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

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

Version: ~ [ linux-5.15-rc1 ] ~ [ linux-5.14.5 ] ~ [ linux-5.13.18 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.66 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.147 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.206 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.246 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.282 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.283 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ 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 #include <linux/blk-cgroup.h>
 36 
 37 #define CREATE_TRACE_POINTS
 38 #include <trace/events/block.h>
 39 
 40 #include "blk.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;
 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 static void blk_clear_congested(struct request_list *rl, int sync)
 67 {
 68 #ifdef CONFIG_CGROUP_WRITEBACK
 69         clear_wb_congested(rl->blkg->wb_congested, sync);
 70 #else
 71         /*
 72          * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
 73          * flip its congestion state for events on other blkcgs.
 74          */
 75         if (rl == &rl->q->root_rl)
 76                 clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
 77 #endif
 78 }
 79 
 80 static void blk_set_congested(struct request_list *rl, int sync)
 81 {
 82 #ifdef CONFIG_CGROUP_WRITEBACK
 83         set_wb_congested(rl->blkg->wb_congested, sync);
 84 #else
 85         /* see blk_clear_congested() */
 86         if (rl == &rl->q->root_rl)
 87                 set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
 88 #endif
 89 }
 90 
 91 void blk_queue_congestion_threshold(struct request_queue *q)
 92 {
 93         int nr;
 94 
 95         nr = q->nr_requests - (q->nr_requests / 8) + 1;
 96         if (nr > q->nr_requests)
 97                 nr = q->nr_requests;
 98         q->nr_congestion_on = nr;
 99 
100         nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
101         if (nr < 1)
102                 nr = 1;
103         q->nr_congestion_off = nr;
104 }
105 
106 /**
107  * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
108  * @bdev:       device
109  *
110  * Locates the passed device's request queue and returns the address of its
111  * backing_dev_info.  This function can only be called if @bdev is opened
112  * and the return value is never NULL.
113  */
114 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
115 {
116         struct request_queue *q = bdev_get_queue(bdev);
117 
118         return &q->backing_dev_info;
119 }
120 EXPORT_SYMBOL(blk_get_backing_dev_info);
121 
122 void blk_rq_init(struct request_queue *q, struct request *rq)
123 {
124         memset(rq, 0, sizeof(*rq));
125 
126         INIT_LIST_HEAD(&rq->queuelist);
127         INIT_LIST_HEAD(&rq->timeout_list);
128         rq->cpu = -1;
129         rq->q = q;
130         rq->__sector = (sector_t) -1;
131         INIT_HLIST_NODE(&rq->hash);
132         RB_CLEAR_NODE(&rq->rb_node);
133         rq->cmd = rq->__cmd;
134         rq->cmd_len = BLK_MAX_CDB;
135         rq->tag = -1;
136         rq->start_time = jiffies;
137         set_start_time_ns(rq);
138         rq->part = NULL;
139 }
140 EXPORT_SYMBOL(blk_rq_init);
141 
142 static void req_bio_endio(struct request *rq, struct bio *bio,
143                           unsigned int nbytes, int error)
144 {
145         if (error)
146                 bio->bi_error = error;
147 
148         if (unlikely(rq->cmd_flags & REQ_QUIET))
149                 bio_set_flag(bio, BIO_QUIET);
150 
151         bio_advance(bio, nbytes);
152 
153         /* don't actually finish bio if it's part of flush sequence */
154         if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
155                 bio_endio(bio);
156 }
157 
158 void blk_dump_rq_flags(struct request *rq, char *msg)
159 {
160         int bit;
161 
162         printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
163                 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
164                 (unsigned long long) rq->cmd_flags);
165 
166         printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
167                (unsigned long long)blk_rq_pos(rq),
168                blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
169         printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
170                rq->bio, rq->biotail, blk_rq_bytes(rq));
171 
172         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
173                 printk(KERN_INFO "  cdb: ");
174                 for (bit = 0; bit < BLK_MAX_CDB; bit++)
175                         printk("%02x ", rq->cmd[bit]);
176                 printk("\n");
177         }
178 }
179 EXPORT_SYMBOL(blk_dump_rq_flags);
180 
181 static void blk_delay_work(struct work_struct *work)
182 {
183         struct request_queue *q;
184 
185         q = container_of(work, struct request_queue, delay_work.work);
186         spin_lock_irq(q->queue_lock);
187         __blk_run_queue(q);
188         spin_unlock_irq(q->queue_lock);
189 }
190 
191 /**
192  * blk_delay_queue - restart queueing after defined interval
193  * @q:          The &struct request_queue in question
194  * @msecs:      Delay in msecs
195  *
196  * Description:
197  *   Sometimes queueing needs to be postponed for a little while, to allow
198  *   resources to come back. This function will make sure that queueing is
199  *   restarted around the specified time. Queue lock must be held.
200  */
201 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
202 {
203         if (likely(!blk_queue_dead(q)))
204                 queue_delayed_work(kblockd_workqueue, &q->delay_work,
205                                    msecs_to_jiffies(msecs));
206 }
207 EXPORT_SYMBOL(blk_delay_queue);
208 
209 /**
210  * blk_start_queue_async - asynchronously restart a previously stopped queue
211  * @q:    The &struct request_queue in question
212  *
213  * Description:
214  *   blk_start_queue_async() will clear the stop flag on the queue, and
215  *   ensure that the request_fn for the queue is run from an async
216  *   context.
217  **/
218 void blk_start_queue_async(struct request_queue *q)
219 {
220         queue_flag_clear(QUEUE_FLAG_STOPPED, q);
221         blk_run_queue_async(q);
222 }
223 EXPORT_SYMBOL(blk_start_queue_async);
224 
225 /**
226  * blk_start_queue - restart a previously stopped queue
227  * @q:    The &struct request_queue in question
228  *
229  * Description:
230  *   blk_start_queue() will clear the stop flag on the queue, and call
231  *   the request_fn for the queue if it was in a stopped state when
232  *   entered. Also see blk_stop_queue(). Queue lock must be held.
233  **/
234 void blk_start_queue(struct request_queue *q)
235 {
236         WARN_ON(!irqs_disabled());
237 
238         queue_flag_clear(QUEUE_FLAG_STOPPED, q);
239         __blk_run_queue(q);
240 }
241 EXPORT_SYMBOL(blk_start_queue);
242 
243 /**
244  * blk_stop_queue - stop a queue
245  * @q:    The &struct request_queue in question
246  *
247  * Description:
248  *   The Linux block layer assumes that a block driver will consume all
249  *   entries on the request queue when the request_fn strategy is called.
250  *   Often this will not happen, because of hardware limitations (queue
251  *   depth settings). If a device driver gets a 'queue full' response,
252  *   or if it simply chooses not to queue more I/O at one point, it can
253  *   call this function to prevent the request_fn from being called until
254  *   the driver has signalled it's ready to go again. This happens by calling
255  *   blk_start_queue() to restart queue operations. Queue lock must be held.
256  **/
257 void blk_stop_queue(struct request_queue *q)
258 {
259         cancel_delayed_work(&q->delay_work);
260         queue_flag_set(QUEUE_FLAG_STOPPED, q);
261 }
262 EXPORT_SYMBOL(blk_stop_queue);
263 
264 /**
265  * blk_sync_queue - cancel any pending callbacks on a queue
266  * @q: the queue
267  *
268  * Description:
269  *     The block layer may perform asynchronous callback activity
270  *     on a queue, such as calling the unplug function after a timeout.
271  *     A block device may call blk_sync_queue to ensure that any
272  *     such activity is cancelled, thus allowing it to release resources
273  *     that the callbacks might use. The caller must already have made sure
274  *     that its ->make_request_fn will not re-add plugging prior to calling
275  *     this function.
276  *
277  *     This function does not cancel any asynchronous activity arising
278  *     out of elevator or throttling code. That would require elevator_exit()
279  *     and blkcg_exit_queue() to be called with queue lock initialized.
280  *
281  */
282 void blk_sync_queue(struct request_queue *q)
283 {
284         del_timer_sync(&q->timeout);
285 
286         if (q->mq_ops) {
287                 struct blk_mq_hw_ctx *hctx;
288                 int i;
289 
290                 queue_for_each_hw_ctx(q, hctx, i) {
291                         cancel_delayed_work_sync(&hctx->run_work);
292                         cancel_delayed_work_sync(&hctx->delay_work);
293                 }
294         } else {
295                 cancel_delayed_work_sync(&q->delay_work);
296         }
297 }
298 EXPORT_SYMBOL(blk_sync_queue);
299 
300 /**
301  * __blk_run_queue_uncond - run a queue whether or not it has been stopped
302  * @q:  The queue to run
303  *
304  * Description:
305  *    Invoke request handling on a queue if there are any pending requests.
306  *    May be used to restart request handling after a request has completed.
307  *    This variant runs the queue whether or not the queue has been
308  *    stopped. Must be called with the queue lock held and interrupts
309  *    disabled. See also @blk_run_queue.
310  */
311 inline void __blk_run_queue_uncond(struct request_queue *q)
312 {
313         if (unlikely(blk_queue_dead(q)))
314                 return;
315 
316         /*
317          * Some request_fn implementations, e.g. scsi_request_fn(), unlock
318          * the queue lock internally. As a result multiple threads may be
319          * running such a request function concurrently. Keep track of the
320          * number of active request_fn invocations such that blk_drain_queue()
321          * can wait until all these request_fn calls have finished.
322          */
323         q->request_fn_active++;
324         q->request_fn(q);
325         q->request_fn_active--;
326 }
327 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
328 
329 /**
330  * __blk_run_queue - run a single device queue
331  * @q:  The queue to run
332  *
333  * Description:
334  *    See @blk_run_queue. This variant must be called with the queue lock
335  *    held and interrupts disabled.
336  */
337 void __blk_run_queue(struct request_queue *q)
338 {
339         if (unlikely(blk_queue_stopped(q)))
340                 return;
341 
342         __blk_run_queue_uncond(q);
343 }
344 EXPORT_SYMBOL(__blk_run_queue);
345 
346 /**
347  * blk_run_queue_async - run a single device queue in workqueue context
348  * @q:  The queue to run
349  *
350  * Description:
351  *    Tells kblockd to perform the equivalent of @blk_run_queue on behalf
352  *    of us. The caller must hold the queue lock.
353  */
354 void blk_run_queue_async(struct request_queue *q)
355 {
356         if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
357                 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
358 }
359 EXPORT_SYMBOL(blk_run_queue_async);
360 
361 /**
362  * blk_run_queue - run a single device queue
363  * @q: The queue to run
364  *
365  * Description:
366  *    Invoke request handling on this queue, if it has pending work to do.
367  *    May be used to restart queueing when a request has completed.
368  */
369 void blk_run_queue(struct request_queue *q)
370 {
371         unsigned long flags;
372 
373         spin_lock_irqsave(q->queue_lock, flags);
374         __blk_run_queue(q);
375         spin_unlock_irqrestore(q->queue_lock, flags);
376 }
377 EXPORT_SYMBOL(blk_run_queue);
378 
379 void blk_put_queue(struct request_queue *q)
380 {
381         kobject_put(&q->kobj);
382 }
383 EXPORT_SYMBOL(blk_put_queue);
384 
385 /**
386  * __blk_drain_queue - drain requests from request_queue
387  * @q: queue to drain
388  * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
389  *
390  * Drain requests from @q.  If @drain_all is set, all requests are drained.
391  * If not, only ELVPRIV requests are drained.  The caller is responsible
392  * for ensuring that no new requests which need to be drained are queued.
393  */
394 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
395         __releases(q->queue_lock)
396         __acquires(q->queue_lock)
397 {
398         int i;
399 
400         lockdep_assert_held(q->queue_lock);
401 
402         while (true) {
403                 bool drain = false;
404 
405                 /*
406                  * The caller might be trying to drain @q before its
407                  * elevator is initialized.
408                  */
409                 if (q->elevator)
410                         elv_drain_elevator(q);
411 
412                 blkcg_drain_queue(q);
413 
414                 /*
415                  * This function might be called on a queue which failed
416                  * driver init after queue creation or is not yet fully
417                  * active yet.  Some drivers (e.g. fd and loop) get unhappy
418                  * in such cases.  Kick queue iff dispatch queue has
419                  * something on it and @q has request_fn set.
420                  */
421                 if (!list_empty(&q->queue_head) && q->request_fn)
422                         __blk_run_queue(q);
423 
424                 drain |= q->nr_rqs_elvpriv;
425                 drain |= q->request_fn_active;
426 
427                 /*
428                  * Unfortunately, requests are queued at and tracked from
429                  * multiple places and there's no single counter which can
430                  * be drained.  Check all the queues and counters.
431                  */
432                 if (drain_all) {
433                         struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
434                         drain |= !list_empty(&q->queue_head);
435                         for (i = 0; i < 2; i++) {
436                                 drain |= q->nr_rqs[i];
437                                 drain |= q->in_flight[i];
438                                 if (fq)
439                                     drain |= !list_empty(&fq->flush_queue[i]);
440                         }
441                 }
442 
443                 if (!drain)
444                         break;
445 
446                 spin_unlock_irq(q->queue_lock);
447 
448                 msleep(10);
449 
450                 spin_lock_irq(q->queue_lock);
451         }
452 
453         /*
454          * With queue marked dead, any woken up waiter will fail the
455          * allocation path, so the wakeup chaining is lost and we're
456          * left with hung waiters. We need to wake up those waiters.
457          */
458         if (q->request_fn) {
459                 struct request_list *rl;
460 
461                 blk_queue_for_each_rl(rl, q)
462                         for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
463                                 wake_up_all(&rl->wait[i]);
464         }
465 }
466 
467 /**
468  * blk_queue_bypass_start - enter queue bypass mode
469  * @q: queue of interest
470  *
471  * In bypass mode, only the dispatch FIFO queue of @q is used.  This
472  * function makes @q enter bypass mode and drains all requests which were
473  * throttled or issued before.  On return, it's guaranteed that no request
474  * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
475  * inside queue or RCU read lock.
476  */
477 void blk_queue_bypass_start(struct request_queue *q)
478 {
479         spin_lock_irq(q->queue_lock);
480         q->bypass_depth++;
481         queue_flag_set(QUEUE_FLAG_BYPASS, q);
482         spin_unlock_irq(q->queue_lock);
483 
484         /*
485          * Queues start drained.  Skip actual draining till init is
486          * complete.  This avoids lenghty delays during queue init which
487          * can happen many times during boot.
