~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/block/blk-core.c

Version: ~ [ linux-5.2-rc1 ] ~ [ linux-5.1.2 ] ~ [ linux-5.0.16 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.43 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.119 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.176 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.179 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.139 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.67 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (C) 1991, 1992 Linus Torvalds
  3  * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
  4  * Elevator latency, (C) 2000  Andrea Arcangeli <andrea@suse.de> SuSE
  5  * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
  6  * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
  7  *      -  July2000
  8  * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
  9  */
 10 
 11 /*
 12  * This handles all read/write requests to block devices
 13  */
 14 #include <linux/kernel.h>
 15 #include <linux/module.h>
 16 #include <linux/backing-dev.h>
 17 #include <linux/bio.h>
 18 #include <linux/blkdev.h>
 19 #include <linux/blk-mq.h>
 20 #include <linux/highmem.h>
 21 #include <linux/mm.h>
 22 #include <linux/kernel_stat.h>
 23 #include <linux/string.h>
 24 #include <linux/init.h>
 25 #include <linux/completion.h>
 26 #include <linux/slab.h>
 27 #include <linux/swap.h>
 28 #include <linux/writeback.h>
 29 #include <linux/task_io_accounting_ops.h>
 30 #include <linux/fault-inject.h>
 31 #include <linux/list_sort.h>
 32 #include <linux/delay.h>
 33 #include <linux/ratelimit.h>
 34 #include <linux/pm_runtime.h>
 35 
 36 #define CREATE_TRACE_POINTS
 37 #include <trace/events/block.h>
 38 
 39 #include "blk.h"
 40 #include "blk-cgroup.h"
 41 #include "blk-mq.h"
 42 
 43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
 44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
 45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
 46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
 47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
 48 
 49 DEFINE_IDA(blk_queue_ida);
 50 
 51 /*
 52  * For the allocated request tables
 53  */
 54 struct kmem_cache *request_cachep = NULL;
 55 
 56 /*
 57  * For queue allocation
 58  */
 59 struct kmem_cache *blk_requestq_cachep;
 60 
 61 /*
 62  * Controlling structure to kblockd
 63  */
 64 static struct workqueue_struct *kblockd_workqueue;
 65 
 66 void blk_queue_congestion_threshold(struct request_queue *q)
 67 {
 68         int nr;
 69 
 70         nr = q->nr_requests - (q->nr_requests / 8) + 1;
 71         if (nr > q->nr_requests)
 72                 nr = q->nr_requests;
 73         q->nr_congestion_on = nr;
 74 
 75         nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
 76         if (nr < 1)
 77                 nr = 1;
 78         q->nr_congestion_off = nr;
 79 }
 80 
 81 /**
 82  * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
 83  * @bdev:       device
 84  *
 85  * Locates the passed device's request queue and returns the address of its
 86  * backing_dev_info.  This function can only be called if @bdev is opened
 87  * and the return value is never NULL.
 88  */
 89 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
 90 {
 91         struct request_queue *q = bdev_get_queue(bdev);
 92 
 93         return &q->backing_dev_info;
 94 }
 95 EXPORT_SYMBOL(blk_get_backing_dev_info);
 96 
 97 void blk_rq_init(struct request_queue *q, struct request *rq)
 98 {
 99         memset(rq, 0, sizeof(*rq));
100 
101         INIT_LIST_HEAD(&rq->queuelist);
102         INIT_LIST_HEAD(&rq->timeout_list);
103         rq->cpu = -1;
104         rq->q = q;
105         rq->__sector = (sector_t) -1;
106         INIT_HLIST_NODE(&rq->hash);
107         RB_CLEAR_NODE(&rq->rb_node);
108         rq->cmd = rq->__cmd;
109         rq->cmd_len = BLK_MAX_CDB;
110         rq->tag = -1;
111         rq->start_time = jiffies;
112         set_start_time_ns(rq);
113         rq->part = NULL;
114 }
115 EXPORT_SYMBOL(blk_rq_init);
116 
117 static void req_bio_endio(struct request *rq, struct bio *bio,
118                           unsigned int nbytes, int error)
119 {
120         if (error)
121                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
122         else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
123                 error = -EIO;
124 
125         if (unlikely(rq->cmd_flags & REQ_QUIET))
126                 set_bit(BIO_QUIET, &bio->bi_flags);
127 
128         bio_advance(bio, nbytes);
129 
130         /* don't actually finish bio if it's part of flush sequence */
131         if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
132                 bio_endio(bio, error);
133 }
134 
135 void blk_dump_rq_flags(struct request *rq, char *msg)
136 {
137         int bit;
138 
139         printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
140                 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
141                 (unsigned long long) rq->cmd_flags);
142 
143         printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
144                (unsigned long long)blk_rq_pos(rq),
145                blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
146         printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
147                rq->bio, rq->biotail, blk_rq_bytes(rq));
148 
149         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
150                 printk(KERN_INFO "  cdb: ");
151                 for (bit = 0; bit < BLK_MAX_CDB; bit++)
152                         printk("%02x ", rq->cmd[bit]);
153                 printk("\n");
154         }
155 }
156 EXPORT_SYMBOL(blk_dump_rq_flags);
157 
158 static void blk_delay_work(struct work_struct *work)
159 {
160         struct request_queue *q;
161 
162         q = container_of(work, struct request_queue, delay_work.work);
163         spin_lock_irq(q->queue_lock);
164         __blk_run_queue(q);
165         spin_unlock_irq(q->queue_lock);
166 }
167 
168 /**
169  * blk_delay_queue - restart queueing after defined interval
170  * @q:          The &struct request_queue in question
171  * @msecs:      Delay in msecs
172  *
173  * Description:
174  *   Sometimes queueing needs to be postponed for a little while, to allow
175  *   resources to come back. This function will make sure that queueing is
176  *   restarted around the specified time. Queue lock must be held.
177  */
178 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
179 {
180         if (likely(!blk_queue_dead(q)))
181                 queue_delayed_work(kblockd_workqueue, &q->delay_work,
182                                    msecs_to_jiffies(msecs));
183 }
184 EXPORT_SYMBOL(blk_delay_queue);
185 
186 /**
187  * blk_start_queue - restart a previously stopped queue
188  * @q:    The &struct request_queue in question
189  *
190  * Description:
191  *   blk_start_queue() will clear the stop flag on the queue, and call
192  *   the request_fn for the queue if it was in a stopped state when
193  *   entered. Also see blk_stop_queue(). Queue lock must be held.
194  **/
195 void blk_start_queue(struct request_queue *q)
196 {
197         WARN_ON(!in_interrupt() && !irqs_disabled());
198 
199         queue_flag_clear(QUEUE_FLAG_STOPPED, q);
200         __blk_run_queue(q);
201 }
202 EXPORT_SYMBOL(blk_start_queue);
203 
204 /**
205  * blk_stop_queue - stop a queue
206  * @q:    The &struct request_queue in question
207  *
208  * Description:
209  *   The Linux block layer assumes that a block driver will consume all
210  *   entries on the request queue when the request_fn strategy is called.
211  *   Often this will not happen, because of hardware limitations (queue
212  *   depth settings). If a device driver gets a 'queue full' response,
213  *   or if it simply chooses not to queue more I/O at one point, it can
214  *   call this function to prevent the request_fn from being called until
215  *   the driver has signalled it's ready to go again. This happens by calling
216  *   blk_start_queue() to restart queue operations. Queue lock must be held.
217  **/
218 void blk_stop_queue(struct request_queue *q)
219 {
220         cancel_delayed_work(&q->delay_work);
221         queue_flag_set(QUEUE_FLAG_STOPPED, q);
222 }
223 EXPORT_SYMBOL(blk_stop_queue);
224 
225 /**
226  * blk_sync_queue - cancel any pending callbacks on a queue
227  * @q: the queue
228  *
229  * Description:
230  *     The block layer may perform asynchronous callback activity
231  *     on a queue, such as calling the unplug function after a timeout.
232  *     A block device may call blk_sync_queue to ensure that any
233  *     such activity is cancelled, thus allowing it to release resources
234  *     that the callbacks might use. The caller must already have made sure
235  *     that its ->make_request_fn will not re-add plugging prior to calling
236  *     this function.
237  *
238  *     This function does not cancel any asynchronous activity arising
239  *     out of elevator or throttling code. That would require elevator_exit()
240  *     and blkcg_exit_queue() to be called with queue lock initialized.
241  *
242  */
243 void blk_sync_queue(struct request_queue *q)
244 {
245         del_timer_sync(&q->timeout);
246 
247         if (q->mq_ops) {
248                 struct blk_mq_hw_ctx *hctx;
249                 int i;
250 
251                 queue_for_each_hw_ctx(q, hctx, i) {
252                         cancel_delayed_work_sync(&hctx->run_work);
253                         cancel_delayed_work_sync(&hctx->delay_work);
254                 }
255         } else {
256                 cancel_delayed_work_sync(&q->delay_work);
257         }
258 }
259 EXPORT_SYMBOL(blk_sync_queue);
260 
261 /**
262  * __blk_run_queue_uncond - run a queue whether or not it has been stopped
263  * @q:  The queue to run
264  *
265  * Description:
266  *    Invoke request handling on a queue if there are any pending requests.
267  *    May be used to restart request handling after a request has completed.
268  *    This variant runs the queue whether or not the queue has been
269  *    stopped. Must be called with the queue lock held and interrupts
270  *    disabled. See also @blk_run_queue.
271  */
272 inline void __blk_run_queue_uncond(struct request_queue *q)
273 {
274         if (unlikely(blk_queue_dead(q)))
275                 return;
276 
277         /*
278          * Some request_fn implementations, e.g. scsi_request_fn(), unlock
279          * the queue lock internally. As a result multiple threads may be
280          * running such a request function concurrently. Keep track of the
281          * number of active request_fn invocations such that blk_drain_queue()
282          * can wait until all these request_fn calls have finished.
283          */
284         q->request_fn_active++;
285         q->request_fn(q);
286         q->request_fn_active--;
287 }
288 
289 /**
290  * __blk_run_queue - run a single device queue
291  * @q:  The queue to run
292  *
293  * Description:
294  *    See @blk_run_queue. This variant must be called with the queue lock
295  *    held and interrupts disabled.
296  */
297 void __blk_run_queue(struct request_queue *q)
298 {
299         if (unlikely(blk_queue_stopped(q)))
300                 return;
301 
302         __blk_run_queue_uncond(q);
303 }
304 EXPORT_SYMBOL(__blk_run_queue);
305 
306 /**
307  * blk_run_queue_async - run a single device queue in workqueue context
308  * @q:  The queue to run
309  *
310  * Description:
311  *    Tells kblockd to perform the equivalent of @blk_run_queue on behalf
312  *    of us. The caller must hold the queue lock.
313  */
314 void blk_run_queue_async(struct request_queue *q)
315 {
316         if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
317                 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
318 }
319 EXPORT_SYMBOL(blk_run_queue_async);
320 
321 /**
322  * blk_run_queue - run a single device queue
323  * @q: The queue to run
324  *
325  * Description:
326  *    Invoke request handling on this queue, if it has pending work to do.
327  *    May be used to restart queueing when a request has completed.
328  */
329 void blk_run_queue(struct request_queue *q)
330 {
331         unsigned long flags;
332 
333         spin_lock_irqsave(q->queue_lock, flags);
334         __blk_run_queue(q);
335         spin_unlock_irqrestore(q->queue_lock, flags);
336 }
337 EXPORT_SYMBOL(blk_run_queue);
338 
339 void blk_put_queue(struct request_queue *q)
340 {
341         kobject_put(&q->kobj);
342 }
343 EXPORT_SYMBOL(blk_put_queue);
344 
345 /**
346  * __blk_drain_queue - drain requests from request_queue
347  * @q: queue to drain
348  * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
349  *
350  * Drain requests from @q.  If @drain_all is set, all requests are drained.
351  * If not, only ELVPRIV requests are drained.  The caller is responsible
352  * for ensuring that no new requests which need to be drained are queued.
353  */
354 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
355         __releases(q->queue_lock)
356         __acquires(q->queue_lock)
357 {
358         int i;
359 
360         lockdep_assert_held(q->queue_lock);
361 
362         while (true) {
363                 bool drain = false;
364 
365                 /*
366                  * The caller might be trying to drain @q before its
367                  * elevator is initialized.
368                  */
369                 if (q->elevator)
370                         elv_drain_elevator(q);
371 
372                 blkcg_drain_queue(q);
373 
374                 /*
375                  * This function might be called on a queue which failed
376                  * driver init after queue creation or is not yet fully
377                  * active yet.  Some drivers (e.g. fd and loop) get unhappy
378                  * in such cases.  Kick queue iff dispatch queue has
379                  * something on it and @q has request_fn set.
380                  */
381                 if (!list_empty(&q->queue_head) && q->request_fn)
382                         __blk_run_queue(q);
383 
384                 drain |= q->nr_rqs_elvpriv;
385                 drain |= q->request_fn_active;
386 
387                 /*
388                  * Unfortunately, requests are queued at and tracked from
389                  * multiple places and there's no single counter which can
390                  * be drained.  Check all the queues and counters.
391                  */
392                 if (drain_all) {
393                         struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
394                         drain |= !list_empty(&q->queue_head);
395                         for (i = 0; i < 2; i++) {
396                                 drain |= q->nr_rqs[i];
397                                 drain |= q->in_flight[i];
398                                 if (fq)
399                                     drain |= !list_empty(&fq->flush_queue[i]);
400                         }
401                 }
402 
403                 if (!drain)
404                         break;
405 
406                 spin_unlock_irq(q->queue_lock);
407 
408                 msleep(10);
409 
410                 spin_lock_irq(q->queue_lock);
411         }
412 
413         /*
414          * With queue marked dead, any woken up waiter will fail the
415          * allocation path, so the wakeup chaining is lost and we're
416          * left with hung waiters. We need to wake up those waiters.
