Merge branch 'for-linus-5.2' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad...
[sfrench/cifs-2.6.git] / drivers / scsi / scsi_lib.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 1999 Eric Youngdale
4  * Copyright (C) 2014 Christoph Hellwig
5  *
6  *  SCSI queueing library.
7  *      Initial versions: Eric Youngdale (eric@andante.org).
8  *                        Based upon conversations with large numbers
9  *                        of people at Linux Expo.
10  */
11
12 #include <linux/bio.h>
13 #include <linux/bitops.h>
14 #include <linux/blkdev.h>
15 #include <linux/completion.h>
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
23 #include <linux/blk-mq.h>
24 #include <linux/ratelimit.h>
25 #include <asm/unaligned.h>
26
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_cmnd.h>
29 #include <scsi/scsi_dbg.h>
30 #include <scsi/scsi_device.h>
31 #include <scsi/scsi_driver.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_host.h>
34 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
35 #include <scsi/scsi_dh.h>
36
37 #include <trace/events/scsi.h>
38
39 #include "scsi_debugfs.h"
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
42
43 /*
44  * Size of integrity metadata is usually small, 1 inline sg should
45  * cover normal cases.
46  */
47 #ifdef CONFIG_ARCH_NO_SG_CHAIN
48 #define  SCSI_INLINE_PROT_SG_CNT  0
49 #define  SCSI_INLINE_SG_CNT  0
50 #else
51 #define  SCSI_INLINE_PROT_SG_CNT  1
52 #define  SCSI_INLINE_SG_CNT  2
53 #endif
54
55 static struct kmem_cache *scsi_sdb_cache;
56 static struct kmem_cache *scsi_sense_cache;
57 static struct kmem_cache *scsi_sense_isadma_cache;
58 static DEFINE_MUTEX(scsi_sense_cache_mutex);
59
60 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
61
62 static inline struct kmem_cache *
63 scsi_select_sense_cache(bool unchecked_isa_dma)
64 {
65         return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
66 }
67
68 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
69                                    unsigned char *sense_buffer)
70 {
71         kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
72                         sense_buffer);
73 }
74
75 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
76         gfp_t gfp_mask, int numa_node)
77 {
78         return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
79                                      gfp_mask, numa_node);
80 }
81
82 int scsi_init_sense_cache(struct Scsi_Host *shost)
83 {
84         struct kmem_cache *cache;
85         int ret = 0;
86
87         cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
88         if (cache)
89                 return 0;
90
91         mutex_lock(&scsi_sense_cache_mutex);
92         if (shost->unchecked_isa_dma) {
93                 scsi_sense_isadma_cache =
94                         kmem_cache_create("scsi_sense_cache(DMA)",
95                                 SCSI_SENSE_BUFFERSIZE, 0,
96                                 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
97                 if (!scsi_sense_isadma_cache)
98                         ret = -ENOMEM;
99         } else {
100                 scsi_sense_cache =
101                         kmem_cache_create_usercopy("scsi_sense_cache",
102                                 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
103                                 0, SCSI_SENSE_BUFFERSIZE, NULL);
104                 if (!scsi_sense_cache)
105                         ret = -ENOMEM;
106         }
107
108         mutex_unlock(&scsi_sense_cache_mutex);
109         return ret;
110 }
111
112 /*
113  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
114  * not change behaviour from the previous unplug mechanism, experimentation
115  * may prove this needs changing.
116  */
117 #define SCSI_QUEUE_DELAY        3
118
119 static void
120 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
121 {
122         struct Scsi_Host *host = cmd->device->host;
123         struct scsi_device *device = cmd->device;
124         struct scsi_target *starget = scsi_target(device);
125
126         /*
127          * Set the appropriate busy bit for the device/host.
128          *
129          * If the host/device isn't busy, assume that something actually
130          * completed, and that we should be able to queue a command now.
131          *
132          * Note that the prior mid-layer assumption that any host could
133          * always queue at least one command is now broken.  The mid-layer
134          * will implement a user specifiable stall (see
135          * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
136          * if a command is requeued with no other commands outstanding
137          * either for the device or for the host.
138          */
139         switch (reason) {
140         case SCSI_MLQUEUE_HOST_BUSY:
141                 atomic_set(&host->host_blocked, host->max_host_blocked);
142                 break;
143         case SCSI_MLQUEUE_DEVICE_BUSY:
144         case SCSI_MLQUEUE_EH_RETRY:
145                 atomic_set(&device->device_blocked,
146                            device->max_device_blocked);
147                 break;
148         case SCSI_MLQUEUE_TARGET_BUSY:
149                 atomic_set(&starget->target_blocked,
150                            starget->max_target_blocked);
151                 break;
152         }
153 }
154
155 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
156 {
157         if (cmd->request->rq_flags & RQF_DONTPREP) {
158                 cmd->request->rq_flags &= ~RQF_DONTPREP;
159                 scsi_mq_uninit_cmd(cmd);
160         } else {
161                 WARN_ON_ONCE(true);
162         }
163         blk_mq_requeue_request(cmd->request, true);
164 }
165
166 /**
167  * __scsi_queue_insert - private queue insertion
168  * @cmd: The SCSI command being requeued
169  * @reason:  The reason for the requeue
170  * @unbusy: Whether the queue should be unbusied
171  *
172  * This is a private queue insertion.  The public interface
173  * scsi_queue_insert() always assumes the queue should be unbusied
174  * because it's always called before the completion.  This function is
175  * for a requeue after completion, which should only occur in this
176  * file.
177  */
178 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
179 {
180         struct scsi_device *device = cmd->device;
181
182         SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
183                 "Inserting command %p into mlqueue\n", cmd));
184
185         scsi_set_blocked(cmd, reason);
186
187         /*
188          * Decrement the counters, since these commands are no longer
189          * active on the host/device.
190          */
191         if (unbusy)
192                 scsi_device_unbusy(device);
193
194         /*
195          * Requeue this command.  It will go before all other commands
196          * that are already in the queue. Schedule requeue work under
197          * lock such that the kblockd_schedule_work() call happens
198          * before blk_cleanup_queue() finishes.
199          */
200         cmd->result = 0;
201
202         blk_mq_requeue_request(cmd->request, true);
203 }
204
205 /*
206  * Function:    scsi_queue_insert()
207  *
208  * Purpose:     Insert a command in the midlevel queue.
209  *
210  * Arguments:   cmd    - command that we are adding to queue.
211  *              reason - why we are inserting command to queue.
212  *
213  * Lock status: Assumed that lock is not held upon entry.
214  *
215  * Returns:     Nothing.
216  *
217  * Notes:       We do this for one of two cases.  Either the host is busy
218  *              and it cannot accept any more commands for the time being,
219  *              or the device returned QUEUE_FULL and can accept no more
220  *              commands.
221  * Notes:       This could be called either from an interrupt context or a
222  *              normal process context.
223  */
224 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
225 {
226         __scsi_queue_insert(cmd, reason, true);
227 }
228
229
230 /**
231  * __scsi_execute - insert request and wait for the result
232  * @sdev:       scsi device
233  * @cmd:        scsi command
234  * @data_direction: data direction
235  * @buffer:     data buffer
236  * @bufflen:    len of buffer
237  * @sense:      optional sense buffer
238  * @sshdr:      optional decoded sense header
239  * @timeout:    request timeout in seconds
240  * @retries:    number of times to retry request
241  * @flags:      flags for ->cmd_flags
242  * @rq_flags:   flags for ->rq_flags
243  * @resid:      optional residual length
244  *
245  * Returns the scsi_cmnd result field if a command was executed, or a negative
246  * Linux error code if we didn't get that far.
247  */
248 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
249                  int data_direction, void *buffer, unsigned bufflen,
250                  unsigned char *sense, struct scsi_sense_hdr *sshdr,
251                  int timeout, int retries, u64 flags, req_flags_t rq_flags,
252                  int *resid)
253 {
254         struct request *req;
255         struct scsi_request *rq;
256         int ret = DRIVER_ERROR << 24;
257
258         req = blk_get_request(sdev->request_queue,
259                         data_direction == DMA_TO_DEVICE ?
260                         REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
261         if (IS_ERR(req))
262                 return ret;
263         rq = scsi_req(req);
264
265         if (bufflen &&  blk_rq_map_kern(sdev->request_queue, req,
266                                         buffer, bufflen, GFP_NOIO))
267                 goto out;
268
269         rq->cmd_len = COMMAND_SIZE(cmd[0]);
270         memcpy(rq->cmd, cmd, rq->cmd_len);
271         rq->retries = retries;
272         req->timeout = timeout;
273         req->cmd_flags |= flags;
274         req->rq_flags |= rq_flags | RQF_QUIET;
275
276         /*
277          * head injection *required* here otherwise quiesce won't work
278          */
279         blk_execute_rq(req->q, NULL, req, 1);
280
281         /*
282          * Some devices (USB mass-storage in particular) may transfer
283          * garbage data together with a residue indicating that the data
284          * is invalid.  Prevent the garbage from being misinterpreted
285          * and prevent security leaks by zeroing out the excess data.
286          */
287         if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
288                 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
289
290         if (resid)
291                 *resid = rq->resid_len;
292         if (sense && rq->sense_len)
293                 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
294         if (sshdr)
295                 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
296         ret = rq->result;
297  out:
298         blk_put_request(req);
299
300         return ret;
301 }
302 EXPORT_SYMBOL(__scsi_execute);
303
304 /*
305  * Function:    scsi_init_cmd_errh()
306  *
307  * Purpose:     Initialize cmd fields related to error handling.
308  *
309  * Arguments:   cmd     - command that is ready to be queued.
310  *
311  * Notes:       This function has the job of initializing a number of
312  *              fields related to error handling.   Typically this will
313  *              be called once for each command, as required.
314  */
315 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
316 {
317         scsi_set_resid(cmd, 0);
318         memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
319         if (cmd->cmd_len == 0)
320                 cmd->cmd_len = scsi_command_size(cmd->cmnd);
321 }
322
323 /*
324  * Decrement the host_busy counter and wake up the error handler if necessary.
325  * Avoid as follows that the error handler is not woken up if shost->host_busy
326  * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
327  * with an RCU read lock in this function to ensure that this function in its
328  * entirety either finishes before scsi_eh_scmd_add() increases the
329  * host_failed counter or that it notices the shost state change made by
330  * scsi_eh_scmd_add().
331  */
332 static void scsi_dec_host_busy(struct Scsi_Host *shost)
333 {
334         unsigned long flags;
335
336         rcu_read_lock();
337         atomic_dec(&shost->host_busy);
338         if (unlikely(scsi_host_in_recovery(shost))) {
339                 spin_lock_irqsave(shost->host_lock, flags);
340                 if (shost->host_failed || shost->host_eh_scheduled)
341                         scsi_eh_wakeup(shost);
342                 spin_unlock_irqrestore(shost->host_lock, flags);
343         }
344         rcu_read_unlock();
345 }
346
347 void scsi_device_unbusy(struct scsi_device *sdev)
348 {
349         struct Scsi_Host *shost = sdev->host;
350         struct scsi_target *starget = scsi_target(sdev);
351
352         scsi_dec_host_busy(shost);
353
354         if (starget->can_queue > 0)
355                 atomic_dec(&starget->target_busy);
356
357         atomic_dec(&sdev->device_busy);
358 }
359
360 static void scsi_kick_queue(struct request_queue *q)
361 {
362         blk_mq_run_hw_queues(q, false);
363 }
364
365 /*
366  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
367  * and call blk_run_queue for all the scsi_devices on the target -
368  * including current_sdev first.
369  *
370  * Called with *no* scsi locks held.
371  */
372 static void scsi_single_lun_run(struct scsi_device *current_sdev)
373 {
374         struct Scsi_Host *shost = current_sdev->host;
375         struct scsi_device *sdev, *tmp;
376         struct scsi_target *starget = scsi_target(current_sdev);
377         unsigned long flags;
378
379         spin_lock_irqsave(shost->host_lock, flags);
380         starget->starget_sdev_user = NULL;
381         spin_unlock_irqrestore(shost->host_lock, flags);
382
383         /*
384          * Call blk_run_queue for all LUNs on the target, starting with
385          * current_sdev. We race with others (to set starget_sdev_user),
386          * but in most cases, we will be first. Ideally, each LU on the
387          * target would get some limited time or requests on the target.