488          */
489         if (blk_queue_init_done(q)) {
490                 spin_lock_irq(q->queue_lock);
491                 __blk_drain_queue(q, false);
492                 spin_unlock_irq(q->queue_lock);
493 
494                 /* ensure blk_queue_bypass() is %true inside RCU read lock */
495                 synchronize_rcu();
496         }
497 }
498 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
499 
500 /**
501  * blk_queue_bypass_end - leave queue bypass mode
502  * @q: queue of interest
503  *
504  * Leave bypass mode and restore the normal queueing behavior.
505  */
506 void blk_queue_bypass_end(struct request_queue *q)
507 {
508         spin_lock_irq(q->queue_lock);
509         if (!--q->bypass_depth)
510                 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
511         WARN_ON_ONCE(q->bypass_depth < 0);
512         spin_unlock_irq(q->queue_lock);
513 }
514 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
515 
516 void blk_set_queue_dying(struct request_queue *q)
517 {
518         queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
519 
520         if (q->mq_ops)
521                 blk_mq_wake_waiters(q);
522         else {
523                 struct request_list *rl;
524 
525                 blk_queue_for_each_rl(rl, q) {
526                         if (rl->rq_pool) {
527                                 wake_up(&rl->wait[BLK_RW_SYNC]);
528                                 wake_up(&rl->wait[BLK_RW_ASYNC]);
529                         }
530                 }
531         }
532 }
533 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
534 
535 /**
536  * blk_cleanup_queue - shutdown a request queue
537  * @q: request queue to shutdown
538  *
539  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
540  * put it.  All future requests will be failed immediately with -ENODEV.
541  */
542 void blk_cleanup_queue(struct request_queue *q)
543 {
544         spinlock_t *lock = q->queue_lock;
545 
546         /* mark @q DYING, no new request or merges will be allowed afterwards */
547         mutex_lock(&q->sysfs_lock);
548         blk_set_queue_dying(q);
549         spin_lock_irq(lock);
550 
551         /*
552          * A dying queue is permanently in bypass mode till released.  Note
553          * that, unlike blk_queue_bypass_start(), we aren't performing
554          * synchronize_rcu() after entering bypass mode to avoid the delay
555          * as some drivers create and destroy a lot of queues while
556          * probing.  This is still safe because blk_release_queue() will be
557          * called only after the queue refcnt drops to zero and nothing,
558          * RCU or not, would be traversing the queue by then.
559          */
560         q->bypass_depth++;
561         queue_flag_set(QUEUE_FLAG_BYPASS, q);
562 
563         queue_flag_set(QUEUE_FLAG_NOMERGES, q);
564         queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
565         queue_flag_set(QUEUE_FLAG_DYING, q);
566         spin_unlock_irq(lock);
567         mutex_unlock(&q->sysfs_lock);
568 
569         /*
570          * Drain all requests queued before DYING marking. Set DEAD flag to
571          * prevent that q->request_fn() gets invoked after draining finished.
572          */
573         blk_freeze_queue(q);
574         spin_lock_irq(lock);
575         if (!q->mq_ops)
576                 __blk_drain_queue(q, true);
577         queue_flag_set(QUEUE_FLAG_DEAD, q);
578         spin_unlock_irq(lock);
579 
580         /* for synchronous bio-based driver finish in-flight integrity i/o */
581         blk_flush_integrity();
582 
583         /* @q won't process any more request, flush async actions */
584         del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
585         blk_sync_queue(q);
586 
587         if (q->mq_ops)
588                 blk_mq_free_queue(q);
589         percpu_ref_exit(&q->q_usage_counter);
590 
591         spin_lock_irq(lock);
592         if (q->queue_lock != &q->__queue_lock)
593                 q->queue_lock = &q->__queue_lock;
594         spin_unlock_irq(lock);
595 
596         bdi_unregister(&q->backing_dev_info);
597 
598         /* @q is and will stay empty, shutdown and put */
599         blk_put_queue(q);
600 }
601 EXPORT_SYMBOL(blk_cleanup_queue);
602 
603 /* Allocate memory local to the request queue */
604 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
605 {
606         int nid = (int)(long)data;
607         return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
608 }
609 
610 static void free_request_struct(void *element, void *unused)
611 {
612         kmem_cache_free(request_cachep, element);
613 }
614 
615 int blk_init_rl(struct request_list *rl, struct request_queue *q,
616                 gfp_t gfp_mask)
617 {
618         if (unlikely(rl->rq_pool))
619                 return 0;
620 
621         rl->q = q;
622         rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
623         rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
624         init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
625         init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
626 
627         rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
628                                           free_request_struct,
629                                           (void *)(long)q->node, gfp_mask,
630                                           q->node);
631         if (!rl->rq_pool)
632                 return -ENOMEM;
633 
634         return 0;
635 }
636 
637 void blk_exit_rl(struct request_list *rl)
638 {
639         if (rl->rq_pool)
640                 mempool_destroy(rl->rq_pool);
641 }
642 
643 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
644 {
645         return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
646 }
647 EXPORT_SYMBOL(blk_alloc_queue);
648 
649 int blk_queue_enter(struct request_queue *q, bool nowait)
650 {
651         while (true) {
652                 int ret;
653 
654                 if (percpu_ref_tryget_live(&q->q_usage_counter))
655                         return 0;
656 
657                 if (nowait)
658                         return -EBUSY;
659 
660                 ret = wait_event_interruptible(q->mq_freeze_wq,
661                                 !atomic_read(&q->mq_freeze_depth) ||
662                                 blk_queue_dying(q));
663                 if (blk_queue_dying(q))
664                         return -ENODEV;
665                 if (ret)
666                         return ret;
667         }
668 }
669 
670 void blk_queue_exit(struct request_queue *q)
671 {
672         percpu_ref_put(&q->q_usage_counter);
673 }
674 
675 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
676 {
677         struct request_queue *q =
678                 container_of(ref, struct request_queue, q_usage_counter);
679 
680         wake_up_all(&q->mq_freeze_wq);
681 }
682 
683 static void blk_rq_timed_out_timer(unsigned long data)
684 {
685         struct request_queue *q = (struct request_queue *)data;
686 
687         kblockd_schedule_work(&q->timeout_work);
688 }
689 
690 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
691 {
692         struct request_queue *q;
693         int err;
694 
695         q = kmem_cache_alloc_node(blk_requestq_cachep,
696                                 gfp_mask | __GFP_ZERO, node_id);
697         if (!q)
698                 return NULL;
699 
700         q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
701         if (q->id < 0)
702                 goto fail_q;
703 
704         q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
705         if (!q->bio_split)
706                 goto fail_id;
707 
708         q->backing_dev_info.ra_pages =
709                         (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
710         q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
711         q->backing_dev_info.name = "block";
712         q->node = node_id;
713 
714         err = bdi_init(&q->backing_dev_info);
715         if (err)
716                 goto fail_split;
717 
718         setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
719                     laptop_mode_timer_fn, (unsigned long) q);
720         setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
721         INIT_LIST_HEAD(&q->queue_head);
722         INIT_LIST_HEAD(&q->timeout_list);
723         INIT_LIST_HEAD(&q->icq_list);
724 #ifdef CONFIG_BLK_CGROUP
725         INIT_LIST_HEAD(&q->blkg_list);
726 #endif
727         INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
728 
729         kobject_init(&q->kobj, &blk_queue_ktype);
730 
731         mutex_init(&q->sysfs_lock);
732         spin_lock_init(&q->__queue_lock);
733 
734         /*
735          * By default initialize queue_lock to internal lock and driver can
736          * override it later if need be.
737          */
738         q->queue_lock = &q->__queue_lock;
739 
740         /*
741          * A queue starts its life with bypass turned on to avoid
742          * unnecessary bypass on/off overhead and nasty surprises during
743          * init.  The initial bypass will be finished when the queue is
744          * registered by blk_register_queue().
745          */
746         q->bypass_depth = 1;
747         __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
748 
749         init_waitqueue_head(&q->mq_freeze_wq);
750 
751         /*
752          * Init percpu_ref in atomic mode so that it's faster to shutdown.
753          * See blk_register_queue() for details.
754          */
755         if (percpu_ref_init(&q->q_usage_counter,
756                                 blk_queue_usage_counter_release,
757                                 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
758                 goto fail_bdi;
759 
760         if (blkcg_init_queue(q))
761                 goto fail_ref;
762 
763         return q;
764 
765 fail_ref:
766         percpu_ref_exit(&q->q_usage_counter);
767 fail_bdi:
768         bdi_destroy(&q->backing_dev_info);
769 fail_split:
770         bioset_free(q->bio_split);
771 fail_id:
772         ida_simple_remove(&blk_queue_ida, q->id);
773 fail_q:
774         kmem_cache_free(blk_requestq_cachep, q);
775         return NULL;
776 }
777 EXPORT_SYMBOL(blk_alloc_queue_node);
778 
779 /**
780  * blk_init_queue  - prepare a request queue for use with a block device
781  * @rfn:  The function to be called to process requests that have been
782  *        placed on the queue.
783  * @lock: Request queue spin lock
784  *
785  * Description:
786  *    If a block device wishes to use the standard request handling procedures,
787  *    which sorts requests and coalesces adjacent requests, then it must
788  *    call blk_init_queue().  The function @rfn will be called when there
789  *    are requests on the queue that need to be processed.  If the device
790  *    supports plugging, then @rfn may not be called immediately when requests
791  *    are available on the queue, but may be called at some time later instead.
792  *    Plugged queues are generally unplugged when a buffer belonging to one
793  *    of the requests on the queue is needed, or due to memory pressure.
794  *
795  *    @rfn is not required, or even expected, to remove all requests off the
796  *    queue, but only as many as it can handle at a time.  If it does leave
797  *    requests on the queue, it is responsible for arranging that the requests
798  *    get dealt with eventually.
799  *
800  *    The queue spin lock must be held while manipulating the requests on the
801  *    request queue; this lock will be taken also from interrupt context, so irq
802  *    disabling is needed for it.
803  *
804  *    Function returns a pointer to the initialized request queue, or %NULL if
805  *    it didn't succeed.
806  *
807  * Note:
808  *    blk_init_queue() must be paired with a blk_cleanup_queue() call
809  *    when the block device is deactivated (such as at module unload).
810  **/
811 
812 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
813 {
814         return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
815 }
816 EXPORT_SYMBOL(blk_init_queue);
817 
818 struct request_queue *
819 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
820 {
821         struct request_queue *uninit_q, *q;
822 
823         uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
824         if (!uninit_q)
825                 return NULL;
826 
827         q = blk_init_allocated_queue(uninit_q, rfn, lock);
828         if (!q)
829                 blk_cleanup_queue(uninit_q);
830 
831         return q;
832 }
833 EXPORT_SYMBOL(blk_init_queue_node);
834 
835 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
836 
837 struct request_queue *
838 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
839                          spinlock_t *lock)
840 {
841         if (!q)
842                 return NULL;
843 
844         q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
845         if (!q->fq)
846                 return NULL;
847 
848         if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
849                 goto fail;
850 
851         INIT_WORK(&q->timeout_work, blk_timeout_work);
852         q->request_fn           = rfn;
853         q->prep_rq_fn           = NULL;
854         q->unprep_rq_fn         = NULL;
855         q->queue_flags          |= QUEUE_FLAG_DEFAULT;
856 
857         /* Override internal queue lock with supplied lock pointer */
858         if (lock)
859                 q->queue_lock           = lock;
860 
861         /*
862          * This also sets hw/phys segments, boundary and size
863          */
864         blk_queue_make_request(q, blk_queue_bio);
865 
866         q->sg_reserved_size = INT_MAX;
867 
868         /* Protect q->elevator from elevator_change */
869         mutex_lock(&q->sysfs_lock);
870 
871         /* init elevator */
872         if (elevator_init(q, NULL)) {
873                 mutex_unlock(&q->sysfs_lock);
874                 goto fail;
875         }
876 
877         mutex_unlock(&q->sysfs_lock);
878 
879         return q;
880 
881 fail:
882         blk_free_flush_queue(q->fq);
883         return NULL;
884 }
885 EXPORT_SYMBOL(blk_init_allocated_queue);
886 
887 bool blk_get_queue(struct request_queue *q)
888 {
889         if (likely(!blk_queue_dying(q))) {
890                 __blk_get_queue(q);
891                 return true;
892         }
893 
894         return false;
895 }
896 EXPORT_SYMBOL(blk_get_queue);
897 
898 static inline void blk_free_request(struct request_list *rl, struct request *rq)
899 {
900         if (rq->cmd_flags & REQ_ELVPRIV) {
901                 elv_put_request(rl->q, rq);
902                 if (rq->elv.icq)
903                         put_io_context(rq->elv.icq->ioc);
904         }
905 
906         mempool_free(rq, rl->rq_pool);
907 }
908 
909 /*
910  * ioc_batching returns true if the ioc is a valid batching request and
911  * should be given priority access to a request.
912  */
913 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
914 {
915         if (!ioc)
916                 return 0;
917 
918         /*
919          * Make sure the process is able to allocate at least 1 request
920          * even if the batch times out, otherwise we could theoretically
921          * lose wakeups.
922          */
923         return ioc->nr_batch_requests == q->nr_batching ||
924                 (ioc->nr_batch_requests > 0
925                 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
926 }
927 
928 /*
929  * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
930  * will cause the process to be a "batcher" on all queues in the system. This
931  * is the behaviour we want though - once it gets a wakeup it should be given
932  * a nice run.
933  */
934 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
935 {
936         if (!ioc || ioc_batching(q, ioc))
937                 return;
938 
939         ioc->nr_batch_requests = q->nr_batching;
940         ioc->last_waited = jiffies;
941 }
942 
943 static void __freed_request(struct request_list *rl, int sync)
944 {
945         struct request_queue *q = rl->q;
946 
947         if (rl->count[sync] < queue_congestion_off_threshold(q))
948                 blk_clear_congested(rl, sync);
949 
950         if (rl->count[sync] + 1 <= q->nr_requests) {
951                 if (waitqueue_active(&rl->wait[sync]))
952                         wake_up(&rl->wait[sync]);
953 
954                 blk_clear_rl_full(rl, sync);
955         }
956 }
957 
958 /*
959  * A request has just been released.  Account for it, update the full and
960  * congestion status, wake up any waiters.   Called under q->queue_lock.