417          */
418         if (q->request_fn) {
419                 struct request_list *rl;
420 
421                 blk_queue_for_each_rl(rl, q)
422                         for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
423                                 wake_up_all(&rl->wait[i]);
424         }
425 }
426 
427 /**
428  * blk_queue_bypass_start - enter queue bypass mode
429  * @q: queue of interest
430  *
431  * In bypass mode, only the dispatch FIFO queue of @q is used.  This
432  * function makes @q enter bypass mode and drains all requests which were
433  * throttled or issued before.  On return, it's guaranteed that no request
434  * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
435  * inside queue or RCU read lock.
436  */
437 void blk_queue_bypass_start(struct request_queue *q)
438 {
439         spin_lock_irq(q->queue_lock);
440         q->bypass_depth++;
441         queue_flag_set(QUEUE_FLAG_BYPASS, q);
442         spin_unlock_irq(q->queue_lock);
443 
444         /*
445          * Queues start drained.  Skip actual draining till init is
446          * complete.  This avoids lenghty delays during queue init which
447          * can happen many times during boot.
448          */
449         if (blk_queue_init_done(q)) {
450                 spin_lock_irq(q->queue_lock);
451                 __blk_drain_queue(q, false);
452                 spin_unlock_irq(q->queue_lock);
453 
454                 /* ensure blk_queue_bypass() is %true inside RCU read lock */
455                 synchronize_rcu();
456         }
457 }
458 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
459 
460 /**
461  * blk_queue_bypass_end - leave queue bypass mode
462  * @q: queue of interest
463  *
464  * Leave bypass mode and restore the normal queueing behavior.
465  */
466 void blk_queue_bypass_end(struct request_queue *q)
467 {
468         spin_lock_irq(q->queue_lock);
469         if (!--q->bypass_depth)
470                 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
471         WARN_ON_ONCE(q->bypass_depth < 0);
472         spin_unlock_irq(q->queue_lock);
473 }
474 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
475 
476 void blk_set_queue_dying(struct request_queue *q)
477 {
478         spin_lock_irq(q->queue_lock);
479         queue_flag_set(QUEUE_FLAG_DYING, q);
480         spin_unlock_irq(q->queue_lock);
481 
482         if (q->mq_ops)
483                 blk_mq_wake_waiters(q);
484         else {
485                 struct request_list *rl;
486 
487                 blk_queue_for_each_rl(rl, q) {
488                         if (rl->rq_pool) {
489                                 wake_up(&rl->wait[BLK_RW_SYNC]);
490                                 wake_up(&rl->wait[BLK_RW_ASYNC]);
491                         }
492                 }
493         }
494 }
495 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
496 
497 /**
498  * blk_cleanup_queue - shutdown a request queue
499  * @q: request queue to shutdown
500  *
501  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
502  * put it.  All future requests will be failed immediately with -ENODEV.
503  */
504 void blk_cleanup_queue(struct request_queue *q)
505 {
506         spinlock_t *lock = q->queue_lock;
507 
508         /* mark @q DYING, no new request or merges will be allowed afterwards */
509         mutex_lock(&q->sysfs_lock);
510         blk_set_queue_dying(q);
511         spin_lock_irq(lock);
512 
513         /*
514          * A dying queue is permanently in bypass mode till released.  Note
515          * that, unlike blk_queue_bypass_start(), we aren't performing
516          * synchronize_rcu() after entering bypass mode to avoid the delay
517          * as some drivers create and destroy a lot of queues while
518          * probing.  This is still safe because blk_release_queue() will be
519          * called only after the queue refcnt drops to zero and nothing,
520          * RCU or not, would be traversing the queue by then.
521          */
522         q->bypass_depth++;
523         queue_flag_set(QUEUE_FLAG_BYPASS, q);
524 
525         queue_flag_set(QUEUE_FLAG_NOMERGES, q);
526         queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
527         queue_flag_set(QUEUE_FLAG_DYING, q);
528         spin_unlock_irq(lock);
529         mutex_unlock(&q->sysfs_lock);
530 
531         /*
532          * Drain all requests queued before DYING marking. Set DEAD flag to
533          * prevent that q->request_fn() gets invoked after draining finished.
534          */
535         if (q->mq_ops) {
536                 blk_mq_freeze_queue(q);
537                 spin_lock_irq(lock);
538         } else {
539                 spin_lock_irq(lock);
540                 __blk_drain_queue(q, true);
541         }
542         queue_flag_set(QUEUE_FLAG_DEAD, q);
543         spin_unlock_irq(lock);
544 
545         /* @q won't process any more request, flush async actions */
546         del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
547         blk_sync_queue(q);
548 
549         if (q->mq_ops)
550                 blk_mq_free_queue(q);
551 
552         spin_lock_irq(lock);
553         if (q->queue_lock != &q->__queue_lock)
554                 q->queue_lock = &q->__queue_lock;
555         spin_unlock_irq(lock);
556 
557         bdi_destroy(&q->backing_dev_info);
558 
559         /* @q is and will stay empty, shutdown and put */
560         blk_put_queue(q);
561 }
562 EXPORT_SYMBOL(blk_cleanup_queue);
563 
564 /* Allocate memory local to the request queue */
565 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
566 {
567         int nid = (int)(long)data;
568         return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
569 }
570 
571 static void free_request_struct(void *element, void *unused)
572 {
573         kmem_cache_free(request_cachep, element);
574 }
575 
576 int blk_init_rl(struct request_list *rl, struct request_queue *q,
577                 gfp_t gfp_mask)
578 {
579         if (unlikely(rl->rq_pool))
580                 return 0;
581 
582         rl->q = q;
583         rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
584         rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
585         init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
586         init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
587 
588         rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
589                                           free_request_struct,
590                                           (void *)(long)q->node, gfp_mask,
591                                           q->node);
592         if (!rl->rq_pool)
593                 return -ENOMEM;
594 
595         return 0;
596 }
597 
598 void blk_exit_rl(struct request_list *rl)
599 {
600         if (rl->rq_pool)
601                 mempool_destroy(rl->rq_pool);
602 }
603 
604 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
605 {
606         return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
607 }
608 EXPORT_SYMBOL(blk_alloc_queue);
609 
610 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
611 {
612         struct request_queue *q;
613         int err;
614 
615         q = kmem_cache_alloc_node(blk_requestq_cachep,
616                                 gfp_mask | __GFP_ZERO, node_id);
617         if (!q)
618                 return NULL;
619 
620         q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
621         if (q->id < 0)
622                 goto fail_q;
623 
624         q->backing_dev_info.ra_pages =
625                         (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
626         q->backing_dev_info.state = 0;
627         q->backing_dev_info.capabilities = 0;
628         q->backing_dev_info.name = "block";
629         q->node = node_id;
630 
631         err = bdi_init(&q->backing_dev_info);
632         if (err)
633                 goto fail_id;
634 
635         setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
636                     laptop_mode_timer_fn, (unsigned long) q);
637         setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
638         INIT_LIST_HEAD(&q->queue_head);
639         INIT_LIST_HEAD(&q->timeout_list);
640         INIT_LIST_HEAD(&q->icq_list);
641 #ifdef CONFIG_BLK_CGROUP
642         INIT_LIST_HEAD(&q->blkg_list);
643 #endif
644         INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
645 
646         kobject_init(&q->kobj, &blk_queue_ktype);
647 
648         mutex_init(&q->sysfs_lock);
649         spin_lock_init(&q->__queue_lock);
650 
651         /*
652          * By default initialize queue_lock to internal lock and driver can
653          * override it later if need be.
654          */
655         q->queue_lock = &q->__queue_lock;
656 
657         /*
658          * A queue starts its life with bypass turned on to avoid
659          * unnecessary bypass on/off overhead and nasty surprises during
660          * init.  The initial bypass will be finished when the queue is
661          * registered by blk_register_queue().
662          */
663         q->bypass_depth = 1;
664         __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
665 
666         init_waitqueue_head(&q->mq_freeze_wq);
667 
668         if (blkcg_init_queue(q))
669                 goto fail_bdi;
670 
671         return q;
672 
673 fail_bdi:
674         bdi_destroy(&q->backing_dev_info);
675 fail_id:
676         ida_simple_remove(&blk_queue_ida, q->id);
677 fail_q:
678         kmem_cache_free(blk_requestq_cachep, q);
679         return NULL;
680 }
681 EXPORT_SYMBOL(blk_alloc_queue_node);
682 
683 /**
684  * blk_init_queue  - prepare a request queue for use with a block device
685  * @rfn:  The function to be called to process requests that have been
686  *        placed on the queue.
687  * @lock: Request queue spin lock
688  *
689  * Description:
690  *    If a block device wishes to use the standard request handling procedures,
691  *    which sorts requests and coalesces adjacent requests, then it must
692  *    call blk_init_queue().  The function @rfn will be called when there
693  *    are requests on the queue that need to be processed.  If the device
694  *    supports plugging, then @rfn may not be called immediately when requests
695  *    are available on the queue, but may be called at some time later instead.
696  *    Plugged queues are generally unplugged when a buffer belonging to one
697  *    of the requests on the queue is needed, or due to memory pressure.
698  *
699  *    @rfn is not required, or even expected, to remove all requests off the
700  *    queue, but only as many as it can handle at a time.  If it does leave
701  *    requests on the queue, it is responsible for arranging that the requests
702  *    get dealt with eventually.
703  *
704  *    The queue spin lock must be held while manipulating the requests on the
705  *    request queue; this lock will be taken also from interrupt context, so irq
706  *    disabling is needed for it.
707  *
708  *    Function returns a pointer to the initialized request queue, or %NULL if
709  *    it didn't succeed.
710  *
711  * Note:
712  *    blk_init_queue() must be paired with a blk_cleanup_queue() call
713  *    when the block device is deactivated (such as at module unload).
714  **/
715 
716 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
717 {
718         return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
719 }
720 EXPORT_SYMBOL(blk_init_queue);
721 
722 struct request_queue *
723 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
724 {
725         struct request_queue *uninit_q, *q;
726 
727         uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
728         if (!uninit_q)
729                 return NULL;
730 
731         q = blk_init_allocated_queue(uninit_q, rfn, lock);
732         if (!q)
733                 blk_cleanup_queue(uninit_q);
734 
735         return q;
736 }
737 EXPORT_SYMBOL(blk_init_queue_node);
738 
739 static void blk_queue_bio(struct request_queue *q, struct bio *bio);
740 
741 struct request_queue *
742 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
743                          spinlock_t *lock)
744 {
745         if (!q)
746                 return NULL;
747 
748         q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
749         if (!q->fq)
750                 return NULL;
751 
752         if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
753                 goto fail;
754 
755         q->request_fn           = rfn;
756         q->prep_rq_fn           = NULL;
757         q->unprep_rq_fn         = NULL;
758         q->queue_flags          |= QUEUE_FLAG_DEFAULT;
759 
760         /* Override internal queue lock with supplied lock pointer */
761         if (lock)
762                 q->queue_lock           = lock;
763 
764         /*
765          * This also sets hw/phys segments, boundary and size
766          */
767         blk_queue_make_request(q, blk_queue_bio);
768 
769         q->sg_reserved_size = INT_MAX;
770 
771         /* Protect q->elevator from elevator_change */
772         mutex_lock(&q->sysfs_lock);
773 
774         /* init elevator */
775         if (elevator_init(q, NULL)) {
776                 mutex_unlock(&q->sysfs_lock);
777                 goto fail;
778         }
779 
780         mutex_unlock(&q->sysfs_lock);
781 
782         return q;
783 
784 fail:
785         blk_free_flush_queue(q->fq);
786         return NULL;
787 }
788 EXPORT_SYMBOL(blk_init_allocated_queue);
789 
790 bool blk_get_queue(struct request_queue *q)
791 {
792         if (likely(!blk_queue_dying(q))) {
793                 __blk_get_queue(q);
794                 return true;
795         }
796 
797         return false;
798 }
799 EXPORT_SYMBOL(blk_get_queue);
800 
801 static inline void blk_free_request(struct request_list *rl, struct request *rq)
802 {
803         if (rq->cmd_flags & REQ_ELVPRIV) {
804                 elv_put_request(rl->q, rq);
805                 if (rq->elv.icq)
806                         put_io_context(rq->elv.icq->ioc);
807         }
808 
809         mempool_free(rq, rl->rq_pool);
810 }
811 
812 /*
813  * ioc_batching returns true if the ioc is a valid batching request and
814  * should be given priority access to a request.
815  */
816 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
817 {
818         if (!ioc)
819                 return 0;
820 
821         /*
822          * Make sure the process is able to allocate at least 1 request
823          * even if the batch times out, otherwise we could theoretically
824          * lose wakeups.
825          */
826         return ioc->nr_batch_requests == q->nr_batching ||
827                 (ioc->nr_batch_requests > 0
828                 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
829 }
830 
831 /*
832  * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
833  * will cause the process to be a "batcher" on all queues in the system. This
834  * is the behaviour we want though - once it gets a wakeup it should be given
835  * a nice run.
836  */
837 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
838 {
839         if (!ioc || ioc_batching(q, ioc))
840                 return;
841 
842         ioc->nr_batch_requests = q->nr_batching;
843         ioc->last_waited = jiffies;
844 }
845 
846 static void __freed_request(struct request_list *rl, int sync)
847 {
848         struct request_queue *q = rl->q;
849 
850         /*
851          * bdi isn't aware of blkcg yet.  As all async IOs end up root
852          * blkcg anyway, just use root blkcg state.
853          */
854         if (rl == &q->root_rl &&
855             rl->count[sync] < queue_congestion_off_threshold(q))
856                 blk_clear_queue_congested(q, sync);
857 
858         if (rl->count[sync] + 1 <= q->nr_requests) {
859                 if (waitqueue_active(&rl->wait[sync]))
860                         wake_up(&rl->wait[sync]);
861 
862                 blk_clear_rl_full(rl, sync);
863         }
864 }
865 
866 /*
867  * A request has just been released.  Account for it, update the full and
868  * congestion status, wake up any waiters.   Called under q->queue_lock.