388          */
389         scsi_kick_queue(current_sdev->request_queue);
390
391         spin_lock_irqsave(shost->host_lock, flags);
392         if (starget->starget_sdev_user)
393                 goto out;
394         list_for_each_entry_safe(sdev, tmp, &starget->devices,
395                         same_target_siblings) {
396                 if (sdev == current_sdev)
397                         continue;
398                 if (scsi_device_get(sdev))
399                         continue;
400
401                 spin_unlock_irqrestore(shost->host_lock, flags);
402                 scsi_kick_queue(sdev->request_queue);
403                 spin_lock_irqsave(shost->host_lock, flags);
404         
405                 scsi_device_put(sdev);
406         }
407  out:
408         spin_unlock_irqrestore(shost->host_lock, flags);
409 }
410
411 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
412 {
413         if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
414                 return true;
415         if (atomic_read(&sdev->device_blocked) > 0)
416                 return true;
417         return false;
418 }
419
420 static inline bool scsi_target_is_busy(struct scsi_target *starget)
421 {
422         if (starget->can_queue > 0) {
423                 if (atomic_read(&starget->target_busy) >= starget->can_queue)
424                         return true;
425                 if (atomic_read(&starget->target_blocked) > 0)
426                         return true;
427         }
428         return false;
429 }
430
431 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
432 {
433         if (shost->can_queue > 0 &&
434             atomic_read(&shost->host_busy) >= shost->can_queue)
435                 return true;
436         if (atomic_read(&shost->host_blocked) > 0)
437                 return true;
438         if (shost->host_self_blocked)
439                 return true;
440         return false;
441 }
442
443 static void scsi_starved_list_run(struct Scsi_Host *shost)
444 {
445         LIST_HEAD(starved_list);
446         struct scsi_device *sdev;
447         unsigned long flags;
448
449         spin_lock_irqsave(shost->host_lock, flags);
450         list_splice_init(&shost->starved_list, &starved_list);
451
452         while (!list_empty(&starved_list)) {
453                 struct request_queue *slq;
454
455                 /*
456                  * As long as shost is accepting commands and we have
457                  * starved queues, call blk_run_queue. scsi_request_fn
458                  * drops the queue_lock and can add us back to the
459                  * starved_list.
460                  *
461                  * host_lock protects the starved_list and starved_entry.
462                  * scsi_request_fn must get the host_lock before checking
463                  * or modifying starved_list or starved_entry.
464                  */
465                 if (scsi_host_is_busy(shost))
466                         break;
467
468                 sdev = list_entry(starved_list.next,
469                                   struct scsi_device, starved_entry);
470                 list_del_init(&sdev->starved_entry);
471                 if (scsi_target_is_busy(scsi_target(sdev))) {
472                         list_move_tail(&sdev->starved_entry,
473                                        &shost->starved_list);
474                         continue;
475                 }
476
477                 /*
478                  * Once we drop the host lock, a racing scsi_remove_device()
479                  * call may remove the sdev from the starved list and destroy
480                  * it and the queue.  Mitigate by taking a reference to the
481                  * queue and never touching the sdev again after we drop the
482                  * host lock.  Note: if __scsi_remove_device() invokes
483                  * blk_cleanup_queue() before the queue is run from this
484                  * function then blk_run_queue() will return immediately since
485                  * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
486                  */
487                 slq = sdev->request_queue;
488                 if (!blk_get_queue(slq))
489                         continue;
490                 spin_unlock_irqrestore(shost->host_lock, flags);
491
492                 scsi_kick_queue(slq);
493                 blk_put_queue(slq);
494
495                 spin_lock_irqsave(shost->host_lock, flags);
496         }
497         /* put any unprocessed entries back */
498         list_splice(&starved_list, &shost->starved_list);
499         spin_unlock_irqrestore(shost->host_lock, flags);
500 }
501
502 /*
503  * Function:   scsi_run_queue()
504  *
505  * Purpose:    Select a proper request queue to serve next
506  *
507  * Arguments:  q       - last request's queue
508  *
509  * Returns:     Nothing
510  *
511  * Notes:      The previous command was completely finished, start
512  *             a new one if possible.
513  */
514 static void scsi_run_queue(struct request_queue *q)
515 {
516         struct scsi_device *sdev = q->queuedata;
517
518         if (scsi_target(sdev)->single_lun)
519                 scsi_single_lun_run(sdev);
520         if (!list_empty(&sdev->host->starved_list))
521                 scsi_starved_list_run(sdev->host);
522
523         blk_mq_run_hw_queues(q, false);
524 }
525
526 void scsi_requeue_run_queue(struct work_struct *work)
527 {
528         struct scsi_device *sdev;
529         struct request_queue *q;
530
531         sdev = container_of(work, struct scsi_device, requeue_work);
532         q = sdev->request_queue;
533         scsi_run_queue(q);
534 }
535
536 void scsi_run_host_queues(struct Scsi_Host *shost)
537 {
538         struct scsi_device *sdev;
539
540         shost_for_each_device(sdev, shost)
541                 scsi_run_queue(sdev->request_queue);
542 }
543
544 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
545 {
546         if (!blk_rq_is_passthrough(cmd->request)) {
547                 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
548
549                 if (drv->uninit_command)
550                         drv->uninit_command(cmd);
551         }
552 }
553
554 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
555 {
556         if (cmd->sdb.table.nents)
557                 sg_free_table_chained(&cmd->sdb.table,
558                                 SCSI_INLINE_SG_CNT);
559         if (scsi_prot_sg_count(cmd))
560                 sg_free_table_chained(&cmd->prot_sdb->table,
561                                 SCSI_INLINE_PROT_SG_CNT);
562 }
563
564 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
565 {
566         scsi_mq_free_sgtables(cmd);
567         scsi_uninit_cmd(cmd);
568         scsi_del_cmd_from_list(cmd);
569 }
570
571 /* Returns false when no more bytes to process, true if there are more */
572 static bool scsi_end_request(struct request *req, blk_status_t error,
573                 unsigned int bytes)
574 {
575         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
576         struct scsi_device *sdev = cmd->device;
577         struct request_queue *q = sdev->request_queue;
578
579         if (blk_update_request(req, error, bytes))
580                 return true;
581
582         if (blk_queue_add_random(q))
583                 add_disk_randomness(req->rq_disk);
584
585         if (!blk_rq_is_scsi(req)) {
586                 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
587                 cmd->flags &= ~SCMD_INITIALIZED;
588         }
589
590         /*
591          * Calling rcu_barrier() is not necessary here because the
592          * SCSI error handler guarantees that the function called by
593          * call_rcu() has been called before scsi_end_request() is
594          * called.
595          */
596         destroy_rcu_head(&cmd->rcu);
597
598         /*
599          * In the MQ case the command gets freed by __blk_mq_end_request,
600          * so we have to do all cleanup that depends on it earlier.
601          *
602          * We also can't kick the queues from irq context, so we
603          * will have to defer it to a workqueue.
604          */
605         scsi_mq_uninit_cmd(cmd);
606
607         /*
608          * queue is still alive, so grab the ref for preventing it
609          * from being cleaned up during running queue.
610          */
611         percpu_ref_get(&q->q_usage_counter);
612
613         __blk_mq_end_request(req, error);
614
615         if (scsi_target(sdev)->single_lun ||
616             !list_empty(&sdev->host->starved_list))
617                 kblockd_schedule_work(&sdev->requeue_work);
618         else
619                 blk_mq_run_hw_queues(q, true);
620
621         percpu_ref_put(&q->q_usage_counter);
622         return false;
623 }
624
625 /**
626  * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
627  * @cmd:        SCSI command
628  * @result:     scsi error code
629  *
630  * Translate a SCSI result code into a blk_status_t value. May reset the host
631  * byte of @cmd->result.
632  */
633 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
634 {
635         switch (host_byte(result)) {
636         case DID_OK:
637                 /*
638                  * Also check the other bytes than the status byte in result
639                  * to handle the case when a SCSI LLD sets result to
640                  * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
641                  */
642                 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
643                         return BLK_STS_OK;
644                 return BLK_STS_IOERR;
645         case DID_TRANSPORT_FAILFAST:
646                 return BLK_STS_TRANSPORT;
647         case DID_TARGET_FAILURE:
648                 set_host_byte(cmd, DID_OK);
649                 return BLK_STS_TARGET;
650         case DID_NEXUS_FAILURE:
651                 set_host_byte(cmd, DID_OK);
652                 return BLK_STS_NEXUS;
653         case DID_ALLOC_FAILURE:
654                 set_host_byte(cmd, DID_OK);
655                 return BLK_STS_NOSPC;
656         case DID_MEDIUM_ERROR:
657                 set_host_byte(cmd, DID_OK);
658                 return BLK_STS_MEDIUM;
659         default:
660                 return BLK_STS_IOERR;
661         }
662 }
663
664 /* Helper for scsi_io_completion() when "reprep" action required. */
665 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
666                                       struct request_queue *q)
667 {
668         /* A new command will be prepared and issued. */
669         scsi_mq_requeue_cmd(cmd);
670 }
671
672 /* Helper for scsi_io_completion() when special action required. */
673 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
674 {
675         struct request_queue *q = cmd->device->request_queue;
676         struct request *req = cmd->request;
677         int level = 0;
678         enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
679               ACTION_DELAYED_RETRY} action;
680         unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
681         struct scsi_sense_hdr sshdr;
682         bool sense_valid;
683         bool sense_current = true;      /* false implies "deferred sense" */
684         blk_status_t blk_stat;
685
686         sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
687         if (sense_valid)
688                 sense_current = !scsi_sense_is_deferred(&sshdr);
689
690         blk_stat = scsi_result_to_blk_status(cmd, result);
691
692         if (host_byte(result) == DID_RESET) {
693                 /* Third party bus reset or reset for error recovery
694                  * reasons.  Just retry the command and see what
695                  * happens.
696                  */
697                 action = ACTION_RETRY;
698         } else if (sense_valid && sense_current) {
699                 switch (sshdr.sense_key) {
700                 case UNIT_ATTENTION:
701                         if (cmd->device->removable) {
702                                 /* Detected disc change.  Set a bit
703                                  * and quietly refuse further access.
704                                  */
705                                 cmd->device->changed = 1;
706                                 action = ACTION_FAIL;
707                         } else {
708                                 /* Must have been a power glitch, or a
709                                  * bus reset.  Could not have been a
710                                  * media change, so we just retry the
711                                  * command and see what happens.
712                                  */
713                                 action = ACTION_RETRY;
714                         }
715                         break;
716                 case ILLEGAL_REQUEST:
717                         /* If we had an ILLEGAL REQUEST returned, then
718                          * we may have performed an unsupported
719                          * command.  The only thing this should be
720                          * would be a ten byte read where only a six
721                          * byte read was supported.  Also, on a system
722                          * where READ CAPACITY failed, we may have
723                          * read past the end of the disk.
724                          */
725                         if ((cmd->device->use_10_for_rw &&
726                             sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
727                             (cmd->cmnd[0] == READ_10 ||
728                              cmd->cmnd[0] == WRITE_10)) {
729                                 /* This will issue a new 6-byte command. */
730                                 cmd->device->use_10_for_rw = 0;
731                                 action = ACTION_REPREP;
732                         } else if (sshdr.asc == 0x10) /* DIX */ {
733                                 action = ACTION_FAIL;
734                                 blk_stat = BLK_STS_PROTECTION;
735                         /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
736                         } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
737                                 action = ACTION_FAIL;
738                                 blk_stat = BLK_STS_TARGET;
739                         } else
740                                 action = ACTION_FAIL;
741                         break;
742                 case ABORTED_COMMAND:
743                         action = ACTION_FAIL;
744                         if (sshdr.asc == 0x10) /* DIF */
745                                 blk_stat = BLK_STS_PROTECTION;
746                         break;
747                 case NOT_READY:
748                         /* If the device is in the process of becoming
749                          * ready, or has a temporary blockage, retry.