961  */
962 static void freed_request(struct request_list *rl, unsigned int flags)
963 {
964         struct request_queue *q = rl->q;
965         int sync = rw_is_sync(flags);
966 
967         q->nr_rqs[sync]--;
968         rl->count[sync]--;
969         if (flags & REQ_ELVPRIV)
970                 q->nr_rqs_elvpriv--;
971 
972         __freed_request(rl, sync);
973 
974         if (unlikely(rl->starved[sync ^ 1]))
975                 __freed_request(rl, sync ^ 1);
976 }
977 
978 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
979 {
980         struct request_list *rl;
981         int on_thresh, off_thresh;
982 
983         spin_lock_irq(q->queue_lock);
984         q->nr_requests = nr;
985         blk_queue_congestion_threshold(q);
986         on_thresh = queue_congestion_on_threshold(q);
987         off_thresh = queue_congestion_off_threshold(q);
988 
989         blk_queue_for_each_rl(rl, q) {
990                 if (rl->count[BLK_RW_SYNC] >= on_thresh)
991                         blk_set_congested(rl, BLK_RW_SYNC);
992                 else if (rl->count[BLK_RW_SYNC] < off_thresh)
993                         blk_clear_congested(rl, BLK_RW_SYNC);
994 
995                 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
996                         blk_set_congested(rl, BLK_RW_ASYNC);
997                 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
998                         blk_clear_congested(rl, BLK_RW_ASYNC);
999 
1000                 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1001                         blk_set_rl_full(rl, BLK_RW_SYNC);
1002                 } else {
1003                         blk_clear_rl_full(rl, BLK_RW_SYNC);
1004                         wake_up(&rl->wait[BLK_RW_SYNC]);
1005                 }
1006 
1007                 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1008                         blk_set_rl_full(rl, BLK_RW_ASYNC);
1009                 } else {
1010                         blk_clear_rl_full(rl, BLK_RW_ASYNC);
1011                         wake_up(&rl->wait[BLK_RW_ASYNC]);
1012                 }
1013         }
1014 
1015         spin_unlock_irq(q->queue_lock);
1016         return 0;
1017 }
1018 
1019 /*
1020  * Determine if elevator data should be initialized when allocating the
1021  * request associated with @bio.
1022  */
1023 static bool blk_rq_should_init_elevator(struct bio *bio)
1024 {
1025         if (!bio)
1026                 return true;
1027 
1028         /*
1029          * Flush requests do not use the elevator so skip initialization.
1030          * This allows a request to share the flush and elevator data.
1031          */
1032         if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
1033                 return false;
1034 
1035         return true;
1036 }
1037 
1038 /**
1039  * rq_ioc - determine io_context for request allocation
1040  * @bio: request being allocated is for this bio (can be %NULL)
1041  *
1042  * Determine io_context to use for request allocation for @bio.  May return
1043  * %NULL if %current->io_context doesn't exist.
1044  */
1045 static struct io_context *rq_ioc(struct bio *bio)
1046 {
1047 #ifdef CONFIG_BLK_CGROUP
1048         if (bio && bio->bi_ioc)
1049                 return bio->bi_ioc;
1050 #endif
1051         return current->io_context;
1052 }
1053 
1054 /**
1055  * __get_request - get a free request
1056  * @rl: request list to allocate from
1057  * @rw_flags: RW and SYNC flags
1058  * @bio: bio to allocate request for (can be %NULL)
1059  * @gfp_mask: allocation mask
1060  *
1061  * Get a free request from @q.  This function may fail under memory
1062  * pressure or if @q is dead.
1063  *
1064  * Must be called with @q->queue_lock held and,
1065  * Returns ERR_PTR on failure, with @q->queue_lock held.
1066  * Returns request pointer on success, with @q->queue_lock *not held*.
1067  */
1068 static struct request *__get_request(struct request_list *rl, int rw_flags,
1069                                      struct bio *bio, gfp_t gfp_mask)
1070 {
1071         struct request_queue *q = rl->q;
1072         struct request *rq;
1073         struct elevator_type *et = q->elevator->type;
1074         struct io_context *ioc = rq_ioc(bio);
1075         struct io_cq *icq = NULL;
1076         const bool is_sync = rw_is_sync(rw_flags) != 0;
1077         int may_queue;
1078 
1079         if (unlikely(blk_queue_dying(q)))
1080                 return ERR_PTR(-ENODEV);
1081 
1082         may_queue = elv_may_queue(q, rw_flags);
1083         if (may_queue == ELV_MQUEUE_NO)
1084                 goto rq_starved;
1085 
1086         if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1087                 if (rl->count[is_sync]+1 >= q->nr_requests) {
1088                         /*
1089                          * The queue will fill after this allocation, so set
1090                          * it as full, and mark this process as "batching".
1091                          * This process will be allowed to complete a batch of
1092                          * requests, others will be blocked.
1093                          */
1094                         if (!blk_rl_full(rl, is_sync)) {
1095                                 ioc_set_batching(q, ioc);
1096                                 blk_set_rl_full(rl, is_sync);
1097                         } else {
1098                                 if (may_queue != ELV_MQUEUE_MUST
1099                                                 && !ioc_batching(q, ioc)) {
1100                                         /*
1101                                          * The queue is full and the allocating
1102                                          * process is not a "batcher", and not
1103                                          * exempted by the IO scheduler
1104                                          */
1105                                         return ERR_PTR(-ENOMEM);
1106                                 }
1107                         }
1108                 }
1109                 blk_set_congested(rl, is_sync);
1110         }
1111 
1112         /*
1113          * Only allow batching queuers to allocate up to 50% over the defined
1114          * limit of requests, otherwise we could have thousands of requests
1115          * allocated with any setting of ->nr_requests
1116          */
1117         if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1118                 return ERR_PTR(-ENOMEM);
1119 
1120         q->nr_rqs[is_sync]++;
1121         rl->count[is_sync]++;
1122         rl->starved[is_sync] = 0;
1123 
1124         /*
1125          * Decide whether the new request will be managed by elevator.  If
1126          * so, mark @rw_flags and increment elvpriv.  Non-zero elvpriv will
1127          * prevent the current elevator from being destroyed until the new
1128          * request is freed.  This guarantees icq's won't be destroyed and
1129          * makes creating new ones safe.
1130          *
1131          * Also, lookup icq while holding queue_lock.  If it doesn't exist,
1132          * it will be created after releasing queue_lock.
1133          */
1134         if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1135                 rw_flags |= REQ_ELVPRIV;
1136                 q->nr_rqs_elvpriv++;
1137                 if (et->icq_cache && ioc)
1138                         icq = ioc_lookup_icq(ioc, q);
1139         }
1140 
1141         if (blk_queue_io_stat(q))
1142                 rw_flags |= REQ_IO_STAT;
1143         spin_unlock_irq(q->queue_lock);
1144 
1145         /* allocate and init request */
1146         rq = mempool_alloc(rl->rq_pool, gfp_mask);
1147         if (!rq)
1148                 goto fail_alloc;
1149 
1150         blk_rq_init(q, rq);
1151         blk_rq_set_rl(rq, rl);
1152         rq->cmd_flags = rw_flags | REQ_ALLOCED;
1153 
1154         /* init elvpriv */
1155         if (rw_flags & REQ_ELVPRIV) {
1156                 if (unlikely(et->icq_cache && !icq)) {
1157                         if (ioc)
1158                                 icq = ioc_create_icq(ioc, q, gfp_mask);
1159                         if (!icq)
1160                                 goto fail_elvpriv;
1161                 }
1162 
1163                 rq->elv.icq = icq;
1164                 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1165                         goto fail_elvpriv;
1166 
1167                 /* @rq->elv.icq holds io_context until @rq is freed */
1168                 if (icq)
1169                         get_io_context(icq->ioc);
1170         }
1171 out:
1172         /*
1173          * ioc may be NULL here, and ioc_batching will be false. That's
1174          * OK, if the queue is under the request limit then requests need
1175          * not count toward the nr_batch_requests limit. There will always
1176          * be some limit enforced by BLK_BATCH_TIME.
1177          */
1178         if (ioc_batching(q, ioc))
1179                 ioc->nr_batch_requests--;
1180 
1181         trace_block_getrq(q, bio, rw_flags & 1);
1182         return rq;
1183 
1184 fail_elvpriv:
1185         /*
1186          * elvpriv init failed.  ioc, icq and elvpriv aren't mempool backed
1187          * and may fail indefinitely under memory pressure and thus
1188          * shouldn't stall IO.  Treat this request as !elvpriv.  This will
1189          * disturb iosched and blkcg but weird is bettern than dead.
1190          */
1191         printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1192                            __func__, dev_name(q->backing_dev_info.dev));
1193 
1194         rq->cmd_flags &= ~REQ_ELVPRIV;
1195         rq->elv.icq = NULL;
1196 
1197         spin_lock_irq(q->queue_lock);
1198         q->nr_rqs_elvpriv--;
1199         spin_unlock_irq(q->queue_lock);
1200         goto out;
1201 
1202 fail_alloc:
1203         /*
1204          * Allocation failed presumably due to memory. Undo anything we
1205          * might have messed up.
1206          *
1207          * Allocating task should really be put onto the front of the wait
1208          * queue, but this is pretty rare.
1209          */
1210         spin_lock_irq(q->queue_lock);
1211         freed_request(rl, rw_flags);
1212 
1213         /*
1214          * in the very unlikely event that allocation failed and no
1215          * requests for this direction was pending, mark us starved so that
1216          * freeing of a request in the other direction will notice
1217          * us. another possible fix would be to split the rq mempool into
1218          * READ and WRITE
1219          */
1220 rq_starved:
1221         if (unlikely(rl->count[is_sync] == 0))
1222                 rl->starved[is_sync] = 1;
1223         return ERR_PTR(-ENOMEM);
1224 }
1225 
1226 /**
1227  * get_request - get a free request
1228  * @q: request_queue to allocate request from
1229  * @rw_flags: RW and SYNC flags
1230  * @bio: bio to allocate request for (can be %NULL)
1231  * @gfp_mask: allocation mask
1232  *
1233  * Get a free request from @q.  If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1234  * this function keeps retrying under memory pressure and fails iff @q is dead.
1235  *
1236  * Must be called with @q->queue_lock held and,
1237  * Returns ERR_PTR on failure, with @q->queue_lock held.
1238  * Returns request pointer on success, with @q->queue_lock *not held*.
1239  */
1240 static struct request *get_request(struct request_queue *q, int rw_flags,
1241                                    struct bio *bio, gfp_t gfp_mask)
1242 {
1243         const bool is_sync = rw_is_sync(rw_flags) != 0;
1244         DEFINE_WAIT(wait);
1245         struct request_list *rl;
1246         struct request *rq;
1247 
1248         rl = blk_get_rl(q, bio);        /* transferred to @rq on success */
1249 retry:
1250         rq = __get_request(rl, rw_flags, bio, gfp_mask);
1251         if (!IS_ERR(rq))
1252                 return rq;
1253 
1254         if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1255                 blk_put_rl(rl);
1256                 return rq;
1257         }
1258 
1259         /* wait on @rl and retry */
1260         prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1261                                   TASK_UNINTERRUPTIBLE);
1262 
1263         trace_block_sleeprq(q, bio, rw_flags & 1);
1264 
1265         spin_unlock_irq(q->queue_lock);
1266         io_schedule();
1267 
1268         /*
1269          * After sleeping, we become a "batching" process and will be able
1270          * to allocate at least one request, and up to a big batch of them
1271          * for a small period time.  See ioc_batching, ioc_set_batching
1272          */
1273         ioc_set_batching(q, current->io_context);
1274 
1275         spin_lock_irq(q->queue_lock);
1276         finish_wait(&rl->wait[is_sync], &wait);
1277 
1278         goto retry;
1279 }
1280 
1281 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1282                 gfp_t gfp_mask)
1283 {
1284         struct request *rq;
1285 
1286         BUG_ON(rw != READ && rw != WRITE);
1287 
1288         /* create ioc upfront */
1289         create_io_context(gfp_mask, q->node);
1290 
1291         spin_lock_irq(q->queue_lock);
1292         rq = get_request(q, rw, NULL, gfp_mask);
1293         if (IS_ERR(rq))
1294                 spin_unlock_irq(q->queue_lock);
1295         /* q->queue_lock is unlocked at this point */
1296 
1297         return rq;
1298 }
1299 
1300 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1301 {
1302         if (q->mq_ops)
1303                 return blk_mq_alloc_request(q, rw,
1304                         (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1305                                 0 : BLK_MQ_REQ_NOWAIT);
1306         else
1307                 return blk_old_get_request(q, rw, gfp_mask);
1308 }
1309 EXPORT_SYMBOL(blk_get_request);
1310 
1311 /**
1312  * blk_make_request - given a bio, allocate a corresponding struct request.
1313  * @q: target request queue
1314  * @bio:  The bio describing the memory mappings that will be submitted for IO.
1315  *        It may be a chained-bio properly constructed by block/bio layer.
1316  * @gfp_mask: gfp flags to be used for memory allocation
1317  *
1318  * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1319  * type commands. Where the struct request needs to be farther initialized by
1320  * the caller. It is passed a &struct bio, which describes the memory info of
1321  * the I/O transfer.
1322  *
1323  * The caller of blk_make_request must make sure that bi_io_vec
1324  * are set to describe the memory buffers. That bio_data_dir() will return
1325  * the needed direction of the request. (And all bio's in the passed bio-chain
1326  * are properly set accordingly)
1327  *
1328  * If called under none-sleepable conditions, mapped bio buffers must not
1329  * need bouncing, by calling the appropriate masked or flagged allocator,
1330  * suitable for the target device. Otherwise the call to blk_queue_bounce will
1331  * BUG.
1332  *
1333  * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1334  * given to how you allocate bios. In particular, you cannot use
1335  * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1336  * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1337  * thus resulting in a deadlock. Alternatively bios should be allocated using
1338  * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
1339  * If possible a big IO should be split into smaller parts when allocation
1340  * fails. Partial allocation should not be an error, or you risk a live-lock.