869  */
870 static void freed_request(struct request_list *rl, unsigned int flags)
871 {
872         struct request_queue *q = rl->q;
873         int sync = rw_is_sync(flags);
874 
875         q->nr_rqs[sync]--;
876         rl->count[sync]--;
877         if (flags & REQ_ELVPRIV)
878                 q->nr_rqs_elvpriv--;
879 
880         __freed_request(rl, sync);
881 
882         if (unlikely(rl->starved[sync ^ 1]))
883                 __freed_request(rl, sync ^ 1);
884 }
885 
886 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
887 {
888         struct request_list *rl;
889 
890         spin_lock_irq(q->queue_lock);
891         q->nr_requests = nr;
892         blk_queue_congestion_threshold(q);
893 
894         /* congestion isn't cgroup aware and follows root blkcg for now */
895         rl = &q->root_rl;
896 
897         if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
898                 blk_set_queue_congested(q, BLK_RW_SYNC);
899         else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
900                 blk_clear_queue_congested(q, BLK_RW_SYNC);
901 
902         if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
903                 blk_set_queue_congested(q, BLK_RW_ASYNC);
904         else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
905                 blk_clear_queue_congested(q, BLK_RW_ASYNC);
906 
907         blk_queue_for_each_rl(rl, q) {
908                 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
909                         blk_set_rl_full(rl, BLK_RW_SYNC);
910                 } else {
911                         blk_clear_rl_full(rl, BLK_RW_SYNC);
912                         wake_up(&rl->wait[BLK_RW_SYNC]);
913                 }
914 
915                 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
916                         blk_set_rl_full(rl, BLK_RW_ASYNC);
917                 } else {
918                         blk_clear_rl_full(rl, BLK_RW_ASYNC);
919                         wake_up(&rl->wait[BLK_RW_ASYNC]);
920                 }
921         }
922 
923         spin_unlock_irq(q->queue_lock);
924         return 0;
925 }
926 
927 /*
928  * Determine if elevator data should be initialized when allocating the
929  * request associated with @bio.
930  */
931 static bool blk_rq_should_init_elevator(struct bio *bio)
932 {
933         if (!bio)
934                 return true;
935 
936         /*
937          * Flush requests do not use the elevator so skip initialization.
938          * This allows a request to share the flush and elevator data.
939          */
940         if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
941                 return false;
942 
943         return true;
944 }
945 
946 /**
947  * rq_ioc - determine io_context for request allocation
948  * @bio: request being allocated is for this bio (can be %NULL)
949  *
950  * Determine io_context to use for request allocation for @bio.  May return
951  * %NULL if %current->io_context doesn't exist.
952  */
953 static struct io_context *rq_ioc(struct bio *bio)
954 {
955 #ifdef CONFIG_BLK_CGROUP
956         if (bio && bio->bi_ioc)
957                 return bio->bi_ioc;
958 #endif
959         return current->io_context;
960 }
961 
962 /**
963  * __get_request - get a free request
964  * @rl: request list to allocate from
965  * @rw_flags: RW and SYNC flags
966  * @bio: bio to allocate request for (can be %NULL)
967  * @gfp_mask: allocation mask
968  *
969  * Get a free request from @q.  This function may fail under memory
970  * pressure or if @q is dead.
971  *
972  * Must be called with @q->queue_lock held and,
973  * Returns ERR_PTR on failure, with @q->queue_lock held.
974  * Returns request pointer on success, with @q->queue_lock *not held*.
975  */
976 static struct request *__get_request(struct request_list *rl, int rw_flags,
977                                      struct bio *bio, gfp_t gfp_mask)
978 {
979         struct request_queue *q = rl->q;
980         struct request *rq;
981         struct elevator_type *et = q->elevator->type;
982         struct io_context *ioc = rq_ioc(bio);
983         struct io_cq *icq = NULL;
984         const bool is_sync = rw_is_sync(rw_flags) != 0;
985         int may_queue;
986 
987         if (unlikely(blk_queue_dying(q)))
988                 return ERR_PTR(-ENODEV);
989 
990         may_queue = elv_may_queue(q, rw_flags);
991         if (may_queue == ELV_MQUEUE_NO)
992                 goto rq_starved;
993 
994         if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
995                 if (rl->count[is_sync]+1 >= q->nr_requests) {
996                         /*
997                          * The queue will fill after this allocation, so set
998                          * it as full, and mark this process as "batching".
999                          * This process will be allowed to complete a batch of
1000                          * requests, others will be blocked.
1001                          */
1002                         if (!blk_rl_full(rl, is_sync)) {
1003                                 ioc_set_batching(q, ioc);
1004                                 blk_set_rl_full(rl, is_sync);
1005                         } else {
1006                                 if (may_queue != ELV_MQUEUE_MUST
1007                                                 && !ioc_batching(q, ioc)) {
1008                                         /*
1009                                          * The queue is full and the allocating
1010                                          * process is not a "batcher", and not
1011                                          * exempted by the IO scheduler
1012                                          */
1013                                         return ERR_PTR(-ENOMEM);
1014                                 }
1015                         }
1016                 }
1017                 /*
1018                  * bdi isn't aware of blkcg yet.  As all async IOs end up
1019                  * root blkcg anyway, just use root blkcg state.
1020                  */
1021                 if (rl == &q->root_rl)
1022                         blk_set_queue_congested(q, is_sync);
1023         }
1024 
1025         /*
1026          * Only allow batching queuers to allocate up to 50% over the defined
1027          * limit of requests, otherwise we could have thousands of requests
1028          * allocated with any setting of ->nr_requests
1029          */
1030         if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1031                 return ERR_PTR(-ENOMEM);
1032 
1033         q->nr_rqs[is_sync]++;
1034         rl->count[is_sync]++;
1035         rl->starved[is_sync] = 0;
1036 
1037         /*
1038          * Decide whether the new request will be managed by elevator.  If
1039          * so, mark @rw_flags and increment elvpriv.  Non-zero elvpriv will
1040          * prevent the current elevator from being destroyed until the new
1041          * request is freed.  This guarantees icq's won't be destroyed and
1042          * makes creating new ones safe.
1043          *
1044          * Also, lookup icq while holding queue_lock.  If it doesn't exist,
1045          * it will be created after releasing queue_lock.
1046          */
1047         if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1048                 rw_flags |= REQ_ELVPRIV;
1049                 q->nr_rqs_elvpriv++;
1050                 if (et->icq_cache && ioc)
1051                         icq = ioc_lookup_icq(ioc, q);
1052         }
1053 
1054         if (blk_queue_io_stat(q))
1055                 rw_flags |= REQ_IO_STAT;
1056         spin_unlock_irq(q->queue_lock);
1057 
1058         /* allocate and init request */
1059         rq = mempool_alloc(rl->rq_pool, gfp_mask);
1060         if (!rq)
1061                 goto fail_alloc;
1062 
1063         blk_rq_init(q, rq);
1064         blk_rq_set_rl(rq, rl);
1065         rq->cmd_flags = rw_flags | REQ_ALLOCED;
1066 
1067         /* init elvpriv */
1068         if (rw_flags & REQ_ELVPRIV) {
1069                 if (unlikely(et->icq_cache && !icq)) {
1070                         if (ioc)
1071                                 icq = ioc_create_icq(ioc, q, gfp_mask);
1072                         if (!icq)
1073                                 goto fail_elvpriv;
1074                 }
1075 
1076                 rq->elv.icq = icq;
1077                 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1078                         goto fail_elvpriv;
1079 
1080                 /* @rq->elv.icq holds io_context until @rq is freed */
1081                 if (icq)
1082                         get_io_context(icq->ioc);
1083         }
1084 out:
1085         /*
1086          * ioc may be NULL here, and ioc_batching will be false. That's
1087          * OK, if the queue is under the request limit then requests need
1088          * not count toward the nr_batch_requests limit. There will always
1089          * be some limit enforced by BLK_BATCH_TIME.
1090          */
1091         if (ioc_batching(q, ioc))
1092                 ioc->nr_batch_requests--;
1093 
1094         trace_block_getrq(q, bio, rw_flags & 1);
1095         return rq;
1096 
1097 fail_elvpriv:
1098         /*
1099          * elvpriv init failed.  ioc, icq and elvpriv aren't mempool backed
1100          * and may fail indefinitely under memory pressure and thus
1101          * shouldn't stall IO.  Treat this request as !elvpriv.  This will
1102          * disturb iosched and blkcg but weird is bettern than dead.
1103          */
1104         printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1105                            __func__, dev_name(q->backing_dev_info.dev));
1106 
1107         rq->cmd_flags &= ~REQ_ELVPRIV;
1108         rq->elv.icq = NULL;
1109 
1110         spin_lock_irq(q->queue_lock);
1111         q->nr_rqs_elvpriv--;
1112         spin_unlock_irq(q->queue_lock);
1113         goto out;
1114 
1115 fail_alloc:
1116         /*
1117          * Allocation failed presumably due to memory. Undo anything we
1118          * might have messed up.
1119          *
1120          * Allocating task should really be put onto the front of the wait
1121          * queue, but this is pretty rare.
1122          */
1123         spin_lock_irq(q->queue_lock);
1124         freed_request(rl, rw_flags);
1125 
1126         /*
1127          * in the very unlikely event that allocation failed and no
1128          * requests for this direction was pending, mark us starved so that
1129          * freeing of a request in the other direction will notice
1130          * us. another possible fix would be to split the rq mempool into
1131          * READ and WRITE
1132          */
1133 rq_starved:
1134         if (unlikely(rl->count[is_sync] == 0))
1135                 rl->starved[is_sync] = 1;
1136         return ERR_PTR(-ENOMEM);
1137 }
1138 
1139 /**
1140  * get_request - get a free request
1141  * @q: request_queue to allocate request from
1142  * @rw_flags: RW and SYNC flags
1143  * @bio: bio to allocate request for (can be %NULL)
1144  * @gfp_mask: allocation mask
1145  *
1146  * Get a free request from @q.  If %__GFP_WAIT is set in @gfp_mask, this
1147  * function keeps retrying under memory pressure and fails iff @q is dead.
1148  *
1149  * Must be called with @q->queue_lock held and,
1150  * Returns ERR_PTR on failure, with @q->queue_lock held.
1151  * Returns request pointer on success, with @q->queue_lock *not held*.
1152  */
1153 static struct request *get_request(struct request_queue *q, int rw_flags,
1154                                    struct bio *bio, gfp_t gfp_mask)
1155 {
1156         const bool is_sync = rw_is_sync(rw_flags) != 0;
1157         DEFINE_WAIT(wait);
1158         struct request_list *rl;
1159         struct request *rq;
1160 
1161         rl = blk_get_rl(q, bio);        /* transferred to @rq on success */
1162 retry:
1163         rq = __get_request(rl, rw_flags, bio, gfp_mask);
1164         if (!IS_ERR(rq))
1165                 return rq;
1166 
1167         if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1168                 blk_put_rl(rl);
1169                 return rq;
1170         }
1171 
1172         /* wait on @rl and retry */
1173         prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1174                                   TASK_UNINTERRUPTIBLE);
1175 
1176         trace_block_sleeprq(q, bio, rw_flags & 1);
1177 
1178         spin_unlock_irq(q->queue_lock);
1179         io_schedule();
1180 
1181         /*
1182          * After sleeping, we become a "batching" process and will be able
1183          * to allocate at least one request, and up to a big batch of them
1184          * for a small period time.  See ioc_batching, ioc_set_batching
1185          */
1186         ioc_set_batching(q, current->io_context);
1187 
1188         spin_lock_irq(q->queue_lock);
1189         finish_wait(&rl->wait[is_sync], &wait);
1190 
1191         goto retry;
1192 }
1193 
1194 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1195                 gfp_t gfp_mask)
1196 {
1197         struct request *rq;
1198 
1199         BUG_ON(rw != READ && rw != WRITE);
1200 
1201         /* create ioc upfront */
1202         create_io_context(gfp_mask, q->node);
1203 
1204         spin_lock_irq(q->queue_lock);
1205         rq = get_request(q, rw, NULL, gfp_mask);
1206         if (IS_ERR(rq))
1207                 spin_unlock_irq(q->queue_lock);
1208         /* q->queue_lock is unlocked at this point */
1209 
1210         return rq;
1211 }
1212 
1213 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1214 {
1215         if (q->mq_ops)
1216                 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1217         else
1218                 return blk_old_get_request(q, rw, gfp_mask);
1219 }
1220 EXPORT_SYMBOL(blk_get_request);
1221 
1222 /**
1223  * blk_make_request - given a bio, allocate a corresponding struct request.
1224  * @q: target request queue
1225  * @bio:  The bio describing the memory mappings that will be submitted for IO.
1226  *        It may be a chained-bio properly constructed by block/bio layer.
1227  * @gfp_mask: gfp flags to be used for memory allocation
1228  *
1229  * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1230  * type commands. Where the struct request needs to be farther initialized by
1231  * the caller. It is passed a &struct bio, which describes the memory info of
1232  * the I/O transfer.
1233  *
1234  * The caller of blk_make_request must make sure that bi_io_vec
1235  * are set to describe the memory buffers. That bio_data_dir() will return
1236  * the needed direction of the request. (And all bio's in the passed bio-chain
1237  * are properly set accordingly)
1238  *
1239  * If called under none-sleepable conditions, mapped bio buffers must not
1240  * need bouncing, by calling the appropriate masked or flagged allocator,
1241  * suitable for the target device. Otherwise the call to blk_queue_bounce will
1242  * BUG.
1243  *
1244  * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1245  * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1246  * anything but the first bio in the chain. Otherwise you risk waiting for IO
1247  * completion of a bio that hasn't been submitted yet, thus resulting in a
1248  * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1249  * of bio_alloc(), as that avoids the mempool deadlock.
