-level driver calling
2324  *              scsi_unblock_requests().
2325  *
2326  *              This is done as an API function so that changes to the
2327  *              internals of the scsi mid-layer won't require wholesale
2328  *              changes to drivers that use this feature.
2329  */
2330 void scsi_unblock_requests(struct Scsi_Host *shost)
2331 {
2332         shost->host_self_blocked = 0;
2333         scsi_run_host_queues(shost);
2334 }
2335 EXPORT_SYMBOL(scsi_unblock_requests);
2336
2337 int __init scsi_init_queue(void)
2338 {
2339         scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2340                                            sizeof(struct scsi_data_buffer),
2341                                            0, 0, NULL);
2342         if (!scsi_sdb_cache) {
2343                 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2344                 return -ENOMEM;
2345         }
2346
2347         return 0;
2348 }
2349
2350 void scsi_exit_queue(void)
2351 {
2352         kmem_cache_destroy(scsi_sense_cache);
2353         kmem_cache_destroy(scsi_sense_isadma_cache);
2354         kmem_cache_destroy(scsi_sdb_cache);
2355 }
2356
2357 /**
2358  *      scsi_mode_select - issue a mode select
2359  *      @sdev:  SCSI device to be queried
2360  *      @pf:    Page format bit (1 == standard, 0 == vendor specific)
2361  *      @sp:    Save page bit (0 == don't save, 1 == save)
2362  *      @modepage: mode page being requested
2363  *      @buffer: request buffer (may not be smaller than eight bytes)
2364  *      @len:   length of request buffer.
2365  *      @timeout: command timeout
2366  *      @retries: number of retries before failing
2367  *      @data: returns a structure abstracting the mode header data
2368  *      @sshdr: place to put sense data (or NULL if no sense to be collected).
2369  *              must be SCSI_SENSE_BUFFERSIZE big.
2370  *
2371  *      Returns zero if successful; negative error number or scsi
2372  *      status on error
2373  *
2374  */
2375 int
2376 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2377                  unsigned char *buffer, int len, int timeout, int retries,
2378                  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2379 {
2380         unsigned char cmd[10];
2381         unsigned char *real_buffer;
2382         int ret;
2383
2384         memset(cmd, 0, sizeof(cmd));
2385         cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2386
2387         if (sdev->use_10_for_ms) {
2388                 if (len > 65535)
2389                         return -EINVAL;
2390                 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2391                 if (!real_buffer)
2392                         return -ENOMEM;
2393                 memcpy(real_buffer + 8, buffer, len);
2394                 len += 8;
2395                 real_buffer[0] = 0;
2396                 real_buffer[1] = 0;
2397                 real_buffer[2] = data->medium_type;
2398                 real_buffer[3] = data->device_specific;
2399                 real_buffer[4] = data->longlba ? 0x01 : 0;
2400                 real_buffer[5] = 0;
2401                 real_buffer[6] = data->block_descriptor_length >> 8;
2402                 real_buffer[7] = data->block_descriptor_length;
2403
2404                 cmd[0] = MODE_SELECT_10;
2405                 cmd[7] = len >> 8;
2406                 cmd[8] = len;
2407         } else {
2408                 if (len > 255 || data->block_descriptor_length > 255 ||
2409                     data->longlba)
2410                         return -EINVAL;
2411
2412                 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2413                 if (!real_buffer)
2414                         return -ENOMEM;
2415                 memcpy(real_buffer + 4, buffer, len);
2416                 len += 4;
2417                 real_buffer[0] = 0;
2418                 real_buffer[1] = data->medium_type;
2419                 real_buffer[2] = data->device_specific;
2420                 real_buffer[3] = data->block_descriptor_length;
2421                 
2422
2423                 cmd[0] = MODE_SELECT;
2424                 cmd[4] = len;
2425         }
2426
2427         ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2428                                sshdr, timeout, retries, NULL);
2429         kfree(real_buffer);
2430         return ret;
2431 }
2432 EXPORT_SYMBOL_GPL(scsi_mode_select);
2433
2434 /**
2435  *      scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2436  *      @sdev:  SCSI device to be queried
2437  *      @dbd:   set if mode sense will allow block descriptors to be returned
2438  *      @modepage: mode page being requested
2439  *      @buffer: request buffer (may not be smaller than eight bytes)
2440  *      @len:   length of request buffer.
2441  *      @timeout: command timeout
2442  *      @retries: number of retries before failing
2443  *      @data: returns a structure abstracting the mode header data
2444  *      @sshdr: place to put sense data (or NULL if no sense to be collected).
2445  *              must be SCSI_SENSE_BUFFERSIZE big.
2446  *
2447  *      Returns zero if unsuccessful, or the header offset (either 4
2448  *      or 8 depending on whether a six or ten byte command was
2449  *      issued) if successful.