1341  */
1342 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1343                                  gfp_t gfp_mask)
1344 {
1345         struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1346 
1347         if (IS_ERR(rq))
1348                 return rq;
1349 
1350         blk_rq_set_block_pc(rq);
1351 
1352         for_each_bio(bio) {
1353                 struct bio *bounce_bio = bio;
1354                 int ret;
1355 
1356                 blk_queue_bounce(q, &bounce_bio);
1357                 ret = blk_rq_append_bio(q, rq, bounce_bio);
1358                 if (unlikely(ret)) {
1359                         blk_put_request(rq);
1360                         return ERR_PTR(ret);
1361                 }
1362         }
1363 
1364         return rq;
1365 }
1366 EXPORT_SYMBOL(blk_make_request);
1367 
1368 /**
1369  * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1370  * @rq:         request to be initialized
1371  *
1372  */
1373 void blk_rq_set_block_pc(struct request *rq)
1374 {
1375         rq->cmd_type = REQ_TYPE_BLOCK_PC;
1376         rq->__data_len = 0;
1377         rq->__sector = (sector_t) -1;
1378         rq->bio = rq->biotail = NULL;
1379         memset(rq->__cmd, 0, sizeof(rq->__cmd));
1380 }
1381 EXPORT_SYMBOL(blk_rq_set_block_pc);
1382 
1383 /**
1384  * blk_requeue_request - put a request back on queue
1385  * @q:          request queue where request should be inserted
1386  * @rq:         request to be inserted
1387  *
1388  * Description:
1389  *    Drivers often keep queueing requests until the hardware cannot accept
1390  *    more, when that condition happens we need to put the request back
1391  *    on the queue. Must be called with queue lock held.
1392  */
1393 void blk_requeue_request(struct request_queue *q, struct request *rq)
1394 {
1395         blk_delete_timer(rq);
1396         blk_clear_rq_complete(rq);
1397         trace_block_rq_requeue(q, rq);
1398 
1399         if (rq->cmd_flags & REQ_QUEUED)
1400                 blk_queue_end_tag(q, rq);
1401 
1402         BUG_ON(blk_queued_rq(rq));
1403 
1404         elv_requeue_request(q, rq);
1405 }
1406 EXPORT_SYMBOL(blk_requeue_request);
1407 
1408 static void add_acct_request(struct request_queue *q, struct request *rq,
1409                              int where)
1410 {
1411         blk_account_io_start(rq, true);
1412         __elv_add_request(q, rq, where);
1413 }
1414 
1415 static void part_round_stats_single(int cpu, struct hd_struct *part,
1416                                     unsigned long now)
1417 {
1418         int inflight;
1419 
1420         if (now == part->stamp)
1421                 return;
1422 
1423         inflight = part_in_flight(part);
1424         if (inflight) {
1425                 __part_stat_add(cpu, part, time_in_queue,
1426                                 inflight * (now - part->stamp));
1427                 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1428         }
1429         part->stamp = now;
1430 }
1431 
1432 /**
1433  * part_round_stats() - Round off the performance stats on a struct disk_stats.
1434  * @cpu: cpu number for stats access
1435  * @part: target partition
1436  *
1437  * The average IO queue length and utilisation statistics are maintained
1438  * by observing the current state of the queue length and the amount of
1439  * time it has been in this state for.
1440  *
1441  * Normally, that accounting is done on IO completion, but that can result
1442  * in more than a second's worth of IO being accounted for within any one
1443  * second, leading to >100% utilisation.  To deal with that, we call this
1444  * function to do a round-off before returning the results when reading
1445  * /proc/diskstats.  This accounts immediately for all queue usage up to
1446  * the current jiffies and restarts the counters again.
1447  */
1448 void part_round_stats(int cpu, struct hd_struct *part)
1449 {
1450         unsigned long now = jiffies;
1451 
1452         if (part->partno)
1453                 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1454         part_round_stats_single(cpu, part, now);
1455 }
1456 EXPORT_SYMBOL_GPL(part_round_stats);
1457 
1458 #ifdef CONFIG_PM
1459 static void blk_pm_put_request(struct request *rq)
1460 {
1461         if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1462                 pm_runtime_mark_last_busy(rq->q->dev);
1463 }
1464 #else
1465 static inline void blk_pm_put_request(struct request *rq) {}
1466 #endif
1467 
1468 /*
1469  * queue lock must be held
1470  */
1471 void __blk_put_request(struct request_queue *q, struct request *req)
1472 {
1473         if (unlikely(!q))
1474                 return;
1475 
1476         if (q->mq_ops) {
1477                 blk_mq_free_request(req);
1478                 return;
1479         }
1480 
1481         blk_pm_put_request(req);
1482 
1483         elv_completed_request(q, req);
1484 
1485         /* this is a bio leak */
1486         WARN_ON(req->bio != NULL);
1487 
1488         /*
1489          * Request may not have originated from ll_rw_blk. if not,
1490          * it didn't come out of our reserved rq pools
1491          */
1492         if (req->cmd_flags & REQ_ALLOCED) {
1493                 unsigned int flags = req->cmd_flags;
1494                 struct request_list *rl = blk_rq_rl(req);
1495 
1496                 BUG_ON(!list_empty(&req->queuelist));
1497                 BUG_ON(ELV_ON_HASH(req));
1498 
1499                 blk_free_request(rl, req);
1500                 freed_request(rl, flags);
1501                 blk_put_rl(rl);
1502         }
1503 }
1504 EXPORT_SYMBOL_GPL(__blk_put_request);
1505 
1506 void blk_put_request(struct request *req)
1507 {
1508         struct request_queue *q = req->q;
1509 
1510         if (q->mq_ops)
1511                 blk_mq_free_request(req);
1512         else {
1513                 unsigned long flags;
1514 
1515                 spin_lock_irqsave(q->queue_lock, flags);
1516                 __blk_put_request(q, req);
1517                 spin_unlock_irqrestore(q->queue_lock, flags);
1518         }
1519 }
1520 EXPORT_SYMBOL(blk_put_request);
1521 
1522 /**
1523  * blk_add_request_payload - add a payload to a request
1524  * @rq: request to update
1525  * @page: page backing the payload
1526  * @len: length of the payload.
1527  *
1528  * This allows to later add a payload to an already submitted request by
1529  * a block driver.  The driver needs to take care of freeing the payload
1530  * itself.
1531  *
1532  * Note that this is a quite horrible hack and nothing but handling of
1533  * discard requests should ever use it.
1534  */
1535 void blk_add_request_payload(struct request *rq, struct page *page,
1536                 unsigned int len)
1537 {
1538         struct bio *bio = rq->bio;
1539 
1540         bio->bi_io_vec->bv_page = page;
1541         bio->bi_io_vec->bv_offset = 0;
1542         bio->bi_io_vec->bv_len = len;
1543 
1544         bio->bi_iter.bi_size = len;
1545         bio->bi_vcnt = 1;
1546         bio->bi_phys_segments = 1;
1547 
1548         rq->__data_len = rq->resid_len = len;
1549         rq->nr_phys_segments = 1;
1550 }
1551 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1552 
1553 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1554                             struct bio *bio)
1555 {
1556         const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1557 
1558         if (!ll_back_merge_fn(q, req, bio))
1559                 return false;
1560 
1561         trace_block_bio_backmerge(q, req, bio);
1562 
1563         if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1564                 blk_rq_set_mixed_merge(req);
1565 
1566         req->biotail->bi_next = bio;
1567         req->biotail = bio;
1568         req->__data_len += bio->bi_iter.bi_size;
1569         req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1570 
1571         blk_account_io_start(req, false);
1572         return true;
1573 }
1574 
1575 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1576                              struct bio *bio)
1577 {
1578         const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1579 
1580         if (!ll_front_merge_fn(q, req, bio))
1581                 return false;
1582 
1583         trace_block_bio_frontmerge(q, req, bio);
1584 
1585         if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1586                 blk_rq_set_mixed_merge(req);
1587 
1588         bio->bi_next = req->bio;
1589         req->bio = bio;
1590 
1591         req->__sector = bio->bi_iter.bi_sector;
1592         req->__data_len += bio->bi_iter.bi_size;
1593         req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1594 
1595         blk_account_io_start(req, false);
1596         return true;
1597 }
1598 
1599 /**
1600  * blk_attempt_plug_merge - try to merge with %current's plugged list
1601  * @q: request_queue new bio is being queued at
1602  * @bio: new bio being queued
1603  * @request_count: out parameter for number of traversed plugged requests
1604  * @same_queue_rq: pointer to &struct request that gets filled in when
1605  * another request associated with @q is found on the plug list
1606  * (optional, may be %NULL)
1607  *
1608  * Determine whether @bio being queued on @q can be merged with a request
1609  * on %current's plugged list.  Returns %true if merge was successful,
1610  * otherwise %false.
1611  *
1612  * Plugging coalesces IOs from the same issuer for the same purpose without
1613  * going through @q->queue_lock.  As such it's more of an issuing mechanism
1614  * than scheduling, and the request, while may have elvpriv data, is not
1615  * added on the elevator at this point.  In addition, we don't have
1616  * reliable access to the elevator outside queue lock.  Only check basic
1617  * merging parameters without querying the elevator.
1618  *
1619  * Caller must ensure !blk_queue_nomerges(q) beforehand.
1620  */
1621 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1622                             unsigned int *request_count,
1623                             struct request **same_queue_rq)
1624 {
1625         struct blk_plug *plug;
1626         struct request *rq;
1627         bool ret = false;
1628         struct list_head *plug_list;
1629 
1630         plug = current->plug;
1631         if (!plug)
1632                 goto out;
1633         *request_count = 0;
1634 
1635         if (q->mq_ops)
1636                 plug_list = &plug->mq_list;
1637         else
1638                 plug_list = &plug->list;
1639 
1640         list_for_each_entry_reverse(rq, plug_list, queuelist) {
1641                 int el_ret;
1642 
1643                 if (rq->q == q) {
1644                         (*request_count)++;
1645                         /*
1646                          * Only blk-mq multiple hardware queues case checks the
1647                          * rq in the same queue, there should be only one such
1648                          * rq in a queue
1649                          **/
1650                         if (same_queue_rq)
1651                                 *same_queue_rq = rq;
1652                 }
1653 
1654                 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1655                         continue;
1656 
1657                 el_ret = blk_try_merge(rq, bio);
1658                 if (el_ret == ELEVATOR_BACK_MERGE) {
1659                         ret = bio_attempt_back_merge(q, rq, bio);
1660                         if (ret)
1661                                 break;
1662                 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1663                         ret = bio_attempt_front_merge(q, rq, bio);
1664                         if (ret)
1665                                 break;
1666                 }
1667         }
1668 out:
1669         return ret;
1670 }
1671 
1672 unsigned int blk_plug_queued_count(struct request_queue *q)
1673 {
1674         struct blk_plug *plug;
1675         struct request *rq;
1676         struct list_head *plug_list;
1677         unsigned int ret = 0;
1678 
1679         plug = current->plug;
1680         if (!plug)
1681                 goto out;
1682 
1683         if (q->mq_ops)
1684                 plug_list = &plug->mq_list;
1685         else
1686                 plug_list = &plug->list;
1687 
1688         list_for_each_entry(rq, plug_list, queuelist) {
1689                 if (rq->q == q)
1690                         ret++;
1691         }
1692 out:
1693         return ret;
1694 }
1695 
1696 void init_request_from_bio(struct request *req, struct bio *bio)
1697 {
1698         req->cmd_type = REQ_TYPE_FS;
1699 
1700         req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1701         if (bio->bi_rw & REQ_RAHEAD)
1702                 req->cmd_flags |= REQ_FAILFAST_MASK;
1703 
1704         req->errors = 0;
1705         req->__sector = bio->bi_iter.bi_sector;
1706         req->ioprio = bio_prio(bio);
1707         blk_rq_bio_prep(req->q, req, bio);
1708 }
1709 
1710 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1711 {
1712         const bool sync = !!(bio->bi_rw & REQ_SYNC);
1713         struct blk_plug *plug;
1714         int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1715         struct request *req;
1716         unsigned int request_count = 0;
1717 
1718         /*
1719          * low level driver can indicate that it wants pages above a
1720          * certain limit bounced to low memory (ie for highmem, or even
1721          * ISA dma in theory)
1722          */
1723         blk_queue_bounce(q, &bio);
1724 
1725         blk_queue_split(q, &bio, q->bio_split);
1726 
1727         if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1728                 bio->bi_error = -EIO;
1729                 bio_endio(bio);
1730                 return BLK_QC_T_NONE;
1731         }
1732 
1733         if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1734                 spin_lock_irq(q->queue_lock);
1735                 where = ELEVATOR_INSERT_FLUSH;
1736                 goto get_rq;
1737         }
1738 
1739         /*
1740          * Check if we can merge with the plugged list before grabbing
1741          * any locks.
1742          */
1743         if (!blk_queue_nomerges(q)) {
1744                 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1745                         return BLK_QC_T_NONE;
1746         } else
1747                 request_count = blk_plug_queued_count(q);
1748 
1749         spin_lock_irq(q->queue_lock);
1750 
1751         el_ret = elv_merge(q, &req, bio);
1752         if (el_ret == ELEVATOR_BACK_MERGE) {
1753                 if (bio_attempt_back_merge(q, req, bio)) {
1754                         elv_bio_merged(q, req, bio);
1755                         if (!attempt_back_merge(q, req))
1756                                 elv_merged_request(q, req, el_ret);
1757                         goto out_unlock;
1758                 }
1759         } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1760                 if (bio_attempt_front_merge(q, req, bio)) {
1761                         elv_bio_merged(q, req, bio);
1762                         if (!attempt_front_merge(q, req))
1763                                 elv_merged_request(q, req, el_ret);
1764                         goto out_unlock;
1765                 }
1766         }
1767 
1768 get_rq:
1769         /*
1770          * This sync check and mask will be re-done in init_request_from_bio(),
1771          * but we need to set it earlier to expose the sync flag to the
1772          * rq allocator and io schedulers.
1773          */
1774         rw_flags = bio_data_dir(bio);
1775         if (sync)
1776                 rw_flags |= REQ_SYNC;
1777 
1778         /*
1779          * Grab a free request. This is might sleep but can not fail.
1780          * Returns with the queue unlocked.