1250  * If possible a big IO should be split into smaller parts when allocation
1251  * fails. Partial allocation should not be an error, or you risk a live-lock.
1252  */
1253 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1254                                  gfp_t gfp_mask)
1255 {
1256         struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1257 
1258         if (IS_ERR(rq))
1259                 return rq;
1260 
1261         blk_rq_set_block_pc(rq);
1262 
1263         for_each_bio(bio) {
1264                 struct bio *bounce_bio = bio;
1265                 int ret;
1266 
1267                 blk_queue_bounce(q, &bounce_bio);
1268                 ret = blk_rq_append_bio(q, rq, bounce_bio);
1269                 if (unlikely(ret)) {
1270                         blk_put_request(rq);
1271                         return ERR_PTR(ret);
1272                 }
1273         }
1274 
1275         return rq;
1276 }
1277 EXPORT_SYMBOL(blk_make_request);
1278 
1279 /**
1280  * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1281  * @rq:         request to be initialized
1282  *
1283  */
1284 void blk_rq_set_block_pc(struct request *rq)
1285 {
1286         rq->cmd_type = REQ_TYPE_BLOCK_PC;
1287         rq->__data_len = 0;
1288         rq->__sector = (sector_t) -1;
1289         rq->bio = rq->biotail = NULL;
1290         memset(rq->__cmd, 0, sizeof(rq->__cmd));
1291 }
1292 EXPORT_SYMBOL(blk_rq_set_block_pc);
1293 
1294 /**
1295  * blk_requeue_request - put a request back on queue
1296  * @q:          request queue where request should be inserted
1297  * @rq:         request to be inserted
1298  *
1299  * Description:
1300  *    Drivers often keep queueing requests until the hardware cannot accept
1301  *    more, when that condition happens we need to put the request back
1302  *    on the queue. Must be called with queue lock held.
1303  */
1304 void blk_requeue_request(struct request_queue *q, struct request *rq)
1305 {
1306         blk_delete_timer(rq);
1307         blk_clear_rq_complete(rq);
1308         trace_block_rq_requeue(q, rq);
1309 
1310         if (rq->cmd_flags & REQ_QUEUED)
1311                 blk_queue_end_tag(q, rq);
1312 
1313         BUG_ON(blk_queued_rq(rq));
1314 
1315         elv_requeue_request(q, rq);
1316 }
1317 EXPORT_SYMBOL(blk_requeue_request);
1318 
1319 static void add_acct_request(struct request_queue *q, struct request *rq,
1320                              int where)
1321 {
1322         blk_account_io_start(rq, true);
1323         __elv_add_request(q, rq, where);
1324 }
1325 
1326 static void part_round_stats_single(int cpu, struct hd_struct *part,
1327                                     unsigned long now)
1328 {
1329         int inflight;
1330 
1331         if (now == part->stamp)
1332                 return;
1333 
1334         inflight = part_in_flight(part);
1335         if (inflight) {
1336                 __part_stat_add(cpu, part, time_in_queue,
1337                                 inflight * (now - part->stamp));
1338                 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1339         }
1340         part->stamp = now;
1341 }
1342 
1343 /**
1344  * part_round_stats() - Round off the performance stats on a struct disk_stats.
1345  * @cpu: cpu number for stats access
1346  * @part: target partition
1347  *
1348  * The average IO queue length and utilisation statistics are maintained
1349  * by observing the current state of the queue length and the amount of
1350  * time it has been in this state for.
1351  *
1352  * Normally, that accounting is done on IO completion, but that can result
1353  * in more than a second's worth of IO being accounted for within any one
1354  * second, leading to >100% utilisation.  To deal with that, we call this
1355  * function to do a round-off before returning the results when reading
1356  * /proc/diskstats.  This accounts immediately for all queue usage up to
1357  * the current jiffies and restarts the counters again.
1358  */
1359 void part_round_stats(int cpu, struct hd_struct *part)
1360 {
1361         unsigned long now = jiffies;
1362 
1363         if (part->partno)
1364                 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1365         part_round_stats_single(cpu, part, now);
1366 }
1367 EXPORT_SYMBOL_GPL(part_round_stats);
1368 
1369 #ifdef CONFIG_PM
1370 static void blk_pm_put_request(struct request *rq)
1371 {
1372         if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1373                 pm_runtime_mark_last_busy(rq->q->dev);
1374 }
1375 #else
1376 static inline void blk_pm_put_request(struct request *rq) {}
1377 #endif
1378 
1379 /*
1380  * queue lock must be held
1381  */
1382 void __blk_put_request(struct request_queue *q, struct request *req)
1383 {
1384         if (unlikely(!q))
1385                 return;
1386 
1387         if (q->mq_ops) {
1388                 blk_mq_free_request(req);
1389                 return;
1390         }
1391 
1392         blk_pm_put_request(req);
1393 
1394         elv_completed_request(q, req);
1395 
1396         /* this is a bio leak */
1397         WARN_ON(req->bio != NULL);
1398 
1399         /*
1400          * Request may not have originated from ll_rw_blk. if not,
1401          * it didn't come out of our reserved rq pools
1402          */
1403         if (req->cmd_flags & REQ_ALLOCED) {
1404                 unsigned int flags = req->cmd_flags;
1405                 struct request_list *rl = blk_rq_rl(req);
1406 
1407                 BUG_ON(!list_empty(&req->queuelist));
1408                 BUG_ON(ELV_ON_HASH(req));
1409 
1410                 blk_free_request(rl, req);
1411                 freed_request(rl, flags);
1412                 blk_put_rl(rl);
1413         }
1414 }
1415 EXPORT_SYMBOL_GPL(__blk_put_request);
1416 
1417 void blk_put_request(struct request *req)
1418 {
1419         struct request_queue *q = req->q;
1420 
1421         if (q->mq_ops)
1422                 blk_mq_free_request(req);
1423         else {
1424                 unsigned long flags;
1425 
1426                 spin_lock_irqsave(q->queue_lock, flags);
1427                 __blk_put_request(q, req);
1428                 spin_unlock_irqrestore(q->queue_lock, flags);
1429         }
1430 }
1431 EXPORT_SYMBOL(blk_put_request);
1432 
1433 /**
1434  * blk_add_request_payload - add a payload to a request
1435  * @rq: request to update
1436  * @page: page backing the payload
1437  * @len: length of the payload.
1438  *
1439  * This allows to later add a payload to an already submitted request by
1440  * a block driver.  The driver needs to take care of freeing the payload
1441  * itself.
1442  *
1443  * Note that this is a quite horrible hack and nothing but handling of
1444  * discard requests should ever use it.
1445  */
1446 void blk_add_request_payload(struct request *rq, struct page *page,
1447                 unsigned int len)
1448 {
1449         struct bio *bio = rq->bio;
1450 
1451         bio->bi_io_vec->bv_page = page;
1452         bio->bi_io_vec->bv_offset = 0;
1453         bio->bi_io_vec->bv_len = len;
1454 
1455         bio->bi_iter.bi_size = len;
1456         bio->bi_vcnt = 1;
1457         bio->bi_phys_segments = 1;
1458 
1459         rq->__data_len = rq->resid_len = len;
1460         rq->nr_phys_segments = 1;
1461 }
1462 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1463 
1464 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1465                             struct bio *bio)
1466 {
1467         const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1468 
1469         if (!ll_back_merge_fn(q, req, bio))
1470                 return false;
1471 
1472         trace_block_bio_backmerge(q, req, bio);
1473 
1474         if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1475                 blk_rq_set_mixed_merge(req);
1476 
1477         req->biotail->bi_next = bio;
1478         req->biotail = bio;
1479         req->__data_len += bio->bi_iter.bi_size;
1480         req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1481 
1482         blk_account_io_start(req, false);
1483         return true;
1484 }
1485 
1486 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1487                              struct bio *bio)
1488 {
1489         const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1490 
1491         if (!ll_front_merge_fn(q, req, bio))
1492                 return false;
1493 
1494         trace_block_bio_frontmerge(q, req, bio);
1495 
1496         if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1497                 blk_rq_set_mixed_merge(req);
1498 
1499         bio->bi_next = req->bio;
1500         req->bio = bio;
1501 
1502         req->__sector = bio->bi_iter.bi_sector;
1503         req->__data_len += bio->bi_iter.bi_size;
1504         req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1505 
1506         blk_account_io_start(req, false);
1507         return true;
1508 }
1509 
1510 /**
1511  * blk_attempt_plug_merge - try to merge with %current's plugged list
1512  * @q: request_queue new bio is being queued at
1513  * @bio: new bio being queued
1514  * @request_count: out parameter for number of traversed plugged requests
1515  *
1516  * Determine whether @bio being queued on @q can be merged with a request
1517  * on %current's plugged list.  Returns %true if merge was successful,
1518  * otherwise %false.
1519  *
1520  * Plugging coalesces IOs from the same issuer for the same purpose without
1521  * going through @q->queue_lock.  As such it's more of an issuing mechanism
1522  * than scheduling, and the request, while may have elvpriv data, is not
1523  * added on the elevator at this point.  In addition, we don't have
1524  * reliable access to the elevator outside queue lock.  Only check basic
1525  * merging parameters without querying the elevator.
1526  *
1527  * Caller must ensure !blk_queue_nomerges(q) beforehand.
1528  */
1529 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1530                             unsigned int *request_count)
1531 {
1532         struct blk_plug *plug;
1533         struct request *rq;
1534         bool ret = false;
1535         struct list_head *plug_list;
1536 
1537         plug = current->plug;
1538         if (!plug)
1539                 goto out;
1540         *request_count = 0;
1541 
1542         if (q->mq_ops)
1543                 plug_list = &plug->mq_list;
1544         else
1545                 plug_list = &plug->list;
1546 
1547         list_for_each_entry_reverse(rq, plug_list, queuelist) {
1548                 int el_ret;
1549 
1550                 if (rq->q == q)
1551                         (*request_count)++;
1552 
1553                 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1554                         continue;
1555 
1556                 el_ret = blk_try_merge(rq, bio);
1557                 if (el_ret == ELEVATOR_BACK_MERGE) {
1558                         ret = bio_attempt_back_merge(q, rq, bio);
1559                         if (ret)
1560                                 break;
1561                 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1562                         ret = bio_attempt_front_merge(q, rq, bio);
1563                         if (ret)
1564                                 break;
1565                 }
1566         }
1567 out:
1568         return ret;
1569 }
1570 
1571 void init_request_from_bio(struct request *req, struct bio *bio)
1572 {
1573         req->cmd_type = REQ_TYPE_FS;
1574 
1575         req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1576         if (bio->bi_rw & REQ_RAHEAD)
1577                 req->cmd_flags |= REQ_FAILFAST_MASK;
1578 
1579         req->errors = 0;
1580         req->__sector = bio->bi_iter.bi_sector;
1581         req->ioprio = bio_prio(bio);
1582         blk_rq_bio_prep(req->q, req, bio);
1583 }
1584 
1585 static void blk_queue_bio(struct request_queue *q, struct bio *bio)
1586 {
1587         const bool sync = !!(bio->bi_rw & REQ_SYNC);
1588         struct blk_plug *plug;
1589         int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1590         struct request *req;
1591         unsigned int request_count = 0;
1592 
1593         /*
1594          * low level driver can indicate that it wants pages above a
1595          * certain limit bounced to low memory (ie for highmem, or even
1596          * ISA dma in theory)
1597          */
1598         blk_queue_bounce(q, &bio);
1599 
1600         if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1601                 bio_endio(bio, -EIO);
1602                 return;
1603         }
1604 
1605         if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1606                 spin_lock_irq(q->queue_lock);
1607                 where = ELEVATOR_INSERT_FLUSH;
1608                 goto get_rq;
1609         }
1610 
1611         /*
1612          * Check if we can merge with the plugged list before grabbing
1613          * any locks.
1614          */
1615         if (!blk_queue_nomerges(q) &&
1616             blk_attempt_plug_merge(q, bio, &request_count))
1617                 return;
1618 
1619         spin_lock_irq(q->queue_lock);
1620 
1621         el_ret = elv_merge(q, &req, bio);
1622         if (el_ret == ELEVATOR_BACK_MERGE) {
1623                 if (bio_attempt_back_merge(q, req, bio)) {
1624                         elv_bio_merged(q, req, bio);
1625                         if (!attempt_back_merge(q, req))
1626                                 elv_merged_request(q, req, el_ret);
1627                         goto out_unlock;
1628                 }
1629         } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1630                 if (bio_attempt_front_merge(q, req, bio)) {
1631                         elv_bio_merged(q, req, bio);
1632                         if (!attempt_front_merge(q, req))
1633                                 elv_merged_request(q, req, el_ret);
1634                         goto out_unlock;
1635                 }
1636         }
1637 
1638 get_rq:
1639         /*
1640          * This sync check and mask will be re-done in init_request_from_bio(),
1641          * but we need to set it earlier to expose the sync flag to the
1642          * rq allocator and io schedulers.
1643          */
1644         rw_flags = bio_data_dir(bio);
1645         if (sync)
1646                 rw_flags |= REQ_SYNC;
1647 
1648         /*
1649          * Grab a free request. This is might sleep but can not fail.
1650          * Returns with the queue unlocked.
1651          */
1652         req = get_request(q, rw_flags, bio, GFP_NOIO);
1653         if (IS_ERR(req)) {
1654                 bio_endio(bio, PTR_ERR(req));   /* @q is dead */
1655                 goto out_unlock;
1656         }
1657 
1658         /*
1659          * After dropping the lock and possibly sleeping here, our request
1660          * may now be mergeable after it had proven unmergeable (above).
1661          * We don't worry about that case for efficiency. It won't happen
1662          * often, and the elevators are able to handle it.