750                          */
751                         if (sshdr.asc == 0x04) {
752                                 switch (sshdr.ascq) {
753                                 case 0x01: /* becoming ready */
754                                 case 0x04: /* format in progress */
755                                 case 0x05: /* rebuild in progress */
756                                 case 0x06: /* recalculation in progress */
757                                 case 0x07: /* operation in progress */
758                                 case 0x08: /* Long write in progress */
759                                 case 0x09: /* self test in progress */
760                                 case 0x14: /* space allocation in progress */
761                                 case 0x1a: /* start stop unit in progress */
762                                 case 0x1b: /* sanitize in progress */
763                                 case 0x1d: /* configuration in progress */
764                                 case 0x24: /* depopulation in progress */
765                                         action = ACTION_DELAYED_RETRY;
766                                         break;
767                                 default:
768                                         action = ACTION_FAIL;
769                                         break;
770                                 }
771                         } else
772                                 action = ACTION_FAIL;
773                         break;
774                 case VOLUME_OVERFLOW:
775                         /* See SSC3rXX or current. */
776                         action = ACTION_FAIL;
777                         break;
778                 default:
779                         action = ACTION_FAIL;
780                         break;
781                 }
782         } else
783                 action = ACTION_FAIL;
784
785         if (action != ACTION_FAIL &&
786             time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
787                 action = ACTION_FAIL;
788
789         switch (action) {
790         case ACTION_FAIL:
791                 /* Give up and fail the remainder of the request */
792                 if (!(req->rq_flags & RQF_QUIET)) {
793                         static DEFINE_RATELIMIT_STATE(_rs,
794                                         DEFAULT_RATELIMIT_INTERVAL,
795                                         DEFAULT_RATELIMIT_BURST);
796
797                         if (unlikely(scsi_logging_level))
798                                 level =
799                                      SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
800                                                     SCSI_LOG_MLCOMPLETE_BITS);
801
802                         /*
803                          * if logging is enabled the failure will be printed
804                          * in scsi_log_completion(), so avoid duplicate messages
805                          */
806                         if (!level && __ratelimit(&_rs)) {
807                                 scsi_print_result(cmd, NULL, FAILED);
808                                 if (driver_byte(result) == DRIVER_SENSE)
809                                         scsi_print_sense(cmd);
810                                 scsi_print_command(cmd);
811                         }
812                 }
813                 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
814                         return;
815                 /*FALLTHRU*/
816         case ACTION_REPREP:
817                 scsi_io_completion_reprep(cmd, q);
818                 break;
819         case ACTION_RETRY:
820                 /* Retry the same command immediately */
821                 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
822                 break;
823         case ACTION_DELAYED_RETRY:
824                 /* Retry the same command after a delay */
825                 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
826                 break;
827         }
828 }
829
830 /*
831  * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
832  * new result that may suppress further error checking. Also modifies
833  * *blk_statp in some cases.
834  */
835 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
836                                         blk_status_t *blk_statp)
837 {
838         bool sense_valid;
839         bool sense_current = true;      /* false implies "deferred sense" */
840         struct request *req = cmd->request;
841         struct scsi_sense_hdr sshdr;
842
843         sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
844         if (sense_valid)
845                 sense_current = !scsi_sense_is_deferred(&sshdr);
846
847         if (blk_rq_is_passthrough(req)) {
848                 if (sense_valid) {
849                         /*
850                          * SG_IO wants current and deferred errors
851                          */
852                         scsi_req(req)->sense_len =
853                                 min(8 + cmd->sense_buffer[7],
854                                     SCSI_SENSE_BUFFERSIZE);
855                 }
856                 if (sense_current)
857                         *blk_statp = scsi_result_to_blk_status(cmd, result);
858         } else if (blk_rq_bytes(req) == 0 && sense_current) {
859                 /*
860                  * Flush commands do not transfers any data, and thus cannot use
861                  * good_bytes != blk_rq_bytes(req) as the signal for an error.
862                  * This sets *blk_statp explicitly for the problem case.
863                  */
864                 *blk_statp = scsi_result_to_blk_status(cmd, result);
865         }
866         /*
867          * Recovered errors need reporting, but they're always treated as
868          * success, so fiddle the result code here.  For passthrough requests
869          * we already took a copy of the original into sreq->result which
870          * is what gets returned to the user
871          */
872         if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
873                 bool do_print = true;
874                 /*
875                  * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
876                  * skip print since caller wants ATA registers. Only occurs
877                  * on SCSI ATA PASS_THROUGH commands when CK_COND=1
878                  */
879                 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
880                         do_print = false;
881                 else if (req->rq_flags & RQF_QUIET)
882                         do_print = false;
883                 if (do_print)
884                         scsi_print_sense(cmd);
885                 result = 0;
886                 /* for passthrough, *blk_statp may be set */
887                 *blk_statp = BLK_STS_OK;
888         }
889         /*
890          * Another corner case: the SCSI status byte is non-zero but 'good'.
891          * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
892          * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
893          * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
894          * intermediate statuses (both obsolete in SAM-4) as good.
895          */
896         if (status_byte(result) && scsi_status_is_good(result)) {
897                 result = 0;
898                 *blk_statp = BLK_STS_OK;
899         }
900         return result;
901 }
902
903 /*
904  * Function:    scsi_io_completion()
905  *
906  * Purpose:     Completion processing for block device I/O requests.
907  *
908  * Arguments:   cmd   - command that is finished.
909  *
910  * Lock status: Assumed that no lock is held upon entry.
911  *
912  * Returns:     Nothing
913  *
914  * Notes:       We will finish off the specified number of sectors.  If we
915  *              are done, the command block will be released and the queue
916  *              function will be goosed.  If we are not done then we have to
917  *              figure out what to do next:
918  *
919  *              a) We can call scsi_requeue_command().  The request
920  *                 will be unprepared and put back on the queue.  Then
921  *                 a new command will be created for it.  This should
922  *                 be used if we made forward progress, or if we want
923  *                 to switch from READ(10) to READ(6) for example.
924  *
925  *              b) We can call __scsi_queue_insert().  The request will
926  *                 be put back on the queue and retried using the same
927  *                 command as before, possibly after a delay.
928  *
929  *              c) We can call scsi_end_request() with blk_stat other than
930  *                 BLK_STS_OK, to fail the remainder of the request.
931  */
932 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
933 {
934         int result = cmd->result;
935         struct request_queue *q = cmd->device->request_queue;
936         struct request *req = cmd->request;
937         blk_status_t blk_stat = BLK_STS_OK;
938
939         if (unlikely(result))   /* a nz result may or may not be an error */
940                 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
941
942         if (unlikely(blk_rq_is_passthrough(req))) {
943                 /*
944                  * scsi_result_to_blk_status may have reset the host_byte
945                  */
946                 scsi_req(req)->result = cmd->result;
947         }
948
949         /*
950          * Next deal with any sectors which we were able to correctly
951          * handle.
952          */
953         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
954                 "%u sectors total, %d bytes done.\n",
955                 blk_rq_sectors(req), good_bytes));
956
957         /*
958          * Next deal with any sectors which we were able to correctly
959          * handle. Failed, zero length commands always need to drop down
960          * to retry code. Fast path should return in this block.
961          */
962         if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
963                 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
964                         return; /* no bytes remaining */
965         }
966
967         /* Kill remainder if no retries. */
968         if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
969                 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
970                         WARN_ONCE(true,
971                             "Bytes remaining after failed, no-retry command");
972                 return;
973         }
974
975         /*
976          * If there had been no error, but we have leftover bytes in the
977          * requeues just queue the command up again.
978          */
979         if (likely(result == 0))
980                 scsi_io_completion_reprep(cmd, q);
981         else
982                 scsi_io_completion_action(cmd, result);
983 }
984
985 static blk_status_t scsi_init_sgtable(struct request *req,
986                 struct scsi_data_buffer *sdb)
987 {
988         int count;
989
990         /*
991          * If sg table allocation fails, requeue request later.
992          */
993         if (unlikely(sg_alloc_table_chained(&sdb->table,
994                         blk_rq_nr_phys_segments(req), sdb->table.sgl,
995                         SCSI_INLINE_SG_CNT)))
996                 return BLK_STS_RESOURCE;
997
998         /* 
999          * Next, walk the list, and fill in the addresses and sizes of
1000          * each segment.
1001          */
1002         count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1003         BUG_ON(count > sdb->table.nents);
1004         sdb->table.nents = count;
1005         sdb->length = blk_rq_payload_bytes(req);
1006         return BLK_STS_OK;
1007 }
1008
1009 /*
1010  * Function:    scsi_init_io()
1011  *
1012  * Purpose:     SCSI I/O initialize function.
1013  *
1014  * Arguments:   cmd   - Command descriptor we wish to initialize
1015  *
1016  * Returns:     BLK_STS_OK on success
1017  *              BLK_STS_RESOURCE if the failure is retryable
1018  *              BLK_STS_IOERR if the failure is fatal
1019  */
1020 blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
1021 {
1022         struct request *rq = cmd->request;
1023         blk_status_t ret;
1024
1025         if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1026                 return BLK_STS_IOERR;
1027
1028         ret = scsi_init_sgtable(rq, &cmd->sdb);
1029         if (ret)
1030                 return ret;
1031
1032         if (blk_integrity_rq(rq)) {
1033                 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1034                 int ivecs, count;
1035
1036                 if (WARN_ON_ONCE(!prot_sdb)) {
1037                         /*
1038                          * This can happen if someone (e.g. multipath)
1039                          * queues a command to a device on an adapter
1040                          * that does not support DIX.
1041                          */
1042                         ret = BLK_STS_IOERR;
1043                         goto out_free_sgtables;
1044                 }
1045
1046                 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1047
1048                 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1049                                 prot_sdb->table.sgl,
1050                                 SCSI_INLINE_PROT_SG_CNT)) {
1051                         ret = BLK_STS_RESOURCE;
1052                         goto out_free_sgtables;
1053                 }
1054
1055                 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1056                                                 prot_sdb->table.sgl);
1057                 BUG_ON(count > ivecs);
1058                 BUG_ON(count > queue_max_integrity_segments(rq->q));
1059
1060                 cmd->prot_sdb = prot_sdb;
1061                 cmd->prot_sdb->table.nents = count;
1062         }
1063
1064         return BLK_STS_OK;
1065 out_free_sgtables:
1066         scsi_mq_free_sgtables(cmd);
1067         return ret;
1068 }
1069 EXPORT_SYMBOL(scsi_init_io);
1070
1071 /**
1072  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1073  * @rq: Request associated with the SCSI command to be initialized.
1074  *
1075  * This function initializes the members of struct scsi_cmnd that must be
1076  * initialized before request processing starts and that won't be
1077  * reinitialized if a SCSI command is requeued.
1078  *
1079  * Called from inside blk_get_request() for pass-through requests and from
1080  * inside scsi_init_command() for filesystem requests.
1081  */
1082 static void scsi_initialize_rq(struct request *rq)
1083 {
1084         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1085
1086         scsi_req_init(&cmd->req);
1087         init_rcu_head(&cmd->rcu);
1088         cmd->jiffies_at_alloc = jiffies;
1089         cmd->retries = 0;
1090 }
1091
1092 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1093 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1094 {
1095         struct scsi_device *sdev = cmd->device;
1096         struct Scsi_Host *shost = sdev->host;
1097         unsigned long flags;
1098
1099         if (shost->use_cmd_list) {
1100                 spin_lock_irqsave(&sdev->list_lock, flags);
1101                 list_add_tail(&cmd->list, &sdev->cmd_list);
1102                 spin_unlock_irqrestore(&sdev->list_lock, flags);
1103         }
1104 }
1105
1106 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1107 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1108 {
1109         struct scsi_device *sdev = cmd->device;
1110         struct Scsi_Host *shost = sdev->host;
1111         unsigned long flags;
1112
1113         if (shost->use_cmd_list) {
1114                 spin_lock_irqsave(&sdev->list_lock, flags);
1115                 BUG_ON(list_empty(&cmd->list));
1116                 list_del_init(&cmd->list);
1117                 spin_unlock_irqrestore(&sdev->list_lock, flags);
1118         }
1119 }
1120
1121 /* Called after a request has been started. */
1122 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1123 {
1124         void *buf = cmd->sense_buffer;
1125         void *prot = cmd->prot_sdb;
1126         struct request *rq = blk_mq_rq_from_pdu(cmd);
1127         unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1128         unsigned long jiffies_at_alloc;
1129         int retries;
1130
1131         if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1132                 flags |= SCMD_INITIALIZED;
1133                 scsi_initialize_rq(rq);
1134         }
1135
1136         jiffies_at_alloc = cmd->jiffies_at_alloc;
1137         retries = cmd->retries;
1138         /* zero out the cmd, except for the embedded scsi_request */
1139         memset((char *)cmd + sizeof(cmd->req), 0,
1140                 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1141
1142         cmd->device = dev;
1143         cmd->sense_buffer = buf;
1144         cmd->prot_sdb = prot;
1145         cmd->flags = flags;
1146         INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1147         cmd->jiffies_at_alloc = jiffies_at_alloc;
1148         cmd->retries = retries;
1149
1150         scsi_add_cmd_to_list(cmd);
1151 }
1152
1153 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1154                 struct request *req)
1155 {
1156         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1157
1158         /*
1159          * Passthrough requests may transfer data, in which case they must
1160          * a bio attached to them.  Or they might contain a SCSI command
1161          * that does not transfer data, in which case they may optionally
1162          * submit a request without an attached bio.