2450  */
2451 int
2452 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2453                   unsigned char *buffer, int len, int timeout, int retries,
2454                   struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2455 {
2456         unsigned char cmd[12];
2457         int use_10_for_ms;
2458         int header_length;
2459         int result, retry_count = retries;
2460         struct scsi_sense_hdr my_sshdr;
2461
2462         memset(data, 0, sizeof(*data));
2463         memset(&cmd[0], 0, 12);
2464         cmd[1] = dbd & 0x18;    /* allows DBD and LLBA bits */
2465         cmd[2] = modepage;
2466
2467         /* caller might not be interested in sense, but we need it */
2468         if (!sshdr)
2469                 sshdr = &my_sshdr;
2470
2471  retry:
2472         use_10_for_ms = sdev->use_10_for_ms;
2473
2474         if (use_10_for_ms) {
2475                 if (len < 8)
2476                         len = 8;
2477
2478                 cmd[0] = MODE_SENSE_10;
2479                 cmd[8] = len;
2480                 header_length = 8;
2481         } else {
2482                 if (len < 4)
2483                         len = 4;
2484
2485                 cmd[0] = MODE_SENSE;
2486                 cmd[4] = len;
2487                 header_length = 4;
2488         }
2489
2490         memset(buffer, 0, len);
2491
2492         result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2493                                   sshdr, timeout, retries, NULL);
2494
2495         /* This code looks awful: what it's doing is making sure an
2496          * ILLEGAL REQUEST sense return identifies the actual command
2497          * byte as the problem.  MODE_SENSE commands can return
2498          * ILLEGAL REQUEST if the code page isn't supported */
2499
2500         if (use_10_for_ms && !scsi_status_is_good(result) &&
2501             (driver_byte(result) & DRIVER_SENSE)) {
2502                 if (scsi_sense_valid(sshdr)) {
2503                         if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2504                             (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2505                                 /* 
2506                                  * Invalid command operation code
2507                                  */
2508                                 sdev->use_10_for_ms = 0;
2509                                 goto retry;
2510                         }
2511                 }
2512         }
2513
2514         if(scsi_status_is_good(result)) {
2515                 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2516                              (modepage == 6 || modepage == 8))) {
2517                         /* Initio breakage? */
2518                         header_length = 0;
2519                         data->length = 13;
2520                         data->medium_type = 0;
2521                         data->device_specific = 0;
2522                         data->longlba = 0;
2523                         data->block_descriptor_length = 0;
2524                 } else if(use_10_for_ms) {
2525                         data->length = buffer[0]*256 + buffer[1] + 2;
2526                         data->medium_type = buffer[2];
2527                         data->device_specific = buffer[3];
2528                         data->longlba = buffer[4] & 0x01;
2529                         data->block_descriptor_length = buffer[6]*256
2530                                 + buffer[7];
2531                 } else {
2532                         data->length = buffer[0] + 1;
2533                         data->medium_type = buffer[1];
2534                         data->device_specific = buffer[2];
2535                         data->block_descriptor_length = buffer[3];
2536                 }
2537                 data->header_length = header_length;
2538         } else if ((status_byte(result) == CHECK_CONDITION) &&
2539                    scsi_sense_valid(sshdr) &&
2540                    sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2541                 retry_count--;
2542                 goto retry;
2543         }
2544
2545         return result;
2546 }
2547 EXPORT_SYMBOL(scsi_mode_sense);
2548
2549 /**
2550  *      scsi_test_unit_ready - test if unit is ready
2551  *      @sdev:  scsi device to change the state of.
2552  *      @timeout: command timeout
2553  *      @retries: number of retries before failing
2554  *      @sshdr: outpout pointer for decoded sense information.
2555  *
2556  *      Returns zero if unsuccessful or an error if TUR failed.  For
2557  *      removable media, UNIT_ATTENTION sets ->changed flag.
2558  **/
2559 int
2560 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2561                      struct scsi_sense_hdr *sshdr)
2562 {
2563         char cmd[] = {
2564                 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2565         };
2566         int result;
2567
2568         /* try to eat the UNIT_ATTENTION if there are enough retries */
2569         do {
2570                 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2571                                           timeout, retries, NULL);
2572                 if (sdev->removable && scsi_sense_valid(sshdr) &&
2573                     sshdr->sense_key == UNIT_ATTENTION)
2574                         sdev->changed = 1;
2575         } while (scsi_sense_valid(sshdr) &&
2576                  sshdr->sense_key == UNIT_ATTENTION && --retries);
2577
2578         return result;
2579 }
2580 EXPORT_SYMBOL(scsi_test_unit_ready);
2581
2582 /**
2583  *      scsi_device_set_state - Take the given device through the device state model.
2584  *      @sdev:  scsi device to change the state of.
2585  *      @state: state to change to.
2586  *
2587  *      Returns zero if successful or an error if the requested
2588  *      transition is illegal.