1663          */
1664         init_request_from_bio(req, bio);
1665 
1666         if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1667                 req->cpu = raw_smp_processor_id();
1668 
1669         plug = current->plug;
1670         if (plug) {
1671                 /*
1672                  * If this is the first request added after a plug, fire
1673                  * of a plug trace.
1674                  */
1675                 if (!request_count)
1676                         trace_block_plug(q);
1677                 else {
1678                         if (request_count >= BLK_MAX_REQUEST_COUNT) {
1679                                 blk_flush_plug_list(plug, false);
1680                                 trace_block_plug(q);
1681                         }
1682                 }
1683                 list_add_tail(&req->queuelist, &plug->list);
1684                 blk_account_io_start(req, true);
1685         } else {
1686                 spin_lock_irq(q->queue_lock);
1687                 add_acct_request(q, req, where);
1688                 __blk_run_queue(q);
1689 out_unlock:
1690                 spin_unlock_irq(q->queue_lock);
1691         }
1692 }
1693 
1694 /*
1695  * If bio->bi_dev is a partition, remap the location
1696  */
1697 static inline void blk_partition_remap(struct bio *bio)
1698 {
1699         struct block_device *bdev = bio->bi_bdev;
1700 
1701         if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1702                 struct hd_struct *p = bdev->bd_part;
1703 
1704                 bio->bi_iter.bi_sector += p->start_sect;
1705                 bio->bi_bdev = bdev->bd_contains;
1706 
1707                 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1708                                       bdev->bd_dev,
1709                                       bio->bi_iter.bi_sector - p->start_sect);
1710         }
1711 }
1712 
1713 static void handle_bad_sector(struct bio *bio)
1714 {
1715         char b[BDEVNAME_SIZE];
1716 
1717         printk(KERN_INFO "attempt to access beyond end of device\n");
1718         printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1719                         bdevname(bio->bi_bdev, b),
1720                         bio->bi_rw,
1721                         (unsigned long long)bio_end_sector(bio),
1722                         (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1723 
1724         set_bit(BIO_EOF, &bio->bi_flags);
1725 }
1726 
1727 #ifdef CONFIG_FAIL_MAKE_REQUEST
1728 
1729 static DECLARE_FAULT_ATTR(fail_make_request);
1730 
1731 static int __init setup_fail_make_request(char *str)
1732 {
1733         return setup_fault_attr(&fail_make_request, str);
1734 }
1735 __setup("fail_make_request=", setup_fail_make_request);
1736 
1737 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1738 {
1739         return part->make_it_fail && should_fail(&fail_make_request, bytes);
1740 }
1741 
1742 static int __init fail_make_request_debugfs(void)
1743 {
1744         struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1745                                                 NULL, &fail_make_request);
1746 
1747         return PTR_ERR_OR_ZERO(dir);
1748 }
1749 
1750 late_initcall(fail_make_request_debugfs);
1751 
1752 #else /* CONFIG_FAIL_MAKE_REQUEST */
1753 
1754 static inline bool should_fail_request(struct hd_struct *part,
1755                                         unsigned int bytes)
1756 {
1757         return false;
1758 }
1759 
1760 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1761 
1762 /*
1763  * Check whether this bio extends beyond the end of the device.
1764  */
1765 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1766 {
1767         sector_t maxsector;
1768 
1769         if (!nr_sectors)
1770                 return 0;
1771 
1772         /* Test device or partition size, when known. */
1773         maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1774         if (maxsector) {
1775                 sector_t sector = bio->bi_iter.bi_sector;
1776 
1777                 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1778                         /*
1779                          * This may well happen - the kernel calls bread()
1780                          * without checking the size of the device, e.g., when
1781                          * mounting a device.
1782                          */
1783                         handle_bad_sector(bio);
1784                         return 1;
1785                 }
1786         }
1787 
1788         return 0;
1789 }
1790 
1791 static noinline_for_stack bool
1792 generic_make_request_checks(struct bio *bio)
1793 {
1794         struct request_queue *q;
1795         int nr_sectors = bio_sectors(bio);
1796         int err = -EIO;
1797         char b[BDEVNAME_SIZE];
1798         struct hd_struct *part;
1799 
1800         might_sleep();
1801 
1802         if (bio_check_eod(bio, nr_sectors))
1803                 goto end_io;
1804 
1805         q = bdev_get_queue(bio->bi_bdev);
1806         if (unlikely(!q)) {
1807                 printk(KERN_ERR
1808                        "generic_make_request: Trying to access "
1809                         "nonexistent block-device %s (%Lu)\n",
1810                         bdevname(bio->bi_bdev, b),
1811                         (long long) bio->bi_iter.bi_sector);
1812                 goto end_io;
1813         }
1814 
1815         if (likely(bio_is_rw(bio) &&
1816                    nr_sectors > queue_max_hw_sectors(q))) {
1817                 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1818                        bdevname(bio->bi_bdev, b),
1819                        bio_sectors(bio),
1820                        queue_max_hw_sectors(q));
1821                 goto end_io;
1822         }
1823 
1824         part = bio->bi_bdev->bd_part;
1825         if (should_fail_request(part, bio->bi_iter.bi_size) ||
1826             should_fail_request(&part_to_disk(part)->part0,
1827                                 bio->bi_iter.bi_size))
1828                 goto end_io;
1829 
1830         /*
1831          * If this device has partitions, remap block n
1832          * of partition p to block n+start(p) of the disk.
1833          */
1834         blk_partition_remap(bio);
1835 
1836         if (bio_check_eod(bio, nr_sectors))
1837                 goto end_io;
1838 
1839         /*
1840          * Filter flush bio's early so that make_request based
1841          * drivers without flush support don't have to worry
1842          * about them.
1843          */
1844         if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1845                 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1846                 if (!nr_sectors) {
1847                         err = 0;
1848                         goto end_io;
1849                 }
1850         }
1851 
1852         if ((bio->bi_rw & REQ_DISCARD) &&
1853             (!blk_queue_discard(q) ||
1854              ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1855                 err = -EOPNOTSUPP;
1856                 goto end_io;
1857         }
1858 
1859         if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1860                 err = -EOPNOTSUPP;
1861                 goto end_io;
1862         }
1863 
1864         /*
1865          * Various block parts want %current->io_context and lazy ioc
1866          * allocation ends up trading a lot of pain for a small amount of
1867          * memory.  Just allocate it upfront.  This may fail and block
1868          * layer knows how to live with it.
1869          */
1870         create_io_context(GFP_ATOMIC, q->node);
1871 
1872         if (blk_throtl_bio(q, bio))
1873                 return false;   /* throttled, will be resubmitted later */
1874 
1875         trace_block_bio_queue(q, bio);
1876         return true;
1877 
1878 end_io:
1879         bio_endio(bio, err);
1880         return false;
1881 }
1882 
1883 /**
1884  * generic_make_request - hand a buffer to its device driver for I/O
1885  * @bio:  The bio describing the location in memory and on the device.
1886  *
1887  * generic_make_request() is used to make I/O requests of block
1888  * devices. It is passed a &struct bio, which describes the I/O that needs
1889  * to be done.
1890  *
1891  * generic_make_request() does not return any status.  The
1892  * success/failure status of the request, along with notification of
1893  * completion, is delivered asynchronously through the bio->bi_end_io
1894  * function described (one day) else where.
1895  *
1896  * The caller of generic_make_request must make sure that bi_io_vec
1897  * are set to describe the memory buffer, and that bi_dev and bi_sector are
1898  * set to describe the device address, and the
1899  * bi_end_io and optionally bi_private are set to describe how
1900  * completion notification should be signaled.
1901  *
1902  * generic_make_request and the drivers it calls may use bi_next if this
1903  * bio happens to be merged with someone else, and may resubmit the bio to
1904  * a lower device by calling into generic_make_request recursively, which
1905  * means the bio should NOT be touched after the call to ->make_request_fn.
1906  */
1907 void generic_make_request(struct bio *bio)
1908 {
1909         struct bio_list bio_list_on_stack;
1910 
1911         if (!generic_make_request_checks(bio))
1912                 return;
1913 
1914         /*
1915          * We only want one ->make_request_fn to be active at a time, else
1916          * stack usage with stacked devices could be a problem.  So use
1917          * current->bio_list to keep a list of requests submited by a
1918          * make_request_fn function.  current->bio_list is also used as a
1919          * flag to say if generic_make_request is currently active in this
1920          * task or not.  If it is NULL, then no make_request is active.  If
1921          * it is non-NULL, then a make_request is active, and new requests
1922          * should be added at the tail
1923          */
1924         if (current->bio_list) {
1925                 bio_list_add(current->bio_list, bio);
1926                 return;
1927         }
1928 
1929         /* following loop may be a bit non-obvious, and so deserves some
1930          * explanation.
1931          * Before entering the loop, bio->bi_next is NULL (as all callers
1932          * ensure that) so we have a list with a single bio.
1933          * We pretend that we have just taken it off a longer list, so
1934          * we assign bio_list to a pointer to the bio_list_on_stack,
1935          * thus initialising the bio_list of new bios to be
1936          * added.  ->make_request() may indeed add some more bios
1937          * through a recursive call to generic_make_request.  If it
1938          * did, we find a non-NULL value in bio_list and re-enter the loop
1939          * from the top.  In this case we really did just take the bio
1940          * of the top of the list (no pretending) and so remove it from
1941          * bio_list, and call into ->make_request() again.
1942          */
1943         BUG_ON(bio->bi_next);
1944         bio_list_init(&bio_list_on_stack);
1945         current->bio_list = &bio_list_on_stack;
1946         do {
1947                 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1948 
1949                 q->make_request_fn(q, bio);
1950 
1951                 bio = bio_list_pop(current->bio_list);
1952         } while (bio);
1953         current->bio_list = NULL; /* deactivate */
1954 }
1955 EXPORT_SYMBOL(generic_make_request);
1956 
1957 /**
1958  * submit_bio - submit a bio to the block device layer for I/O
1959  * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1960  * @bio: The &struct bio which describes the I/O
1961  *
1962  * submit_bio() is very similar in purpose to generic_make_request(), and
1963  * uses that function to do most of the work. Both are fairly rough
1964  * interfaces; @bio must be presetup and ready for I/O.
1965  *
1966  */
1967 void submit_bio(int rw, struct bio *bio)
1968 {
1969         bio->bi_rw |= rw;
1970 
1971         /*
1972          * If it's a regular read/write or a barrier with data attached,
1973          * go through the normal accounting stuff before submission.
1974          */
1975         if (bio_has_data(bio)) {
1976                 unsigned int count;
1977 
1978                 if (unlikely(rw & REQ_WRITE_SAME))
1979                         count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1980                 else
1981                         count = bio_sectors(bio);
1982 
1983                 if (rw & WRITE) {
1984                         count_vm_events(PGPGOUT, count);
1985                 } else {
1986                         task_io_account_read(bio->bi_iter.bi_size);
1987                         count_vm_events(PGPGIN, count);
1988                 }
1989 
1990                 if (unlikely(block_dump)) {
1991                         char b[BDEVNAME_SIZE];
1992                         printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1993                         current->comm, task_pid_nr(current),
1994                                 (rw & WRITE) ? "WRITE" : "READ",
1995                                 (unsigned long long)bio->bi_iter.bi_sector,
1996                                 bdevname(bio->bi_bdev, b),
1997                                 count);
1998                 }
1999         }
2000 
2001         generic_make_request(bio);
2002 }
2003 EXPORT_SYMBOL(submit_bio);
2004 
2005 /**
2006  * blk_rq_check_limits - Helper function to check a request for the queue limit
2007  * @q:  the queue
2008  * @rq: the request being checked
2009  *
2010  * Description:
2011  *    @rq may have been made based on weaker limitations of upper-level queues
2012  *    in request stacking drivers, and it may violate the limitation of @q.
2013  *    Since the block layer and the underlying device driver trust @rq
2014  *    after it is inserted to @q, it should be checked against @q before
2015  *    the insertion using this generic function.
2016  *
2017  *    This function should also be useful for request stacking drivers
2018  *    in some cases below, so export this function.
2019  *    Request stacking drivers like request-based dm may change the queue
2020  *    limits while requests are in the queue (e.g. dm's table swapping).
2021  *    Such request stacking drivers should check those requests against
2022  *    the new queue limits again when they dispatch those requests,
2023  *    although such checkings are also done against the old queue limits
2024  *    when submitting requests.
2025  */
2026 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
2027 {
2028         if (!rq_mergeable(rq))
2029                 return 0;
2030 
2031         if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2032                 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2033                 return -EIO;
2034         }
2035 
2036         /*
2037          * queue's settings related to segment counting like q->bounce_pfn
2038          * may differ from that of other stacking queues.
2039          * Recalculate it to check the request correctly on this queue's
2040          * limitation.
2041          */
2042         blk_recalc_rq_segments(rq);
2043         if (rq->nr_phys_segments > queue_max_segments(q)) {
2044                 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2045                 return -EIO;
2046         }
2047 
2048         return 0;
2049 }
2050 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2051 
2052 /**
2053  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2054  * @q:  the queue to submit the request
2055  * @rq: the request being queued
2056  */
2057 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2058 {
2059         unsigned long flags;
2060         int where = ELEVATOR_INSERT_BACK;
2061 
2062         if (blk_rq_check_limits(q, rq))
2063                 return -EIO;
2064 
2065         if (rq->rq_disk &&
2066             should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2067                 return -EIO;
2068 
2069         if (q->mq_ops) {
2070                 if (blk_queue_io_stat(q))
2071                         blk_account_io_start(rq, true);
2072                 blk_mq_insert_request(rq, false, true, false);
2073                 return 0;
2074         }
2075 
2076         spin_lock_irqsave(q->queue_lock, flags);
2077         if (unlikely(blk_queue_dying(q))) {
2078                 spin_unlock_irqrestore(q->queue_lock, flags);
2079                 return -ENODEV;
2080         }
2081 
2082         /*
2083          * Submitting request must be dequeued before calling this function
2084          * because it will be linked to another request_queue
2085          */
2086         BUG_ON(blk_queued_rq(rq));
2087 
2088         if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2089                 where = ELEVATOR_INSERT_FLUSH;
2090 
2091         add_acct_request(q, rq, where);
2092         if (where == ELEVATOR_INSERT_FLUSH)
2093                 __blk_run_queue(q);
2094         spin_unlock_irqrestore(q->queue_lock, flags);
2095 
2096         return 0;
2097 }
2098 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2099 
2100 /**
2101  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2102  * @rq: request to examine
2103  *
2104  * Description:
2105  *     A request could be merge of IOs which require different failure
2106  *     handling.  This function determines the number of bytes which
2107  *     can be failed from the beginning of the request without
2108  *     crossing into area which need to be retried further.