1163          */
1164         if (req->bio) {
1165                 blk_status_t ret = scsi_init_io(cmd);
1166                 if (unlikely(ret != BLK_STS_OK))
1167                         return ret;
1168         } else {
1169                 BUG_ON(blk_rq_bytes(req));
1170
1171                 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1172         }
1173
1174         cmd->cmd_len = scsi_req(req)->cmd_len;
1175         cmd->cmnd = scsi_req(req)->cmd;
1176         cmd->transfersize = blk_rq_bytes(req);
1177         cmd->allowed = scsi_req(req)->retries;
1178         return BLK_STS_OK;
1179 }
1180
1181 /*
1182  * Setup a normal block command.  These are simple request from filesystems
1183  * that still need to be translated to SCSI CDBs from the ULD.
1184  */
1185 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1186                 struct request *req)
1187 {
1188         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1189
1190         if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1191                 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1192                 if (ret != BLK_STS_OK)
1193                         return ret;
1194         }
1195
1196         cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1197         memset(cmd->cmnd, 0, BLK_MAX_CDB);
1198         return scsi_cmd_to_driver(cmd)->init_command(cmd);
1199 }
1200
1201 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1202                 struct request *req)
1203 {
1204         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1205
1206         if (!blk_rq_bytes(req))
1207                 cmd->sc_data_direction = DMA_NONE;
1208         else if (rq_data_dir(req) == WRITE)
1209                 cmd->sc_data_direction = DMA_TO_DEVICE;
1210         else
1211                 cmd->sc_data_direction = DMA_FROM_DEVICE;
1212
1213         if (blk_rq_is_scsi(req))
1214                 return scsi_setup_scsi_cmnd(sdev, req);
1215         else
1216                 return scsi_setup_fs_cmnd(sdev, req);
1217 }
1218
1219 static blk_status_t
1220 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1221 {
1222         switch (sdev->sdev_state) {
1223         case SDEV_OFFLINE:
1224         case SDEV_TRANSPORT_OFFLINE:
1225                 /*
1226                  * If the device is offline we refuse to process any
1227                  * commands.  The device must be brought online
1228                  * before trying any recovery commands.
1229                  */
1230                 sdev_printk(KERN_ERR, sdev,
1231                             "rejecting I/O to offline device\n");
1232                 return BLK_STS_IOERR;
1233         case SDEV_DEL:
1234                 /*
1235                  * If the device is fully deleted, we refuse to
1236                  * process any commands as well.
1237                  */
1238                 sdev_printk(KERN_ERR, sdev,
1239                             "rejecting I/O to dead device\n");
1240                 return BLK_STS_IOERR;
1241         case SDEV_BLOCK:
1242         case SDEV_CREATED_BLOCK:
1243                 return BLK_STS_RESOURCE;
1244         case SDEV_QUIESCE:
1245                 /*
1246                  * If the devices is blocked we defer normal commands.
1247                  */
1248                 if (req && !(req->rq_flags & RQF_PREEMPT))
1249                         return BLK_STS_RESOURCE;
1250                 return BLK_STS_OK;
1251         default:
1252                 /*
1253                  * For any other not fully online state we only allow
1254                  * special commands.  In particular any user initiated
1255                  * command is not allowed.
1256                  */
1257                 if (req && !(req->rq_flags & RQF_PREEMPT))
1258                         return BLK_STS_IOERR;
1259                 return BLK_STS_OK;
1260         }
1261 }
1262
1263 /*
1264  * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1265  * return 0.
1266  *
1267  * Called with the queue_lock held.
1268  */
1269 static inline int scsi_dev_queue_ready(struct request_queue *q,
1270                                   struct scsi_device *sdev)
1271 {
1272         unsigned int busy;
1273
1274         busy = atomic_inc_return(&sdev->device_busy) - 1;
1275         if (atomic_read(&sdev->device_blocked)) {
1276                 if (busy)
1277                         goto out_dec;
1278
1279                 /*
1280                  * unblock after device_blocked iterates to zero
1281                  */
1282                 if (atomic_dec_return(&sdev->device_blocked) > 0)
1283                         goto out_dec;
1284                 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1285                                    "unblocking device at zero depth\n"));
1286         }
1287
1288         if (busy >= sdev->queue_depth)
1289                 goto out_dec;
1290
1291         return 1;
1292 out_dec:
1293         atomic_dec(&sdev->device_busy);
1294         return 0;
1295 }
1296
1297 /*
1298  * scsi_target_queue_ready: checks if there we can send commands to target
1299  * @sdev: scsi device on starget to check.
1300  */
1301 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1302                                            struct scsi_device *sdev)
1303 {
1304         struct scsi_target *starget = scsi_target(sdev);
1305         unsigned int busy;
1306
1307         if (starget->single_lun) {
1308                 spin_lock_irq(shost->host_lock);
1309                 if (starget->starget_sdev_user &&
1310                     starget->starget_sdev_user != sdev) {
1311                         spin_unlock_irq(shost->host_lock);
1312                         return 0;
1313                 }
1314                 starget->starget_sdev_user = sdev;
1315                 spin_unlock_irq(shost->host_lock);
1316         }
1317
1318         if (starget->can_queue <= 0)
1319                 return 1;
1320
1321         busy = atomic_inc_return(&starget->target_busy) - 1;
1322         if (atomic_read(&starget->target_blocked) > 0) {
1323                 if (busy)
1324                         goto starved;
1325
1326                 /*
1327                  * unblock after target_blocked iterates to zero
1328                  */
1329                 if (atomic_dec_return(&starget->target_blocked) > 0)
1330                         goto out_dec;
1331
1332                 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1333                                  "unblocking target at zero depth\n"));
1334         }
1335
1336         if (busy >= starget->can_queue)
1337                 goto starved;
1338
1339         return 1;
1340
1341 starved:
1342         spin_lock_irq(shost->host_lock);
1343         list_move_tail(&sdev->starved_entry, &shost->starved_list);
1344         spin_unlock_irq(shost->host_lock);
1345 out_dec:
1346         if (starget->can_queue > 0)
1347                 atomic_dec(&starget->target_busy);
1348         return 0;
1349 }
1350
1351 /*
1352  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1353  * return 0. We must end up running the queue again whenever 0 is
1354  * returned, else IO can hang.
1355  */
1356 static inline int scsi_host_queue_ready(struct request_queue *q,
1357                                    struct Scsi_Host *shost,
1358                                    struct scsi_device *sdev)
1359 {
1360         unsigned int busy;
1361
1362         if (scsi_host_in_recovery(shost))
1363                 return 0;
1364
1365         busy = atomic_inc_return(&shost->host_busy) - 1;
1366         if (atomic_read(&shost->host_blocked) > 0) {
1367                 if (busy)
1368                         goto starved;
1369
1370                 /*
1371                  * unblock after host_blocked iterates to zero
1372                  */
1373                 if (atomic_dec_return(&shost->host_blocked) > 0)
1374                         goto out_dec;
1375
1376                 SCSI_LOG_MLQUEUE(3,
1377                         shost_printk(KERN_INFO, shost,
1378                                      "unblocking host at zero depth\n"));
1379         }
1380
1381         if (shost->can_queue > 0 && busy >= shost->can_queue)
1382                 goto starved;
1383         if (shost->host_self_blocked)
1384                 goto starved;
1385
1386         /* We're OK to process the command, so we can't be starved */
1387         if (!list_empty(&sdev->starved_entry)) {
1388                 spin_lock_irq(shost->host_lock);
1389                 if (!list_empty(&sdev->starved_entry))
1390                         list_del_init(&sdev->starved_entry);
1391                 spin_unlock_irq(shost->host_lock);
1392         }
1393
1394         return 1;
1395
1396 starved:
1397         spin_lock_irq(shost->host_lock);
1398         if (list_empty(&sdev->starved_entry))
1399                 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1400         spin_unlock_irq(shost->host_lock);
1401 out_dec:
1402         scsi_dec_host_busy(shost);
1403         return 0;
1404 }
1405
1406 /*
1407  * Busy state exporting function for request stacking drivers.
1408  *
1409  * For efficiency, no lock is taken to check the busy state of
1410  * shost/starget/sdev, since the returned value is not guaranteed and
1411  * may be changed after request stacking drivers call the function,
1412  * regardless of taking lock or not.
1413  *
1414  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1415  * needs to return 'not busy'. Otherwise, request stacking drivers
1416  * may hold requests forever.
1417  */
1418 static bool scsi_mq_lld_busy(struct request_queue *q)
1419 {
1420         struct scsi_device *sdev = q->queuedata;
1421         struct Scsi_Host *shost;
1422
1423         if (blk_queue_dying(q))
1424                 return false;
1425
1426         shost = sdev->host;
1427
1428         /*
1429          * Ignore host/starget busy state.
1430          * Since block layer does not have a concept of fairness across
1431          * multiple queues, congestion of host/starget needs to be handled
1432          * in SCSI layer.
1433          */
1434         if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1435                 return true;
1436
1437         return false;
1438 }
1439
1440 static void scsi_softirq_done(struct request *rq)
1441 {
1442         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1443         unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1444         int disposition;
1445
1446         INIT_LIST_HEAD(&cmd->eh_entry);
1447
1448         atomic_inc(&cmd->device->iodone_cnt);
1449         if (cmd->result)
1450                 atomic_inc(&cmd->device->ioerr_cnt);
1451
1452         disposition = scsi_decide_disposition(cmd);
1453         if (disposition != SUCCESS &&
1454             time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1455                 sdev_printk(KERN_ERR, cmd->device,
1456                             "timing out command, waited %lus\n",
1457                             wait_for/HZ);
1458                 disposition = SUCCESS;
1459         }
1460
1461         scsi_log_completion(cmd, disposition);
1462
1463         switch (disposition) {
1464                 case SUCCESS:
1465                         scsi_finish_command(cmd);
1466                         break;
1467                 case NEEDS_RETRY:
1468                         scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1469                         break;
1470                 case ADD_TO_MLQUEUE:
1471                         scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1472                         break;
1473                 default:
1474                         scsi_eh_scmd_add(cmd);
1475                         break;
1476         }
1477 }
1478
1479 /**
1480  * scsi_dispatch_command - Dispatch a command to the low-level driver.
1481  * @cmd: command block we are dispatching.
1482  *
1483  * Return: nonzero return request was rejected and device's queue needs to be
1484  * plugged.
1485  */
1486 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1487 {
1488         struct Scsi_Host *host = cmd->device->host;
1489         int rtn = 0;
1490
1491         atomic_inc(&cmd->device->iorequest_cnt);
1492
1493         /* check if the device is still usable */
1494         if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1495                 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1496                  * returns an immediate error upwards, and signals
1497                  * that the device is no longer present */
1498                 cmd->result = DID_NO_CONNECT << 16;
1499                 goto done;
1500         }
1501
1502         /* Check to see if the scsi lld made this device blocked. */
1503         if (unlikely(scsi_device_blocked(cmd->device))) {
1504                 /*
1505                  * in blocked state, the command is just put back on
1506                  * the device queue.  The suspend state has already
1507                  * blocked the queue so future requests should not
1508                  * occur until the device transitions out of the
1509                  * suspend state.
1510                  */
1511                 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1512                         "queuecommand : device blocked\n"));
1513                 return SCSI_MLQUEUE_DEVICE_BUSY;
1514         }
1515
1516         /* Store the LUN value in cmnd, if needed. */
1517         if (cmd->device->lun_in_cdb)
1518                 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1519                                (cmd->device->lun << 5 & 0xe0);
1520
1521         scsi_log_send(cmd);
1522
1523         /*
1524          * Before we queue this command, check if the command
1525          * length exceeds what the host adapter can handle.