2589  */
2590 int
2591 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2592 {
2593         enum scsi_device_state oldstate = sdev->sdev_state;
2594
2595         if (state == oldstate)
2596                 return 0;
2597
2598         switch (state) {
2599         case SDEV_CREATED:
2600                 switch (oldstate) {
2601                 case SDEV_CREATED_BLOCK:
2602                         break;
2603                 default:
2604                         goto illegal;
2605                 }
2606                 break;
2607                         
2608         case SDEV_RUNNING:
2609                 switch (oldstate) {
2610                 case SDEV_CREATED:
2611                 case SDEV_OFFLINE:
2612                 case SDEV_TRANSPORT_OFFLINE:
2613                 case SDEV_QUIESCE:
2614                 case SDEV_BLOCK:
2615                         break;
2616                 default:
2617                         goto illegal;
2618                 }
2619                 break;
2620
2621         case SDEV_QUIESCE:
2622                 switch (oldstate) {
2623                 case SDEV_RUNNING:
2624                 case SDEV_OFFLINE:
2625                 case SDEV_TRANSPORT_OFFLINE:
2626                         break;
2627                 default:
2628                         goto illegal;
2629                 }
2630                 break;
2631
2632         case SDEV_OFFLINE:
2633         case SDEV_TRANSPORT_OFFLINE:
2634                 switch (oldstate) {
2635                 case SDEV_CREATED:
2636                 case SDEV_RUNNING:
2637                 case SDEV_QUIESCE:
2638                 case SDEV_BLOCK:
2639                         break;
2640                 default:
2641                         goto illegal;
2642                 }
2643                 break;
2644
2645         case SDEV_BLOCK:
2646                 switch (oldstate) {
2647                 case SDEV_RUNNING:
2648                 case SDEV_CREATED_BLOCK:
2649                         break;
2650                 default:
2651                         goto illegal;
2652                 }
2653                 break;
2654
2655         case SDEV_CREATED_BLOCK:
2656                 switch (oldstate) {
2657                 case SDEV_CREATED:
2658                         break;
2659                 default:
2660                         goto illegal;
2661                 }
2662                 break;
2663
2664         case SDEV_CANCEL:
2665                 switch (oldstate) {
2666                 case SDEV_CREATED:
2667                 case SDEV_RUNNING:
2668                 case SDEV_QUIESCE:
2669                 case SDEV_OFFLINE:
2670                 case SDEV_TRANSPORT_OFFLINE:
2671                         break;
2672                 default:
2673                         goto illegal;
2674                 }
2675                 break;
2676
2677         case SDEV_DEL:
2678                 switch (oldstate) {
2679                 case SDEV_CREATED:
2680                 case SDEV_RUNNING:
2681                 case SDEV_OFFLINE:
2682                 case SDEV_TRANSPORT_OFFLINE:
2683                 case SDEV_CANCEL:
2684                 case SDEV_BLOCK:
2685                 case SDEV_CREATED_BLOCK:
2686                         break;
2687                 default:
2688                         goto illegal;
2689                 }
2690                 break;
2691
2692         }
2693         sdev->sdev_state = state;
2694         sysfs_notify(&sdev->sdev_gendev.kobj, NULL, "state");
2695         return 0;
2696
2697  illegal:
2698         SCSI_LOG_ERROR_RECOVERY(1,
2699                                 sdev_printk(KERN_ERR, sdev,
2700                                             "Illegal state transition %s->%s",
2701                                             scsi_device_state_name(oldstate),
2702                                             scsi_device_state_name(state))
2703                                 );
2704         return -EINVAL;
2705 }
2706 EXPORT_SYMBOL(scsi_device_set_state);
2707
2708 /**
2709  *      sdev_evt_emit - emit a single SCSI device uevent
2710  *      @sdev: associated SCSI device
2711  *      @evt: event to emit
2712  *
2713  *      Send a single uevent (scsi_event) to the associated scsi_device.
2714  */
2715 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2716 {
2717         int idx = 0;
2718         char *envp[3];
2719
2720         switch (evt->evt_type) {
2721         case SDEV_EVT_MEDIA_CHANGE:
2722                 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2723                 break;
2724         case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2725                 scsi_rescan_device(&sdev->sdev_gendev);
2726                 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2727                 break;
2728         case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2729                 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2730                 break;
2731         case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2732                envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2733                 break;
2734         case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2735                 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2736                 break;
2737         case SDEV_EVT_LUN_CHANGE_REPORTED:
2738                 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2739                 break;
2740         case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2741                 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2742                 break;
2743         default:
2744                 /* do nothing */
2745                 break;
2746         }
2747
2748         envp[idx++] = NULL;
2749
2750         kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2751 }
2752
2753 /**
2754  *      sdev_evt_thread - send a uevent for each scsi event
2755  *      @work: work struct for scsi_device
2756  *
2757  *      Dispatch queued events to their associated scsi_device kobjects
2758  *      as uevents.