1781          */
1782         req = get_request(q, rw_flags, bio, GFP_NOIO);
1783         if (IS_ERR(req)) {
1784                 bio->bi_error = PTR_ERR(req);
1785                 bio_endio(bio);
1786                 goto out_unlock;
1787         }
1788 
1789         /*
1790          * After dropping the lock and possibly sleeping here, our request
1791          * may now be mergeable after it had proven unmergeable (above).
1792          * We don't worry about that case for efficiency. It won't happen
1793          * often, and the elevators are able to handle it.
1794          */
1795         init_request_from_bio(req, bio);
1796 
1797         if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1798                 req->cpu = raw_smp_processor_id();
1799 
1800         plug = current->plug;
1801         if (plug) {
1802                 /*
1803                  * If this is the first request added after a plug, fire
1804                  * of a plug trace.
1805                  */
1806                 if (!request_count)
1807                         trace_block_plug(q);
1808                 else {
1809                         if (request_count >= BLK_MAX_REQUEST_COUNT) {
1810                                 blk_flush_plug_list(plug, false);
1811                                 trace_block_plug(q);
1812                         }
1813                 }
1814                 list_add_tail(&req->queuelist, &plug->list);
1815                 blk_account_io_start(req, true);
1816         } else {
1817                 spin_lock_irq(q->queue_lock);
1818                 add_acct_request(q, req, where);
1819                 __blk_run_queue(q);
1820 out_unlock:
1821                 spin_unlock_irq(q->queue_lock);
1822         }
1823 
1824         return BLK_QC_T_NONE;
1825 }
1826 
1827 /*
1828  * If bio->bi_dev is a partition, remap the location
1829  */
1830 static inline void blk_partition_remap(struct bio *bio)
1831 {
1832         struct block_device *bdev = bio->bi_bdev;
1833 
1834         if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1835                 struct hd_struct *p = bdev->bd_part;
1836 
1837                 bio->bi_iter.bi_sector += p->start_sect;
1838                 bio->bi_bdev = bdev->bd_contains;
1839 
1840                 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1841                                       bdev->bd_dev,
1842                                       bio->bi_iter.bi_sector - p->start_sect);
1843         }
1844 }
1845 
1846 static void handle_bad_sector(struct bio *bio)
1847 {
1848         char b[BDEVNAME_SIZE];
1849 
1850         printk(KERN_INFO "attempt to access beyond end of device\n");
1851         printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1852                         bdevname(bio->bi_bdev, b),
1853                         bio->bi_rw,
1854                         (unsigned long long)bio_end_sector(bio),
1855                         (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1856 }
1857 
1858 #ifdef CONFIG_FAIL_MAKE_REQUEST
1859 
1860 static DECLARE_FAULT_ATTR(fail_make_request);
1861 
1862 static int __init setup_fail_make_request(char *str)
1863 {
1864         return setup_fault_attr(&fail_make_request, str);
1865 }
1866 __setup("fail_make_request=", setup_fail_make_request);
1867 
1868 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1869 {
1870         return part->make_it_fail && should_fail(&fail_make_request, bytes);
1871 }
1872 
1873 static int __init fail_make_request_debugfs(void)
1874 {
1875         struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1876                                                 NULL, &fail_make_request);
1877 
1878         return PTR_ERR_OR_ZERO(dir);
1879 }
1880 
1881 late_initcall(fail_make_request_debugfs);
1882 
1883 #else /* CONFIG_FAIL_MAKE_REQUEST */
1884 
1885 static inline bool should_fail_request(struct hd_struct *part,
1886                                         unsigned int bytes)
1887 {
1888         return false;
1889 }
1890 
1891 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1892 
1893 /*
1894  * Check whether this bio extends beyond the end of the device.
1895  */
1896 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1897 {
1898         sector_t maxsector;
1899 
1900         if (!nr_sectors)
1901                 return 0;
1902 
1903         /* Test device or partition size, when known. */
1904         maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1905         if (maxsector) {
1906                 sector_t sector = bio->bi_iter.bi_sector;
1907 
1908                 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1909                         /*
1910                          * This may well happen - the kernel calls bread()
1911                          * without checking the size of the device, e.g., when
1912                          * mounting a device.
1913                          */
1914                         handle_bad_sector(bio);
1915                         return 1;
1916                 }
1917         }
1918 
1919         return 0;
1920 }
1921 
1922 static noinline_for_stack bool
1923 generic_make_request_checks(struct bio *bio)
1924 {
1925         struct request_queue *q;
1926         int nr_sectors = bio_sectors(bio);
1927         int err = -EIO;
1928         char b[BDEVNAME_SIZE];
1929         struct hd_struct *part;
1930 
1931         might_sleep();
1932 
1933         if (bio_check_eod(bio, nr_sectors))
1934                 goto end_io;
1935 
1936         q = bdev_get_queue(bio->bi_bdev);
1937         if (unlikely(!q)) {
1938                 printk(KERN_ERR
1939                        "generic_make_request: Trying to access "
1940                         "nonexistent block-device %s (%Lu)\n",
1941                         bdevname(bio->bi_bdev, b),
1942                         (long long) bio->bi_iter.bi_sector);
1943                 goto end_io;
1944         }
1945 
1946         part = bio->bi_bdev->bd_part;
1947         if (should_fail_request(part, bio->bi_iter.bi_size) ||
1948             should_fail_request(&part_to_disk(part)->part0,
1949                                 bio->bi_iter.bi_size))
1950                 goto end_io;
1951 
1952         /*
1953          * If this device has partitions, remap block n
1954          * of partition p to block n+start(p) of the disk.
1955          */
1956         blk_partition_remap(bio);
1957 
1958         if (bio_check_eod(bio, nr_sectors))
1959                 goto end_io;
1960 
1961         /*
1962          * Filter flush bio's early so that make_request based
1963          * drivers without flush support don't have to worry
1964          * about them.
1965          */
1966         if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1967                 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1968                 if (!nr_sectors) {
1969                         err = 0;
1970                         goto end_io;
1971                 }
1972         }
1973 
1974         if ((bio->bi_rw & REQ_DISCARD) &&
1975             (!blk_queue_discard(q) ||
1976              ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1977                 err = -EOPNOTSUPP;
1978                 goto end_io;
1979         }
1980 
1981         if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1982                 err = -EOPNOTSUPP;
1983                 goto end_io;
1984         }
1985 
1986         /*
1987          * Various block parts want %current->io_context and lazy ioc
1988          * allocation ends up trading a lot of pain for a small amount of
1989          * memory.  Just allocate it upfront.  This may fail and block
1990          * layer knows how to live with it.
1991          */
1992         create_io_context(GFP_ATOMIC, q->node);
1993 
1994         if (!blkcg_bio_issue_check(q, bio))
1995                 return false;
1996 
1997         trace_block_bio_queue(q, bio);
1998         return true;
1999 
2000 end_io:
2001         bio->bi_error = err;
2002         bio_endio(bio);
2003         return false;
2004 }
2005 
2006 /**
2007  * generic_make_request - hand a buffer to its device driver for I/O
2008  * @bio:  The bio describing the location in memory and on the device.
2009  *
2010  * generic_make_request() is used to make I/O requests of block
2011  * devices. It is passed a &struct bio, which describes the I/O that needs
2012  * to be done.
2013  *
2014  * generic_make_request() does not return any status.  The
2015  * success/failure status of the request, along with notification of
2016  * completion, is delivered asynchronously through the bio->bi_end_io
2017  * function described (one day) else where.
2018  *
2019  * The caller of generic_make_request must make sure that bi_io_vec
2020  * are set to describe the memory buffer, and that bi_dev and bi_sector are
2021  * set to describe the device address, and the
2022  * bi_end_io and optionally bi_private are set to describe how
2023  * completion notification should be signaled.
2024  *
2025  * generic_make_request and the drivers it calls may use bi_next if this
2026  * bio happens to be merged with someone else, and may resubmit the bio to
2027  * a lower device by calling into generic_make_request recursively, which
2028  * means the bio should NOT be touched after the call to ->make_request_fn.
2029  */
2030 blk_qc_t generic_make_request(struct bio *bio)
2031 {
2032         struct bio_list bio_list_on_stack;
2033         blk_qc_t ret = BLK_QC_T_NONE;
2034 
2035         if (!generic_make_request_checks(bio))
2036                 goto out;
2037 
2038         /*
2039          * We only want one ->make_request_fn to be active at a time, else
2040          * stack usage with stacked devices could be a problem.  So use
2041          * current->bio_list to keep a list of requests submited by a
2042          * make_request_fn function.  current->bio_list is also used as a
2043          * flag to say if generic_make_request is currently active in this
2044          * task or not.  If it is NULL, then no make_request is active.  If
2045          * it is non-NULL, then a make_request is active, and new requests
2046          * should be added at the tail
2047          */
2048         if (current->bio_list) {
2049                 bio_list_add(current->bio_list, bio);
2050                 goto out;
2051         }
2052 
2053         /* following loop may be a bit non-obvious, and so deserves some
2054          * explanation.
2055          * Before entering the loop, bio->bi_next is NULL (as all callers
2056          * ensure that) so we have a list with a single bio.
2057          * We pretend that we have just taken it off a longer list, so
2058          * we assign bio_list to a pointer to the bio_list_on_stack,
2059          * thus initialising the bio_list of new bios to be
2060          * added.  ->make_request() may indeed add some more bios
2061          * through a recursive call to generic_make_request.  If it
2062          * did, we find a non-NULL value in bio_list and re-enter the loop
2063          * from the top.  In this case we really did just take the bio
2064          * of the top of the list (no pretending) and so remove it from
2065          * bio_list, and call into ->make_request() again.
2066          */
2067         BUG_ON(bio->bi_next);
2068         bio_list_init(&bio_list_on_stack);
2069         current->bio_list = &bio_list_on_stack;
2070         do {
2071                 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2072 
2073                 if (likely(blk_queue_enter(q, false) == 0)) {
2074                         ret = q->make_request_fn(q, bio);
2075 
2076                         blk_queue_exit(q);
2077 
2078                         bio = bio_list_pop(current->bio_list);
2079                 } else {
2080                         struct bio *bio_next = bio_list_pop(current->bio_list);
2081 
2082                         bio_io_error(bio);
2083                         bio = bio_next;
2084                 }
2085         } while (bio);
2086         current->bio_list = NULL; /* deactivate */
2087 
2088 out:
2089         return ret;
2090 }
2091 EXPORT_SYMBOL(generic_make_request);
2092 
2093 /**
2094  * submit_bio - submit a bio to the block device layer for I/O
2095  * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2096  * @bio: The &struct bio which describes the I/O
2097  *
2098  * submit_bio() is very similar in purpose to generic_make_request(), and
2099  * uses that function to do most of the work. Both are fairly rough
2100  * interfaces; @bio must be presetup and ready for I/O.
2101  *
2102  */
2103 blk_qc_t submit_bio(int rw, struct bio *bio)
2104 {
2105         bio->bi_rw |= rw;
2106 
2107         /*
2108          * If it's a regular read/write or a barrier with data attached,
2109          * go through the normal accounting stuff before submission.
2110          */
2111         if (bio_has_data(bio)) {
2112                 unsigned int count;
2113 
2114                 if (unlikely(rw & REQ_WRITE_SAME))
2115                         count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2116                 else
2117                         count = bio_sectors(bio);
2118 
2119                 if (rw & WRITE) {
2120                         count_vm_events(PGPGOUT, count);
2121                 } else {
2122                         task_io_account_read(bio->bi_iter.bi_size);
2123                         count_vm_events(PGPGIN, count);
2124                 }
2125 
2126                 if (unlikely(block_dump)) {
2127                         char b[BDEVNAME_SIZE];
2128                         printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2129                         current->comm, task_pid_nr(current),
2130                                 (rw & WRITE) ? "WRITE" : "READ",
2131                                 (unsigned long long)bio->bi_iter.bi_sector,
2132                                 bdevname(bio->bi_bdev, b),
2133                                 count);
2134                 }
2135         }
2136 
2137         return generic_make_request(bio);
2138 }
2139 EXPORT_SYMBOL(submit_bio);
2140 
2141 /**
2142  * blk_cloned_rq_check_limits - Helper function to check a cloned request
2143  *                              for new the queue limits
2144  * @q:  the queue
2145  * @rq: the request being checked
2146  *
2147  * Description:
2148  *    @rq may have been made based on weaker limitations of upper-level queues
2149  *    in request stacking drivers, and it may violate the limitation of @q.
2150  *    Since the block layer and the underlying device driver trust @rq
2151  *    after it is inserted to @q, it should be checked against @q before
2152  *    the insertion using this generic function.
2153  *
2154  *    Request stacking drivers like request-based dm may change the queue
2155  *    limits when retrying requests on other queues. Those requests need
2156  *    to be checked against the new queue limits again during dispatch.
2157  */
2158 static int blk_cloned_rq_check_limits(struct request_queue *q,
2159                                       struct request *rq)
2160 {
2161         if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2162                 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2163                 return -EIO;
2164         }
2165 
2166         /*
2167          * queue's settings related to segment counting like q->bounce_pfn
2168          * may differ from that of other stacking queues.
2169          * Recalculate it to check the request correctly on this queue's
2170          * limitation.
2171          */
2172         blk_recalc_rq_segments(rq);
2173         if (rq->nr_phys_segments > queue_max_segments(q)) {
2174                 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2175                 return -EIO;
2176         }
2177 
2178         return 0;
2179 }
2180 
2181 /**
2182  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2183  * @q:  the queue to submit the request
2184  * @rq: the request being queued
2185  */
2186 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2187 {
2188         unsigned long flags;
2189         int where = ELEVATOR_INSERT_BACK;
2190 
2191         if (blk_cloned_rq_check_limits(q, rq))
2192                 return -EIO;
2193 
2194         if (rq->rq_disk &&
2195             should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2196                 return -EIO;
2197 
2198         if (q->mq_ops) {
2199                 if (blk_queue_io_stat(q))
2200                         blk_account_io_start(rq, true);
2201                 blk_mq_insert_request(rq, false, true, false);
2202                 return 0;
2203         }
2204 
2205         spin_lock_irqsave(q->queue_lock, flags);
2206         if (unlikely(blk_queue_dying(q))) {
2207                 spin_unlock_irqrestore(q->queue_lock, flags);
2208                 return -ENODEV;
2209         }
2210 
2211         /*
2212          * Submitting request must be dequeued before calling this function
2213          * because it will be linked to another request_queue
2214          */
2215         BUG_ON(blk_queued_rq(rq));
2216 
2217         if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2218                 where = ELEVATOR_INSERT_FLUSH;
2219 
2220         add_acct_request(q, rq, where);
2221         if (where == ELEVATOR_INSERT_FLUSH)
2222                 __blk_run_queue(q);
2223         spin_unlock_irqrestore(q->queue_lock, flags);
2224 
2225         return 0;
2226 }
2227 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2228 
2229 /**
2230  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2231  * @rq: request to examine
2232  *
2233  * Description:
2234  *     A request could be merge of IOs which require different failure
2235  *     handling.  This function determines the number of bytes which
2236  *     can be failed from the beginning of the request without
2237  *     crossing into area which need to be retried further.