1526          */
1527         if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1528                 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1529                                "queuecommand : command too long. "
1530                                "cdb_size=%d host->max_cmd_len=%d\n",
1531                                cmd->cmd_len, cmd->device->host->max_cmd_len));
1532                 cmd->result = (DID_ABORT << 16);
1533                 goto done;
1534         }
1535
1536         if (unlikely(host->shost_state == SHOST_DEL)) {
1537                 cmd->result = (DID_NO_CONNECT << 16);
1538                 goto done;
1539
1540         }
1541
1542         trace_scsi_dispatch_cmd_start(cmd);
1543         rtn = host->hostt->queuecommand(host, cmd);
1544         if (rtn) {
1545                 trace_scsi_dispatch_cmd_error(cmd, rtn);
1546                 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1547                     rtn != SCSI_MLQUEUE_TARGET_BUSY)
1548                         rtn = SCSI_MLQUEUE_HOST_BUSY;
1549
1550                 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1551                         "queuecommand : request rejected\n"));
1552         }
1553
1554         return rtn;
1555  done:
1556         cmd->scsi_done(cmd);
1557         return 0;
1558 }
1559
1560 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1561 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1562 {
1563         return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1564                 sizeof(struct scatterlist);
1565 }
1566
1567 static blk_status_t scsi_mq_prep_fn(struct request *req)
1568 {
1569         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1570         struct scsi_device *sdev = req->q->queuedata;
1571         struct Scsi_Host *shost = sdev->host;
1572         struct scatterlist *sg;
1573
1574         scsi_init_command(sdev, cmd);
1575
1576         cmd->request = req;
1577         cmd->tag = req->tag;
1578         cmd->prot_op = SCSI_PROT_NORMAL;
1579
1580         sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1581         cmd->sdb.table.sgl = sg;
1582
1583         if (scsi_host_get_prot(shost)) {
1584                 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1585
1586                 cmd->prot_sdb->table.sgl =
1587                         (struct scatterlist *)(cmd->prot_sdb + 1);
1588         }
1589
1590         blk_mq_start_request(req);
1591
1592         return scsi_setup_cmnd(sdev, req);
1593 }
1594
1595 static void scsi_mq_done(struct scsi_cmnd *cmd)
1596 {
1597         if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1598                 return;
1599         trace_scsi_dispatch_cmd_done(cmd);
1600
1601         /*
1602          * If the block layer didn't complete the request due to a timeout
1603          * injection, scsi must clear its internal completed state so that the
1604          * timeout handler will see it needs to escalate its own error
1605          * recovery.
1606          */
1607         if (unlikely(!blk_mq_complete_request(cmd->request)))
1608                 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1609 }
1610
1611 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1612 {
1613         struct request_queue *q = hctx->queue;
1614         struct scsi_device *sdev = q->queuedata;
1615
1616         atomic_dec(&sdev->device_busy);
1617 }
1618
1619 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1620 {
1621         struct request_queue *q = hctx->queue;
1622         struct scsi_device *sdev = q->queuedata;
1623
1624         if (scsi_dev_queue_ready(q, sdev))
1625                 return true;
1626
1627         if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1628                 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1629         return false;
1630 }
1631
1632 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1633                          const struct blk_mq_queue_data *bd)
1634 {
1635         struct request *req = bd->rq;
1636         struct request_queue *q = req->q;
1637         struct scsi_device *sdev = q->queuedata;
1638         struct Scsi_Host *shost = sdev->host;
1639         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1640         blk_status_t ret;
1641         int reason;
1642
1643         /*
1644          * If the device is not in running state we will reject some or all
1645          * commands.
1646          */
1647         if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1648                 ret = scsi_prep_state_check(sdev, req);
1649                 if (ret != BLK_STS_OK)
1650                         goto out_put_budget;
1651         }
1652
1653         ret = BLK_STS_RESOURCE;
1654         if (!scsi_target_queue_ready(shost, sdev))
1655                 goto out_put_budget;
1656         if (!scsi_host_queue_ready(q, shost, sdev))
1657                 goto out_dec_target_busy;
1658
1659         if (!(req->rq_flags & RQF_DONTPREP)) {
1660                 ret = scsi_mq_prep_fn(req);
1661                 if (ret != BLK_STS_OK)
1662                         goto out_dec_host_busy;
1663                 req->rq_flags |= RQF_DONTPREP;
1664         } else {
1665                 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1666                 blk_mq_start_request(req);
1667         }
1668
1669         if (sdev->simple_tags)
1670                 cmd->flags |= SCMD_TAGGED;
1671         else
1672                 cmd->flags &= ~SCMD_TAGGED;
1673
1674         scsi_init_cmd_errh(cmd);
1675         cmd->scsi_done = scsi_mq_done;
1676
1677         reason = scsi_dispatch_cmd(cmd);
1678         if (reason) {
1679                 scsi_set_blocked(cmd, reason);
1680                 ret = BLK_STS_RESOURCE;
1681                 goto out_dec_host_busy;
1682         }
1683
1684         return BLK_STS_OK;
1685
1686 out_dec_host_busy:
1687         scsi_dec_host_busy(shost);
1688 out_dec_target_busy:
1689         if (scsi_target(sdev)->can_queue > 0)
1690                 atomic_dec(&scsi_target(sdev)->target_busy);
1691 out_put_budget:
1692         scsi_mq_put_budget(hctx);
1693         switch (ret) {
1694         case BLK_STS_OK:
1695                 break;
1696         case BLK_STS_RESOURCE:
1697                 if (atomic_read(&sdev->device_busy) ||
1698                     scsi_device_blocked(sdev))
1699                         ret = BLK_STS_DEV_RESOURCE;
1700                 break;
1701         default:
1702                 if (unlikely(!scsi_device_online(sdev)))
1703                         scsi_req(req)->result = DID_NO_CONNECT << 16;
1704                 else
1705                         scsi_req(req)->result = DID_ERROR << 16;
1706                 /*
1707                  * Make sure to release all allocated resources when
1708                  * we hit an error, as we will never see this command
1709                  * again.
1710                  */
1711                 if (req->rq_flags & RQF_DONTPREP)
1712                         scsi_mq_uninit_cmd(cmd);
1713                 break;
1714         }
1715         return ret;
1716 }
1717
1718 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1719                 bool reserved)
1720 {
1721         if (reserved)
1722                 return BLK_EH_RESET_TIMER;
1723         return scsi_times_out(req);
1724 }
1725
1726 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1727                                 unsigned int hctx_idx, unsigned int numa_node)
1728 {
1729         struct Scsi_Host *shost = set->driver_data;
1730         const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1731         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1732         struct scatterlist *sg;
1733
1734         if (unchecked_isa_dma)
1735                 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1736         cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1737                                                     GFP_KERNEL, numa_node);
1738         if (!cmd->sense_buffer)
1739                 return -ENOMEM;
1740         cmd->req.sense = cmd->sense_buffer;
1741
1742         if (scsi_host_get_prot(shost)) {
1743                 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1744                         shost->hostt->cmd_size;
1745                 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1746         }
1747
1748         return 0;
1749 }
1750
1751 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1752                                  unsigned int hctx_idx)
1753 {
1754         struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1755
1756         scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1757                                cmd->sense_buffer);
1758 }
1759
1760 static int scsi_map_queues(struct blk_mq_tag_set *set)
1761 {
1762         struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1763
1764         if (shost->hostt->map_queues)
1765                 return shost->hostt->map_queues(shost);
1766         return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1767 }
1768
1769 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1770 {
1771         struct device *dev = shost->dma_dev;
1772
1773         /*
1774          * this limit is imposed by hardware restrictions
1775          */
1776         blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1777                                         SG_MAX_SEGMENTS));
1778
1779         if (scsi_host_prot_dma(shost)) {
1780                 shost->sg_prot_tablesize =
1781                         min_not_zero(shost->sg_prot_tablesize,
1782                                      (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1783                 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1784                 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1785         }
1786
1787         blk_queue_max_hw_sectors(q, shost->max_sectors);
1788         if (shost->unchecked_isa_dma)
1789                 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1790         blk_queue_segment_boundary(q, shost->dma_boundary);
1791         dma_set_seg_boundary(dev, shost->dma_boundary);
1792
1793         blk_queue_max_segment_size(q, shost->max_segment_size);
1794         dma_set_max_seg_size(dev, shost->max_segment_size);
1795
1796         /*
1797          * Set a reasonable default alignment:  The larger of 32-byte (dword),
1798          * which is a common minimum for HBAs, and the minimum DMA alignment,
1799          * which is set by the platform.
1800          *
1801          * Devices that require a bigger alignment can increase it later.
1802          */
1803         blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1804 }
1805 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1806
1807 static const struct blk_mq_ops scsi_mq_ops = {
1808         .get_budget     = scsi_mq_get_budget,
1809         .put_budget     = scsi_mq_put_budget,
1810         .queue_rq       = scsi_queue_rq,
1811         .complete       = scsi_softirq_done,
1812         .timeout        = scsi_timeout,
1813 #ifdef CONFIG_BLK_DEBUG_FS
1814         .show_rq        = scsi_show_rq,
1815 #endif
1816         .init_request   = scsi_mq_init_request,
1817         .exit_request   = scsi_mq_exit_request,
1818         .initialize_rq_fn = scsi_initialize_rq,
1819         .busy           = scsi_mq_lld_busy,
1820         .map_queues     = scsi_map_queues,
1821 };
1822
1823 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1824 {
1825         sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1826         if (IS_ERR(sdev->request_queue))
1827                 return NULL;
1828
1829         sdev->request_queue->queuedata = sdev;
1830         __scsi_init_queue(sdev->host, sdev->request_queue);
1831         blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1832         return sdev->request_queue;
1833 }
1834
1835 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1836 {
1837         unsigned int cmd_size, sgl_size;
1838
1839         sgl_size = scsi_mq_inline_sgl_size(shost);
1840         cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1841         if (scsi_host_get_prot(shost))
1842                 cmd_size += sizeof(struct scsi_data_buffer) +
1843                         sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1844
1845         memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1846         shost->tag_set.ops = &scsi_mq_ops;
1847         shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1848         shost->tag_set.queue_depth = shost->can_queue;
1849         shost->tag_set.cmd_size = cmd_size;
1850         shost->tag_set.numa_node = NUMA_NO_NODE;
1851         shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1852         shost->tag_set.flags |=
1853                 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1854         shost->tag_set.driver_data = shost;
1855
1856         return blk_mq_alloc_tag_set(&shost->tag_set);
1857 }
1858
1859 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1860 {
1861         blk_mq_free_tag_set(&shost->tag_set);
1862 }
1863
1864 /**
1865  * scsi_device_from_queue - return sdev associated with a request_queue
1866  * @q: The request queue to return the sdev from
1867  *
1868  * Return the sdev associated with a request queue or NULL if the
1869  * request_queue does not reference a SCSI device.
1870  */
1871 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1872 {
1873         struct scsi_device *sdev = NULL;
1874
1875         if (q->mq_ops == &scsi_mq_ops)
1876                 sdev = q->queuedata;
1877         if (!sdev || !get_device(&sdev->sdev_gendev))
1878                 sdev = NULL;
1879
1880         return sdev;
1881 }
1882 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1883
1884 /*
1885  * Function:    scsi_block_requests()
1886  *
1887  * Purpose:     Utility function used by low-level drivers to prevent further
1888  *              commands from being queued to the device.
1889  *
1890  * Arguments:   shost       - Host in question
1891  *
1892  * Returns:     Nothing
1893  *
1894  * Lock status: No locks are assumed held.
1895  *
1896  * Notes:       There is no timer nor any other means by which the requests
1897  *              get unblocked other than the low-level driver calling
1898  *              scsi_unblock_requests().
1899  */
1900 void scsi_block_requests(struct Scsi_Host *shost)
1901 {
1902         shost->host_self_blocked = 1;
1903 }
1904 EXPORT_SYMBOL(scsi_block_requests);
1905
1906 /*
1907  * Function:    scsi_unblock_requests()
1908  *
1909  * Purpose:     Utility function used by low-level drivers to allow further
1910  *              commands from being queued to the device.
1911  *
1912  * Arguments:   shost       - Host in question
1913  *
1914  * Returns:     Nothing
1915  *
1916  * Lock status: No locks are assumed held.