2109  *
2110  * Return:
2111  *     The number of bytes to fail.
2112  *
2113  * Context:
2114  *     queue_lock must be held.
2115  */
2116 unsigned int blk_rq_err_bytes(const struct request *rq)
2117 {
2118         unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2119         unsigned int bytes = 0;
2120         struct bio *bio;
2121 
2122         if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2123                 return blk_rq_bytes(rq);
2124 
2125         /*
2126          * Currently the only 'mixing' which can happen is between
2127          * different fastfail types.  We can safely fail portions
2128          * which have all the failfast bits that the first one has -
2129          * the ones which are at least as eager to fail as the first
2130          * one.
2131          */
2132         for (bio = rq->bio; bio; bio = bio->bi_next) {
2133                 if ((bio->bi_rw & ff) != ff)
2134                         break;
2135                 bytes += bio->bi_iter.bi_size;
2136         }
2137 
2138         /* this could lead to infinite loop */
2139         BUG_ON(blk_rq_bytes(rq) && !bytes);
2140         return bytes;
2141 }
2142 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2143 
2144 void blk_account_io_completion(struct request *req, unsigned int bytes)
2145 {
2146         if (blk_do_io_stat(req)) {
2147                 const int rw = rq_data_dir(req);
2148                 struct hd_struct *part;
2149                 int cpu;
2150 
2151                 cpu = part_stat_lock();
2152                 part = req->part;
2153                 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2154                 part_stat_unlock();
2155         }
2156 }
2157 
2158 void blk_account_io_done(struct request *req)
2159 {
2160         /*
2161          * Account IO completion.  flush_rq isn't accounted as a
2162          * normal IO on queueing nor completion.  Accounting the
2163          * containing request is enough.
2164          */
2165         if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2166                 unsigned long duration = jiffies - req->start_time;
2167                 const int rw = rq_data_dir(req);
2168                 struct hd_struct *part;
2169                 int cpu;
2170 
2171                 cpu = part_stat_lock();
2172                 part = req->part;
2173 
2174                 part_stat_inc(cpu, part, ios[rw]);
2175                 part_stat_add(cpu, part, ticks[rw], duration);
2176                 part_round_stats(cpu, part);
2177                 part_dec_in_flight(part, rw);
2178 
2179                 hd_struct_put(part);
2180                 part_stat_unlock();
2181         }
2182 }
2183 
2184 #ifdef CONFIG_PM
2185 /*
2186  * Don't process normal requests when queue is suspended
2187  * or in the process of suspending/resuming
2188  */
2189 static struct request *blk_pm_peek_request(struct request_queue *q,
2190                                            struct request *rq)
2191 {
2192         if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2193             (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2194                 return NULL;
2195         else
2196                 return rq;
2197 }
2198 #else
2199 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2200                                                   struct request *rq)
2201 {
2202         return rq;
2203 }
2204 #endif
2205 
2206 void blk_account_io_start(struct request *rq, bool new_io)
2207 {
2208         struct hd_struct *part;
2209         int rw = rq_data_dir(rq);
2210         int cpu;
2211 
2212         if (!blk_do_io_stat(rq))
2213                 return;
2214 
2215         cpu = part_stat_lock();
2216 
2217         if (!new_io) {
2218                 part = rq->part;
2219                 part_stat_inc(cpu, part, merges[rw]);
2220         } else {
2221                 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2222                 if (!hd_struct_try_get(part)) {
2223                         /*
2224                          * The partition is already being removed,
2225                          * the request will be accounted on the disk only
2226                          *
2227                          * We take a reference on disk->part0 although that
2228                          * partition will never be deleted, so we can treat
2229                          * it as any other partition.
2230                          */
2231                         part = &rq->rq_disk->part0;
2232                         hd_struct_get(part);
2233                 }
2234                 part_round_stats(cpu, part);
2235                 part_inc_in_flight(part, rw);
2236                 rq->part = part;
2237         }
2238 
2239         part_stat_unlock();
2240 }
2241 
2242 /**
2243  * blk_peek_request - peek at the top of a request queue
2244  * @q: request queue to peek at
2245  *
2246  * Description:
2247  *     Return the request at the top of @q.  The returned request
2248  *     should be started using blk_start_request() before LLD starts
2249  *     processing it.
2250  *
2251  * Return:
2252  *     Pointer to the request at the top of @q if available.  Null
2253  *     otherwise.
2254  *
2255  * Context:
2256  *     queue_lock must be held.
2257  */
2258 struct request *blk_peek_request(struct request_queue *q)
2259 {
2260         struct request *rq;
2261         int ret;
2262 
2263         while ((rq = __elv_next_request(q)) != NULL) {
2264 
2265                 rq = blk_pm_peek_request(q, rq);
2266                 if (!rq)
2267                         break;
2268 
2269                 if (!(rq->cmd_flags & REQ_STARTED)) {
2270                         /*
2271                          * This is the first time the device driver
2272                          * sees this request (possibly after
2273                          * requeueing).  Notify IO scheduler.
2274                          */
2275                         if (rq->cmd_flags & REQ_SORTED)
2276                                 elv_activate_rq(q, rq);
2277 
2278                         /*
2279                          * just mark as started even if we don't start
2280                          * it, a request that has been delayed should
2281                          * not be passed by new incoming requests
2282                          */
2283                         rq->cmd_flags |= REQ_STARTED;
2284                         trace_block_rq_issue(q, rq);
2285                 }
2286 
2287                 if (!q->boundary_rq || q->boundary_rq == rq) {
2288                         q->end_sector = rq_end_sector(rq);
2289                         q->boundary_rq = NULL;
2290                 }
2291 
2292                 if (rq->cmd_flags & REQ_DONTPREP)
2293                         break;
2294 
2295                 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2296                         /*
2297                          * make sure space for the drain appears we
2298                          * know we can do this because max_hw_segments
2299                          * has been adjusted to be one fewer than the
2300                          * device can handle
2301                          */
2302                         rq->nr_phys_segments++;
2303                 }
2304 
2305                 if (!q->prep_rq_fn)
2306                         break;
2307 
2308                 ret = q->prep_rq_fn(q, rq);
2309                 if (ret == BLKPREP_OK) {
2310                         break;
2311                 } else if (ret == BLKPREP_DEFER) {
2312                         /*
2313                          * the request may have been (partially) prepped.
2314                          * we need to keep this request in the front to
2315                          * avoid resource deadlock.  REQ_STARTED will
2316                          * prevent other fs requests from passing this one.
2317                          */
2318                         if (q->dma_drain_size && blk_rq_bytes(rq) &&
2319                             !(rq->cmd_flags & REQ_DONTPREP)) {
2320                                 /*
2321                                  * remove the space for the drain we added
2322                                  * so that we don't add it again
2323                                  */
2324                                 --rq->nr_phys_segments;
2325                         }
2326 
2327                         rq = NULL;
2328                         break;
2329                 } else if (ret == BLKPREP_KILL) {
2330                         rq->cmd_flags |= REQ_QUIET;
2331                         /*
2332                          * Mark this request as started so we don't trigger
2333                          * any debug logic in the end I/O path.
2334                          */
2335                         blk_start_request(rq);
2336                         __blk_end_request_all(rq, -EIO);
2337                 } else {
2338                         printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2339                         break;
2340                 }
2341         }
2342 
2343         return rq;
2344 }
2345 EXPORT_SYMBOL(blk_peek_request);
2346 
2347 void blk_dequeue_request(struct request *rq)
2348 {
2349         struct request_queue *q = rq->q;
2350 
2351         BUG_ON(list_empty(&rq->queuelist));
2352         BUG_ON(ELV_ON_HASH(rq));
2353 
2354         list_del_init(&rq->queuelist);
2355 
2356         /*
2357          * the time frame between a request being removed from the lists
2358          * and to it is freed is accounted as io that is in progress at
2359          * the driver side.
2360          */
2361         if (blk_account_rq(rq)) {
2362                 q->in_flight[rq_is_sync(rq)]++;
2363                 set_io_start_time_ns(rq);
2364         }
2365 }
2366 
2367 /**
2368  * blk_start_request - start request processing on the driver
2369  * @req: request to dequeue
2370  *
2371  * Description:
2372  *     Dequeue @req and start timeout timer on it.  This hands off the
2373  *     request to the driver.
2374  *
2375  *     Block internal functions which don't want to start timer should
2376  *     call blk_dequeue_request().
2377  *
2378  * Context:
2379  *     queue_lock must be held.
2380  */
2381 void blk_start_request(struct request *req)
2382 {
2383         blk_dequeue_request(req);
2384 
2385         /*
2386          * We are now handing the request to the hardware, initialize
2387          * resid_len to full count and add the timeout handler.
2388          */
2389         req->resid_len = blk_rq_bytes(req);
2390         if (unlikely(blk_bidi_rq(req)))
2391                 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2392 
2393         BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2394         blk_add_timer(req);
2395 }
2396 EXPORT_SYMBOL(blk_start_request);
2397 
2398 /**
2399  * blk_fetch_request - fetch a request from a request queue
2400  * @q: request queue to fetch a request from
2401  *
2402  * Description:
2403  *     Return the request at the top of @q.  The request is started on
2404  *     return and LLD can start processing it immediately.
2405  *
2406  * Return:
2407  *     Pointer to the request at the top of @q if available.  Null
2408  *     otherwise.
2409  *
2410  * Context:
2411  *     queue_lock must be held.
2412  */
2413 struct request *blk_fetch_request(struct request_queue *q)
2414 {
2415         struct request *rq;
2416 
2417         rq = blk_peek_request(q);
2418         if (rq)
2419                 blk_start_request(rq);
2420         return rq;
2421 }
2422 EXPORT_SYMBOL(blk_fetch_request);
2423 
2424 /**
2425  * blk_update_request - Special helper function for request stacking drivers
2426  * @req:      the request being processed
2427  * @error:    %0 for success, < %0 for error
2428  * @nr_bytes: number of bytes to complete @req
2429  *
2430  * Description:
2431  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
2432  *     the request structure even if @req doesn't have leftover.
2433  *     If @req has leftover, sets it up for the next range of segments.
2434  *
2435  *     This special helper function is only for request stacking drivers
2436  *     (e.g. request-based dm) so that they can handle partial completion.
2437  *     Actual device drivers should use blk_end_request instead.
2438  *
2439  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2440  *     %false return from this function.
2441  *
2442  * Return:
2443  *     %false - this request doesn't have any more data
2444  *     %true  - this request has more data
2445  **/
2446 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2447 {
2448         int total_bytes;
2449 
2450         trace_block_rq_complete(req->q, req, nr_bytes);
2451 
2452         if (!req->bio)
2453                 return false;
2454 
2455         /*
2456          * For fs requests, rq is just carrier of independent bio's
2457          * and each partial completion should be handled separately.
2458          * Reset per-request error on each partial completion.
2459          *
2460          * TODO: tj: This is too subtle.  It would be better to let
2461          * low level drivers do what they see fit.
2462          */
2463         if (req->cmd_type == REQ_TYPE_FS)
2464                 req->errors = 0;
2465 
2466         if (error && req->cmd_type == REQ_TYPE_FS &&
2467             !(req->cmd_flags & REQ_QUIET)) {
2468                 char *error_type;
2469 
2470                 switch (error) {
2471                 case -ENOLINK:
2472                         error_type = "recoverable transport";
2473                         break;
2474                 case -EREMOTEIO:
2475                         error_type = "critical target";
2476                         break;
2477                 case -EBADE:
2478                         error_type = "critical nexus";
2479                         break;
2480                 case -ETIMEDOUT:
2481                         error_type = "timeout";
2482                         break;
2483                 case -ENOSPC:
2484                         error_type = "critical space allocation";
2485                         break;
2486                 case -ENODATA:
2487                         error_type = "critical medium";
2488                         break;
2489                 case -EIO:
2490                 default:
2491                         error_type = "I/O";
2492                         break;
2493                 }
2494                 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2495                                    __func__, error_type, req->rq_disk ?
2496                                    req->rq_disk->disk_name : "?",
2497                                    (unsigned long long)blk_rq_pos(req));
2498 
2499         }
2500 
2501         blk_account_io_completion(req, nr_bytes);
2502 
2503         total_bytes = 0;
2504         while (req->bio) {
2505                 struct bio *bio = req->bio;
2506                 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2507 
2508                 if (bio_bytes == bio->bi_iter.bi_size)
2509                         req->bio = bio->bi_next;
2510 
2511                 req_bio_endio(req, bio, bio_bytes, error);
2512 
2513                 total_bytes += bio_bytes;
2514                 nr_bytes -= bio_bytes;
2515 
2516                 if (!nr_bytes)
2517                         break;
2518         }
2519 
2520         /*
2521          * completely done
2522          */
2523         if (!req->bio) {
2524                 /*
2525                  * Reset counters so that the request stacking driver
2526                  * can find how many bytes remain in the request
2527                  * later.