2238  *
2239  * Return:
2240  *     The number of bytes to fail.
2241  *
2242  * Context:
2243  *     queue_lock must be held.
2244  */
2245 unsigned int blk_rq_err_bytes(const struct request *rq)
2246 {
2247         unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2248         unsigned int bytes = 0;
2249         struct bio *bio;
2250 
2251         if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2252                 return blk_rq_bytes(rq);
2253 
2254         /*
2255          * Currently the only 'mixing' which can happen is between
2256          * different fastfail types.  We can safely fail portions
2257          * which have all the failfast bits that the first one has -
2258          * the ones which are at least as eager to fail as the first
2259          * one.
2260          */
2261         for (bio = rq->bio; bio; bio = bio->bi_next) {
2262                 if ((bio->bi_rw & ff) != ff)
2263                         break;
2264                 bytes += bio->bi_iter.bi_size;
2265         }
2266 
2267         /* this could lead to infinite loop */
2268         BUG_ON(blk_rq_bytes(rq) && !bytes);
2269         return bytes;
2270 }
2271 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2272 
2273 void blk_account_io_completion(struct request *req, unsigned int bytes)
2274 {
2275         if (blk_do_io_stat(req)) {
2276                 const int rw = rq_data_dir(req);
2277                 struct hd_struct *part;
2278                 int cpu;
2279 
2280                 cpu = part_stat_lock();
2281                 part = req->part;
2282                 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2283                 part_stat_unlock();
2284         }
2285 }
2286 
2287 void blk_account_io_done(struct request *req)
2288 {
2289         /*
2290          * Account IO completion.  flush_rq isn't accounted as a
2291          * normal IO on queueing nor completion.  Accounting the
2292          * containing request is enough.
2293          */
2294         if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2295                 unsigned long duration = jiffies - req->start_time;
2296                 const int rw = rq_data_dir(req);
2297                 struct hd_struct *part;
2298                 int cpu;
2299 
2300                 cpu = part_stat_lock();
2301                 part = req->part;
2302 
2303                 part_stat_inc(cpu, part, ios[rw]);
2304                 part_stat_add(cpu, part, ticks[rw], duration);
2305                 part_round_stats(cpu, part);
2306                 part_dec_in_flight(part, rw);
2307 
2308                 hd_struct_put(part);
2309                 part_stat_unlock();
2310         }
2311 }
2312 
2313 #ifdef CONFIG_PM
2314 /*
2315  * Don't process normal requests when queue is suspended
2316  * or in the process of suspending/resuming
2317  */
2318 static struct request *blk_pm_peek_request(struct request_queue *q,
2319                                            struct request *rq)
2320 {
2321         if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2322             (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2323                 return NULL;
2324         else
2325                 return rq;
2326 }
2327 #else
2328 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2329                                                   struct request *rq)
2330 {
2331         return rq;
2332 }
2333 #endif
2334 
2335 void blk_account_io_start(struct request *rq, bool new_io)
2336 {
2337         struct hd_struct *part;
2338         int rw = rq_data_dir(rq);
2339         int cpu;
2340 
2341         if (!blk_do_io_stat(rq))
2342                 return;
2343 
2344         cpu = part_stat_lock();
2345 
2346         if (!new_io) {
2347                 part = rq->part;
2348                 part_stat_inc(cpu, part, merges[rw]);
2349         } else {
2350                 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2351                 if (!hd_struct_try_get(part)) {
2352                         /*
2353                          * The partition is already being removed,
2354                          * the request will be accounted on the disk only
2355                          *
2356                          * We take a reference on disk->part0 although that
2357                          * partition will never be deleted, so we can treat
2358                          * it as any other partition.
2359                          */
2360                         part = &rq->rq_disk->part0;
2361                         hd_struct_get(part);
2362                 }
2363                 part_round_stats(cpu, part);
2364                 part_inc_in_flight(part, rw);
2365                 rq->part = part;
2366         }
2367 
2368         part_stat_unlock();
2369 }
2370 
2371 /**
2372  * blk_peek_request - peek at the top of a request queue
2373  * @q: request queue to peek at
2374  *
2375  * Description:
2376  *     Return the request at the top of @q.  The returned request
2377  *     should be started using blk_start_request() before LLD starts
2378  *     processing it.
2379  *
2380  * Return:
2381  *     Pointer to the request at the top of @q if available.  Null
2382  *     otherwise.
2383  *
2384  * Context:
2385  *     queue_lock must be held.
2386  */
2387 struct request *blk_peek_request(struct request_queue *q)
2388 {
2389         struct request *rq;
2390         int ret;
2391 
2392         while ((rq = __elv_next_request(q)) != NULL) {
2393 
2394                 rq = blk_pm_peek_request(q, rq);
2395                 if (!rq)
2396                         break;
2397 
2398                 if (!(rq->cmd_flags & REQ_STARTED)) {
2399                         /*
2400                          * This is the first time the device driver
2401                          * sees this request (possibly after
2402                          * requeueing).  Notify IO scheduler.
2403                          */
2404                         if (rq->cmd_flags & REQ_SORTED)
2405                                 elv_activate_rq(q, rq);
2406 
2407                         /*
2408                          * just mark as started even if we don't start
2409                          * it, a request that has been delayed should
2410                          * not be passed by new incoming requests
2411                          */
2412                         rq->cmd_flags |= REQ_STARTED;
2413                         trace_block_rq_issue(q, rq);
2414                 }
2415 
2416                 if (!q->boundary_rq || q->boundary_rq == rq) {
2417                         q->end_sector = rq_end_sector(rq);
2418                         q->boundary_rq = NULL;
2419                 }
2420 
2421                 if (rq->cmd_flags & REQ_DONTPREP)
2422                         break;
2423 
2424                 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2425                         /*
2426                          * make sure space for the drain appears we
2427                          * know we can do this because max_hw_segments
2428                          * has been adjusted to be one fewer than the
2429                          * device can handle
2430                          */
2431                         rq->nr_phys_segments++;
2432                 }
2433 
2434                 if (!q->prep_rq_fn)
2435                         break;
2436 
2437                 ret = q->prep_rq_fn(q, rq);
2438                 if (ret == BLKPREP_OK) {
2439                         break;
2440                 } else if (ret == BLKPREP_DEFER) {
2441                         /*
2442                          * the request may have been (partially) prepped.
2443                          * we need to keep this request in the front to
2444                          * avoid resource deadlock.  REQ_STARTED will
2445                          * prevent other fs requests from passing this one.
2446                          */
2447                         if (q->dma_drain_size && blk_rq_bytes(rq) &&
2448                             !(rq->cmd_flags & REQ_DONTPREP)) {
2449                                 /*
2450                                  * remove the space for the drain we added
2451                                  * so that we don't add it again
2452                                  */
2453                                 --rq->nr_phys_segments;
2454                         }
2455 
2456                         rq = NULL;
2457                         break;
2458                 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2459                         int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2460 
2461                         rq->cmd_flags |= REQ_QUIET;
2462                         /*
2463                          * Mark this request as started so we don't trigger
2464                          * any debug logic in the end I/O path.
2465                          */
2466                         blk_start_request(rq);
2467                         __blk_end_request_all(rq, err);
2468                 } else {
2469                         printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2470                         break;
2471                 }
2472         }
2473 
2474         return rq;
2475 }
2476 EXPORT_SYMBOL(blk_peek_request);
2477 
2478 void blk_dequeue_request(struct request *rq)
2479 {
2480         struct request_queue *q = rq->q;
2481 
2482         BUG_ON(list_empty(&rq->queuelist));
2483         BUG_ON(ELV_ON_HASH(rq));
2484 
2485         list_del_init(&rq->queuelist);
2486 
2487         /*
2488          * the time frame between a request being removed from the lists
2489          * and to it is freed is accounted as io that is in progress at
2490          * the driver side.
2491          */
2492         if (blk_account_rq(rq)) {
2493                 q->in_flight[rq_is_sync(rq)]++;
2494                 set_io_start_time_ns(rq);
2495         }
2496 }
2497 
2498 /**
2499  * blk_start_request - start request processing on the driver
2500  * @req: request to dequeue
2501  *
2502  * Description:
2503  *     Dequeue @req and start timeout timer on it.  This hands off the
2504  *     request to the driver.
2505  *
2506  *     Block internal functions which don't want to start timer should
2507  *     call blk_dequeue_request().
2508  *
2509  * Context:
2510  *     queue_lock must be held.
2511  */
2512 void blk_start_request(struct request *req)
2513 {
2514         blk_dequeue_request(req);
2515 
2516         /*
2517          * We are now handing the request to the hardware, initialize
2518          * resid_len to full count and add the timeout handler.
2519          */
2520         req->resid_len = blk_rq_bytes(req);
2521         if (unlikely(blk_bidi_rq(req)))
2522                 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2523 
2524         BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2525         blk_add_timer(req);
2526 }
2527 EXPORT_SYMBOL(blk_start_request);
2528 
2529 /**
2530  * blk_fetch_request - fetch a request from a request queue
2531  * @q: request queue to fetch a request from
2532  *
2533  * Description:
2534  *     Return the request at the top of @q.  The request is started on
2535  *     return and LLD can start processing it immediately.
2536  *
2537  * Return:
2538  *     Pointer to the request at the top of @q if available.  Null
2539  *     otherwise.
2540  *
2541  * Context:
2542  *     queue_lock must be held.
2543  */
2544 struct request *blk_fetch_request(struct request_queue *q)
2545 {
2546         struct request *rq;
2547 
2548         rq = blk_peek_request(q);
2549         if (rq)
2550                 blk_start_request(rq);
2551         return rq;
2552 }
2553 EXPORT_SYMBOL(blk_fetch_request);
2554 
2555 /**
2556  * blk_update_request - Special helper function for request stacking drivers
2557  * @req:      the request being processed
2558  * @error:    %0 for success, < %0 for error
2559  * @nr_bytes: number of bytes to complete @req
2560  *
2561  * Description:
2562  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
2563  *     the request structure even if @req doesn't have leftover.
2564  *     If @req has leftover, sets it up for the next range of segments.
2565  *
2566  *     This special helper function is only for request stacking drivers
2567  *     (e.g. request-based dm) so that they can handle partial completion.
2568  *     Actual device drivers should use blk_end_request instead.
2569  *
2570  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2571  *     %false return from this function.
2572  *
2573  * Return:
2574  *     %false - this request doesn't have any more data
2575  *     %true  - this request has more data
2576  **/
2577 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2578 {
2579         int total_bytes;
2580 
2581         trace_block_rq_complete(req->q, req, nr_bytes);
2582 
2583         if (!req->bio)
2584                 return false;
2585 
2586         /*
2587          * For fs requests, rq is just carrier of independent bio's
2588          * and each partial completion should be handled separately.
2589          * Reset per-request error on each partial completion.
2590          *
2591          * TODO: tj: This is too subtle.  It would be better to let
2592          * low level drivers do what they see fit.
2593          */
2594         if (req->cmd_type == REQ_TYPE_FS)
2595                 req->errors = 0;
2596 
2597         if (error && req->cmd_type == REQ_TYPE_FS &&
2598             !(req->cmd_flags & REQ_QUIET)) {
2599                 char *error_type;
2600 
2601                 switch (error) {
2602                 case -ENOLINK:
2603                         error_type = "recoverable transport";
2604                         break;
2605                 case -EREMOTEIO:
2606                         error_type = "critical target";
2607                         break;
2608                 case -EBADE:
2609                         error_type = "critical nexus";
2610                         break;
2611                 case -ETIMEDOUT:
2612                         error_type = "timeout";
2613                         break;
2614                 case -ENOSPC:
2615                         error_type = "critical space allocation";
2616                         break;
2617                 case -ENODATA:
2618                         error_type = "critical medium";
2619                         break;
2620                 case -EIO:
2621                 default:
2622                         error_type = "I/O";
2623                         break;
2624                 }
2625                 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2626                                    __func__, error_type, req->rq_disk ?
2627                                    req->rq_disk->disk_name : "?",
2628                                    (unsigned long long)blk_rq_pos(req));
2629 
2630         }
2631 
2632         blk_account_io_completion(req, nr_bytes);
2633 
2634         total_bytes = 0;
2635         while (req->bio) {
2636                 struct bio *bio = req->bio;
2637                 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2638 
2639                 if (bio_bytes == bio->bi_iter.bi_size)
2640                         req->bio = bio->bi_next;
2641 
2642                 req_bio_endio(req, bio, bio_bytes, error);
2643 
2644                 total_bytes += bio_bytes;
2645                 nr_bytes -= bio_bytes;
2646 
2647                 if (!nr_bytes)
2648                         break;
2649         }
2650 
2651         /*
2652          * completely done
2653          */
2654         if (!req->bio) {
2655                 /*
2656                  * Reset counters so that the request stacking driver
2657                  * can find how many bytes remain in the request
2658                  * later.
2659                  */
2660                 req->__data_len = 0;
2661                 return false;
2662         }
2663 
2664         req->__data_len -= total_bytes;
2665 
2666         /* update sector only for requests with clear definition of sector */
2667         if (req->cmd_type == REQ_TYPE_FS)
2668                 req->__sector += total_bytes >> 9;
2669 
2670         /* mixed attributes always follow the first bio */
2671         if (req->cmd_flags & REQ_MIXED_MERGE) {
2672                 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2673                 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2674         }
2675 
2676         /*
2677          * If total number of sectors is less than the first segment
2678          * size, something has gone terribly wrong.