1917  *
1918  * Notes:       There is no timer nor any other means by which the requests
1919  *              get unblocked other than the low-level driver calling
1920  *              scsi_unblock_requests().
1921  *
1922  *              This is done as an API function so that changes to the
1923  *              internals of the scsi mid-layer won't require wholesale
1924  *              changes to drivers that use this feature.
1925  */
1926 void scsi_unblock_requests(struct Scsi_Host *shost)
1927 {
1928         shost->host_self_blocked = 0;
1929         scsi_run_host_queues(shost);
1930 }
1931 EXPORT_SYMBOL(scsi_unblock_requests);
1932
1933 int __init scsi_init_queue(void)
1934 {
1935         scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1936                                            sizeof(struct scsi_data_buffer),
1937                                            0, 0, NULL);
1938         if (!scsi_sdb_cache) {
1939                 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1940                 return -ENOMEM;
1941         }
1942
1943         return 0;
1944 }
1945
1946 void scsi_exit_queue(void)
1947 {
1948         kmem_cache_destroy(scsi_sense_cache);
1949         kmem_cache_destroy(scsi_sense_isadma_cache);
1950         kmem_cache_destroy(scsi_sdb_cache);
1951 }
1952
1953 /**
1954  *      scsi_mode_select - issue a mode select
1955  *      @sdev:  SCSI device to be queried
1956  *      @pf:    Page format bit (1 == standard, 0 == vendor specific)
1957  *      @sp:    Save page bit (0 == don't save, 1 == save)
1958  *      @modepage: mode page being requested
1959  *      @buffer: request buffer (may not be smaller than eight bytes)
1960  *      @len:   length of request buffer.
1961  *      @timeout: command timeout
1962  *      @retries: number of retries before failing
1963  *      @data: returns a structure abstracting the mode header data
1964  *      @sshdr: place to put sense data (or NULL if no sense to be collected).
1965  *              must be SCSI_SENSE_BUFFERSIZE big.
1966  *
1967  *      Returns zero if successful; negative error number or scsi
1968  *      status on error
1969  *
1970  */
1971 int
1972 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1973                  unsigned char *buffer, int len, int timeout, int retries,
1974                  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1975 {
1976         unsigned char cmd[10];
1977         unsigned char *real_buffer;
1978         int ret;
1979
1980         memset(cmd, 0, sizeof(cmd));
1981         cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1982
1983         if (sdev->use_10_for_ms) {
1984                 if (len > 65535)
1985                         return -EINVAL;
1986                 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1987                 if (!real_buffer)
1988                         return -ENOMEM;
1989                 memcpy(real_buffer + 8, buffer, len);
1990                 len += 8;
1991                 real_buffer[0] = 0;
1992                 real_buffer[1] = 0;
1993                 real_buffer[2] = data->medium_type;
1994                 real_buffer[3] = data->device_specific;
1995                 real_buffer[4] = data->longlba ? 0x01 : 0;
1996                 real_buffer[5] = 0;
1997                 real_buffer[6] = data->block_descriptor_length >> 8;
1998                 real_buffer[7] = data->block_descriptor_length;
1999
2000                 cmd[0] = MODE_SELECT_10;
2001                 cmd[7] = len >> 8;
2002                 cmd[8] = len;
2003         } else {
2004                 if (len > 255 || data->block_descriptor_length > 255 ||
2005                     data->longlba)
2006                         return -EINVAL;
2007
2008                 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2009                 if (!real_buffer)
2010                         return -ENOMEM;
2011                 memcpy(real_buffer + 4, buffer, len);
2012                 len += 4;
2013                 real_buffer[0] = 0;
2014                 real_buffer[1] = data->medium_type;
2015                 real_buffer[2] = data->device_specific;
2016                 real_buffer[3] = data->block_descriptor_length;
2017                 
2018
2019                 cmd[0] = MODE_SELECT;
2020                 cmd[4] = len;
2021         }
2022
2023         ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2024                                sshdr, timeout, retries, NULL);
2025         kfree(real_buffer);
2026         return ret;
2027 }
2028 EXPORT_SYMBOL_GPL(scsi_mode_select);
2029
2030 /**
2031  *      scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2032  *      @sdev:  SCSI device to be queried
2033  *      @dbd:   set if mode sense will allow block descriptors to be returned
2034  *      @modepage: mode page being requested
2035  *      @buffer: request buffer (may not be smaller than eight bytes)
2036  *      @len:   length of request buffer.
2037  *      @timeout: command timeout
2038  *      @retries: number of retries before failing
2039  *      @data: returns a structure abstracting the mode header data
2040  *      @sshdr: place to put sense data (or NULL if no sense to be collected).
2041  *              must be SCSI_SENSE_BUFFERSIZE big.
2042  *
2043  *      Returns zero if unsuccessful, or the header offset (either 4
2044  *      or 8 depending on whether a six or ten byte command was
2045  *      issued) if successful.
2046  */
2047 int
2048 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2049                   unsigned char *buffer, int len, int timeout, int retries,
2050                   struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2051 {
2052         unsigned char cmd[12];
2053         int use_10_for_ms;
2054         int header_length;
2055         int result, retry_count = retries;
2056         struct scsi_sense_hdr my_sshdr;
2057
2058         memset(data, 0, sizeof(*data));
2059         memset(&cmd[0], 0, 12);
2060         cmd[1] = dbd & 0x18;    /* allows DBD and LLBA bits */
2061         cmd[2] = modepage;
2062
2063         /* caller might not be interested in sense, but we need it */
2064         if (!sshdr)
2065                 sshdr = &my_sshdr;
2066
2067  retry:
2068         use_10_for_ms = sdev->use_10_for_ms;
2069
2070         if (use_10_for_ms) {
2071                 if (len < 8)
2072                         len = 8;
2073
2074                 cmd[0] = MODE_SENSE_10;
2075                 cmd[8] = len;
2076                 header_length = 8;
2077         } else {
2078                 if (len < 4)
2079                         len = 4;
2080
2081                 cmd[0] = MODE_SENSE;
2082                 cmd[4] = len;
2083                 header_length = 4;
2084         }
2085
2086         memset(buffer, 0, len);
2087
2088         result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2089                                   sshdr, timeout, retries, NULL);
2090
2091         /* This code looks awful: what it's doing is making sure an
2092          * ILLEGAL REQUEST sense return identifies the actual command
2093          * byte as the problem.  MODE_SENSE commands can return
2094          * ILLEGAL REQUEST if the code page isn't supported */
2095
2096         if (use_10_for_ms && !scsi_status_is_good(result) &&
2097             driver_byte(result) == DRIVER_SENSE) {
2098                 if (scsi_sense_valid(sshdr)) {
2099                         if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2100                             (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2101                                 /* 
2102                                  * Invalid command operation code
2103                                  */
2104                                 sdev->use_10_for_ms = 0;
2105                                 goto retry;
2106                         }
2107                 }
2108         }
2109
2110         if(scsi_status_is_good(result)) {
2111                 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2112                              (modepage == 6 || modepage == 8))) {
2113                         /* Initio breakage? */
2114                         header_length = 0;
2115                         data->length = 13;
2116                         data->medium_type = 0;
2117                         data->device_specific = 0;
2118                         data->longlba = 0;
2119                         data->block_descriptor_length = 0;
2120                 } else if(use_10_for_ms) {
2121                         data->length = buffer[0]*256 + buffer[1] + 2;
2122                         data->medium_type = buffer[2];
2123                         data->device_specific = buffer[3];
2124                         data->longlba = buffer[4] & 0x01;
2125                         data->block_descriptor_length = buffer[6]*256
2126                                 + buffer[7];
2127                 } else {
2128                         data->length = buffer[0] + 1;
2129                         data->medium_type = buffer[1];
2130                         data->device_specific = buffer[2];
2131                         data->block_descriptor_length = buffer[3];
2132                 }
2133                 data->header_length = header_length;
2134         } else if ((status_byte(result) == CHECK_CONDITION) &&
2135                    scsi_sense_valid(sshdr) &&
2136                    sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2137                 retry_count--;
2138                 goto retry;
2139         }
2140
2141         return result;
2142 }
2143 EXPORT_SYMBOL(scsi_mode_sense);
2144
2145 /**
2146  *      scsi_test_unit_ready - test if unit is ready
2147  *      @sdev:  scsi device to change the state of.
2148  *      @timeout: command timeout
2149  *      @retries: number of retries before failing
2150  *      @sshdr: outpout pointer for decoded sense information.
2151  *
2152  *      Returns zero if unsuccessful or an error if TUR failed.  For
2153  *      removable media, UNIT_ATTENTION sets ->changed flag.
2154  **/
2155 int
2156 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2157                      struct scsi_sense_hdr *sshdr)
2158 {
2159         char cmd[] = {
2160                 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2161         };
2162         int result;
2163
2164         /* try to eat the UNIT_ATTENTION if there are enough retries */
2165         do {
2166                 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2167                                           timeout, 1, NULL);
2168                 if (sdev->removable && scsi_sense_valid(sshdr) &&
2169                     sshdr->sense_key == UNIT_ATTENTION)
2170                         sdev->changed = 1;
2171         } while (scsi_sense_valid(sshdr) &&
2172                  sshdr->sense_key == UNIT_ATTENTION && --retries);
2173
2174         return result;
2175 }
2176 EXPORT_SYMBOL(scsi_test_unit_ready);
2177
2178 /**
2179  *      scsi_device_set_state - Take the given device through the device state model.
2180  *      @sdev:  scsi device to change the state of.
2181  *      @state: state to change to.
2182  *
2183  *      Returns zero if successful or an error if the requested
2184  *      transition is illegal.
2185  */
2186 int
2187 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2188 {
2189         enum scsi_device_state oldstate = sdev->sdev_state;
2190
2191         if (state == oldstate)
2192                 return 0;
2193
2194         switch (state) {
2195         case SDEV_CREATED:
2196                 switch (oldstate) {
2197                 case SDEV_CREATED_BLOCK:
2198                         break;
2199                 default:
2200                         goto illegal;
2201                 }
2202                 break;
2203                         
2204         case SDEV_RUNNING:
2205                 switch (oldstate) {
2206                 case SDEV_CREATED:
2207                 case SDEV_OFFLINE:
2208                 case SDEV_TRANSPORT_OFFLINE:
2209                 case SDEV_QUIESCE:
2210                 case SDEV_BLOCK:
2211                         break;
2212                 default:
2213                         goto illegal;
2214                 }
2215                 break;
2216
2217         case SDEV_QUIESCE:
2218                 switch (oldstate) {
2219                 case SDEV_RUNNING:
2220                 case SDEV_OFFLINE:
2221                 case SDEV_TRANSPORT_OFFLINE:
2222                         break;
2223                 default:
2224                         goto illegal;
2225                 }
2226                 break;
2227
2228         case SDEV_OFFLINE:
2229         case SDEV_TRANSPORT_OFFLINE:
2230                 switch (oldstate) {
2231                 case SDEV_CREATED:
2232                 case SDEV_RUNNING:
2233                 case SDEV_QUIESCE:
2234                 case SDEV_BLOCK:
2235                         break;
2236                 default:
2237                         goto illegal;
2238                 }
2239                 break;
2240
2241         case SDEV_BLOCK:
2242                 switch (oldstate) {
2243                 case SDEV_RUNNING:
2244                 case SDEV_CREATED_BLOCK:
2245                 case SDEV_OFFLINE:
2246                         break;
2247                 default:
2248                         goto illegal;
2249                 }
2250                 break;
2251
2252         case SDEV_CREATED_BLOCK:
2253                 switch (oldstate) {
2254                 case SDEV_CREATED:
2255                         break;
2256                 default:
2257                         goto illegal;
2258                 }
2259                 break;
2260
2261         case SDEV_CANCEL:
2262                 switch (oldstate) {
2263                 case SDEV_CREATED:
2264                 case SDEV_RUNNING:
2265                 case SDEV_QUIESCE:
2266                 case SDEV_OFFLINE:
2267                 case SDEV_TRANSPORT_OFFLINE:
2268                         break;
2269                 default:
2270                         goto illegal;
2271                 }
2272                 break;
2273
2274         case SDEV_DEL:
2275                 switch (oldstate) {
2276                 case SDEV_CREATED:
2277                 case SDEV_RUNNING:
2278                 case SDEV_OFFLINE:
2279                 case SDEV_TRANSPORT_OFFLINE:
2280                 case SDEV_CANCEL:
2281                 case SDEV_BLOCK:
2282                 case SDEV_CREATED_BLOCK:
2283                         break;
2284                 default:
2285                         goto illegal;
2286                 }
2287                 break;
2288
2289         }
2290         sdev->sdev_state = state;
2291         return 0;
2292
2293  illegal:
2294         SCSI_LOG_ERROR_RECOVERY(1,
2295                                 sdev_printk(KERN_ERR, sdev,
2296                                             "Illegal state transition %s->%s",
2297                                             scsi_device_state_name(oldstate),
2298                                             scsi_device_state_name(state))
2299                                 );
2300         return -EINVAL;
2301 }
2302 EXPORT_SYMBOL(scsi_device_set_state);
2303
2304 /**
2305  *      sdev_evt_emit - emit a single SCSI device uevent
2306  *      @sdev: associated SCSI device
2307  *      @evt: event to emit
2308  *
2309  *      Send a single uevent (scsi_event) to the associated scsi_device.