2759  */
2760 void scsi_evt_thread(struct work_struct *work)
2761 {
2762         struct scsi_device *sdev;
2763         enum scsi_device_event evt_type;
2764         LIST_HEAD(event_list);
2765
2766         sdev = container_of(work, struct scsi_device, event_work);
2767
2768         for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2769                 if (test_and_clear_bit(evt_type, sdev->pending_events))
2770                         sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2771
2772         while (1) {
2773                 struct scsi_event *evt;
2774                 struct list_head *this, *tmp;
2775                 unsigned long flags;
2776
2777                 spin_lock_irqsave(&sdev->list_lock, flags);
2778                 list_splice_init(&sdev->event_list, &event_list);
2779                 spin_unlock_irqrestore(&sdev->list_lock, flags);
2780
2781                 if (list_empty(&event_list))
2782                         break;
2783
2784                 list_for_each_safe(this, tmp, &event_list) {
2785                         evt = list_entry(this, struct scsi_event, node);
2786                         list_del(&evt->node);
2787                         scsi_evt_emit(sdev, evt);
2788                         kfree(evt);
2789                 }
2790         }
2791 }
2792
2793 /**
2794  *      sdev_evt_send - send asserted event to uevent thread
2795  *      @sdev: scsi_device event occurred on
2796  *      @evt: event to send
2797  *
2798  *      Assert scsi device event asynchronously.
2799  */
2800 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2801 {
2802         unsigned long flags;
2803
2804 #if 0
2805         /* FIXME: currently this check eliminates all media change events
2806          * for polled devices.  Need to update to discriminate between AN
2807          * and polled events */
2808         if (!test_bit(evt->evt_type, sdev->supported_events)) {
2809                 kfree(evt);
2810                 return;
2811         }
2812 #endif
2813
2814         spin_lock_irqsave(&sdev->list_lock, flags);
2815         list_add_tail(&evt->node, &sdev->event_list);
2816         schedule_work(&sdev->event_work);
2817         spin_unlock_irqrestore(&sdev->list_lock, flags);
2818 }
2819 EXPORT_SYMBOL_GPL(sdev_evt_send);
2820
2821 /**
2822  *      sdev_evt_alloc - allocate a new scsi event
2823  *      @evt_type: type of event to allocate
2824  *      @gfpflags: GFP flags for allocation
2825  *
2826  *      Allocates and returns a new scsi_event.
2827  */
2828 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2829                                   gfp_t gfpflags)
2830 {
2831         struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2832         if (!evt)
2833                 return NULL;
2834
2835         evt->evt_type = evt_type;
2836         INIT_LIST_HEAD(&evt->node);
2837
2838         /* evt_type-specific initialization, if any */
2839         switch (evt_type) {
2840         case SDEV_EVT_MEDIA_CHANGE:
2841         case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2842         case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2843         case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2844         case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2845         case SDEV_EVT_LUN_CHANGE_REPORTED:
2846         case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2847         default:
2848                 /* do nothing */
2849                 break;
2850         }
2851
2852         return evt;
2853 }
2854 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2855
2856 /**
2857  *      sdev_evt_send_simple - send asserted event to uevent thread
2858  *      @sdev: scsi_device event occurred on
2859  *      @evt_type: type of event to send
2860  *      @gfpflags: GFP flags for allocation
2861  *
2862  *      Assert scsi device event asynchronously, given an event type.
2863  */
2864 void sdev_evt_send_simple(struct scsi_device *sdev,
2865                           enum scsi_device_event evt_type, gfp_t gfpflags)
2866 {
2867         struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2868         if (!evt) {
2869                 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2870                             evt_type);
2871                 return;
2872         }
2873
2874         sdev_evt_send(sdev, evt);
2875 }
2876 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2877
2878 /**
2879  * scsi_request_fn_active() - number of kernel threads inside scsi_request_fn()
2880  * @sdev: SCSI device to count the number of scsi_request_fn() callers for.
2881  */
2882 static int scsi_request_fn_active(struct scsi_device *sdev)
2883 {
2884         struct request_queue *q = sdev->request_queue;
2885         int request_fn_active;
2886
2887         WARN_ON_ONCE(sdev->host->use_blk_mq);
2888
2889         spin_lock_irq(q->queue_lock);
2890         request_fn_active = q->request_fn_active;
2891         spin_unlock_irq(q->queue_lock);
2892
2893         return request_fn_active;
2894 }
2895
2896 /**
2897  * scsi_wait_for_queuecommand() - wait for ongoing queuecommand() calls
2898  * @sdev: SCSI device pointer.
2899  *
2900  * Wait until the ongoing shost->hostt->queuecommand() calls that are
2901  * invoked from scsi_request_fn() have finished.
2902  */
2903 static void scsi_wait_for_queuecommand(struct scsi_device *sdev)
2904 {
2905         WARN_ON_ONCE(sdev->host->use_blk_mq);
2906
2907         while (scsi_request_fn_active(sdev))
2908                 msleep(20);
2909 }
2910
2911 /**
2912  *      scsi_device_quiesce - Block user issued commands.
2913  *      @sdev:  scsi device to quiesce.