2679          */
2680         if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2681                 blk_dump_rq_flags(req, "request botched");
2682                 req->__data_len = blk_rq_cur_bytes(req);
2683         }
2684 
2685         /* recalculate the number of segments */
2686         blk_recalc_rq_segments(req);
2687 
2688         return true;
2689 }
2690 EXPORT_SYMBOL_GPL(blk_update_request);
2691 
2692 static bool blk_update_bidi_request(struct request *rq, int error,
2693                                     unsigned int nr_bytes,
2694                                     unsigned int bidi_bytes)
2695 {
2696         if (blk_update_request(rq, error, nr_bytes))
2697                 return true;
2698 
2699         /* Bidi request must be completed as a whole */
2700         if (unlikely(blk_bidi_rq(rq)) &&
2701             blk_update_request(rq->next_rq, error, bidi_bytes))
2702                 return true;
2703 
2704         if (blk_queue_add_random(rq->q))
2705                 add_disk_randomness(rq->rq_disk);
2706 
2707         return false;
2708 }
2709 
2710 /**
2711  * blk_unprep_request - unprepare a request
2712  * @req:        the request
2713  *
2714  * This function makes a request ready for complete resubmission (or
2715  * completion).  It happens only after all error handling is complete,
2716  * so represents the appropriate moment to deallocate any resources
2717  * that were allocated to the request in the prep_rq_fn.  The queue
2718  * lock is held when calling this.
2719  */
2720 void blk_unprep_request(struct request *req)
2721 {
2722         struct request_queue *q = req->q;
2723 
2724         req->cmd_flags &= ~REQ_DONTPREP;
2725         if (q->unprep_rq_fn)
2726                 q->unprep_rq_fn(q, req);
2727 }
2728 EXPORT_SYMBOL_GPL(blk_unprep_request);
2729 
2730 /*
2731  * queue lock must be held
2732  */
2733 void blk_finish_request(struct request *req, int error)
2734 {
2735         if (req->cmd_flags & REQ_QUEUED)
2736                 blk_queue_end_tag(req->q, req);
2737 
2738         BUG_ON(blk_queued_rq(req));
2739 
2740         if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2741                 laptop_io_completion(&req->q->backing_dev_info);
2742 
2743         blk_delete_timer(req);
2744 
2745         if (req->cmd_flags & REQ_DONTPREP)
2746                 blk_unprep_request(req);
2747 
2748         blk_account_io_done(req);
2749 
2750         if (req->end_io)
2751                 req->end_io(req, error);
2752         else {
2753                 if (blk_bidi_rq(req))
2754                         __blk_put_request(req->next_rq->q, req->next_rq);
2755 
2756                 __blk_put_request(req->q, req);
2757         }
2758 }
2759 EXPORT_SYMBOL(blk_finish_request);
2760 
2761 /**
2762  * blk_end_bidi_request - Complete a bidi request
2763  * @rq:         the request to complete
2764  * @error:      %0 for success, < %0 for error
2765  * @nr_bytes:   number of bytes to complete @rq
2766  * @bidi_bytes: number of bytes to complete @rq->next_rq
2767  *
2768  * Description:
2769  *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2770  *     Drivers that supports bidi can safely call this member for any
2771  *     type of request, bidi or uni.  In the later case @bidi_bytes is
2772  *     just ignored.
2773  *
2774  * Return:
2775  *     %false - we are done with this request
2776  *     %true  - still buffers pending for this request
2777  **/
2778 static bool blk_end_bidi_request(struct request *rq, int error,
2779                                  unsigned int nr_bytes, unsigned int bidi_bytes)
2780 {
2781         struct request_queue *q = rq->q;
2782         unsigned long flags;
2783 
2784         if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2785                 return true;
2786 
2787         spin_lock_irqsave(q->queue_lock, flags);
2788         blk_finish_request(rq, error);
2789         spin_unlock_irqrestore(q->queue_lock, flags);
2790 
2791         return false;
2792 }
2793 
2794 /**
2795  * __blk_end_bidi_request - Complete a bidi request with queue lock held
2796  * @rq:         the request to complete
2797  * @error:      %0 for success, < %0 for error
2798  * @nr_bytes:   number of bytes to complete @rq
2799  * @bidi_bytes: number of bytes to complete @rq->next_rq
2800  *
2801  * Description:
2802  *     Identical to blk_end_bidi_request() except that queue lock is
2803  *     assumed to be locked on entry and remains so on return.
2804  *
2805  * Return:
2806  *     %false - we are done with this request
2807  *     %true  - still buffers pending for this request
2808  **/
2809 bool __blk_end_bidi_request(struct request *rq, int error,
2810                                    unsigned int nr_bytes, unsigned int bidi_bytes)
2811 {
2812         if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2813                 return true;
2814 
2815         blk_finish_request(rq, error);
2816 
2817         return false;
2818 }
2819 
2820 /**
2821  * blk_end_request - Helper function for drivers to complete the request.
2822  * @rq:       the request being processed
2823  * @error:    %0 for success, < %0 for error
2824  * @nr_bytes: number of bytes to complete
2825  *
2826  * Description:
2827  *     Ends I/O on a number of bytes attached to @rq.
2828  *     If @rq has leftover, sets it up for the next range of segments.
2829  *
2830  * Return:
2831  *     %false - we are done with this request
2832  *     %true  - still buffers pending for this request
2833  **/
2834 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2835 {
2836         return blk_end_bidi_request(rq, error, nr_bytes, 0);
2837 }
2838 EXPORT_SYMBOL(blk_end_request);
2839 
2840 /**
2841  * blk_end_request_all - Helper function for drives to finish the request.
2842  * @rq: the request to finish
2843  * @error: %0 for success, < %0 for error
2844  *
2845  * Description:
2846  *     Completely finish @rq.
2847  */
2848 void blk_end_request_all(struct request *rq, int error)
2849 {
2850         bool pending;
2851         unsigned int bidi_bytes = 0;
2852 
2853         if (unlikely(blk_bidi_rq(rq)))
2854                 bidi_bytes = blk_rq_bytes(rq->next_rq);
2855 
2856         pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2857         BUG_ON(pending);
2858 }
2859 EXPORT_SYMBOL(blk_end_request_all);
2860 
2861 /**
2862  * blk_end_request_cur - Helper function to finish the current request chunk.
2863  * @rq: the request to finish the current chunk for
2864  * @error: %0 for success, < %0 for error
2865  *
2866  * Description:
2867  *     Complete the current consecutively mapped chunk from @rq.
2868  *
2869  * Return:
2870  *     %false - we are done with this request
2871  *     %true  - still buffers pending for this request
2872  */
2873 bool blk_end_request_cur(struct request *rq, int error)
2874 {
2875         return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2876 }
2877 EXPORT_SYMBOL(blk_end_request_cur);
2878 
2879 /**
2880  * blk_end_request_err - Finish a request till the next failure boundary.
2881  * @rq: the request to finish till the next failure boundary for
2882  * @error: must be negative errno
2883  *
2884  * Description:
2885  *     Complete @rq till the next failure boundary.
2886  *
2887  * Return:
2888  *     %false - we are done with this request
2889  *     %true  - still buffers pending for this request
2890  */
2891 bool blk_end_request_err(struct request *rq, int error)
2892 {
2893         WARN_ON(error >= 0);
2894         return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2895 }
2896 EXPORT_SYMBOL_GPL(blk_end_request_err);
2897 
2898 /**
2899  * __blk_end_request - Helper function for drivers to complete the request.
2900  * @rq:       the request being processed
2901  * @error:    %0 for success, < %0 for error
2902  * @nr_bytes: number of bytes to complete
2903  *
2904  * Description:
2905  *     Must be called with queue lock held unlike blk_end_request().
2906  *
2907  * Return:
2908  *     %false - we are done with this request
2909  *     %true  - still buffers pending for this request
2910  **/
2911 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2912 {
2913         return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2914 }
2915 EXPORT_SYMBOL(__blk_end_request);
2916 
2917 /**
2918  * __blk_end_request_all - Helper function for drives to finish the request.
2919  * @rq: the request to finish
2920  * @error: %0 for success, < %0 for error
2921  *
2922  * Description:
2923  *     Completely finish @rq.  Must be called with queue lock held.
2924  */
2925 void __blk_end_request_all(struct request *rq, int error)
2926 {
2927         bool pending;
2928         unsigned int bidi_bytes = 0;
2929 
2930         if (unlikely(blk_bidi_rq(rq)))
2931                 bidi_bytes = blk_rq_bytes(rq->next_rq);
2932 
2933         pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2934         BUG_ON(pending);
2935 }
2936 EXPORT_SYMBOL(__blk_end_request_all);
2937 
2938 /**
2939  * __blk_end_request_cur - Helper function to finish the current request chunk.
2940  * @rq: the request to finish the current chunk for
2941  * @error: %0 for success, < %0 for error
2942  *
2943  * Description:
2944  *     Complete the current consecutively mapped chunk from @rq.  Must
2945  *     be called with queue lock held.
2946  *
2947  * Return:
2948  *     %false - we are done with this request
2949  *     %true  - still buffers pending for this request
2950  */
2951 bool __blk_end_request_cur(struct request *rq, int error)
2952 {
2953         return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2954 }
2955 EXPORT_SYMBOL(__blk_end_request_cur);
2956 
2957 /**
2958  * __blk_end_request_err - Finish a request till the next failure boundary.
2959  * @rq: the request to finish till the next failure boundary for
2960  * @error: must be negative errno
2961  *
2962  * Description:
2963  *     Complete @rq till the next failure boundary.  Must be called
2964  *     with queue lock held.
2965  *
2966  * Return:
2967  *     %false - we are done with this request
2968  *     %true  - still buffers pending for this request
2969  */
2970 bool __blk_end_request_err(struct request *rq, int error)
2971 {
2972         WARN_ON(error >= 0);
2973         return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2974 }
2975 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2976 
2977 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2978                      struct bio *bio)
2979 {
2980         /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2981         rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2982 
2983         if (bio_has_data(bio))
2984                 rq->nr_phys_segments = bio_phys_segments(q, bio);
2985 
2986         rq->__data_len = bio->bi_iter.bi_size;
2987         rq->bio = rq->biotail = bio;
2988 
2989         if (bio->bi_bdev)
2990                 rq->rq_disk = bio->bi_bdev->bd_disk;
2991 }
2992 
2993 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2994 /**
2995  * rq_flush_dcache_pages - Helper function to flush all pages in a request
2996  * @rq: the request to be flushed
2997  *
2998  * Description:
2999  *     Flush all pages in @rq.
3000  */
3001 void rq_flush_dcache_pages(struct request *rq)
3002 {
3003         struct req_iterator iter;
3004         struct bio_vec bvec;
3005 
3006         rq_for_each_segment(bvec, rq, iter)
3007                 flush_dcache_page(bvec.bv_page);
3008 }
3009 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3010 #endif
3011 
3012 /**
3013  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3014  * @q : the queue of the device being checked
3015  *
3016  * Description:
3017  *    Check if underlying low-level drivers of a device are busy.
3018  *    If the drivers want to export their busy state, they must set own
3019  *    exporting function using blk_queue_lld_busy() first.
3020  *
3021  *    Basically, this function is used only by request stacking drivers
3022  *    to stop dispatching requests to underlying devices when underlying
3023  *    devices are busy.  This behavior helps more I/O merging on the queue
3024  *    of the request stacking driver and prevents I/O throughput regression
3025  *    on burst I/O load.
3026  *
3027  * Return:
3028  *    0 - Not busy (The request stacking driver should dispatch request)
3029  *    1 - Busy (The request stacking driver should stop dispatching request)
3030  */
3031 int blk_lld_busy(struct request_queue *q)
3032 {
3033         if (q->lld_busy_fn)
3034                 return q->lld_busy_fn(q);
3035 
3036         return 0;
3037 }
3038 EXPORT_SYMBOL_GPL(blk_lld_busy);
3039 
3040 /**
3041  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3042  * @rq: the clone request to be cleaned up
3043  *
3044  * Description:
3045  *     Free all bios in @rq for a cloned request.
3046  */
3047 void blk_rq_unprep_clone(struct request *rq)
3048 {
3049         struct bio *bio;
3050 
3051         while ((bio = rq->bio) != NULL) {
3052                 rq->bio = bio->bi_next;
3053 
3054                 bio_put(bio);
3055         }
3056 }
3057 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3058 
3059 /*
3060  * Copy attributes of the original request to the clone request.
3061  * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3062  */
3063 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3064 {
3065         dst->cpu = src->cpu;
3066         dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
3067         dst->cmd_type = src->cmd_type;
3068         dst->__sector = blk_rq_pos(src);
3069         dst->__data_len = blk_rq_bytes(src);
3070         dst->nr_phys_segments = src->nr_phys_segments;
3071         dst->ioprio = src->ioprio;
3072         dst->extra_len = src->extra_len;
3073 }
3074 
3075 /**
3076  * blk_rq_prep_clone - Helper function to setup clone request
3077  * @rq: the request to be setup
3078  * @rq_src: original request to be cloned
3079  * @bs: bio_set that bios for clone are allocated from
3080  * @gfp_mask: memory allocation mask for bio
3081  * @bio_ctr: setup function to be called for each clone bio.
3082  *           Returns %0 for success, non %0 for failure.
3083  * @data: private data to be passed to @bio_ctr
3084  *
3085  * Description:
3086  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3087  *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3088  *     are not copied, and copying such parts is the caller's responsibility.
3089  *     Also, pages which the original bios are pointing to are not copied
3090  *     and the cloned bios just point same pages.
3091  *     So cloned bios must be completed before original bios, which means
3092  *     the caller must complete @rq before @rq_src.