2310  */
2311 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2312 {
2313         int idx = 0;
2314         char *envp[3];
2315
2316         switch (evt->evt_type) {
2317         case SDEV_EVT_MEDIA_CHANGE:
2318                 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2319                 break;
2320         case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2321                 scsi_rescan_device(&sdev->sdev_gendev);
2322                 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2323                 break;
2324         case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2325                 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2326                 break;
2327         case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2328                envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2329                 break;
2330         case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2331                 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2332                 break;
2333         case SDEV_EVT_LUN_CHANGE_REPORTED:
2334                 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2335                 break;
2336         case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2337                 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2338                 break;
2339         case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2340                 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2341                 break;
2342         default:
2343                 /* do nothing */
2344                 break;
2345         }
2346
2347         envp[idx++] = NULL;
2348
2349         kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2350 }
2351
2352 /**
2353  *      sdev_evt_thread - send a uevent for each scsi event
2354  *      @work: work struct for scsi_device
2355  *
2356  *      Dispatch queued events to their associated scsi_device kobjects
2357  *      as uevents.
2358  */
2359 void scsi_evt_thread(struct work_struct *work)
2360 {
2361         struct scsi_device *sdev;
2362         enum scsi_device_event evt_type;
2363         LIST_HEAD(event_list);
2364
2365         sdev = container_of(work, struct scsi_device, event_work);
2366
2367         for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2368                 if (test_and_clear_bit(evt_type, sdev->pending_events))
2369                         sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2370
2371         while (1) {
2372                 struct scsi_event *evt;
2373                 struct list_head *this, *tmp;
2374                 unsigned long flags;
2375
2376                 spin_lock_irqsave(&sdev->list_lock, flags);
2377                 list_splice_init(&sdev->event_list, &event_list);
2378                 spin_unlock_irqrestore(&sdev->list_lock, flags);
2379
2380                 if (list_empty(&event_list))
2381                         break;
2382
2383                 list_for_each_safe(this, tmp, &event_list) {
2384                         evt = list_entry(this, struct scsi_event, node);
2385                         list_del(&evt->node);
2386                         scsi_evt_emit(sdev, evt);
2387                         kfree(evt);
2388                 }
2389         }
2390 }
2391
2392 /**
2393  *      sdev_evt_send - send asserted event to uevent thread
2394  *      @sdev: scsi_device event occurred on
2395  *      @evt: event to send
2396  *
2397  *      Assert scsi device event asynchronously.
2398  */
2399 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2400 {
2401         unsigned long flags;
2402
2403 #if 0
2404         /* FIXME: currently this check eliminates all media change events
2405          * for polled devices.  Need to update to discriminate between AN
2406          * and polled events */
2407         if (!test_bit(evt->evt_type, sdev->supported_events)) {
2408                 kfree(evt);
2409                 return;
2410         }
2411 #endif
2412
2413         spin_lock_irqsave(&sdev->list_lock, flags);
2414         list_add_tail(&evt->node, &sdev->event_list);
2415         schedule_work(&sdev->event_work);
2416         spin_unlock_irqrestore(&sdev->list_lock, flags);
2417 }
2418 EXPORT_SYMBOL_GPL(sdev_evt_send);
2419
2420 /**
2421  *      sdev_evt_alloc - allocate a new scsi event
2422  *      @evt_type: type of event to allocate
2423  *      @gfpflags: GFP flags for allocation
2424  *
2425  *      Allocates and returns a new scsi_event.
2426  */
2427 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2428                                   gfp_t gfpflags)
2429 {
2430         struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2431         if (!evt)
2432                 return NULL;
2433
2434         evt->evt_type = evt_type;
2435         INIT_LIST_HEAD(&evt->node);
2436
2437         /* evt_type-specific initialization, if any */
2438         switch (evt_type) {
2439         case SDEV_EVT_MEDIA_CHANGE:
2440         case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2441         case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2442         case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2443         case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2444         case SDEV_EVT_LUN_CHANGE_REPORTED:
2445         case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2446         case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2447         default:
2448                 /* do nothing */
2449                 break;
2450         }
2451
2452         return evt;
2453 }
2454 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2455
2456 /**
2457  *      sdev_evt_send_simple - send asserted event to uevent thread
2458  *      @sdev: scsi_device event occurred on
2459  *      @evt_type: type of event to send
2460  *      @gfpflags: GFP flags for allocation
2461  *
2462  *      Assert scsi device event asynchronously, given an event type.
2463  */
2464 void sdev_evt_send_simple(struct scsi_device *sdev,
2465                           enum scsi_device_event evt_type, gfp_t gfpflags)
2466 {
2467         struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2468         if (!evt) {
2469                 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2470                             evt_type);
2471                 return;
2472         }
2473
2474         sdev_evt_send(sdev, evt);
2475 }
2476 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2477
2478 /**
2479  *      scsi_device_quiesce - Block user issued commands.
2480  *      @sdev:  scsi device to quiesce.
2481  *
2482  *      This works by trying to transition to the SDEV_QUIESCE state
2483  *      (which must be a legal transition).  When the device is in this
2484  *      state, only special requests will be accepted, all others will
2485  *      be deferred.  Since special requests may also be requeued requests,
2486  *      a successful return doesn't guarantee the device will be 
2487  *      totally quiescent.
2488  *
2489  *      Must be called with user context, may sleep.
2490  *
2491  *      Returns zero if unsuccessful or an error if not.
2492  */
2493 int
2494 scsi_device_quiesce(struct scsi_device *sdev)
2495 {
2496         struct request_queue *q = sdev->request_queue;
2497         int err;
2498
2499         /*
2500          * It is allowed to call scsi_device_quiesce() multiple times from
2501          * the same context but concurrent scsi_device_quiesce() calls are
2502          * not allowed.
2503          */
2504         WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2505
2506         if (sdev->quiesced_by == current)
2507                 return 0;
2508
2509         blk_set_pm_only(q);
2510
2511         blk_mq_freeze_queue(q);
2512         /*
2513          * Ensure that the effect of blk_set_pm_only() will be visible
2514          * for percpu_ref_tryget() callers that occur after the queue
2515          * unfreeze even if the queue was already frozen before this function
2516          * was called. See also https://lwn.net/Articles/573497/.
2517          */
2518         synchronize_rcu();
2519         blk_mq_unfreeze_queue(q);
2520
2521         mutex_lock(&sdev->state_mutex);
2522         err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2523         if (err == 0)
2524                 sdev->quiesced_by = current;
2525         else
2526                 blk_clear_pm_only(q);
2527         mutex_unlock(&sdev->state_mutex);
2528
2529         return err;
2530 }
2531 EXPORT_SYMBOL(scsi_device_quiesce);
2532
2533 /**
2534  *      scsi_device_resume - Restart user issued commands to a quiesced device.
2535  *      @sdev:  scsi device to resume.
2536  *
2537  *      Moves the device from quiesced back to running and restarts the
2538  *      queues.
2539  *
2540  *      Must be called with user context, may sleep.
2541  */
2542 void scsi_device_resume(struct scsi_device *sdev)
2543 {
2544         /* check if the device state was mutated prior to resume, and if
2545          * so assume the state is being managed elsewhere (for example
2546          * device deleted during suspend)
2547          */
2548         mutex_lock(&sdev->state_mutex);
2549         if (sdev->quiesced_by) {
2550                 sdev->quiesced_by = NULL;
2551                 blk_clear_pm_only(sdev->request_queue);
2552         }
2553         if (sdev->sdev_state == SDEV_QUIESCE)
2554                 scsi_device_set_state(sdev, SDEV_RUNNING);
2555         mutex_unlock(&sdev->state_mutex);
2556 }
2557 EXPORT_SYMBOL(scsi_device_resume);
2558
2559 static void
2560 device_quiesce_fn(struct scsi_device *sdev, void *data)
2561 {
2562         scsi_device_quiesce(sdev);
2563 }
2564
2565 void
2566 scsi_target_quiesce(struct scsi_target *starget)
2567 {
2568         starget_for_each_device(starget, NULL, device_quiesce_fn);
2569 }
2570 EXPORT_SYMBOL(scsi_target_quiesce);
2571
2572 static void
2573 device_resume_fn(struct scsi_device *sdev, void *data)
2574 {
2575         scsi_device_resume(sdev);
2576 }
2577
2578 void
2579 scsi_target_resume(struct scsi_target *starget)
2580 {
2581         starget_for_each_device(starget, NULL, device_resume_fn);
2582 }
2583 EXPORT_SYMBOL(scsi_target_resume);
2584
2585 /**
2586  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2587  * @sdev: device to block
2588  *
2589  * Pause SCSI command processing on the specified device. Does not sleep.
2590  *
2591  * Returns zero if successful or a negative error code upon failure.
2592  *
2593  * Notes:
2594  * This routine transitions the device to the SDEV_BLOCK state (which must be
2595  * a legal transition). When the device is in this state, command processing
2596  * is paused until the device leaves the SDEV_BLOCK state. See also
2597  * scsi_internal_device_unblock_nowait().
2598  */
2599 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2600 {
2601         struct request_queue *q = sdev->request_queue;
2602         int err = 0;
2603
2604         err = scsi_device_set_state(sdev, SDEV_BLOCK);
2605         if (err) {
2606                 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2607
2608                 if (err)
2609                         return err;
2610         }
2611
2612         /* 
2613          * The device has transitioned to SDEV_BLOCK.  Stop the
2614          * block layer from calling the midlayer with this device's
2615          * request queue. 
2616          */
2617         blk_mq_quiesce_queue_nowait(q);
2618         return 0;
2619 }
2620 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2621
2622 /**
2623  * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2624  * @sdev: device to block
2625  *
2626  * Pause SCSI command processing on the specified device and wait until all
2627  * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2628  *
2629  * Returns zero if successful or a negative error code upon failure.
2630  *
2631  * Note:
2632  * This routine transitions the device to the SDEV_BLOCK state (which must be
2633  * a legal transition). When the device is in this state, command processing
2634  * is paused until the device leaves the SDEV_BLOCK state. See also
2635  * scsi_internal_device_unblock().
2636  */
2637 static int scsi_internal_device_block(struct scsi_device *sdev)
2638 {
2639         struct request_queue *q = sdev->request_queue;
2640         int err;
2641
2642         mutex_lock(&sdev->state_mutex);
2643         err = scsi_internal_device_block_nowait(sdev);
2644         if (err == 0)
2645                 blk_mq_quiesce_queue(q);
2646         mutex_unlock(&sdev->state_mutex);
2647
2648         return err;
2649 }
2650  
2651 void scsi_start_queue(struct scsi_device *sdev)
2652 {
2653         struct request_queue *q = sdev->request_queue;
2654
2655         blk_mq_unquiesce_queue(q);
2656 }
2657
2658 /**
2659  * scsi_internal_device_unblock_nowait - resume a device after a block request
2660  * @sdev:       device to resume
2661  * @new_state:  state to set the device to after unblocking
2662  *
2663  * Restart the device queue for a previously suspended SCSI device. Does not
2664  * sleep.
2665  *
2666  * Returns zero if successful or a negative error code upon failure.
2667  *
2668  * Notes:
2669  * This routine transitions the device to the SDEV_RUNNING state or to one of
2670  * the offline states (which must be a legal transition) allowing the midlayer
2671  * to goose the queue for this device.
2672  */
2673 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2674                                         enum scsi_device_state new_state)
2675 {
2676         /*
2677          * Try to transition the scsi device to SDEV_RUNNING or one of the
2678          * offlined states and goose the device queue if successful.