2528                  */
2529                 req->__data_len = 0;
2530                 return false;
2531         }
2532 
2533         req->__data_len -= total_bytes;
2534 
2535         /* update sector only for requests with clear definition of sector */
2536         if (req->cmd_type == REQ_TYPE_FS)
2537                 req->__sector += total_bytes >> 9;
2538 
2539         /* mixed attributes always follow the first bio */
2540         if (req->cmd_flags & REQ_MIXED_MERGE) {
2541                 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2542                 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2543         }
2544 
2545         /*
2546          * If total number of sectors is less than the first segment
2547          * size, something has gone terribly wrong.
2548          */
2549         if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2550                 blk_dump_rq_flags(req, "request botched");
2551                 req->__data_len = blk_rq_cur_bytes(req);
2552         }
2553 
2554         /* recalculate the number of segments */
2555         blk_recalc_rq_segments(req);
2556 
2557         return true;
2558 }
2559 EXPORT_SYMBOL_GPL(blk_update_request);
2560 
2561 static bool blk_update_bidi_request(struct request *rq, int error,
2562                                     unsigned int nr_bytes,
2563                                     unsigned int bidi_bytes)
2564 {
2565         if (blk_update_request(rq, error, nr_bytes))
2566                 return true;
2567 
2568         /* Bidi request must be completed as a whole */
2569         if (unlikely(blk_bidi_rq(rq)) &&
2570             blk_update_request(rq->next_rq, error, bidi_bytes))
2571                 return true;
2572 
2573         if (blk_queue_add_random(rq->q))
2574                 add_disk_randomness(rq->rq_disk);
2575 
2576         return false;
2577 }
2578 
2579 /**
2580  * blk_unprep_request - unprepare a request
2581  * @req:        the request
2582  *
2583  * This function makes a request ready for complete resubmission (or
2584  * completion).  It happens only after all error handling is complete,
2585  * so represents the appropriate moment to deallocate any resources
2586  * that were allocated to the request in the prep_rq_fn.  The queue
2587  * lock is held when calling this.
2588  */
2589 void blk_unprep_request(struct request *req)
2590 {
2591         struct request_queue *q = req->q;
2592 
2593         req->cmd_flags &= ~REQ_DONTPREP;
2594         if (q->unprep_rq_fn)
2595                 q->unprep_rq_fn(q, req);
2596 }
2597 EXPORT_SYMBOL_GPL(blk_unprep_request);
2598 
2599 /*
2600  * queue lock must be held
2601  */
2602 void blk_finish_request(struct request *req, int error)
2603 {
2604         if (req->cmd_flags & REQ_QUEUED)
2605                 blk_queue_end_tag(req->q, req);
2606 
2607         BUG_ON(blk_queued_rq(req));
2608 
2609         if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2610                 laptop_io_completion(&req->q->backing_dev_info);
2611 
2612         blk_delete_timer(req);
2613 
2614         if (req->cmd_flags & REQ_DONTPREP)
2615                 blk_unprep_request(req);
2616 
2617         blk_account_io_done(req);
2618 
2619         if (req->end_io)
2620                 req->end_io(req, error);
2621         else {
2622                 if (blk_bidi_rq(req))
2623                         __blk_put_request(req->next_rq->q, req->next_rq);
2624 
2625                 __blk_put_request(req->q, req);
2626         }
2627 }
2628 EXPORT_SYMBOL(blk_finish_request);
2629 
2630 /**
2631  * blk_end_bidi_request - Complete a bidi request
2632  * @rq:         the request to complete
2633  * @error:      %0 for success, < %0 for error
2634  * @nr_bytes:   number of bytes to complete @rq
2635  * @bidi_bytes: number of bytes to complete @rq->next_rq
2636  *
2637  * Description:
2638  *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2639  *     Drivers that supports bidi can safely call this member for any
2640  *     type of request, bidi or uni.  In the later case @bidi_bytes is
2641  *     just ignored.
2642  *
2643  * Return:
2644  *     %false - we are done with this request
2645  *     %true  - still buffers pending for this request
2646  **/
2647 static bool blk_end_bidi_request(struct request *rq, int error,
2648                                  unsigned int nr_bytes, unsigned int bidi_bytes)
2649 {
2650         struct request_queue *q = rq->q;
2651         unsigned long flags;
2652 
2653         if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2654                 return true;
2655 
2656         spin_lock_irqsave(q->queue_lock, flags);
2657         blk_finish_request(rq, error);
2658         spin_unlock_irqrestore(q->queue_lock, flags);
2659 
2660         return false;
2661 }
2662 
2663 /**
2664  * __blk_end_bidi_request - Complete a bidi request with queue lock held
2665  * @rq:         the request to complete
2666  * @error:      %0 for success, < %0 for error
2667  * @nr_bytes:   number of bytes to complete @rq
2668  * @bidi_bytes: number of bytes to complete @rq->next_rq
2669  *
2670  * Description:
2671  *     Identical to blk_end_bidi_request() except that queue lock is
2672  *     assumed to be locked on entry and remains so on return.
2673  *
2674  * Return:
2675  *     %false - we are done with this request
2676  *     %true  - still buffers pending for this request
2677  **/
2678 bool __blk_end_bidi_request(struct request *rq, int error,
2679                                    unsigned int nr_bytes, unsigned int bidi_bytes)
2680 {
2681         if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2682                 return true;
2683 
2684         blk_finish_request(rq, error);
2685 
2686         return false;
2687 }
2688 
2689 /**
2690  * blk_end_request - Helper function for drivers to complete the request.
2691  * @rq:       the request being processed
2692  * @error:    %0 for success, < %0 for error
2693  * @nr_bytes: number of bytes to complete
2694  *
2695  * Description:
2696  *     Ends I/O on a number of bytes attached to @rq.
2697  *     If @rq has leftover, sets it up for the next range of segments.
2698  *
2699  * Return:
2700  *     %false - we are done with this request
2701  *     %true  - still buffers pending for this request
2702  **/
2703 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2704 {
2705         return blk_end_bidi_request(rq, error, nr_bytes, 0);
2706 }
2707 EXPORT_SYMBOL(blk_end_request);
2708 
2709 /**
2710  * blk_end_request_all - Helper function for drives to finish the request.
2711  * @rq: the request to finish
2712  * @error: %0 for success, < %0 for error
2713  *
2714  * Description:
2715  *     Completely finish @rq.
2716  */
2717 void blk_end_request_all(struct request *rq, int error)
2718 {
2719         bool pending;
2720         unsigned int bidi_bytes = 0;
2721 
2722         if (unlikely(blk_bidi_rq(rq)))
2723                 bidi_bytes = blk_rq_bytes(rq->next_rq);
2724 
2725         pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2726         BUG_ON(pending);
2727 }
2728 EXPORT_SYMBOL(blk_end_request_all);
2729 
2730 /**
2731  * blk_end_request_cur - Helper function to finish the current request chunk.
2732  * @rq: the request to finish the current chunk for
2733  * @error: %0 for success, < %0 for error
2734  *
2735  * Description:
2736  *     Complete the current consecutively mapped chunk from @rq.
2737  *
2738  * Return:
2739  *     %false - we are done with this request
2740  *     %true  - still buffers pending for this request
2741  */
2742 bool blk_end_request_cur(struct request *rq, int error)
2743 {
2744         return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2745 }
2746 EXPORT_SYMBOL(blk_end_request_cur);
2747 
2748 /**
2749  * blk_end_request_err - Finish a request till the next failure boundary.
2750  * @rq: the request to finish till the next failure boundary for
2751  * @error: must be negative errno
2752  *
2753  * Description:
2754  *     Complete @rq till the next failure boundary.
2755  *
2756  * Return:
2757  *     %false - we are done with this request
2758  *     %true  - still buffers pending for this request
2759  */
2760 bool blk_end_request_err(struct request *rq, int error)
2761 {
2762         WARN_ON(error >= 0);
2763         return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2764 }
2765 EXPORT_SYMBOL_GPL(blk_end_request_err);
2766 
2767 /**
2768  * __blk_end_request - Helper function for drivers to complete the request.
2769  * @rq:       the request being processed
2770  * @error:    %0 for success, < %0 for error
2771  * @nr_bytes: number of bytes to complete
2772  *
2773  * Description:
2774  *     Must be called with queue lock held unlike blk_end_request().
2775  *
2776  * Return:
2777  *     %false - we are done with this request
2778  *     %true  - still buffers pending for this request
2779  **/
2780 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2781 {
2782         return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2783 }
2784 EXPORT_SYMBOL(__blk_end_request);
2785 
2786 /**
2787  * __blk_end_request_all - Helper function for drives to finish the request.
2788  * @rq: the request to finish
2789  * @error: %0 for success, < %0 for error
2790  *
2791  * Description:
2792  *     Completely finish @rq.  Must be called with queue lock held.
2793  */
2794 void __blk_end_request_all(struct request *rq, int error)
2795 {
2796         bool pending;
2797         unsigned int bidi_bytes = 0;
2798 
2799         if (unlikely(blk_bidi_rq(rq)))
2800                 bidi_bytes = blk_rq_bytes(rq->next_rq);
2801 
2802         pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2803         BUG_ON(pending);
2804 }
2805 EXPORT_SYMBOL(__blk_end_request_all);
2806 
2807 /**
2808  * __blk_end_request_cur - Helper function to finish the current request chunk.
2809  * @rq: the request to finish the current chunk for
2810  * @error: %0 for success, < %0 for error
2811  *
2812  * Description:
2813  *     Complete the current consecutively mapped chunk from @rq.  Must
2814  *     be called with queue lock held.
2815  *
2816  * Return:
2817  *     %false - we are done with this request
2818  *     %true  - still buffers pending for this request
2819  */
2820 bool __blk_end_request_cur(struct request *rq, int error)
2821 {
2822         return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2823 }
2824 EXPORT_SYMBOL(__blk_end_request_cur);
2825 
2826 /**
2827  * __blk_end_request_err - Finish a request till the next failure boundary.
2828  * @rq: the request to finish till the next failure boundary for
2829  * @error: must be negative errno
2830  *
2831  * Description:
2832  *     Complete @rq till the next failure boundary.  Must be called
2833  *     with queue lock held.
2834  *
2835  * Return:
2836  *     %false - we are done with this request
2837  *     %true  - still buffers pending for this request
2838  */
2839 bool __blk_end_request_err(struct request *rq, int error)
2840 {
2841         WARN_ON(error >= 0);
2842         return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2843 }
2844 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2845 
2846 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2847                      struct bio *bio)
2848 {
2849         /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2850         rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2851 
2852         if (bio_has_data(bio))
2853                 rq->nr_phys_segments = bio_phys_segments(q, bio);
2854 
2855         rq->__data_len = bio->bi_iter.bi_size;
2856         rq->bio = rq->biotail = bio;
2857 
2858         if (bio->bi_bdev)
2859                 rq->rq_disk = bio->bi_bdev->bd_disk;
2860 }
2861 
2862 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2863 /**
2864  * rq_flush_dcache_pages - Helper function to flush all pages in a request
2865  * @rq: the request to be flushed
2866  *
2867  * Description:
2868  *     Flush all pages in @rq.
2869  */
2870 void rq_flush_dcache_pages(struct request *rq)
2871 {
2872         struct req_iterator iter;
2873         struct bio_vec bvec;
2874 
2875         rq_for_each_segment(bvec, rq, iter)
2876                 flush_dcache_page(bvec.bv_page);
2877 }
2878 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2879 #endif
2880 
2881 /**
2882  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2883  * @q : the queue of the device being checked
2884  *
2885  * Description:
2886  *    Check if underlying low-level drivers of a device are busy.
2887  *    If the drivers want to export their busy state, they must set own
2888  *    exporting function using blk_queue_lld_busy() first.
2889  *
2890  *    Basically, this function is used only by request stacking drivers
2891  *    to stop dispatching requests to underlying devices when underlying
2892  *    devices are busy.  This behavior helps more I/O merging on the queue
2893  *    of the request stacking driver and prevents I/O throughput regression
2894  *    on burst I/O load.
2895  *
2896  * Return:
2897  *    0 - Not busy (The request stacking driver should dispatch request)
2898  *    1 - Busy (The request stacking driver should stop dispatching request)
2899  */
2900 int blk_lld_busy(struct request_queue *q)
2901 {
2902         if (q->lld_busy_fn)
2903                 return q->lld_busy_fn(q);
2904 
2905         return 0;
2906 }
2907 EXPORT_SYMBOL_GPL(blk_lld_busy);
2908 
2909 /**
2910  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2911  * @rq: the clone request to be cleaned up
2912  *
2913  * Description:
2914  *     Free all bios in @rq for a cloned request.
2915  */
2916 void blk_rq_unprep_clone(struct request *rq)
2917 {
2918         struct bio *bio;
2919 
2920         while ((bio = rq->bio) != NULL) {
2921                 rq->bio = bio->bi_next;
2922 
2923                 bio_put(bio);
2924         }
2925 }
2926 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2927 
2928 /*
2929  * Copy attributes of the original request to the clone request.
2930  * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2931  */
2932 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2933 {
2934         dst->cpu = src->cpu;
2935         dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2936         dst->cmd_type = src->cmd_type;
2937         dst->__sector = blk_rq_pos(src);
2938         dst->__data_len = blk_rq_bytes(src);
2939         dst->nr_phys_segments = src->nr_phys_segments;
2940         dst->ioprio = src->ioprio;
2941         dst->extra_len = src->extra_len;
2942 }
2943 
2944 /**
2945  * blk_rq_prep_clone - Helper function to setup clone request
2946  * @rq: the request to be setup
2947  * @rq_src: original request to be cloned
2948  * @bs: bio_set that bios for clone are allocated from
2949  * @gfp_mask: memory allocation mask for bio
2950  * @bio_ctr: setup function to be called for each clone bio.
2951  *           Returns %0 for success, non %0 for failure.
2952  * @data: private data to be passed to @bio_ctr
2953  *
2954  * Description:
2955  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2956  *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
2957  *     are not copied, and copying such parts is the caller's responsibility.