2914  *
2915  *      This works by trying to transition to the SDEV_QUIESCE state
2916  *      (which must be a legal transition).  When the device is in this
2917  *      state, only special requests will be accepted, all others will
2918  *      be deferred.  Since special requests may also be requeued requests,
2919  *      a successful return doesn't guarantee the device will be 
2920  *      totally quiescent.
2921  *
2922  *      Must be called with user context, may sleep.
2923  *
2924  *      Returns zero if unsuccessful or an error if not.
2925  */
2926 int
2927 scsi_device_quiesce(struct scsi_device *sdev)
2928 {
2929         int err;
2930
2931         mutex_lock(&sdev->state_mutex);
2932         err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2933         mutex_unlock(&sdev->state_mutex);
2934
2935         if (err)
2936                 return err;
2937
2938         scsi_run_queue(sdev->request_queue);
2939         while (atomic_read(&sdev->device_busy)) {
2940                 msleep_interruptible(200);
2941                 scsi_run_queue(sdev->request_queue);
2942         }
2943         return 0;
2944 }
2945 EXPORT_SYMBOL(scsi_device_quiesce);
2946
2947 /**
2948  *      scsi_device_resume - Restart user issued commands to a quiesced device.
2949  *      @sdev:  scsi device to resume.
2950  *
2951  *      Moves the device from quiesced back to running and restarts the
2952  *      queues.
2953  *
2954  *      Must be called with user context, may sleep.
2955  */
2956 void scsi_device_resume(struct scsi_device *sdev)
2957 {
2958         /* check if the device state was mutated prior to resume, and if
2959          * so assume the state is being managed elsewhere (for example
2960          * device deleted during suspend)
2961          */
2962         mutex_lock(&sdev->state_mutex);
2963         if (sdev->sdev_state == SDEV_QUIESCE &&
2964             scsi_device_set_state(sdev, SDEV_RUNNING) == 0)
2965                 scsi_run_queue(sdev->request_queue);
2966         mutex_unlock(&sdev->state_mutex);
2967 }
2968 EXPORT_SYMBOL(scsi_device_resume);
2969
2970 static void
2971 device_quiesce_fn(struct scsi_device *sdev, void *data)
2972 {
2973         scsi_device_quiesce(sdev);
2974 }
2975
2976 void
2977 scsi_target_quiesce(struct scsi_target *starget)
2978 {
2979         starget_for_each_device(starget, NULL, device_quiesce_fn);
2980 }
2981 EXPORT_SYMBOL(scsi_target_quiesce);
2982
2983 static void
2984 device_resume_fn(struct scsi_device *sdev, void *data)
2985 {
2986         scsi_device_resume(sdev);
2987 }
2988
2989 void
2990 scsi_target_resume(struct scsi_target *starget)
2991 {
2992         starget_for_each_device(starget, NULL, device_resume_fn);
2993 }
2994 EXPORT_SYMBOL(scsi_target_resume);
2995
2996 /**
2997  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2998  * @sdev: device to block
2999  *
3000  * Pause SCSI command processing on the specified device. Does not sleep.
3001  *
3002  * Returns zero if successful or a negative error code upon failure.
3003  *
3004  * Notes:
3005  * This routine transitions the device to the SDEV_BLOCK state (which must be
3006  * a legal transition). When the device is in this state, command processing
3007  * is paused until the device leaves the SDEV_BLOCK state. See also
3008  * scsi_internal_device_unblock_nowait().
3009  */
3010 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
3011 {
3012         struct request_queue *q = sdev->request_queue;
3013         unsigned long flags;
3014         int err = 0;
3015
3016         err = scsi_device_set_state(sdev, SDEV_BLOCK);
3017         if (err) {
3018                 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
3019
3020                 if (err)
3021                         return err;
3022         }
3023
3024         /* 
3025          * The device has transitioned to SDEV_BLOCK.  Stop the
3026          * block layer from calling the midlayer with this device's
3027          * request queue. 
3028          */
3029         if (q->mq_ops) {
3030                 blk_mq_quiesce_queue_nowait(q);
3031         } else {
3032                 spin_lock_irqsave(q->queue_lock, flags);
3033                 blk_stop_queue(q);
3034                 spin_unlock_irqrestore(q->queue_lock, flags);
3035         }
3036
3037         return 0;
3038 }
3039 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
3040
3041 /**
3042  * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
3043  * @sdev: device to block
3044  *
3045  * Pause SCSI command processing on the specified device and wait until all
3046  * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
3047  *
3048  * Returns zero if successful or a negative error code upon failure.
3049  *
3050  * Note:
3051  * This routine transitions the device to the SDEV_BLOCK state (which must be
3052  * a legal transition). When the device is in this state, command processing
3053  * is paused until the device leaves the SDEV_BLOCK state. See also
3054  * scsi_internal_device_unblock().
3055  *
3056  * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
3057  * scsi_internal_device_block() has blocked a SCSI device and also
3058  * remove the rport mutex lock and unlock calls from srp_queuecommand().