3093  */
3094 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3095                       struct bio_set *bs, gfp_t gfp_mask,
3096                       int (*bio_ctr)(struct bio *, struct bio *, void *),
3097                       void *data)
3098 {
3099         struct bio *bio, *bio_src;
3100 
3101         if (!bs)
3102                 bs = fs_bio_set;
3103 
3104         __rq_for_each_bio(bio_src, rq_src) {
3105                 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3106                 if (!bio)
3107                         goto free_and_out;
3108 
3109                 if (bio_ctr && bio_ctr(bio, bio_src, data))
3110                         goto free_and_out;
3111 
3112                 if (rq->bio) {
3113                         rq->biotail->bi_next = bio;
3114                         rq->biotail = bio;
3115                 } else
3116                         rq->bio = rq->biotail = bio;
3117         }
3118 
3119         __blk_rq_prep_clone(rq, rq_src);
3120 
3121         return 0;
3122 
3123 free_and_out:
3124         if (bio)
3125                 bio_put(bio);
3126         blk_rq_unprep_clone(rq);
3127 
3128         return -ENOMEM;
3129 }
3130 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3131 
3132 int kblockd_schedule_work(struct work_struct *work)
3133 {
3134         return queue_work(kblockd_workqueue, work);
3135 }
3136 EXPORT_SYMBOL(kblockd_schedule_work);
3137 
3138 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3139                                   unsigned long delay)
3140 {
3141         return queue_delayed_work(kblockd_workqueue, dwork, delay);
3142 }
3143 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3144 
3145 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3146                                      unsigned long delay)
3147 {
3148         return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3149 }
3150 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3151 
3152 /**
3153  * blk_start_plug - initialize blk_plug and track it inside the task_struct
3154  * @plug:       The &struct blk_plug that needs to be initialized
3155  *
3156  * Description:
3157  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
3158  *   pending I/O should the task end up blocking between blk_start_plug() and
3159  *   blk_finish_plug(). This is important from a performance perspective, but
3160  *   also ensures that we don't deadlock. For instance, if the task is blocking
3161  *   for a memory allocation, memory reclaim could end up wanting to free a
3162  *   page belonging to that request that is currently residing in our private
3163  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
3164  *   this kind of deadlock.
3165  */
3166 void blk_start_plug(struct blk_plug *plug)
3167 {
3168         struct task_struct *tsk = current;
3169 
3170         /*
3171          * If this is a nested plug, don't actually assign it.
3172          */
3173         if (tsk->plug)
3174                 return;
3175 
3176         INIT_LIST_HEAD(&plug->list);
3177         INIT_LIST_HEAD(&plug->mq_list);
3178         INIT_LIST_HEAD(&plug->cb_list);
3179         /*
3180          * Store ordering should not be needed here, since a potential
3181          * preempt will imply a full memory barrier
3182          */
3183         tsk->plug = plug;
3184 }
3185 EXPORT_SYMBOL(blk_start_plug);
3186 
3187 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3188 {
3189         struct request *rqa = container_of(a, struct request, queuelist);
3190         struct request *rqb = container_of(b, struct request, queuelist);
3191 
3192         return !(rqa->q < rqb->q ||
3193                 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3194 }
3195 
3196 /*
3197  * If 'from_schedule' is true, then postpone the dispatch of requests
3198  * until a safe kblockd context. We due this to avoid accidental big
3199  * additional stack usage in driver dispatch, in places where the originally
3200  * plugger did not intend it.
3201  */
3202 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3203                             bool from_schedule)
3204         __releases(q->queue_lock)
3205 {
3206         trace_block_unplug(q, depth, !from_schedule);
3207 
3208         if (from_schedule)
3209                 blk_run_queue_async(q);
3210         else
3211                 __blk_run_queue(q);
3212         spin_unlock(q->queue_lock);
3213 }
3214 
3215 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3216 {
3217         LIST_HEAD(callbacks);
3218 
3219         while (!list_empty(&plug->cb_list)) {
3220                 list_splice_init(&plug->cb_list, &callbacks);
3221 
3222                 while (!list_empty(&callbacks)) {
3223                         struct blk_plug_cb *cb = list_first_entry(&callbacks,
3224                                                           struct blk_plug_cb,
3225                                                           list);
3226                         list_del(&cb->list);
3227                         cb->callback(cb, from_schedule);
3228                 }
3229         }
3230 }
3231 
3232 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3233                                       int size)
3234 {
3235         struct blk_plug *plug = current->plug;
3236         struct blk_plug_cb *cb;
3237 
3238         if (!plug)
3239                 return NULL;
3240 
3241         list_for_each_entry(cb, &plug->cb_list, list)
3242                 if (cb->callback == unplug && cb->data == data)
3243                         return cb;
3244 
3245         /* Not currently on the callback list */
3246         BUG_ON(size < sizeof(*cb));
3247         cb = kzalloc(size, GFP_ATOMIC);
3248         if (cb) {
3249                 cb->data = data;
3250                 cb->callback = unplug;
3251                 list_add(&cb->list, &plug->cb_list);
3252         }
3253         return cb;
3254 }
3255 EXPORT_SYMBOL(blk_check_plugged);
3256 
3257 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3258 {
3259         struct request_queue *q;
3260         unsigned long flags;
3261         struct request *rq;
3262         LIST_HEAD(list);
3263         unsigned int depth;
3264 
3265         flush_plug_callbacks(plug, from_schedule);
3266 
3267         if (!list_empty(&plug->mq_list))
3268                 blk_mq_flush_plug_list(plug, from_schedule);
3269 
3270         if (list_empty(&plug->list))
3271                 return;
3272 
3273         list_splice_init(&plug->list, &list);
3274 
3275         list_sort(NULL, &list, plug_rq_cmp);
3276 
3277         q = NULL;
3278         depth = 0;
3279 
3280         /*
3281          * Save and disable interrupts here, to avoid doing it for every
3282          * queue lock we have to take.
3283          */
3284         local_irq_save(flags);
3285         while (!list_empty(&list)) {
3286                 rq = list_entry_rq(list.next);
3287                 list_del_init(&rq->queuelist);
3288                 BUG_ON(!rq->q);
3289                 if (rq->q != q) {
3290                         /*
3291                          * This drops the queue lock
3292                          */
3293                         if (q)
3294                                 queue_unplugged(q, depth, from_schedule);
3295                         q = rq->q;
3296                         depth = 0;
3297                         spin_lock(q->queue_lock);
3298                 }
3299 
3300                 /*
3301                  * Short-circuit if @q is dead
3302                  */
3303                 if (unlikely(blk_queue_dying(q))) {
3304                         __blk_end_request_all(rq, -ENODEV);
3305                         continue;
3306                 }
3307 
3308                 /*
3309                  * rq is already accounted, so use raw insert
3310                  */
3311                 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3312                         __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3313                 else
3314                         __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3315 
3316                 depth++;
3317         }
3318 
3319         /*
3320          * This drops the queue lock
3321          */
3322         if (q)
3323                 queue_unplugged(q, depth, from_schedule);
3324 
3325         local_irq_restore(flags);
3326 }
3327 
3328 void blk_finish_plug(struct blk_plug *plug)
3329 {
3330         if (plug != current->plug)
3331                 return;
3332         blk_flush_plug_list(plug, false);
3333 
3334         current->plug = NULL;
3335 }
3336 EXPORT_SYMBOL(blk_finish_plug);
3337 
3338 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3339 {
3340         struct blk_plug *plug;
3341         long state;
3342 
3343         if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3344             !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3345                 return false;
3346 
3347         plug = current->plug;
3348         if (plug)
3349                 blk_flush_plug_list(plug, false);
3350 
3351         state = current->state;
3352         while (!need_resched()) {
3353                 unsigned int queue_num = blk_qc_t_to_queue_num(cookie);
3354                 struct blk_mq_hw_ctx *hctx = q->queue_hw_ctx[queue_num];
3355                 int ret;
3356 
3357                 hctx->poll_invoked++;
3358 
3359                 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3360                 if (ret > 0) {
3361                         hctx->poll_success++;
3362                         set_current_state(TASK_RUNNING);
3363                         return true;
3364                 }
3365 
3366                 if (signal_pending_state(state, current))
3367                         set_current_state(TASK_RUNNING);
3368 
3369                 if (current->state == TASK_RUNNING)
3370                         return true;
3371                 if (ret < 0)
3372                         break;
3373                 cpu_relax();
3374         }
3375 
3376         return false;
3377 }
3378 
3379 #ifdef CONFIG_PM
3380 /**
3381  * blk_pm_runtime_init - Block layer runtime PM initialization routine
3382  * @q: the queue of the device
3383  * @dev: the device the queue belongs to
3384  *
3385  * Description:
3386  *    Initialize runtime-PM-related fields for @q and start auto suspend for
3387  *    @dev. Drivers that want to take advantage of request-based runtime PM
3388  *    should call this function after @dev has been initialized, and its
3389  *    request queue @q has been allocated, and runtime PM for it can not happen
3390  *    yet(either due to disabled/forbidden or its usage_count > 0). In most
3391  *    cases, driver should call this function before any I/O has taken place.
3392  *
3393  *    This function takes care of setting up using auto suspend for the device,
3394  *    the autosuspend delay is set to -1 to make runtime suspend impossible
3395  *    until an updated value is either set by user or by driver. Drivers do
3396  *    not need to touch other autosuspend settings.
3397  *
3398  *    The block layer runtime PM is request based, so only works for drivers
3399  *    that use request as their IO unit instead of those directly use bio's.
3400  */
3401 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3402 {
3403         q->dev = dev;
3404         q->rpm_status = RPM_ACTIVE;
3405         pm_runtime_set_autosuspend_delay(q->dev, -1);
3406         pm_runtime_use_autosuspend(q->dev);
3407 }
3408 EXPORT_SYMBOL(blk_pm_runtime_init);
3409 
3410 /**
3411  * blk_pre_runtime_suspend - Pre runtime suspend check
3412  * @q: the queue of the device
3413  *
3414  * Description:
3415  *    This function will check if runtime suspend is allowed for the device
3416  *    by examining if there are any requests pending in the queue. If there
3417  *    are requests pending, the device can not be runtime suspended; otherwise,
3418  *    the queue's status will be updated to SUSPENDING and the driver can
3419  *    proceed to suspend the device.
3420  *
3421  *    For the not allowed case, we mark last busy for the device so that
3422  *    runtime PM core will try to autosuspend it some time later.
3423  *
3424  *    This function should be called near the start of the device's
3425  *    runtime_suspend callback.
3426  *
3427  * Return:
3428  *    0         - OK to runtime suspend the device
3429  *    -EBUSY    - Device should not be runtime suspended
3430  */
3431 int blk_pre_runtime_suspend(struct request_queue *q)
3432 {
3433         int ret = 0;
3434 
3435         if (!q->dev)
3436                 return ret;
3437 
3438         spin_lock_irq(q->queue_lock);
3439         if (q->nr_pending) {
3440                 ret = -EBUSY;
3441                 pm_runtime_mark_last_busy(q->dev);
3442         } else {
3443                 q->rpm_status = RPM_SUSPENDING;
3444         }
3445         spin_unlock_irq(q->queue_lock);
3446         return ret;
3447 }
3448 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3449 
3450 /**
3451  * blk_post_runtime_suspend - Post runtime suspend processing
3452  * @q: the queue of the device
3453  * @err: return value of the device's runtime_suspend function
3454  *
3455  * Description:
3456  *    Update the queue's runtime status according to the return value of the
3457  *    device's runtime suspend function and mark last busy for the device so
3458  *    that PM core will try to auto suspend the device at a later time.
3459  *
3460  *    This function should be called near the end of the device's
3461  *    runtime_suspend callback.
3462  */
3463 void blk_post_runtime_suspend(struct request_queue *q, int err)
3464 {
3465         if (!q->dev)
3466                 return;
3467 
3468         spin_lock_irq(q->queue_lock);
3469         if (!err) {
3470                 q->rpm_status = RPM_SUSPENDED;
3471         } else {
3472                 q->rpm_status = RPM_ACTIVE;
3473                 pm_runtime_mark_last_busy(q->dev);
3474         }
3475         spin_unlock_irq(q->queue_lock);
3476 }
3477 EXPORT_SYMBOL(blk_post_runtime_suspend);
3478 
3479 /**
3480  * blk_pre_runtime_resume - Pre runtime resume processing
3481  * @q: the queue of the device
3482  *
3483  * Description:
3484  *    Update the queue's runtime status to RESUMING in preparation for the
3485  *    runtime resume of the device.
3486  *
3487  *    This function should be called near the start of the device's
3488  *    runtime_resume callback.
3489  */
3490 void blk_pre_runtime_resume(struct request_queue *q)
3491 {
3492         if (!q->dev)
3493                 return;
3494 
3495         spin_lock_irq(q->queue_lock);
3496         q->rpm_status = RPM_RESUMING;
3497         spin_unlock_irq(q->queue_lock);
3498 }
3499 EXPORT_SYMBOL(blk_pre_runtime_resume);
3500 
3501 /**
3502  * blk_post_runtime_resume - Post runtime resume processing
3503  * @q: the queue of the device
3504  * @err: return value of the device's runtime_resume function
3505  *
3506  * Description:
3507  *    Update the queue's runtime status according to the return value of the
3508  *    device's runtime_resume function. If it is successfully resumed, process
3509  *    the requests that are queued into the device's queue when it is resuming
3510  *    and then mark last busy and initiate autosuspend for it.
3511  *
3512  *    This function should be called near the end of the device's
3513  *    runtime_resume callback.
3514  */
3515 void blk_post_runtime_resume(struct request_queue *q, int err)
3516 {
3517         if (!q->dev)
3518                 return;
3519 
3520         spin_lock_irq(q->queue_lock);
3521         if (!err) {
3522                 q->rpm_status = RPM_ACTIVE;
3523                 __blk_run_queue(q);
3524                 pm_runtime_mark_last_busy(q->dev);
3525                 pm_request_autosuspend(q->dev);
3526         } else {
3527                 q->rpm_status = RPM_SUSPENDED;
3528         }
3529         spin_unlock_irq(q->queue_lock);
3530 }
3531 EXPORT_SYMBOL(blk_post_runtime_resume);
3532 #endif
3533 
3534 int __init blk_dev_init(void)
3535 {
3536         BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3537                         FIELD_SIZEOF(struct request, cmd_flags));
3538 
3539         /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3540         kblockd_workqueue = alloc_workqueue("kblockd",
3541                                             WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3542         if (!kblockd_workqueue)
3543                 panic("Failed to create kblockd\n");
3544 
3545         request_cachep = kmem_cache_create("blkdev_requests",
3546                         sizeof(struct request), 0, SLAB_PANIC, NULL);
3547 
3548         blk_requestq_cachep = kmem_cache_create("request_queue",
3549                         sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3550 
3551         return 0;
3552 }
3553 

~ [ 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