2679          */
2680         switch (sdev->sdev_state) {
2681         case SDEV_BLOCK:
2682         case SDEV_TRANSPORT_OFFLINE:
2683                 sdev->sdev_state = new_state;
2684                 break;
2685         case SDEV_CREATED_BLOCK:
2686                 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2687                     new_state == SDEV_OFFLINE)
2688                         sdev->sdev_state = new_state;
2689                 else
2690                         sdev->sdev_state = SDEV_CREATED;
2691                 break;
2692         case SDEV_CANCEL:
2693         case SDEV_OFFLINE:
2694                 break;
2695         default:
2696                 return -EINVAL;
2697         }
2698         scsi_start_queue(sdev);
2699
2700         return 0;
2701 }
2702 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2703
2704 /**
2705  * scsi_internal_device_unblock - resume a device after a block request
2706  * @sdev:       device to resume
2707  * @new_state:  state to set the device to after unblocking
2708  *
2709  * Restart the device queue for a previously suspended SCSI device. May sleep.
2710  *
2711  * Returns zero if successful or a negative error code upon failure.
2712  *
2713  * Notes:
2714  * This routine transitions the device to the SDEV_RUNNING state or to one of
2715  * the offline states (which must be a legal transition) allowing the midlayer
2716  * to goose the queue for this device.
2717  */
2718 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2719                                         enum scsi_device_state new_state)
2720 {
2721         int ret;
2722
2723         mutex_lock(&sdev->state_mutex);
2724         ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2725         mutex_unlock(&sdev->state_mutex);
2726
2727         return ret;
2728 }
2729
2730 static void
2731 device_block(struct scsi_device *sdev, void *data)
2732 {
2733         scsi_internal_device_block(sdev);
2734 }
2735
2736 static int
2737 target_block(struct device *dev, void *data)
2738 {
2739         if (scsi_is_target_device(dev))
2740                 starget_for_each_device(to_scsi_target(dev), NULL,
2741                                         device_block);
2742         return 0;
2743 }
2744
2745 void
2746 scsi_target_block(struct device *dev)
2747 {
2748         if (scsi_is_target_device(dev))
2749                 starget_for_each_device(to_scsi_target(dev), NULL,
2750                                         device_block);
2751         else
2752                 device_for_each_child(dev, NULL, target_block);
2753 }
2754 EXPORT_SYMBOL_GPL(scsi_target_block);
2755
2756 static void
2757 device_unblock(struct scsi_device *sdev, void *data)
2758 {
2759         scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2760 }
2761
2762 static int
2763 target_unblock(struct device *dev, void *data)
2764 {
2765         if (scsi_is_target_device(dev))
2766                 starget_for_each_device(to_scsi_target(dev), data,
2767                                         device_unblock);
2768         return 0;
2769 }
2770
2771 void
2772 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2773 {
2774         if (scsi_is_target_device(dev))
2775                 starget_for_each_device(to_scsi_target(dev), &new_state,
2776                                         device_unblock);
2777         else
2778                 device_for_each_child(dev, &new_state, target_unblock);
2779 }
2780 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2781
2782 /**
2783  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2784  * @sgl:        scatter-gather list
2785  * @sg_count:   number of segments in sg
2786  * @offset:     offset in bytes into sg, on return offset into the mapped area
2787  * @len:        bytes to map, on return number of bytes mapped
2788  *
2789  * Returns virtual address of the start of the mapped page
2790  */
2791 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2792                           size_t *offset, size_t *len)
2793 {
2794         int i;
2795         size_t sg_len = 0, len_complete = 0;
2796         struct scatterlist *sg;
2797         struct page *page;
2798
2799         WARN_ON(!irqs_disabled());
2800
2801         for_each_sg(sgl, sg, sg_count, i) {
2802                 len_complete = sg_len; /* Complete sg-entries */
2803                 sg_len += sg->length;
2804                 if (sg_len > *offset)
2805                         break;
2806         }
2807
2808         if (unlikely(i == sg_count)) {
2809                 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2810                         "elements %d\n",
2811                        __func__, sg_len, *offset, sg_count);
2812                 WARN_ON(1);
2813                 return NULL;
2814         }
2815
2816         /* Offset starting from the beginning of first page in this sg-entry */
2817         *offset = *offset - len_complete + sg->offset;
2818
2819         /* Assumption: contiguous pages can be accessed as "page + i" */
2820         page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2821         *offset &= ~PAGE_MASK;
2822
2823         /* Bytes in this sg-entry from *offset to the end of the page */
2824         sg_len = PAGE_SIZE - *offset;
2825         if (*len > sg_len)
2826                 *len = sg_len;
2827
2828         return kmap_atomic(page);
2829 }
2830 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2831
2832 /**
2833  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2834  * @virt:       virtual address to be unmapped
2835  */
2836 void scsi_kunmap_atomic_sg(void *virt)
2837 {
2838         kunmap_atomic(virt);
2839 }
2840 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2841
2842 void sdev_disable_disk_events(struct scsi_device *sdev)
2843 {
2844         atomic_inc(&sdev->disk_events_disable_depth);
2845 }
2846 EXPORT_SYMBOL(sdev_disable_disk_events);
2847
2848 void sdev_enable_disk_events(struct scsi_device *sdev)
2849 {
2850         if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2851                 return;
2852         atomic_dec(&sdev->disk_events_disable_depth);
2853 }
2854 EXPORT_SYMBOL(sdev_enable_disk_events);
2855
2856 /**
2857  * scsi_vpd_lun_id - return a unique device identification
2858  * @sdev: SCSI device
2859  * @id:   buffer for the identification
2860  * @id_len:  length of the buffer
2861  *
2862  * Copies a unique device identification into @id based
2863  * on the information in the VPD page 0x83 of the device.
2864  * The string will be formatted as a SCSI name string.
2865  *
2866  * Returns the length of the identification or error on failure.
2867  * If the identifier is longer than the supplied buffer the actual
2868  * identifier length is returned and the buffer is not zero-padded.
2869  */
2870 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2871 {
2872         u8 cur_id_type = 0xff;
2873         u8 cur_id_size = 0;
2874         const unsigned char *d, *cur_id_str;
2875         const struct scsi_vpd *vpd_pg83;
2876         int id_size = -EINVAL;
2877
2878         rcu_read_lock();
2879         vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2880         if (!vpd_pg83) {
2881                 rcu_read_unlock();
2882                 return -ENXIO;
2883         }
2884
2885         /*
2886          * Look for the correct descriptor.
2887          * Order of preference for lun descriptor:
2888          * - SCSI name string
2889          * - NAA IEEE Registered Extended
2890          * - EUI-64 based 16-byte
2891          * - EUI-64 based 12-byte
2892          * - NAA IEEE Registered
2893          * - NAA IEEE Extended
2894          * - T10 Vendor ID
2895          * as longer descriptors reduce the likelyhood
2896          * of identification clashes.
2897          */
2898
2899         /* The id string must be at least 20 bytes + terminating NULL byte */
2900         if (id_len < 21) {
2901                 rcu_read_unlock();
2902                 return -EINVAL;
2903         }
2904
2905         memset(id, 0, id_len);
2906         d = vpd_pg83->data + 4;
2907         while (d < vpd_pg83->data + vpd_pg83->len) {
2908                 /* Skip designators not referring to the LUN */
2909                 if ((d[1] & 0x30) != 0x00)
2910                         goto next_desig;
2911
2912                 switch (d[1] & 0xf) {
2913                 case 0x1:
2914                         /* T10 Vendor ID */
2915                         if (cur_id_size > d[3])
2916                                 break;
2917                         /* Prefer anything */
2918                         if (cur_id_type > 0x01 && cur_id_type != 0xff)
2919                                 break;
2920                         cur_id_size = d[3];
2921                         if (cur_id_size + 4 > id_len)
2922                                 cur_id_size = id_len - 4;
2923                         cur_id_str = d + 4;
2924                         cur_id_type = d[1] & 0xf;
2925                         id_size = snprintf(id, id_len, "t10.%*pE",
2926                                            cur_id_size, cur_id_str);
2927                         break;
2928                 case 0x2:
2929                         /* EUI-64 */
2930                         if (cur_id_size > d[3])
2931                                 break;
2932                         /* Prefer NAA IEEE Registered Extended */
2933                         if (cur_id_type == 0x3 &&
2934                             cur_id_size == d[3])
2935                                 break;
2936                         cur_id_size = d[3];
2937                         cur_id_str = d + 4;
2938                         cur_id_type = d[1] & 0xf;
2939                         switch (cur_id_size) {
2940                         case 8:
2941                                 id_size = snprintf(id, id_len,
2942                                                    "eui.%8phN",
2943                                                    cur_id_str);
2944                                 break;
2945                         case 12:
2946                                 id_size = snprintf(id, id_len,
2947                                                    "eui.%12phN",
2948                                                    cur_id_str);
2949                                 break;
2950                         case 16:
2951                                 id_size = snprintf(id, id_len,
2952                                                    "eui.%16phN",
2953                                                    cur_id_str);
2954                                 break;
2955                         default:
2956                                 cur_id_size = 0;
2957                                 break;
2958                         }
2959                         break;
2960                 case 0x3:
2961                         /* NAA */
2962                         if (cur_id_size > d[3])
2963                                 break;
2964                         cur_id_size = d[3];
2965                         cur_id_str = d + 4;
2966                         cur_id_type = d[1] & 0xf;
2967                         switch (cur_id_size) {
2968                         case 8:
2969                                 id_size = snprintf(id, id_len,
2970                                                    "naa.%8phN",
2971                                                    cur_id_str);
2972                                 break;
2973                         case 16:
2974                                 id_size = snprintf(id, id_len,
2975                                                    "naa.%16phN",
2976                                                    cur_id_str);
2977                                 break;
2978                         default:
2979                                 cur_id_size = 0;
2980                                 break;
2981                         }
2982                         break;
2983                 case 0x8:
2984                         /* SCSI name string */
2985                         if (cur_id_size + 4 > d[3])
2986                                 break;
2987                         /* Prefer others for truncated descriptor */
2988                         if (cur_id_size && d[3] > id_len)
2989                                 break;
2990                         cur_id_size = id_size = d[3];
2991                         cur_id_str = d + 4;
2992                         cur_id_type = d[1] & 0xf;
2993                         if (cur_id_size >= id_len)
2994                                 cur_id_size = id_len - 1;
2995                         memcpy(id, cur_id_str, cur_id_size);
2996                         /* Decrease priority for truncated descriptor */
2997                         if (cur_id_size != id_size)
2998                                 cur_id_size = 6;
2999                         break;
3000                 default:
3001                         break;
3002                 }
3003 next_desig:
3004                 d += d[3] + 4;
3005         }
3006         rcu_read_unlock();
3007
3008         return id_size;
3009 }
3010 EXPORT_SYMBOL(scsi_vpd_lun_id);
3011
3012 /*
3013  * scsi_vpd_tpg_id - return a target port group identifier
3014  * @sdev: SCSI device
3015  *
3016  * Returns the Target Port Group identifier from the information
3017  * froom VPD page 0x83 of the device.
3018  *
3019  * Returns the identifier or error on failure.
3020  */
3021 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3022 {
3023         const unsigned char *d;
3024         const struct scsi_vpd *vpd_pg83;
3025         int group_id = -EAGAIN, rel_port = -1;
3026
3027         rcu_read_lock();
3028         vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3029         if (!vpd_pg83) {
3030                 rcu_read_unlock();
3031                 return -ENXIO;
3032         }
3033
3034         d = vpd_pg83->data + 4;
3035         while (d < vpd_pg83->data + vpd_pg83->len) {
3036                 switch (d[1] & 0xf) {
3037                 case 0x4:
3038                         /* Relative target port */
3039                         rel_port = get_unaligned_be16(&d[6]);
3040                         break;
3041                 case 0x5:
3042                         /* Target port group */
3043                         group_id = get_unaligned_be16(&d[6]);
3044                         break;
3045                 default:
3046                         break;
3047                 }
3048                 d += d[3] + 4;
3049         }
3050         rcu_read_unlock();
3051
3052         if (group_id >= 0 && rel_id && rel_port != -1)
3053                 *rel_id = rel_port;
3054
3055         return group_id;
3056 }
3057 EXPORT_SYMBOL(scsi_vpd_tpg_id);