3059  */
3060 static int scsi_internal_device_block(struct scsi_device *sdev)
3061 {
3062         struct request_queue *q = sdev->request_queue;
3063         int err;
3064
3065         mutex_lock(&sdev->state_mutex);
3066         err = scsi_internal_device_block_nowait(sdev);
3067         if (err == 0) {
3068                 if (q->mq_ops)
3069                         blk_mq_quiesce_queue(q);
3070                 else
3071                         scsi_wait_for_queuecommand(sdev);
3072         }
3073         mutex_unlock(&sdev->state_mutex);
3074
3075         return err;
3076 }
3077  
3078 void scsi_start_queue(struct scsi_device *sdev)
3079 {
3080         struct request_queue *q = sdev->request_queue;
3081         unsigned long flags;
3082
3083         if (q->mq_ops) {
3084                 blk_mq_unquiesce_queue(q);
3085         } else {
3086                 spin_lock_irqsave(q->queue_lock, flags);
3087                 blk_start_queue(q);
3088                 spin_unlock_irqrestore(q->queue_lock, flags);
3089         }
3090 }
3091
3092 /**
3093  * scsi_internal_device_unblock_nowait - resume a device after a block request
3094  * @sdev:       device to resume
3095  * @new_state:  state to set the device to after unblocking
3096  *
3097  * Restart the device queue for a previously suspended SCSI device. Does not
3098  * sleep.
3099  *
3100  * Returns zero if successful or a negative error code upon failure.
3101  *
3102  * Notes:
3103  * This routine transitions the device to the SDEV_RUNNING state or to one of
3104  * the offline states (which must be a legal transition) allowing the midlayer
3105  * to goose the queue for this device.
3106  */
3107 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
3108                                         enum scsi_device_state new_state)
3109 {
3110         /*
3111          * Try to transition the scsi device to SDEV_RUNNING or one of the
3112          * offlined states and goose the device queue if successful.
3113          */
3114         switch (sdev->sdev_state) {
3115         case SDEV_BLOCK:
3116         case SDEV_TRANSPORT_OFFLINE:
3117                 sdev->sdev_state = new_state;
3118                 sysfs_notify(&sdev->sdev_gendev.kobj, NULL, "state");
3119                 break;
3120         case SDEV_CREATED_BLOCK:
3121                 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3122                     new_state == SDEV_OFFLINE)
3123                         sdev->sdev_state = new_state;
3124                 else
3125                         sdev->sdev_state = SDEV_CREATED;
3126                 sysfs_notify(&sdev->sdev_gendev.kobj, NULL, "state");
3127                 break;
3128         case SDEV_CANCEL:
3129         case SDEV_OFFLINE:
3130                 break;
3131         default:
3132                 return -EINVAL;
3133         }
3134         scsi_start_queue(sdev);
3135
3136         return 0;
3137 }
3138 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
3139
3140 /**
3141  * scsi_internal_device_unblock - resume a device after a block request
3142  * @sdev:       device to resume
3143  * @new_state:  state to set the device to after unblocking
3144  *
3145  * Restart the device queue for a previously suspended SCSI device. May sleep.
3146  *
3147  * Returns zero if successful or a negative error code upon failure.
3148  *
3149  * Notes:
3150  * This routine transitions the device to the SDEV_RUNNING state or to one of
3151  * the offline states (which must be a legal transition) allowing the midlayer
3152  * to goose the queue for this device.
3153  */
3154 static int scsi_internal_device_unblock(struct scsi_device *sdev,
3155                                         enum scsi_device_state new_state)
3156 {
3157         int ret;
3158
3159         mutex_lock(&sdev->state_mutex);
3160         ret = scsi_internal_device_unblock_nowait(sdev, new_state);
3161         mutex_unlock(&sdev->state_mutex);
3162
3163         return ret;
3164 }
3165
3166 static void
3167 device_block(struct scsi_device *sdev, void *data)
3168 {
3169         scsi_internal_device_block(sdev);
3170 }
3171
3172 static int
3173 target_block(struct device *dev, void *data)
3174 {
3175         if (scsi_is_target_device(dev))
3176                 starget_for_each_device(to_scsi_target(dev), NULL,
3177                                         device_block);
3178         return 0;
3179 }
3180
3181 void
3182 scsi_target_block(struct device *dev)
3183 {
3184         if (scsi_is_target_device(dev))
3185                 starget_for_each_device(to_scsi_target(dev), NULL,
3186                                         device_block);
3187         else
3188                 device_for_each_child(dev, NULL, target_block);
3189 }
3190 EXPORT_SYMBOL_GPL(scsi_target_block);
3191
3192 static void
3193 device_unblock(struct scsi_device *sdev, void *data)
3194 {
3195         scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3196 }
3197
3198 static int
3199 target_unblock(struct device *dev, void *data)
3200 {
3201         if (scsi_is_target_device(dev))
3202                 starget_for_each_device(to_scsi_target(dev), data,
3203                                         device_unblock);
3204         return 0;
3205 }
3206
3207 void
3208 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3209 {
3210         if (scsi_is_target_device(dev))
3211                 starget_for_each_device(to_scsi_target(dev), &new_state,
3212                                         device_unblock);
3213         else
3214                 device_for_each_child(dev, &new_state, target_unblock);
3215 }
3216 EXPORT_SYMBOL_GPL(scsi_target_unblock);
3217
3218 /**
3219  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3220  * @sgl:        scatter-gather list
3221  * @sg_count:   number of segments in sg
3222  * @offset:     offset in bytes into sg, on return offset into the mapped area
3223  * @len:        bytes to map, on return number of bytes mapped
3224  *
3225  * Returns virtual address of the start of the mapped page
3226  */
3227 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3228                           size_t *offset, size_t *len)
3229 {
3230         int i;
3231         size_t sg_len = 0, len_complete = 0;
3232         struct scatterlist *sg;
3233         struct page *page;
3234
3235         WARN_ON(!irqs_disabled());
3236
3237         for_each_sg(sgl, sg, sg_count, i) {
3238                 len_complete = sg_len; /* Complete sg-entries */
3239                 sg_len += sg->length;
3240                 if (sg_len > *offset)
3241                         break;
3242         }
3243
3244         if (unlikely(i == sg_count)) {
3245                 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3246                         "elements %d\n",
3247                        __func__, sg_len, *offset, sg_count);
3248                 WARN_ON(1);
3249                 return NULL;
3250         }
3251
3252         /* Offset starting from the beginning of first page in this sg-entry */
3253         *offset = *offset - len_complete + sg->offset;
3254
3255         /* Assumption: contiguous pages can be accessed as "page + i" */
3256         page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3257         *offset &= ~PAGE_MASK;
3258
3259         /* Bytes in this sg-entry from *offset to the end of the page */
3260         sg_len = PAGE_SIZE - *offset;
3261         if (*len > sg_len)
3262                 *len = sg_len;
3263
3264         return kmap_atomic(page);
3265 }
3266 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3267
3268 /**
3269  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3270  * @virt:       virtual address to be unmapped
3271  */
3272 void scsi_kunmap_atomic_sg(void *virt)
3273 {
3274         kunmap_atomic(virt);
3275 }
3276 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3277
3278 void sdev_disable_disk_events(struct scsi_device *sdev)
3279 {
3280         atomic_inc(&sdev->disk_events_disable_depth);
3281 }
3282 EXPORT_SYMBOL(sdev_disable_disk_events);
3283
3284 void sdev_enable_disk_events(struct scsi_device *sdev)
3285 {
3286         if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3287                 return;
3288         atomic_dec(&sdev->disk_events_disable_depth);
3289 }
3290 EXPORT_SYMBOL(sdev_enable_disk_events);
3291
3292 /**
3293  * scsi_vpd_lun_id - return a unique device identification
3294  * @sdev: SCSI device
3295  * @id:   buffer for the identification
3296  * @id_len:  length of the buffer
3297  *
3298  * Copies a unique device identification into @id based
3299  * on the information in the VPD page 0x83 of the device.
3300  * The string will be formatted as a SCSI name string.
3301  *
3302  * Returns the length of the identification or error on failure.
3303  * If the identifier is longer than the supplied buffer the actual
3304  * identifier length is returned and the buffer is not zero-padded.
3305  */
3306 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3307 {
3308         u8 cur_id_type = 0xff;
3309         u8 cur_id_size = 0;
3310         const unsigned char *d, *cur_id_str;
3311         const struct scsi_vpd *vpd_pg83;
3312         int id_size = -EINVAL;
3313
3314         rcu_read_lock();
3315         vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3316         if (!vpd_pg83) {
3317                 rcu_read_unlock();
3318                 return -ENXIO;
3319         }
3320
3321         /*
3322          * Look for the correct descriptor.
3323          * Order of preference for lun descriptor:
3324          * - SCSI name string
3325          * - NAA IEEE Registered Extended
3326          * - EUI-64 based 16-byte
3327          * - EUI-64 based 12-byte
3328          * - NAA IEEE Registered
3329          * - NAA IEEE Extended
3330          * - T10 Vendor ID
3331          * as longer descriptors reduce the likelyhood
3332          * of identification clashes.
3333          */
3334
3335         /* The id string must be at least 20 bytes + terminating NULL byte */
3336         if (id_len < 21) {
3337                 rcu_read_unlock();
3338                 return -EINVAL;
3339         }
3340
3341         memset(id, 0, id_len);
3342         d = vpd_pg83->data + 4;
3343         while (d < vpd_pg83->data + vpd_pg83->len) {
3344                 /* Skip designators not referring to the LUN */
3345                 if ((d[1] & 0x30) != 0x00)
3346                         goto next_desig;
3347
3348                 switch (d[1] & 0xf) {
3349                 case 0x1:
3350                         /* T10 Vendor ID */
3351                         if (cur_id_size > d[3])
3352                                 break;
3353                         /* Prefer anything */
3354                         if (cur_id_type > 0x01 && cur_id_type != 0xff)
3355                                 break;
3356                         cur_id_size = d[3];
3357                         if (cur_id_size + 4 > id_len)
3358                                 cur_id_size = id_len - 4;
3359                         cur_id_str = d + 4;
3360                         cur_id_type = d[1] & 0xf;
3361                         id_size = snprintf(id, id_len, "t10.%*pE",
3362                                            cur_id_size, cur_id_str);
3363                         break;
3364                 case 0x2:
3365                         /* EUI-64 */
3366                         if (cur_id_size > d[3])
3367                                 break;
3368                         /* Prefer NAA IEEE Registered Extended */
3369                         if (cur_id_type == 0x3 &&
3370                             cur_id_size == d[3])
3371                                 break;
3372                         cur_id_size = d[3];
3373                         cur_id_str = d + 4;
3374                         cur_id_type = d[1] & 0xf;
3375                         switch (cur_id_size) {
3376                         case 8:
3377                                 id_size = snprintf(id, id_len,
3378                                                    "eui.%8phN",
3379                                                    cur_id_str);
3380                                 break;
3381                         case 12:
3382                                 id_size = snprintf(id, id_len,
3383                                                    "eui.%12phN",
3384                                                    cur_id_str);
3385                                 break;
3386                         case 16:
3387                                 id_size = snprintf(id, id_len,
3388                                                    "eui.%16phN",
3389                                                    cur_id_str);
3390                                 break;
3391                         default:
3392                                 cur_id_size = 0;
3393                                 break;
3394                         }
3395                         break;
3396                 case 0x3:
3397                         /* NAA */
3398                         if (cur_id_size > d[3])
3399                                 break;
3400                         cur_id_size = d[3];
3401                         cur_id_str = d + 4;
3402                         cur_id_type = d[1] & 0xf;
3403                         switch (cur_id_size) {
3404                         case 8:
3405                                 id_size = snprintf(id, id_len,
3406                                                    "naa.%8phN",
3407                                                    cur_id_str);
3408                                 break;
3409                         case 16:
3410                                 id_size = snprintf(id, id_len,
3411                                                    "naa.%16phN",
3412                                                    cur_id_str);
3413                                 break;
3414                         default:
3415                                 cur_id_size = 0;
3416                                 break;
3417                         }
3418                         break;
3419                 case 0x8:
3420                         /* SCSI name string */
3421                         if (cur_id_size + 4 > d[3])
3422                                 break;
3423                         /* Prefer others for truncated descriptor */
3424                         if (cur_id_size && d[3] > id_len)
3425                                 break;
3426                         cur_id_size = id_size = d[3];
3427                         cur_id_str = d + 4;
3428                         cur_id_type = d[1] & 0xf;
3429                         if (cur_id_size >= id_len)
3430                                 cur_id_size = id_len - 1;
3431                         memcpy(id, cur_id_str, cur_id_size);
3432                         /* Decrease priority for truncated descriptor */
3433                         if (cur_id_size != id_size)
3434                                 cur_id_size = 6;
3435                         break;
3436                 default:
3437                         break;
3438                 }
3439 next_desig:
3440                 d += d[3] + 4;
3441         }
3442         rcu_read_unlock();
3443
3444         return id_size;
3445 }
3446 EXPORT_SYMBOL(scsi_vpd_lun_id);
3447
3448 /*
3449  * scsi_vpd_tpg_id - return a target port group identifier
3450  * @sdev: SCSI device
3451  *
3452  * Returns the Target Port Group identifier from the information
3453  * froom VPD page 0x83 of the device.
3454  *
3455  * Returns the identifier or error on failure.
3456  */
3457 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3458 {
3459         const unsigned char *d;
3460         const struct scsi_vpd *vpd_pg83;
3461         int group_id = -EAGAIN, rel_port = -1;
3462
3463         rcu_read_lock();
3464         vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3465         if (!vpd_pg83) {
3466                 rcu_read_unlock();
3467                 return -ENXIO;
3468         }
3469
3470         d = vpd_pg83->data + 4;
3471         while (d < vpd_pg83->data + vpd_pg83->len) {
3472                 switch (d[1] & 0xf) {
3473                 case 0x4:
3474                         /* Relative target port */
3475                         rel_port = get_unaligned_be16(&d[6]);
3476                         break;
3477                 case 0x5:
3478                         /* Target port group */
3479                         group_id = get_unaligned_be16(&d[6]);
3480                         break;
3481                 default:
3482                         break;
3483                 }
3484                 d += d[3] + 4;
3485         }
3486         rcu_read_unlock();
3487
3488         if (group_id >= 0 && rel_id && rel_port != -1)
3489                 *rel_id = rel_port;
3490
3491         return group_id;
3492 }
3493 EXPORT_SYMBOL(scsi_vpd_tpg_id);