2958  *     Also, pages which the original bios are pointing to are not copied
2959  *     and the cloned bios just point same pages.
2960  *     So cloned bios must be completed before original bios, which means
2961  *     the caller must complete @rq before @rq_src.
2962  */
2963 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2964                       struct bio_set *bs, gfp_t gfp_mask,
2965                       int (*bio_ctr)(struct bio *, struct bio *, void *),
2966                       void *data)
2967 {
2968         struct bio *bio, *bio_src;
2969 
2970         if (!bs)
2971                 bs = fs_bio_set;
2972 
2973         __rq_for_each_bio(bio_src, rq_src) {
2974                 bio = bio_clone_fast(bio_src, gfp_mask, bs);
2975                 if (!bio)
2976                         goto free_and_out;
2977 
2978                 if (bio_ctr && bio_ctr(bio, bio_src, data))
2979                         goto free_and_out;
2980 
2981                 if (rq->bio) {
2982                         rq->biotail->bi_next = bio;
2983                         rq->biotail = bio;
2984                 } else
2985                         rq->bio = rq->biotail = bio;
2986         }
2987 
2988         __blk_rq_prep_clone(rq, rq_src);
2989 
2990         return 0;
2991 
2992 free_and_out:
2993         if (bio)
2994                 bio_put(bio);
2995         blk_rq_unprep_clone(rq);
2996 
2997         return -ENOMEM;
2998 }
2999 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3000 
3001 int kblockd_schedule_work(struct work_struct *work)
3002 {
3003         return queue_work(kblockd_workqueue, work);
3004 }
3005 EXPORT_SYMBOL(kblockd_schedule_work);
3006 
3007 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3008                                   unsigned long delay)
3009 {
3010         return queue_delayed_work(kblockd_workqueue, dwork, delay);
3011 }
3012 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3013 
3014 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3015                                      unsigned long delay)
3016 {
3017         return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3018 }
3019 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3020 
3021 /**
3022  * blk_start_plug - initialize blk_plug and track it inside the task_struct
3023  * @plug:       The &struct blk_plug that needs to be initialized
3024  *
3025  * Description:
3026  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
3027  *   pending I/O should the task end up blocking between blk_start_plug() and
3028  *   blk_finish_plug(). This is important from a performance perspective, but
3029  *   also ensures that we don't deadlock. For instance, if the task is blocking
3030  *   for a memory allocation, memory reclaim could end up wanting to free a
3031  *   page belonging to that request that is currently residing in our private
3032  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
3033  *   this kind of deadlock.
3034  */
3035 void blk_start_plug(struct blk_plug *plug)
3036 {
3037         struct task_struct *tsk = current;
3038 
3039         INIT_LIST_HEAD(&plug->list);
3040         INIT_LIST_HEAD(&plug->mq_list);
3041         INIT_LIST_HEAD(&plug->cb_list);
3042 
3043         /*
3044          * If this is a nested plug, don't actually assign it. It will be
3045          * flushed on its own.
3046          */
3047         if (!tsk->plug) {
3048                 /*
3049                  * Store ordering should not be needed here, since a potential
3050                  * preempt will imply a full memory barrier
3051                  */
3052                 tsk->plug = plug;
3053         }
3054 }
3055 EXPORT_SYMBOL(blk_start_plug);
3056 
3057 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3058 {
3059         struct request *rqa = container_of(a, struct request, queuelist);
3060         struct request *rqb = container_of(b, struct request, queuelist);
3061 
3062         return !(rqa->q < rqb->q ||
3063                 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3064 }
3065 
3066 /*
3067  * If 'from_schedule' is true, then postpone the dispatch of requests
3068  * until a safe kblockd context. We due this to avoid accidental big
3069  * additional stack usage in driver dispatch, in places where the originally
3070  * plugger did not intend it.
3071  */
3072 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3073                             bool from_schedule)
3074         __releases(q->queue_lock)
3075 {
3076         trace_block_unplug(q, depth, !from_schedule);
3077 
3078         if (from_schedule)
3079                 blk_run_queue_async(q);
3080         else
3081                 __blk_run_queue(q);
3082         spin_unlock(q->queue_lock);
3083 }
3084 
3085 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3086 {
3087         LIST_HEAD(callbacks);
3088 
3089         while (!list_empty(&plug->cb_list)) {
3090                 list_splice_init(&plug->cb_list, &callbacks);
3091 
3092                 while (!list_empty(&callbacks)) {
3093                         struct blk_plug_cb *cb = list_first_entry(&callbacks,
3094                                                           struct blk_plug_cb,
3095                                                           list);
3096                         list_del(&cb->list);
3097                         cb->callback(cb, from_schedule);
3098                 }
3099         }
3100 }
3101 
3102 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3103                                       int size)
3104 {
3105         struct blk_plug *plug = current->plug;
3106         struct blk_plug_cb *cb;
3107 
3108         if (!plug)
3109                 return NULL;
3110 
3111         list_for_each_entry(cb, &plug->cb_list, list)
3112                 if (cb->callback == unplug && cb->data == data)
3113                         return cb;
3114 
3115         /* Not currently on the callback list */
3116         BUG_ON(size < sizeof(*cb));
3117         cb = kzalloc(size, GFP_ATOMIC);
3118         if (cb) {
3119                 cb->data = data;
3120                 cb->callback = unplug;
3121                 list_add(&cb->list, &plug->cb_list);
3122         }
3123         return cb;
3124 }
3125 EXPORT_SYMBOL(blk_check_plugged);
3126 
3127 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3128 {
3129         struct request_queue *q;
3130         unsigned long flags;
3131         struct request *rq;
3132         LIST_HEAD(list);
3133         unsigned int depth;
3134 
3135         flush_plug_callbacks(plug, from_schedule);
3136 
3137         if (!list_empty(&plug->mq_list))
3138                 blk_mq_flush_plug_list(plug, from_schedule);
3139 
3140         if (list_empty(&plug->list))
3141                 return;
3142 
3143         list_splice_init(&plug->list, &list);
3144 
3145         list_sort(NULL, &list, plug_rq_cmp);
3146 
3147         q = NULL;
3148         depth = 0;
3149 
3150         /*
3151          * Save and disable interrupts here, to avoid doing it for every
3152          * queue lock we have to take.
3153          */
3154         local_irq_save(flags);
3155         while (!list_empty(&list)) {
3156                 rq = list_entry_rq(list.next);
3157                 list_del_init(&rq->queuelist);
3158                 BUG_ON(!rq->q);
3159                 if (rq->q != q) {
3160                         /*
3161                          * This drops the queue lock
3162                          */
3163                         if (q)
3164                                 queue_unplugged(q, depth, from_schedule);
3165                         q = rq->q;
3166                         depth = 0;
3167                         spin_lock(q->queue_lock);
3168                 }
3169 
3170                 /*
3171                  * Short-circuit if @q is dead
3172                  */
3173                 if (unlikely(blk_queue_dying(q))) {
3174                         __blk_end_request_all(rq, -ENODEV);
3175                         continue;
3176                 }
3177 
3178                 /*
3179                  * rq is already accounted, so use raw insert
3180                  */
3181                 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3182                         __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3183                 else
3184                         __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3185 
3186                 depth++;
3187         }
3188 
3189         /*
3190          * This drops the queue lock
3191          */
3192         if (q)
3193                 queue_unplugged(q, depth, from_schedule);
3194 
3195         local_irq_restore(flags);
3196 }
3197 
3198 void blk_finish_plug(struct blk_plug *plug)
3199 {
3200         blk_flush_plug_list(plug, false);
3201 
3202         if (plug == current->plug)
3203                 current->plug = NULL;
3204 }
3205 EXPORT_SYMBOL(blk_finish_plug);
3206 
3207 #ifdef CONFIG_PM
3208 /**
3209  * blk_pm_runtime_init - Block layer runtime PM initialization routine
3210  * @q: the queue of the device
3211  * @dev: the device the queue belongs to
3212  *
3213  * Description:
3214  *    Initialize runtime-PM-related fields for @q and start auto suspend for
3215  *    @dev. Drivers that want to take advantage of request-based runtime PM
3216  *    should call this function after @dev has been initialized, and its
3217  *    request queue @q has been allocated, and runtime PM for it can not happen
3218  *    yet(either due to disabled/forbidden or its usage_count > 0). In most
3219  *    cases, driver should call this function before any I/O has taken place.
3220  *
3221  *    This function takes care of setting up using auto suspend for the device,
3222  *    the autosuspend delay is set to -1 to make runtime suspend impossible
3223  *    until an updated value is either set by user or by driver. Drivers do
3224  *    not need to touch other autosuspend settings.
3225  *
3226  *    The block layer runtime PM is request based, so only works for drivers
3227  *    that use request as their IO unit instead of those directly use bio's.
3228  */
3229 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3230 {
3231         q->dev = dev;
3232         q->rpm_status = RPM_ACTIVE;
3233         pm_runtime_set_autosuspend_delay(q->dev, -1);
3234         pm_runtime_use_autosuspend(q->dev);
3235 }
3236 EXPORT_SYMBOL(blk_pm_runtime_init);
3237 
3238 /**
3239  * blk_pre_runtime_suspend - Pre runtime suspend check
3240  * @q: the queue of the device
3241  *
3242  * Description:
3243  *    This function will check if runtime suspend is allowed for the device
3244  *    by examining if there are any requests pending in the queue. If there
3245  *    are requests pending, the device can not be runtime suspended; otherwise,
3246  *    the queue's status will be updated to SUSPENDING and the driver can
3247  *    proceed to suspend the device.
3248  *
3249  *    For the not allowed case, we mark last busy for the device so that
3250  *    runtime PM core will try to autosuspend it some time later.
3251  *
3252  *    This function should be called near the start of the device's
3253  *    runtime_suspend callback.
3254  *
3255  * Return:
3256  *    0         - OK to runtime suspend the device
3257  *    -EBUSY    - Device should not be runtime suspended
3258  */
3259 int blk_pre_runtime_suspend(struct request_queue *q)
3260 {
3261         int ret = 0;
3262 
3263         spin_lock_irq(q->queue_lock);
3264         if (q->nr_pending) {
3265                 ret = -EBUSY;
3266                 pm_runtime_mark_last_busy(q->dev);
3267         } else {
3268                 q->rpm_status = RPM_SUSPENDING;
3269         }
3270         spin_unlock_irq(q->queue_lock);
3271         return ret;
3272 }
3273 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3274 
3275 /**
3276  * blk_post_runtime_suspend - Post runtime suspend processing
3277  * @q: the queue of the device
3278  * @err: return value of the device's runtime_suspend function
3279  *
3280  * Description:
3281  *    Update the queue's runtime status according to the return value of the
3282  *    device's runtime suspend function and mark last busy for the device so
3283  *    that PM core will try to auto suspend the device at a later time.
3284  *
3285  *    This function should be called near the end of the device's
3286  *    runtime_suspend callback.
3287  */
3288 void blk_post_runtime_suspend(struct request_queue *q, int err)
3289 {
3290         spin_lock_irq(q->queue_lock);
3291         if (!err) {
3292                 q->rpm_status = RPM_SUSPENDED;
3293         } else {
3294                 q->rpm_status = RPM_ACTIVE;
3295                 pm_runtime_mark_last_busy(q->dev);
3296         }
3297         spin_unlock_irq(q->queue_lock);
3298 }
3299 EXPORT_SYMBOL(blk_post_runtime_suspend);
3300 
3301 /**
3302  * blk_pre_runtime_resume - Pre runtime resume processing
3303  * @q: the queue of the device
3304  *
3305  * Description:
3306  *    Update the queue's runtime status to RESUMING in preparation for the
3307  *    runtime resume of the device.
3308  *
3309  *    This function should be called near the start of the device's
3310  *    runtime_resume callback.
3311  */
3312 void blk_pre_runtime_resume(struct request_queue *q)
3313 {
3314         spin_lock_irq(q->queue_lock);
3315         q->rpm_status = RPM_RESUMING;
3316         spin_unlock_irq(q->queue_lock);
3317 }
3318 EXPORT_SYMBOL(blk_pre_runtime_resume);
3319 
3320 /**
3321  * blk_post_runtime_resume - Post runtime resume processing
3322  * @q: the queue of the device
3323  * @err: return value of the device's runtime_resume function
3324  *
3325  * Description:
3326  *    Update the queue's runtime status according to the return value of the
3327  *    device's runtime_resume function. If it is successfully resumed, process
3328  *    the requests that are queued into the device's queue when it is resuming
3329  *    and then mark last busy and initiate autosuspend for it.
3330  *
3331  *    This function should be called near the end of the device's
3332  *    runtime_resume callback.
3333  */
3334 void blk_post_runtime_resume(struct request_queue *q, int err)
3335 {
3336         spin_lock_irq(q->queue_lock);
3337         if (!err) {
3338                 q->rpm_status = RPM_ACTIVE;
3339                 __blk_run_queue(q);
3340                 pm_runtime_mark_last_busy(q->dev);
3341                 pm_request_autosuspend(q->dev);
3342         } else {
3343                 q->rpm_status = RPM_SUSPENDED;
3344         }
3345         spin_unlock_irq(q->queue_lock);
3346 }
3347 EXPORT_SYMBOL(blk_post_runtime_resume);
3348 #endif
3349 
3350 int __init blk_dev_init(void)
3351 {
3352         BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3353                         sizeof(((struct request *)0)->cmd_flags));
3354 
3355         /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3356         kblockd_workqueue = alloc_workqueue("kblockd",
3357                                             WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3358         if (!kblockd_workqueue)
3359                 panic("Failed to create kblockd\n");
3360 
3361         request_cachep = kmem_cache_create("blkdev_requests",
3362                         sizeof(struct request), 0, SLAB_PANIC, NULL);
3363 
3364         blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3365                         sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3366 
3367         return 0;
3368 }
3369 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp