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

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

Version: ~ [ linux-5.6 ] ~ [ linux-5.5.13 ] ~ [ linux-5.4.28 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.113 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.174 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.217 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.217 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.82 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ 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/highmem.h>
 20 #include <linux/mm.h>
 21 #include <linux/kernel_stat.h>
 22 #include <linux/string.h>
 23 #include <linux/init.h>
 24 #include <linux/completion.h>
 25 #include <linux/slab.h>
 26 #include <linux/swap.h>
 27 #include <linux/writeback.h>
 28 #include <linux/task_io_accounting_ops.h>
 29 #include <linux/fault-inject.h>
 30 #include <linux/list_sort.h>
 31 #include <linux/delay.h>
 32 
 33 #define CREATE_TRACE_POINTS
 34 #include <trace/events/block.h>
 35 
 36 #include "blk.h"
 37 
 38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
 39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
 40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
 41 
 42 /*
 43  * For the allocated request tables
 44  */
 45 static struct kmem_cache *request_cachep;
 46 
 47 /*
 48  * For queue allocation
 49  */
 50 struct kmem_cache *blk_requestq_cachep;
 51 
 52 /*
 53  * Controlling structure to kblockd
 54  */
 55 static struct workqueue_struct *kblockd_workqueue;
 56 
 57 static void drive_stat_acct(struct request *rq, int new_io)
 58 {
 59         struct hd_struct *part;
 60         int rw = rq_data_dir(rq);
 61         int cpu;
 62 
 63         if (!blk_do_io_stat(rq))
 64                 return;
 65 
 66         cpu = part_stat_lock();
 67 
 68         if (!new_io) {
 69                 part = rq->part;
 70                 part_stat_inc(cpu, part, merges[rw]);
 71         } else {
 72                 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
 73                 if (!hd_struct_try_get(part)) {
 74                         /*
 75                          * The partition is already being removed,
 76                          * the request will be accounted on the disk only
 77                          *
 78                          * We take a reference on disk->part0 although that
 79                          * partition will never be deleted, so we can treat
 80                          * it as any other partition.
 81                          */
 82                         part = &rq->rq_disk->part0;
 83                         hd_struct_get(part);
 84                 }
 85                 part_round_stats(cpu, part);
 86                 part_inc_in_flight(part, rw);
 87                 rq->part = part;
 88         }
 89 
 90         part_stat_unlock();
 91 }
 92 
 93 void blk_queue_congestion_threshold(struct request_queue *q)
 94 {
 95         int nr;
 96 
 97         nr = q->nr_requests - (q->nr_requests / 8) + 1;
 98         if (nr > q->nr_requests)
 99                 nr = q->nr_requests;
100         q->nr_congestion_on = nr;
101 
102         nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
103         if (nr < 1)
104                 nr = 1;
105         q->nr_congestion_off = nr;
106 }
107 
108 /**
109  * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
110  * @bdev:       device
111  *
112  * Locates the passed device's request queue and returns the address of its
113  * backing_dev_info
114  *
115  * Will return NULL if the request queue cannot be located.
116  */
117 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
118 {
119         struct backing_dev_info *ret = NULL;
120         struct request_queue *q = bdev_get_queue(bdev);
121 
122         if (q)
123                 ret = &q->backing_dev_info;
124         return ret;
125 }
126 EXPORT_SYMBOL(blk_get_backing_dev_info);
127 
128 void blk_rq_init(struct request_queue *q, struct request *rq)
129 {
130         memset(rq, 0, sizeof(*rq));
131 
132         INIT_LIST_HEAD(&rq->queuelist);
133         INIT_LIST_HEAD(&rq->timeout_list);
134         rq->cpu = -1;
135         rq->q = q;
136         rq->__sector = (sector_t) -1;
137         INIT_HLIST_NODE(&rq->hash);
138         RB_CLEAR_NODE(&rq->rb_node);
139         rq->cmd = rq->__cmd;
140         rq->cmd_len = BLK_MAX_CDB;
141         rq->tag = -1;
142         rq->ref_count = 1;
143         rq->start_time = jiffies;
144         set_start_time_ns(rq);
145         rq->part = NULL;
146 }
147 EXPORT_SYMBOL(blk_rq_init);
148 
149 static void req_bio_endio(struct request *rq, struct bio *bio,
150                           unsigned int nbytes, int error)
151 {
152         if (error)
153                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
154         else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
155                 error = -EIO;
156 
157         if (unlikely(nbytes > bio->bi_size)) {
158                 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
159                        __func__, nbytes, bio->bi_size);
160                 nbytes = bio->bi_size;
161         }
162 
163         if (unlikely(rq->cmd_flags & REQ_QUIET))
164                 set_bit(BIO_QUIET, &bio->bi_flags);
165 
166         bio->bi_size -= nbytes;
167         bio->bi_sector += (nbytes >> 9);
168 
169         if (bio_integrity(bio))
170                 bio_integrity_advance(bio, nbytes);
171 
172         /* don't actually finish bio if it's part of flush sequence */
173         if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
174                 bio_endio(bio, error);
175 }
176 
177 void blk_dump_rq_flags(struct request *rq, char *msg)
178 {
179         int bit;
180 
181         printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
182                 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
183                 rq->cmd_flags);
184 
185         printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
186                (unsigned long long)blk_rq_pos(rq),
187                blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
188         printk(KERN_INFO "  bio %p, biotail %p, buffer %p, len %u\n",
189                rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
190 
191         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
192                 printk(KERN_INFO "  cdb: ");
193                 for (bit = 0; bit < BLK_MAX_CDB; bit++)
194                         printk("%02x ", rq->cmd[bit]);
195                 printk("\n");
196         }
197 }
198 EXPORT_SYMBOL(blk_dump_rq_flags);
199 
200 static void blk_delay_work(struct work_struct *work)
201 {
202         struct request_queue *q;
203 
204         q = container_of(work, struct request_queue, delay_work.work);
205         spin_lock_irq(q->queue_lock);
206         __blk_run_queue(q);
207         spin_unlock_irq(q->queue_lock);
208 }
209 
210 /**
211  * blk_delay_queue - restart queueing after defined interval
212  * @q:          The &struct request_queue in question
213  * @msecs:      Delay in msecs
214  *
215  * Description:
216  *   Sometimes queueing needs to be postponed for a little while, to allow
217  *   resources to come back. This function will make sure that queueing is
218  *   restarted around the specified time.
219  */
220 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
221 {
222         queue_delayed_work(kblockd_workqueue, &q->delay_work,
223                                 msecs_to_jiffies(msecs));
224 }
225 EXPORT_SYMBOL(blk_delay_queue);
226 
227 /**
228  * blk_start_queue - restart a previously stopped queue
229  * @q:    The &struct request_queue in question
230  *
231  * Description:
232  *   blk_start_queue() will clear the stop flag on the queue, and call
233  *   the request_fn for the queue if it was in a stopped state when
234  *   entered. Also see blk_stop_queue(). Queue lock must be held.
235  **/
236 void blk_start_queue(struct request_queue *q)
237 {
238         WARN_ON(!in_interrupt() && !irqs_disabled());
239 
240         queue_flag_clear(QUEUE_FLAG_STOPPED, q);
241         __blk_run_queue(q);
242 }
243 EXPORT_SYMBOL(blk_start_queue);
244 
245 /**
246  * blk_stop_queue - stop a queue
247  * @q:    The &struct request_queue in question
248  *
249  * Description:
250  *   The Linux block layer assumes that a block driver will consume all
251  *   entries on the request queue when the request_fn strategy is called.
252  *   Often this will not happen, because of hardware limitations (queue
253  *   depth settings). If a device driver gets a 'queue full' response,
254  *   or if it simply chooses not to queue more I/O at one point, it can
255  *   call this function to prevent the request_fn from being called until
256  *   the driver has signalled it's ready to go again. This happens by calling
257  *   blk_start_queue() to restart queue operations. Queue lock must be held.
258  **/
259 void blk_stop_queue(struct request_queue *q)
260 {
261         __cancel_delayed_work(&q->delay_work);
262         queue_flag_set(QUEUE_FLAG_STOPPED, q);
263 }
264 EXPORT_SYMBOL(blk_stop_queue);
265 
266 /**
267  * blk_sync_queue - cancel any pending callbacks on a queue
268  * @q: the queue
269  *
270  * Description:
271  *     The block layer may perform asynchronous callback activity
272  *     on a queue, such as calling the unplug function after a timeout.
273  *     A block device may call blk_sync_queue to ensure that any
274  *     such activity is cancelled, thus allowing it to release resources
275  *     that the callbacks might use. The caller must already have made sure
276  *     that its ->make_request_fn will not re-add plugging prior to calling
277  *     this function.
278  *
279  *     This function does not cancel any asynchronous activity arising
280  *     out of elevator or throttling code. That would require elevaotor_exit()
281  *     and blk_throtl_exit() to be called with queue lock initialized.
282  *
283  */
284 void blk_sync_queue(struct request_queue *q)
285 {
286         del_timer_sync(&q->timeout);
287         cancel_delayed_work_sync(&q->delay_work);
288 }
289 EXPORT_SYMBOL(blk_sync_queue);
290 
291 /**
292  * __blk_run_queue - run a single device queue
293  * @q:  The queue to run
294  *
295  * Description:
296  *    See @blk_run_queue. This variant must be called with the queue lock
297  *    held and interrupts disabled.
298  */
299 void __blk_run_queue(struct request_queue *q)
300 {
301         if (unlikely(blk_queue_stopped(q)))
302                 return;
303 
304         q->request_fn(q);
305 }
306 EXPORT_SYMBOL(__blk_run_queue);
307 
308 /**
309  * blk_run_queue_async - run a single device queue in workqueue context
310  * @q:  The queue to run
311  *
312  * Description:
313  *    Tells kblockd to perform the equivalent of @blk_run_queue on behalf
314  *    of us.
315  */
316 void blk_run_queue_async(struct request_queue *q)
317 {
318         if (likely(!blk_queue_stopped(q))) {
319                 __cancel_delayed_work(&q->delay_work);
320                 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
321         }
322 }
323 EXPORT_SYMBOL(blk_run_queue_async);
324 
325 /**
326  * blk_run_queue - run a single device queue
327  * @q: The queue to run
328  *
329  * Description:
330  *    Invoke request handling on this queue, if it has pending work to do.
331  *    May be used to restart queueing when a request has completed.
332  */
333 void blk_run_queue(struct request_queue *q)
334 {
335         unsigned long flags;
336 
337         spin_lock_irqsave(q->queue_lock, flags);
338         __blk_run_queue(q);
339         spin_unlock_irqrestore(q->queue_lock, flags);
340 }
341 EXPORT_SYMBOL(blk_run_queue);
342 
343 void blk_put_queue(struct request_queue *q)
344 {
345         kobject_put(&q->kobj);
346 }
347 EXPORT_SYMBOL(blk_put_queue);
348 
349 /**
350  * blk_drain_queue - drain requests from request_queue
351  * @q: queue to drain
352  * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
353  *
354  * Drain requests from @q.  If @drain_all is set, all requests are drained.
355  * If not, only ELVPRIV requests are drained.  The caller is responsible
356  * for ensuring that no new requests which need to be drained are queued.
357  */
358 void blk_drain_queue(struct request_queue *q, bool drain_all)
359 {
360         while (true) {
361                 int nr_rqs;
362 
363                 spin_lock_irq(q->queue_lock);
364 
365                 elv_drain_elevator(q);
366                 if (drain_all)
367                         blk_throtl_drain(q);
368 
369                 /*
370                  * This function might be called on a queue which failed
371                  * driver init after queue creation.  Some drivers
372                  * (e.g. fd) get unhappy in such cases.  Kick queue iff
373                  * dispatch queue has something on it.
374                  */
375                 if (!list_empty(&q->queue_head))
376                         __blk_run_queue(q);
377 
378                 if (drain_all)
379                         nr_rqs = q->rq.count[0] + q->rq.count[1];
380                 else
381                         nr_rqs = q->rq.elvpriv;
382 
383                 spin_unlock_irq(q->queue_lock);
384 
385                 if (!nr_rqs)
386                         break;
387                 msleep(10);
388         }
389 }
390 
391 /**
392  * blk_cleanup_queue - shutdown a request queue
393  * @q: request queue to shutdown
394  *
395  * Mark @q DEAD, drain all pending requests, destroy and put it.  All
396  * future requests will be failed immediately with -ENODEV.
397  */
398 void blk_cleanup_queue(struct request_queue *q)
399 {
400         spinlock_t *lock = q->queue_lock;
401 
402         /* mark @q DEAD, no new request or merges will be allowed afterwards */
403         mutex_lock(&q->sysfs_lock);
404         queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
405 
406         spin_lock_irq(lock);
407         queue_flag_set(QUEUE_FLAG_NOMERGES, q);
408         queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
409         queue_flag_set(QUEUE_FLAG_DEAD, q);
410 
411         if (q->queue_lock != &q->__queue_lock)
412                 q->queue_lock = &q->__queue_lock;
413 
414         spin_unlock_irq(lock);
415         mutex_unlock(&q->sysfs_lock);
416 
417         /*
418          * Drain all requests queued before DEAD marking.  The caller might
419          * be trying to tear down @q before its elevator is initialized, in
420          * which case we don't want to call into draining.
421          */
422         if (q->elevator)
423                 blk_drain_queue(q, true);
424 
425         /* @q won't process any more request, flush async actions */
426         del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
427         blk_sync_queue(q);
428 
429         /* @q is and will stay empty, shutdown and put */
430         blk_put_queue(q);
431 }
432 EXPORT_SYMBOL(blk_cleanup_queue);
433 
434 static int blk_init_free_list(struct request_queue *q)
435 {
436         struct request_list *rl = &q->rq;
437 
438         if (unlikely(rl->rq_pool))
439                 return 0;
440 
441         rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
442         rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
443         rl->elvpriv = 0;
444         init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
445         init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
446 
447         rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
448                                 mempool_free_slab, request_cachep, q->node);
449 
450         if (!rl->rq_pool)
451                 return -ENOMEM;
452 
453         return 0;
454 }
455 
456 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
457 {
458         return blk_alloc_queue_node(gfp_mask, -1);
459 }
460 EXPORT_SYMBOL(blk_alloc_queue);
461 
462 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
463 {
464         struct request_queue *q;
465         int err;
466 
467         q = kmem_cache_alloc_node(blk_requestq_cachep,
468                                 gfp_mask | __GFP_ZERO, node_id);
469         if (!q)
470                 return NULL;
471 
472         q->backing_dev_info.ra_pages =
473                         (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
474         q->backing_dev_info.state = 0;
475         q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
476         q->backing_dev_info.name = "block";
477         q->node = node_id;
478 
479         err = bdi_init(&q->backing_dev_info);
480         if (err) {
481                 kmem_cache_free(blk_requestq_cachep, q);
482                 return NULL;
483         }
484 
485         if (blk_throtl_init(q)) {
486                 bdi_destroy(&q->backing_dev_info);
487                 kmem_cache_free(blk_requestq_cachep, q);
488                 return NULL;
489         }
490 
491         setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
492                     laptop_mode_timer_fn, (unsigned long) q);
493         setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
494         INIT_LIST_HEAD(&q->timeout_list);
495         INIT_LIST_HEAD(&q->flush_queue[0]);
496         INIT_LIST_HEAD(&q->flush_queue[1]);
497         INIT_LIST_HEAD(&q->flush_data_in_flight);
498         INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
499 
500         kobject_init(&q->kobj, &blk_queue_ktype);
501 
502 #ifdef CONFIG_BLK_DEV_IO_TRACE
503         mutex_init(&q->blk_trace_mutex);
504 #endif
505         mutex_init(&q->sysfs_lock);
506         spin_lock_init(&q->__queue_lock);
507 
508         /*
509          * By default initialize queue_lock to internal lock and driver can
510          * override it later if need be.
511          */
512         q->queue_lock = &q->__queue_lock;
513 
514         return q;
515 }
516 EXPORT_SYMBOL(blk_alloc_queue_node);
517 
518 /**
519  * blk_init_queue  - prepare a request queue for use with a block device
520  * @rfn:  The function to be called to process requests that have been
521  *        placed on the queue.
522  * @lock: Request queue spin lock
523  *
524  * Description:
525  *    If a block device wishes to use the standard request handling procedures,
526  *    which sorts requests and coalesces adjacent requests, then it must
527  *    call blk_init_queue().  The function @rfn will be called when there
528  *    are requests on the queue that need to be processed.  If the device
529  *    supports plugging, then @rfn may not be called immediately when requests
530  *    are available on the queue, but may be called at some time later instead.
531  *    Plugged queues are generally unplugged when a buffer belonging to one
532  *    of the requests on the queue is needed, or due to memory pressure.
533  *
534  *    @rfn is not required, or even expected, to remove all requests off the
535  *    queue, but only as many as it can handle at a time.  If it does leave
536  *    requests on the queue, it is responsible for arranging that the requests
537  *    get dealt with eventually.
538  *
539  *    The queue spin lock must be held while manipulating the requests on the
540  *    request queue; this lock will be taken also from interrupt context, so irq
541  *    disabling is needed for it.
542  *
543  *    Function returns a pointer to the initialized request queue, or %NULL if
544  *    it didn't succeed.
545  *
546  * Note:
547  *    blk_init_queue() must be paired with a blk_cleanup_queue() call
548  *    when the block device is deactivated (such as at module unload).
549  **/
550 
551 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
552 {
553         return blk_init_queue_node(rfn, lock, -1);
554 }
555 EXPORT_SYMBOL(blk_init_queue);
556 
557 struct request_queue *
558 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
559 {
560         struct request_queue *uninit_q, *q;
561 
562         uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
563         if (!uninit_q)
564                 return NULL;
565 
566         q = blk_init_allocated_queue(uninit_q, rfn, lock);
567         if (!q)
568                 blk_cleanup_queue(uninit_q);
569 
570         return q;
571 }
572 EXPORT_SYMBOL(blk_init_queue_node);
573 
574 struct request_queue *
575 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
576                          spinlock_t *lock)
577 {
578         if (!q)
579                 return NULL;
580 
581         if (blk_init_free_list(q))
582                 return NULL;
583 
584         q->request_fn           = rfn;
585         q->prep_rq_fn           = NULL;
586         q->unprep_rq_fn         = NULL;
587         q->queue_flags          = QUEUE_FLAG_DEFAULT;
588 
589         /* Override internal queue lock with supplied lock pointer */
590         if (lock)
591                 q->queue_lock           = lock;
592 
593         /*
594          * This also sets hw/phys segments, boundary and size
595          */
596         blk_queue_make_request(q, blk_queue_bio);
597 
598         q->sg_reserved_size = INT_MAX;
599 
600         /*
601          * all done
602          */
603         if (!elevator_init(q, NULL)) {
604                 blk_queue_congestion_threshold(q);
605                 return q;
606         }
607 
608         return NULL;
609 }
610 EXPORT_SYMBOL(blk_init_allocated_queue);
611 
612 int blk_get_queue(struct request_queue *q)
613 {
614         if (likely(!blk_queue_dead(q))) {
615                 kobject_get(&q->kobj);
616                 return 0;
617         }
618 
619         return 1;
620 }
621 EXPORT_SYMBOL(blk_get_queue);
622 
623 static inline void blk_free_request(struct request_queue *q, struct request *rq)
624 {
625         if (rq->cmd_flags & REQ_ELVPRIV)
626                 elv_put_request(q, rq);
627         mempool_free(rq, q->rq.rq_pool);
628 }
629 
630 static struct request *
631 blk_alloc_request(struct request_queue *q, unsigned int flags, gfp_t gfp_mask)
632 {
633         struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
634 
635         if (!rq)
636                 return NULL;
637 
638         blk_rq_init(q, rq);
639 
640         rq->cmd_flags = flags | REQ_ALLOCED;
641 
642         if ((flags & REQ_ELVPRIV) &&
643             unlikely(elv_set_request(q, rq, gfp_mask))) {
644                 mempool_free(rq, q->rq.rq_pool);
645                 return NULL;
646         }
647 
648         return rq;
649 }
650 
651 /*
652  * ioc_batching returns true if the ioc is a valid batching request and
653  * should be given priority access to a request.
654  */
655 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
656 {
657         if (!ioc)
658                 return 0;
659 
660         /*
661          * Make sure the process is able to allocate at least 1 request
662          * even if the batch times out, otherwise we could theoretically
663          * lose wakeups.
664          */
665         return ioc->nr_batch_requests == q->nr_batching ||
666                 (ioc->nr_batch_requests > 0
667                 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
668 }
669 
670 /*
671  * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
672  * will cause the process to be a "batcher" on all queues in the system. This
673  * is the behaviour we want though - once it gets a wakeup it should be given
674  * a nice run.
675  */
676 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
677 {
678         if (!ioc || ioc_batching(q, ioc))
679                 return;
680 
681         ioc->nr_batch_requests = q->nr_batching;
682         ioc->last_waited = jiffies;
683 }
684 
685 static void __freed_request(struct request_queue *q, int sync)
686 {
687         struct request_list *rl = &q->rq;
688 
689         if (rl->count[sync] < queue_congestion_off_threshold(q))
690                 blk_clear_queue_congested(q, sync);
691 
692         if (rl->count[sync] + 1 <= q->nr_requests) {
693                 if (waitqueue_active(&rl->wait[sync]))
694                         wake_up(&rl->wait[sync]);
695 
696                 blk_clear_queue_full(q, sync);
697         }
698 }
699 
700 /*
701  * A request has just been released.  Account for it, update the full and
702  * congestion status, wake up any waiters.   Called under q->queue_lock.
703  */
704 static void freed_request(struct request_queue *q, unsigned int flags)
705 {
706         struct request_list *rl = &q->rq;
707         int sync = rw_is_sync(flags);
708 
709         rl->count[sync]--;
710         if (flags & REQ_ELVPRIV)
711                 rl->elvpriv--;
712 
713         __freed_request(q, sync);
714 
715         if (unlikely(rl->starved[sync ^ 1]))
716                 __freed_request(q, sync ^ 1);
717 }
718 
719 /*
720  * Determine if elevator data should be initialized when allocating the
721  * request associated with @bio.
722  */
723 static bool blk_rq_should_init_elevator(struct bio *bio)
724 {
725         if (!bio)
726                 return true;
727 
728         /*
729          * Flush requests do not use the elevator so skip initialization.
730          * This allows a request to share the flush and elevator data.
731          */
732         if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
733                 return false;
734 
735         return true;
736 }
737 
738 /**
739  * get_request - get a free request
740  * @q: request_queue to allocate request from
741  * @rw_flags: RW and SYNC flags
742  * @bio: bio to allocate request for (can be %NULL)
743  * @gfp_mask: allocation mask
744  *
745  * Get a free request from @q.  This function may fail under memory
746  * pressure or if @q is dead.
747  *
748  * Must be callled with @q->queue_lock held and,
749  * Returns %NULL on failure, with @q->queue_lock held.
750  * Returns !%NULL on success, with @q->queue_lock *not held*.
751  */
752 static struct request *get_request(struct request_queue *q, int rw_flags,
753                                    struct bio *bio, gfp_t gfp_mask)
754 {
755         struct request *rq = NULL;
756         struct request_list *rl = &q->rq;
757         struct io_context *ioc = NULL;
758         const bool is_sync = rw_is_sync(rw_flags) != 0;
759         int may_queue;
760 
761         if (unlikely(blk_queue_dead(q)))
762                 return NULL;
763 
764         may_queue = elv_may_queue(q, rw_flags);
765         if (may_queue == ELV_MQUEUE_NO)
766                 goto rq_starved;
767 
768         if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
769                 if (rl->count[is_sync]+1 >= q->nr_requests) {
770                         ioc = current_io_context(GFP_ATOMIC, q->node);
771                         /*
772                          * The queue will fill after this allocation, so set
773                          * it as full, and mark this process as "batching".
774                          * This process will be allowed to complete a batch of
775                          * requests, others will be blocked.
776                          */
777                         if (!blk_queue_full(q, is_sync)) {
778                                 ioc_set_batching(q, ioc);
779                                 blk_set_queue_full(q, is_sync);
780                         } else {
781                                 if (may_queue != ELV_MQUEUE_MUST
782                                                 && !ioc_batching(q, ioc)) {
783                                         /*
784                                          * The queue is full and the allocating
785                                          * process is not a "batcher", and not
786                                          * exempted by the IO scheduler
787                                          */
788                                         goto out;
789                                 }
790                         }
791                 }
792                 blk_set_queue_congested(q, is_sync);
793         }
794 
795         /*
796          * Only allow batching queuers to allocate up to 50% over the defined
797          * limit of requests, otherwise we could have thousands of requests
798          * allocated with any setting of ->nr_requests
799          */
800         if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
801                 goto out;
802 
803         rl->count[is_sync]++;
804         rl->starved[is_sync] = 0;
805 
806         if (blk_rq_should_init_elevator(bio) &&
807             !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
808                 rw_flags |= REQ_ELVPRIV;
809                 rl->elvpriv++;
810         }
811 
812         if (blk_queue_io_stat(q))
813                 rw_flags |= REQ_IO_STAT;
814         spin_unlock_irq(q->queue_lock);
815 
816         rq = blk_alloc_request(q, rw_flags, gfp_mask);
817         if (unlikely(!rq)) {
818                 /*
819                  * Allocation failed presumably due to memory. Undo anything
820                  * we might have messed up.
821                  *
822                  * Allocating task should really be put onto the front of the
823                  * wait queue, but this is pretty rare.
824                  */
825                 spin_lock_irq(q->queue_lock);
826                 freed_request(q, rw_flags);
827 
828                 /*
829                  * in the very unlikely event that allocation failed and no
830                  * requests for this direction was pending, mark us starved
831                  * so that freeing of a request in the other direction will
832                  * notice us. another possible fix would be to split the
833                  * rq mempool into READ and WRITE
834                  */
835 rq_starved:
836                 if (unlikely(rl->count[is_sync] == 0))
837                         rl->starved[is_sync] = 1;
838 
839                 goto out;
840         }
841 
842         /*
843          * ioc may be NULL here, and ioc_batching will be false. That's
844          * OK, if the queue is under the request limit then requests need
845          * not count toward the nr_batch_requests limit. There will always
846          * be some limit enforced by BLK_BATCH_TIME.
847          */
848         if (ioc_batching(q, ioc))
849                 ioc->nr_batch_requests--;
850 
851         trace_block_getrq(q, bio, rw_flags & 1);
852 out:
853         return rq;
854 }
855 
856 /**
857  * get_request_wait - get a free request with retry
858  * @q: request_queue to allocate request from
859  * @rw_flags: RW and SYNC flags
860  * @bio: bio to allocate request for (can be %NULL)
861  *
862  * Get a free request from @q.  This function keeps retrying under memory
863  * pressure and fails iff @q is dead.
864  *
865  * Must be callled with @q->queue_lock held and,
866  * Returns %NULL on failure, with @q->queue_lock held.
867  * Returns !%NULL on success, with @q->queue_lock *not held*.
868  */
869 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
870                                         struct bio *bio)
871 {
872         const bool is_sync = rw_is_sync(rw_flags) != 0;
873         struct request *rq;
874 
875         rq = get_request(q, rw_flags, bio, GFP_NOIO);
876         while (!rq) {
877                 DEFINE_WAIT(wait);
878                 struct io_context *ioc;
879                 struct request_list *rl = &q->rq;
880 
881                 if (unlikely(blk_queue_dead(q)))
882                         return NULL;
883 
884                 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
885                                 TASK_UNINTERRUPTIBLE);
886 
887                 trace_block_sleeprq(q, bio, rw_flags & 1);
888 
889                 spin_unlock_irq(q->queue_lock);
890                 io_schedule();
891 
892                 /*
893                  * After sleeping, we become a "batching" process and
894                  * will be able to allocate at least one request, and
895                  * up to a big batch of them for a small period time.
896                  * See ioc_batching, ioc_set_batching
897                  */
898                 ioc = current_io_context(GFP_NOIO, q->node);
899                 ioc_set_batching(q, ioc);
900 
901                 spin_lock_irq(q->queue_lock);
902                 finish_wait(&rl->wait[is_sync], &wait);
903 
904                 rq = get_request(q, rw_flags, bio, GFP_NOIO);
905         };
906 
907         return rq;
908 }
909 
910 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
911 {
912         struct request *rq;
913 
914         BUG_ON(rw != READ && rw != WRITE);
915 
916         spin_lock_irq(q->queue_lock);
917         if (gfp_mask & __GFP_WAIT)
918                 rq = get_request_wait(q, rw, NULL);
919         else
920                 rq = get_request(q, rw, NULL, gfp_mask);
921         if (!rq)
922                 spin_unlock_irq(q->queue_lock);
923         /* q->queue_lock is unlocked at this point */
924 
925         return rq;
926 }
927 EXPORT_SYMBOL(blk_get_request);
928 
929 /**
930  * blk_make_request - given a bio, allocate a corresponding struct request.
931  * @q: target request queue
932  * @bio:  The bio describing the memory mappings that will be submitted for IO.
933  *        It may be a chained-bio properly constructed by block/bio layer.
934  * @gfp_mask: gfp flags to be used for memory allocation
935  *
936  * blk_make_request is the parallel of generic_make_request for BLOCK_PC
937  * type commands. Where the struct request needs to be farther initialized by
938  * the caller. It is passed a &struct bio, which describes the memory info of
939  * the I/O transfer.
940  *
941  * The caller of blk_make_request must make sure that bi_io_vec
942  * are set to describe the memory buffers. That bio_data_dir() will return
943  * the needed direction of the request. (And all bio's in the passed bio-chain
944  * are properly set accordingly)
945  *
946  * If called under none-sleepable conditions, mapped bio buffers must not
947  * need bouncing, by calling the appropriate masked or flagged allocator,
948  * suitable for the target device. Otherwise the call to blk_queue_bounce will
949  * BUG.
950  *
951  * WARNING: When allocating/cloning a bio-chain, careful consideration should be
952  * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
953  * anything but the first bio in the chain. Otherwise you risk waiting for IO
954  * completion of a bio that hasn't been submitted yet, thus resulting in a
955  * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
956  * of bio_alloc(), as that avoids the mempool deadlock.
957  * If possible a big IO should be split into smaller parts when allocation
958  * fails. Partial allocation should not be an error, or you risk a live-lock.
959  */
960 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
961                                  gfp_t gfp_mask)
962 {
963         struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
964 
965         if (unlikely(!rq))
966                 return ERR_PTR(-ENOMEM);
967 
968         for_each_bio(bio) {
969                 struct bio *bounce_bio = bio;
970                 int ret;
971 
972                 blk_queue_bounce(q, &bounce_bio);
973                 ret = blk_rq_append_bio(q, rq, bounce_bio);
974                 if (unlikely(ret)) {
975                         blk_put_request(rq);
976                         return ERR_PTR(ret);
977                 }
978         }
979 
980         return rq;
981 }
982 EXPORT_SYMBOL(blk_make_request);
983 
984 /**
985  * blk_requeue_request - put a request back on queue
986  * @q:          request queue where request should be inserted
987  * @rq:         request to be inserted
988  *
989  * Description:
990  *    Drivers often keep queueing requests until the hardware cannot accept
991  *    more, when that condition happens we need to put the request back
992  *    on the queue. Must be called with queue lock held.
993  */
994 void blk_requeue_request(struct request_queue *q, struct request *rq)
995 {
996         blk_delete_timer(rq);
997         blk_clear_rq_complete(rq);
998         trace_block_rq_requeue(q, rq);
999 
1000         if (blk_rq_tagged(rq))
1001                 blk_queue_end_tag(q, rq);
1002 
1003         BUG_ON(blk_queued_rq(rq));
1004 
1005         elv_requeue_request(q, rq);
1006 }
1007 EXPORT_SYMBOL(blk_requeue_request);
1008 
1009 static void add_acct_request(struct request_queue *q, struct request *rq,
1010                              int where)
1011 {
1012         drive_stat_acct(rq, 1);
1013         __elv_add_request(q, rq, where);
1014 }
1015 
1016 /**
1017  * blk_insert_request - insert a special request into a request queue
1018  * @q:          request queue where request should be inserted
1019  * @rq:         request to be inserted
1020  * @at_head:    insert request at head or tail of queue
1021  * @data:       private data
1022  *
1023  * Description:
1024  *    Many block devices need to execute commands asynchronously, so they don't
1025  *    block the whole kernel from preemption during request execution.  This is
1026  *    accomplished normally by inserting aritficial requests tagged as
1027  *    REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1028  *    be scheduled for actual execution by the request queue.
1029  *
1030  *    We have the option of inserting the head or the tail of the queue.
1031  *    Typically we use the tail for new ioctls and so forth.  We use the head
1032  *    of the queue for things like a QUEUE_FULL message from a device, or a
1033  *    host that is unable to accept a particular command.
1034  */
1035 void blk_insert_request(struct request_queue *q, struct request *rq,
1036                         int at_head, void *data)
1037 {
1038         int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
1039         unsigned long flags;
1040 
1041         /*
1042          * tell I/O scheduler that this isn't a regular read/write (ie it
1043          * must not attempt merges on this) and that it acts as a soft
1044          * barrier
1045          */
1046         rq->cmd_type = REQ_TYPE_SPECIAL;
1047 
1048         rq->special = data;
1049 
1050         spin_lock_irqsave(q->queue_lock, flags);
1051 
1052         /*
1053          * If command is tagged, release the tag
1054          */
1055         if (blk_rq_tagged(rq))
1056                 blk_queue_end_tag(q, rq);
1057 
1058         add_acct_request(q, rq, where);
1059         __blk_run_queue(q);
1060         spin_unlock_irqrestore(q->queue_lock, flags);
1061 }
1062 EXPORT_SYMBOL(blk_insert_request);
1063 
1064 static void part_round_stats_single(int cpu, struct hd_struct *part,
1065                                     unsigned long now)
1066 {
1067         if (now == part->stamp)
1068                 return;
1069 
1070         if (part_in_flight(part)) {
1071                 __part_stat_add(cpu, part, time_in_queue,
1072                                 part_in_flight(part) * (now - part->stamp));
1073                 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1074         }
1075         part->stamp = now;
1076 }
1077 
1078 /**
1079  * part_round_stats() - Round off the performance stats on a struct disk_stats.
1080  * @cpu: cpu number for stats access
1081  * @part: target partition
1082  *
1083  * The average IO queue length and utilisation statistics are maintained
1084  * by observing the current state of the queue length and the amount of
1085  * time it has been in this state for.
1086  *
1087  * Normally, that accounting is done on IO completion, but that can result
1088  * in more than a second's worth of IO being accounted for within any one
1089  * second, leading to >100% utilisation.  To deal with that, we call this
1090  * function to do a round-off before returning the results when reading
1091  * /proc/diskstats.  This accounts immediately for all queue usage up to
1092  * the current jiffies and restarts the counters again.
1093  */
1094 void part_round_stats(int cpu, struct hd_struct *part)
1095 {
1096         unsigned long now = jiffies;
1097 
1098         if (part->partno)
1099                 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1100         part_round_stats_single(cpu, part, now);
1101 }
1102 EXPORT_SYMBOL_GPL(part_round_stats);
1103 
1104 /*
1105  * queue lock must be held
1106  */
1107 void __blk_put_request(struct request_queue *q, struct request *req)
1108 {
1109         if (unlikely(!q))
1110                 return;
1111         if (unlikely(--req->ref_count))
1112                 return;
1113 
1114         elv_completed_request(q, req);
1115 
1116         /* this is a bio leak */
1117         WARN_ON(req->bio != NULL);
1118 
1119         /*
1120          * Request may not have originated from ll_rw_blk. if not,
1121          * it didn't come out of our reserved rq pools
1122          */
1123         if (req->cmd_flags & REQ_ALLOCED) {
1124                 unsigned int flags = req->cmd_flags;
1125 
1126                 BUG_ON(!list_empty(&req->queuelist));
1127                 BUG_ON(!hlist_unhashed(&req->hash));
1128 
1129                 blk_free_request(q, req);
1130                 freed_request(q, flags);
1131         }
1132 }
1133 EXPORT_SYMBOL_GPL(__blk_put_request);
1134 
1135 void blk_put_request(struct request *req)
1136 {
1137         unsigned long flags;
1138         struct request_queue *q = req->q;
1139 
1140         spin_lock_irqsave(q->queue_lock, flags);
1141         __blk_put_request(q, req);
1142         spin_unlock_irqrestore(q->queue_lock, flags);
1143 }
1144 EXPORT_SYMBOL(blk_put_request);
1145 
1146 /**
1147  * blk_add_request_payload - add a payload to a request
1148  * @rq: request to update
1149  * @page: page backing the payload
1150  * @len: length of the payload.
1151  *
1152  * This allows to later add a payload to an already submitted request by
1153  * a block driver.  The driver needs to take care of freeing the payload
1154  * itself.
1155  *
1156  * Note that this is a quite horrible hack and nothing but handling of
1157  * discard requests should ever use it.
1158  */
1159 void blk_add_request_payload(struct request *rq, struct page *page,
1160                 unsigned int len)
1161 {
1162         struct bio *bio = rq->bio;
1163 
1164         bio->bi_io_vec->bv_page = page;
1165         bio->bi_io_vec->bv_offset = 0;
1166         bio->bi_io_vec->bv_len = len;
1167 
1168         bio->bi_size = len;
1169         bio->bi_vcnt = 1;
1170         bio->bi_phys_segments = 1;
1171 
1172         rq->__data_len = rq->resid_len = len;
1173         rq->nr_phys_segments = 1;
1174         rq->buffer = bio_data(bio);
1175 }
1176 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1177 
1178 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1179                                    struct bio *bio)
1180 {
1181         const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1182 
1183         if (!ll_back_merge_fn(q, req, bio))
1184                 return false;
1185 
1186         trace_block_bio_backmerge(q, bio);
1187 
1188         if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1189                 blk_rq_set_mixed_merge(req);
1190 
1191         req->biotail->bi_next = bio;
1192         req->biotail = bio;
1193         req->__data_len += bio->bi_size;
1194         req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1195 
1196         drive_stat_acct(req, 0);
1197         elv_bio_merged(q, req, bio);
1198         return true;
1199 }
1200 
1201 static bool bio_attempt_front_merge(struct request_queue *q,
1202                                     struct request *req, struct bio *bio)
1203 {
1204         const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1205 
1206         if (!ll_front_merge_fn(q, req, bio))
1207                 return false;
1208 
1209         trace_block_bio_frontmerge(q, bio);
1210 
1211         if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1212                 blk_rq_set_mixed_merge(req);
1213 
1214         bio->bi_next = req->bio;
1215         req->bio = bio;
1216 
1217         /*
1218          * may not be valid. if the low level driver said
1219          * it didn't need a bounce buffer then it better
1220          * not touch req->buffer either...
1221          */
1222         req->buffer = bio_data(bio);
1223         req->__sector = bio->bi_sector;
1224         req->__data_len += bio->bi_size;
1225         req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1226 
1227         drive_stat_acct(req, 0);
1228         elv_bio_merged(q, req, bio);
1229         return true;
1230 }
1231 
1232 /**
1233  * attempt_plug_merge - try to merge with %current's plugged list
1234  * @q: request_queue new bio is being queued at
1235  * @bio: new bio being queued
1236  * @request_count: out parameter for number of traversed plugged requests
1237  *
1238  * Determine whether @bio being queued on @q can be merged with a request
1239  * on %current's plugged list.  Returns %true if merge was successful,
1240  * otherwise %false.
1241  *
1242  * This function is called without @q->queue_lock; however, elevator is
1243  * accessed iff there already are requests on the plugged list which in
1244  * turn guarantees validity of the elevator.
1245  *
1246  * Note that, on successful merge, elevator operation
1247  * elevator_bio_merged_fn() will be called without queue lock.  Elevator
1248  * must be ready for this.
1249  */
1250 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1251                                unsigned int *request_count)
1252 {
1253         struct blk_plug *plug;
1254         struct request *rq;
1255         bool ret = false;
1256 
1257         plug = current->plug;
1258         if (!plug)
1259                 goto out;
1260         *request_count = 0;
1261 
1262         list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1263                 int el_ret;
1264 
1265                 (*request_count)++;
1266 
1267                 if (rq->q != q)
1268                         continue;
1269 
1270                 el_ret = elv_try_merge(rq, bio);
1271                 if (el_ret == ELEVATOR_BACK_MERGE) {
1272                         ret = bio_attempt_back_merge(q, rq, bio);
1273                         if (ret)
1274                                 break;
1275                 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1276                         ret = bio_attempt_front_merge(q, rq, bio);
1277                         if (ret)
1278                                 break;
1279                 }
1280         }
1281 out:
1282         return ret;
1283 }
1284 
1285 void init_request_from_bio(struct request *req, struct bio *bio)
1286 {
1287         req->cmd_type = REQ_TYPE_FS;
1288 
1289         req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1290         if (bio->bi_rw & REQ_RAHEAD)
1291                 req->cmd_flags |= REQ_FAILFAST_MASK;
1292 
1293         req->errors = 0;
1294         req->__sector = bio->bi_sector;
1295         req->ioprio = bio_prio(bio);
1296         blk_rq_bio_prep(req->q, req, bio);
1297 }
1298 
1299 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1300 {
1301         const bool sync = !!(bio->bi_rw & REQ_SYNC);
1302         struct blk_plug *plug;
1303         int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1304         struct request *req;
1305         unsigned int request_count = 0;
1306 
1307         /*
1308          * low level driver can indicate that it wants pages above a
1309          * certain limit bounced to low memory (ie for highmem, or even
1310          * ISA dma in theory)
1311          */
1312         blk_queue_bounce(q, &bio);
1313 
1314         if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1315                 spin_lock_irq(q->queue_lock);
1316                 where = ELEVATOR_INSERT_FLUSH;
1317                 goto get_rq;
1318         }
1319 
1320         /*
1321          * Check if we can merge with the plugged list before grabbing
1322          * any locks.
1323          */
1324         if (attempt_plug_merge(q, bio, &request_count))
1325                 return;
1326 
1327         spin_lock_irq(q->queue_lock);
1328 
1329         el_ret = elv_merge(q, &req, bio);
1330         if (el_ret == ELEVATOR_BACK_MERGE) {
1331                 if (bio_attempt_back_merge(q, req, bio)) {
1332                         if (!attempt_back_merge(q, req))
1333                                 elv_merged_request(q, req, el_ret);
1334                         goto out_unlock;
1335                 }
1336         } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1337                 if (bio_attempt_front_merge(q, req, bio)) {
1338                         if (!attempt_front_merge(q, req))
1339                                 elv_merged_request(q, req, el_ret);
1340                         goto out_unlock;
1341                 }
1342         }
1343 
1344 get_rq:
1345         /*
1346          * This sync check and mask will be re-done in init_request_from_bio(),
1347          * but we need to set it earlier to expose the sync flag to the
1348          * rq allocator and io schedulers.
1349          */
1350         rw_flags = bio_data_dir(bio);
1351         if (sync)
1352                 rw_flags |= REQ_SYNC;
1353 
1354         /*
1355          * Grab a free request. This is might sleep but can not fail.
1356          * Returns with the queue unlocked.
1357          */
1358         req = get_request_wait(q, rw_flags, bio);
1359         if (unlikely(!req)) {
1360                 bio_endio(bio, -ENODEV);        /* @q is dead */
1361                 goto out_unlock;
1362         }
1363 
1364         /*
1365          * After dropping the lock and possibly sleeping here, our request
1366          * may now be mergeable after it had proven unmergeable (above).
1367          * We don't worry about that case for efficiency. It won't happen
1368          * often, and the elevators are able to handle it.
1369          */
1370         init_request_from_bio(req, bio);
1371 
1372         if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1373                 req->cpu = raw_smp_processor_id();
1374 
1375         plug = current->plug;
1376         if (plug) {
1377                 /*
1378                  * If this is the first request added after a plug, fire
1379                  * of a plug trace. If others have been added before, check
1380                  * if we have multiple devices in this plug. If so, make a
1381                  * note to sort the list before dispatch.
1382                  */
1383                 if (list_empty(&plug->list))
1384                         trace_block_plug(q);
1385                 else {
1386                         if (!plug->should_sort) {
1387                                 struct request *__rq;
1388 
1389                                 __rq = list_entry_rq(plug->list.prev);
1390                                 if (__rq->q != q)
1391                                         plug->should_sort = 1;
1392                         }
1393                         if (request_count >= BLK_MAX_REQUEST_COUNT) {
1394                                 blk_flush_plug_list(plug, false);
1395                                 trace_block_plug(q);
1396                         }
1397                 }
1398                 list_add_tail(&req->queuelist, &plug->list);
1399                 drive_stat_acct(req, 1);
1400         } else {
1401                 spin_lock_irq(q->queue_lock);
1402                 add_acct_request(q, req, where);
1403                 __blk_run_queue(q);
1404 out_unlock:
1405                 spin_unlock_irq(q->queue_lock);
1406         }
1407 }
1408 EXPORT_SYMBOL_GPL(blk_queue_bio);       /* for device mapper only */
1409 
1410 /*
1411  * If bio->bi_dev is a partition, remap the location
1412  */
1413 static inline void blk_partition_remap(struct bio *bio)
1414 {
1415         struct block_device *bdev = bio->bi_bdev;
1416 
1417         if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1418                 struct hd_struct *p = bdev->bd_part;
1419 
1420                 bio->bi_sector += p->start_sect;
1421                 bio->bi_bdev = bdev->bd_contains;
1422 
1423                 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1424                                       bdev->bd_dev,
1425                                       bio->bi_sector - p->start_sect);
1426         }
1427 }
1428 
1429 static void handle_bad_sector(struct bio *bio)
1430 {
1431         char b[BDEVNAME_SIZE];
1432 
1433         printk(KERN_INFO "attempt to access beyond end of device\n");
1434         printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1435                         bdevname(bio->bi_bdev, b),
1436                         bio->bi_rw,
1437                         (unsigned long long)bio->bi_sector + bio_sectors(bio),
1438                         (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1439 
1440         set_bit(BIO_EOF, &bio->bi_flags);
1441 }
1442 
1443 #ifdef CONFIG_FAIL_MAKE_REQUEST
1444 
1445 static DECLARE_FAULT_ATTR(fail_make_request);
1446 
1447 static int __init setup_fail_make_request(char *str)
1448 {
1449         return setup_fault_attr(&fail_make_request, str);
1450 }
1451 __setup("fail_make_request=", setup_fail_make_request);
1452 
1453 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1454 {
1455         return part->make_it_fail && should_fail(&fail_make_request, bytes);
1456 }
1457 
1458 static int __init fail_make_request_debugfs(void)
1459 {
1460         struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1461                                                 NULL, &fail_make_request);
1462 
1463         return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1464 }
1465 
1466 late_initcall(fail_make_request_debugfs);
1467 
1468 #else /* CONFIG_FAIL_MAKE_REQUEST */
1469 
1470 static inline bool should_fail_request(struct hd_struct *part,
1471                                         unsigned int bytes)
1472 {
1473         return false;
1474 }
1475 
1476 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1477 
1478 /*
1479  * Check whether this bio extends beyond the end of the device.
1480  */
1481 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1482 {
1483         sector_t maxsector;
1484 
1485         if (!nr_sectors)
1486                 return 0;
1487 
1488         /* Test device or partition size, when known. */
1489         maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1490         if (maxsector) {
1491                 sector_t sector = bio->bi_sector;
1492 
1493                 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1494                         /*
1495                          * This may well happen - the kernel calls bread()
1496                          * without checking the size of the device, e.g., when
1497                          * mounting a device.
1498                          */
1499                         handle_bad_sector(bio);
1500                         return 1;
1501                 }
1502         }
1503 
1504         return 0;
1505 }
1506 
1507 static noinline_for_stack bool
1508 generic_make_request_checks(struct bio *bio)
1509 {
1510         struct request_queue *q;
1511         int nr_sectors = bio_sectors(bio);
1512         int err = -EIO;
1513         char b[BDEVNAME_SIZE];
1514         struct hd_struct *part;
1515 
1516         might_sleep();
1517 
1518         if (bio_check_eod(bio, nr_sectors))
1519                 goto end_io;
1520 
1521         q = bdev_get_queue(bio->bi_bdev);
1522         if (unlikely(!q)) {
1523                 printk(KERN_ERR
1524                        "generic_make_request: Trying to access "
1525                         "nonexistent block-device %s (%Lu)\n",
1526                         bdevname(bio->bi_bdev, b),
1527                         (long long) bio->bi_sector);
1528                 goto end_io;
1529         }
1530 
1531         if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1532                      nr_sectors > queue_max_hw_sectors(q))) {
1533                 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1534                        bdevname(bio->bi_bdev, b),
1535                        bio_sectors(bio),
1536                        queue_max_hw_sectors(q));
1537                 goto end_io;
1538         }
1539 
1540         part = bio->bi_bdev->bd_part;
1541         if (should_fail_request(part, bio->bi_size) ||
1542             should_fail_request(&part_to_disk(part)->part0,
1543                                 bio->bi_size))
1544                 goto end_io;
1545 
1546         /*
1547          * If this device has partitions, remap block n
1548          * of partition p to block n+start(p) of the disk.
1549          */
1550         blk_partition_remap(bio);
1551 
1552         if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1553                 goto end_io;
1554 
1555         if (bio_check_eod(bio, nr_sectors))
1556                 goto end_io;
1557 
1558         /*
1559          * Filter flush bio's early so that make_request based
1560          * drivers without flush support don't have to worry
1561          * about them.
1562          */
1563         if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1564                 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1565                 if (!nr_sectors) {
1566                         err = 0;
1567                         goto end_io;
1568                 }
1569         }
1570 
1571         if ((bio->bi_rw & REQ_DISCARD) &&
1572             (!blk_queue_discard(q) ||
1573              ((bio->bi_rw & REQ_SECURE) &&
1574               !blk_queue_secdiscard(q)))) {
1575                 err = -EOPNOTSUPP;
1576                 goto end_io;
1577         }
1578 
1579         if (blk_throtl_bio(q, bio))
1580                 return false;   /* throttled, will be resubmitted later */
1581 
1582         trace_block_bio_queue(q, bio);
1583         return true;
1584 
1585 end_io:
1586         bio_endio(bio, err);
1587         return false;
1588 }
1589 
1590 /**
1591  * generic_make_request - hand a buffer to its device driver for I/O
1592  * @bio:  The bio describing the location in memory and on the device.
1593  *
1594  * generic_make_request() is used to make I/O requests of block
1595  * devices. It is passed a &struct bio, which describes the I/O that needs
1596  * to be done.
1597  *
1598  * generic_make_request() does not return any status.  The
1599  * success/failure status of the request, along with notification of
1600  * completion, is delivered asynchronously through the bio->bi_end_io
1601  * function described (one day) else where.
1602  *
1603  * The caller of generic_make_request must make sure that bi_io_vec
1604  * are set to describe the memory buffer, and that bi_dev and bi_sector are
1605  * set to describe the device address, and the
1606  * bi_end_io and optionally bi_private are set to describe how
1607  * completion notification should be signaled.
1608  *
1609  * generic_make_request and the drivers it calls may use bi_next if this
1610  * bio happens to be merged with someone else, and may resubmit the bio to
1611  * a lower device by calling into generic_make_request recursively, which
1612  * means the bio should NOT be touched after the call to ->make_request_fn.
1613  */
1614 void generic_make_request(struct bio *bio)
1615 {
1616         struct bio_list bio_list_on_stack;
1617 
1618         if (!generic_make_request_checks(bio))
1619                 return;
1620 
1621         /*
1622          * We only want one ->make_request_fn to be active at a time, else
1623          * stack usage with stacked devices could be a problem.  So use
1624          * current->bio_list to keep a list of requests submited by a
1625          * make_request_fn function.  current->bio_list is also used as a
1626          * flag to say if generic_make_request is currently active in this
1627          * task or not.  If it is NULL, then no make_request is active.  If
1628          * it is non-NULL, then a make_request is active, and new requests
1629          * should be added at the tail
1630          */
1631         if (current->bio_list) {
1632                 bio_list_add(current->bio_list, bio);
1633                 return;
1634         }
1635 
1636         /* following loop may be a bit non-obvious, and so deserves some
1637          * explanation.
1638          * Before entering the loop, bio->bi_next is NULL (as all callers
1639          * ensure that) so we have a list with a single bio.
1640          * We pretend that we have just taken it off a longer list, so
1641          * we assign bio_list to a pointer to the bio_list_on_stack,
1642          * thus initialising the bio_list of new bios to be
1643          * added.  ->make_request() may indeed add some more bios
1644          * through a recursive call to generic_make_request.  If it
1645          * did, we find a non-NULL value in bio_list and re-enter the loop
1646          * from the top.  In this case we really did just take the bio
1647          * of the top of the list (no pretending) and so remove it from
1648          * bio_list, and call into ->make_request() again.
1649          */
1650         BUG_ON(bio->bi_next);
1651         bio_list_init(&bio_list_on_stack);
1652         current->bio_list = &bio_list_on_stack;
1653         do {
1654                 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1655 
1656                 q->make_request_fn(q, bio);
1657 
1658                 bio = bio_list_pop(current->bio_list);
1659         } while (bio);
1660         current->bio_list = NULL; /* deactivate */
1661 }
1662 EXPORT_SYMBOL(generic_make_request);
1663 
1664 /**
1665  * submit_bio - submit a bio to the block device layer for I/O
1666  * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1667  * @bio: The &struct bio which describes the I/O
1668  *
1669  * submit_bio() is very similar in purpose to generic_make_request(), and
1670  * uses that function to do most of the work. Both are fairly rough
1671  * interfaces; @bio must be presetup and ready for I/O.
1672  *
1673  */
1674 void submit_bio(int rw, struct bio *bio)
1675 {
1676         int count = bio_sectors(bio);
1677 
1678         bio->bi_rw |= rw;
1679 
1680         /*
1681          * If it's a regular read/write or a barrier with data attached,
1682          * go through the normal accounting stuff before submission.
1683          */
1684         if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1685                 if (rw & WRITE) {
1686                         count_vm_events(PGPGOUT, count);
1687                 } else {
1688                         task_io_account_read(bio->bi_size);
1689                         count_vm_events(PGPGIN, count);
1690                 }
1691 
1692                 if (unlikely(block_dump)) {
1693                         char b[BDEVNAME_SIZE];
1694                         printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1695                         current->comm, task_pid_nr(current),
1696                                 (rw & WRITE) ? "WRITE" : "READ",
1697                                 (unsigned long long)bio->bi_sector,
1698                                 bdevname(bio->bi_bdev, b),
1699                                 count);
1700                 }
1701         }
1702 
1703         generic_make_request(bio);
1704 }
1705 EXPORT_SYMBOL(submit_bio);
1706 
1707 /**
1708  * blk_rq_check_limits - Helper function to check a request for the queue limit
1709  * @q:  the queue
1710  * @rq: the request being checked
1711  *
1712  * Description:
1713  *    @rq may have been made based on weaker limitations of upper-level queues
1714  *    in request stacking drivers, and it may violate the limitation of @q.
1715  *    Since the block layer and the underlying device driver trust @rq
1716  *    after it is inserted to @q, it should be checked against @q before
1717  *    the insertion using this generic function.
1718  *
1719  *    This function should also be useful for request stacking drivers
1720  *    in some cases below, so export this function.
1721  *    Request stacking drivers like request-based dm may change the queue
1722  *    limits while requests are in the queue (e.g. dm's table swapping).
1723  *    Such request stacking drivers should check those requests agaist
1724  *    the new queue limits again when they dispatch those requests,
1725  *    although such checkings are also done against the old queue limits
1726  *    when submitting requests.
1727  */
1728 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1729 {
1730         if (rq->cmd_flags & REQ_DISCARD)
1731                 return 0;
1732 
1733         if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1734             blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1735                 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1736                 return -EIO;
1737         }
1738 
1739         /*
1740          * queue's settings related to segment counting like q->bounce_pfn
1741          * may differ from that of other stacking queues.
1742          * Recalculate it to check the request correctly on this queue's
1743          * limitation.
1744          */
1745         blk_recalc_rq_segments(rq);
1746         if (rq->nr_phys_segments > queue_max_segments(q)) {
1747                 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1748                 return -EIO;
1749         }
1750 
1751         return 0;
1752 }
1753 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1754 
1755 /**
1756  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1757  * @q:  the queue to submit the request
1758  * @rq: the request being queued
1759  */
1760 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1761 {
1762         unsigned long flags;
1763         int where = ELEVATOR_INSERT_BACK;
1764 
1765         if (blk_rq_check_limits(q, rq))
1766                 return -EIO;
1767 
1768         if (rq->rq_disk &&
1769             should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1770                 return -EIO;
1771 
1772         spin_lock_irqsave(q->queue_lock, flags);
1773 
1774         /*
1775          * Submitting request must be dequeued before calling this function
1776          * because it will be linked to another request_queue
1777          */
1778         BUG_ON(blk_queued_rq(rq));
1779 
1780         if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1781                 where = ELEVATOR_INSERT_FLUSH;
1782 
1783         add_acct_request(q, rq, where);
1784         if (where == ELEVATOR_INSERT_FLUSH)
1785                 __blk_run_queue(q);
1786         spin_unlock_irqrestore(q->queue_lock, flags);
1787 
1788         return 0;
1789 }
1790 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1791 
1792 /**
1793  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1794  * @rq: request to examine
1795  *
1796  * Description:
1797  *     A request could be merge of IOs which require different failure
1798  *     handling.  This function determines the number of bytes which
1799  *     can be failed from the beginning of the request without
1800  *     crossing into area which need to be retried further.
1801  *
1802  * Return:
1803  *     The number of bytes to fail.
1804  *
1805  * Context:
1806  *     queue_lock must be held.
1807  */
1808 unsigned int blk_rq_err_bytes(const struct request *rq)
1809 {
1810         unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1811         unsigned int bytes = 0;
1812         struct bio *bio;
1813 
1814         if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1815                 return blk_rq_bytes(rq);
1816 
1817         /*
1818          * Currently the only 'mixing' which can happen is between
1819          * different fastfail types.  We can safely fail portions
1820          * which have all the failfast bits that the first one has -
1821          * the ones which are at least as eager to fail as the first
1822          * one.
1823          */
1824         for (bio = rq->bio; bio; bio = bio->bi_next) {
1825                 if ((bio->bi_rw & ff) != ff)
1826                         break;
1827                 bytes += bio->bi_size;
1828         }
1829 
1830         /* this could lead to infinite loop */
1831         BUG_ON(blk_rq_bytes(rq) && !bytes);
1832         return bytes;
1833 }
1834 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1835 
1836 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1837 {
1838         if (blk_do_io_stat(req)) {
1839                 const int rw = rq_data_dir(req);
1840                 struct hd_struct *part;
1841                 int cpu;
1842 
1843                 cpu = part_stat_lock();
1844                 part = req->part;
1845                 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1846                 part_stat_unlock();
1847         }
1848 }
1849 
1850 static void blk_account_io_done(struct request *req)
1851 {
1852         /*
1853          * Account IO completion.  flush_rq isn't accounted as a
1854          * normal IO on queueing nor completion.  Accounting the
1855          * containing request is enough.
1856          */
1857         if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1858                 unsigned long duration = jiffies - req->start_time;
1859                 const int rw = rq_data_dir(req);
1860                 struct hd_struct *part;
1861                 int cpu;
1862 
1863                 cpu = part_stat_lock();
1864                 part = req->part;
1865 
1866                 part_stat_inc(cpu, part, ios[rw]);
1867                 part_stat_add(cpu, part, ticks[rw], duration);
1868                 part_round_stats(cpu, part);
1869                 part_dec_in_flight(part, rw);
1870 
1871                 hd_struct_put(part);
1872                 part_stat_unlock();
1873         }
1874 }
1875 
1876 /**
1877  * blk_peek_request - peek at the top of a request queue
1878  * @q: request queue to peek at
1879  *
1880  * Description:
1881  *     Return the request at the top of @q.  The returned request
1882  *     should be started using blk_start_request() before LLD starts
1883  *     processing it.
1884  *
1885  * Return:
1886  *     Pointer to the request at the top of @q if available.  Null
1887  *     otherwise.
1888  *
1889  * Context:
1890  *     queue_lock must be held.
1891  */
1892 struct request *blk_peek_request(struct request_queue *q)
1893 {
1894         struct request *rq;
1895         int ret;
1896 
1897         while ((rq = __elv_next_request(q)) != NULL) {
1898                 if (!(rq->cmd_flags & REQ_STARTED)) {
1899                         /*
1900                          * This is the first time the device driver
1901                          * sees this request (possibly after
1902                          * requeueing).  Notify IO scheduler.
1903                          */
1904                         if (rq->cmd_flags & REQ_SORTED)
1905                                 elv_activate_rq(q, rq);
1906 
1907                         /*
1908                          * just mark as started even if we don't start
1909                          * it, a request that has been delayed should
1910                          * not be passed by new incoming requests
1911                          */
1912                         rq->cmd_flags |= REQ_STARTED;
1913                         trace_block_rq_issue(q, rq);
1914                 }
1915 
1916                 if (!q->boundary_rq || q->boundary_rq == rq) {
1917                         q->end_sector = rq_end_sector(rq);
1918                         q->boundary_rq = NULL;
1919                 }
1920 
1921                 if (rq->cmd_flags & REQ_DONTPREP)
1922                         break;
1923 
1924                 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1925                         /*
1926                          * make sure space for the drain appears we
1927                          * know we can do this because max_hw_segments
1928                          * has been adjusted to be one fewer than the
1929                          * device can handle
1930                          */
1931                         rq->nr_phys_segments++;
1932                 }
1933 
1934                 if (!q->prep_rq_fn)
1935                         break;
1936 
1937                 ret = q->prep_rq_fn(q, rq);
1938                 if (ret == BLKPREP_OK) {
1939                         break;
1940                 } else if (ret == BLKPREP_DEFER) {
1941                         /*
1942                          * the request may have been (partially) prepped.
1943                          * we need to keep this request in the front to
1944                          * avoid resource deadlock.  REQ_STARTED will
1945                          * prevent other fs requests from passing this one.
1946                          */
1947                         if (q->dma_drain_size && blk_rq_bytes(rq) &&
1948                             !(rq->cmd_flags & REQ_DONTPREP)) {
1949                                 /*
1950                                  * remove the space for the drain we added
1951                                  * so that we don't add it again
1952                                  */
1953                                 --rq->nr_phys_segments;
1954                         }
1955 
1956                         rq = NULL;
1957                         break;
1958                 } else if (ret == BLKPREP_KILL) {
1959                         rq->cmd_flags |= REQ_QUIET;
1960                         /*
1961                          * Mark this request as started so we don't trigger
1962                          * any debug logic in the end I/O path.
1963                          */
1964                         blk_start_request(rq);
1965                         __blk_end_request_all(rq, -EIO);
1966                 } else {
1967                         printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1968                         break;
1969                 }
1970         }
1971 
1972         return rq;
1973 }
1974 EXPORT_SYMBOL(blk_peek_request);
1975 
1976 void blk_dequeue_request(struct request *rq)
1977 {
1978         struct request_queue *q = rq->q;
1979 
1980         BUG_ON(list_empty(&rq->queuelist));
1981         BUG_ON(ELV_ON_HASH(rq));
1982 
1983         list_del_init(&rq->queuelist);
1984 
1985         /*
1986          * the time frame between a request being removed from the lists
1987          * and to it is freed is accounted as io that is in progress at
1988          * the driver side.
1989          */
1990         if (blk_account_rq(rq)) {
1991                 q->in_flight[rq_is_sync(rq)]++;
1992                 set_io_start_time_ns(rq);
1993         }
1994 }
1995 
1996 /**
1997  * blk_start_request - start request processing on the driver
1998  * @req: request to dequeue
1999  *
2000  * Description:
2001  *     Dequeue @req and start timeout timer on it.  This hands off the
2002  *     request to the driver.
2003  *
2004  *     Block internal functions which don't want to start timer should
2005  *     call blk_dequeue_request().
2006  *
2007  * Context:
2008  *     queue_lock must be held.
2009  */
2010 void blk_start_request(struct request *req)
2011 {
2012         blk_dequeue_request(req);
2013 
2014         /*
2015          * We are now handing the request to the hardware, initialize
2016          * resid_len to full count and add the timeout handler.
2017          */
2018         req->resid_len = blk_rq_bytes(req);
2019         if (unlikely(blk_bidi_rq(req)))
2020                 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2021 
2022         BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2023         blk_add_timer(req);
2024 }
2025 EXPORT_SYMBOL(blk_start_request);
2026 
2027 /**
2028  * blk_fetch_request - fetch a request from a request queue
2029  * @q: request queue to fetch a request from
2030  *
2031  * Description:
2032  *     Return the request at the top of @q.  The request is started on
2033  *     return and LLD can start processing it immediately.
2034  *
2035  * Return:
2036  *     Pointer to the request at the top of @q if available.  Null
2037  *     otherwise.
2038  *
2039  * Context:
2040  *     queue_lock must be held.
2041  */
2042 struct request *blk_fetch_request(struct request_queue *q)
2043 {
2044         struct request *rq;
2045 
2046         rq = blk_peek_request(q);
2047         if (rq)
2048                 blk_start_request(rq);
2049         return rq;
2050 }
2051 EXPORT_SYMBOL(blk_fetch_request);
2052 
2053 /**
2054  * blk_update_request - Special helper function for request stacking drivers
2055  * @req:      the request being processed
2056  * @error:    %0 for success, < %0 for error
2057  * @nr_bytes: number of bytes to complete @req
2058  *
2059  * Description:
2060  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
2061  *     the request structure even if @req doesn't have leftover.
2062  *     If @req has leftover, sets it up for the next range of segments.
2063  *
2064  *     This special helper function is only for request stacking drivers
2065  *     (e.g. request-based dm) so that they can handle partial completion.
2066  *     Actual device drivers should use blk_end_request instead.
2067  *
2068  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2069  *     %false return from this function.
2070  *
2071  * Return:
2072  *     %false - this request doesn't have any more data
2073  *     %true  - this request has more data
2074  **/
2075 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2076 {
2077         int total_bytes, bio_nbytes, next_idx = 0;
2078         struct bio *bio;
2079 
2080         if (!req->bio)
2081                 return false;
2082 
2083         trace_block_rq_complete(req->q, req, nr_bytes);
2084 
2085         /*
2086          * For fs requests, rq is just carrier of independent bio's
2087          * and each partial completion should be handled separately.
2088          * Reset per-request error on each partial completion.
2089          *
2090          * TODO: tj: This is too subtle.  It would be better to let
2091          * low level drivers do what they see fit.
2092          */
2093         if (req->cmd_type == REQ_TYPE_FS)
2094                 req->errors = 0;
2095 
2096         if (error && req->cmd_type == REQ_TYPE_FS &&
2097             !(req->cmd_flags & REQ_QUIET)) {
2098                 char *error_type;
2099 
2100                 switch (error) {
2101                 case -ENOLINK:
2102                         error_type = "recoverable transport";
2103                         break;
2104                 case -EREMOTEIO:
2105                         error_type = "critical target";
2106                         break;
2107                 case -EBADE:
2108                         error_type = "critical nexus";
2109                         break;
2110                 case -EIO:
2111                 default:
2112                         error_type = "I/O";
2113                         break;
2114                 }
2115                 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2116                        error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2117                        (unsigned long long)blk_rq_pos(req));
2118         }
2119 
2120         blk_account_io_completion(req, nr_bytes);
2121 
2122         total_bytes = bio_nbytes = 0;
2123         while ((bio = req->bio) != NULL) {
2124                 int nbytes;
2125 
2126                 if (nr_bytes >= bio->bi_size) {
2127                         req->bio = bio->bi_next;
2128                         nbytes = bio->bi_size;
2129                         req_bio_endio(req, bio, nbytes, error);
2130                         next_idx = 0;
2131                         bio_nbytes = 0;
2132                 } else {
2133                         int idx = bio->bi_idx + next_idx;
2134 
2135                         if (unlikely(idx >= bio->bi_vcnt)) {
2136                                 blk_dump_rq_flags(req, "__end_that");
2137                                 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2138                                        __func__, idx, bio->bi_vcnt);
2139                                 break;
2140                         }
2141 
2142                         nbytes = bio_iovec_idx(bio, idx)->bv_len;
2143                         BIO_BUG_ON(nbytes > bio->bi_size);
2144 
2145                         /*
2146                          * not a complete bvec done
2147                          */
2148                         if (unlikely(nbytes > nr_bytes)) {
2149                                 bio_nbytes += nr_bytes;
2150                                 total_bytes += nr_bytes;
2151                                 break;
2152                         }
2153 
2154                         /*
2155                          * advance to the next vector
2156                          */
2157                         next_idx++;
2158                         bio_nbytes += nbytes;
2159                 }
2160 
2161                 total_bytes += nbytes;
2162                 nr_bytes -= nbytes;
2163 
2164                 bio = req->bio;
2165                 if (bio) {
2166                         /*
2167                          * end more in this run, or just return 'not-done'
2168                          */
2169                         if (unlikely(nr_bytes <= 0))
2170                                 break;
2171                 }
2172         }
2173 
2174         /*
2175          * completely done
2176          */
2177         if (!req->bio) {
2178                 /*
2179                  * Reset counters so that the request stacking driver
2180                  * can find how many bytes remain in the request
2181                  * later.
2182                  */
2183                 req->__data_len = 0;
2184                 return false;
2185         }
2186 
2187         /*
2188          * if the request wasn't completed, update state
2189          */
2190         if (bio_nbytes) {
2191                 req_bio_endio(req, bio, bio_nbytes, error);
2192                 bio->bi_idx += next_idx;
2193                 bio_iovec(bio)->bv_offset += nr_bytes;
2194                 bio_iovec(bio)->bv_len -= nr_bytes;
2195         }
2196 
2197         req->__data_len -= total_bytes;
2198         req->buffer = bio_data(req->bio);
2199 
2200         /* update sector only for requests with clear definition of sector */
2201         if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2202                 req->__sector += total_bytes >> 9;
2203 
2204         /* mixed attributes always follow the first bio */
2205         if (req->cmd_flags & REQ_MIXED_MERGE) {
2206                 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2207                 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2208         }
2209 
2210         /*
2211          * If total number of sectors is less than the first segment
2212          * size, something has gone terribly wrong.
2213          */
2214         if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2215                 blk_dump_rq_flags(req, "request botched");
2216                 req->__data_len = blk_rq_cur_bytes(req);
2217         }
2218 
2219         /* recalculate the number of segments */
2220         blk_recalc_rq_segments(req);
2221 
2222         return true;
2223 }
2224 EXPORT_SYMBOL_GPL(blk_update_request);
2225 
2226 static bool blk_update_bidi_request(struct request *rq, int error,
2227                                     unsigned int nr_bytes,
2228                                     unsigned int bidi_bytes)
2229 {
2230         if (blk_update_request(rq, error, nr_bytes))
2231                 return true;
2232 
2233         /* Bidi request must be completed as a whole */
2234         if (unlikely(blk_bidi_rq(rq)) &&
2235             blk_update_request(rq->next_rq, error, bidi_bytes))
2236                 return true;
2237 
2238         if (blk_queue_add_random(rq->q))
2239                 add_disk_randomness(rq->rq_disk);
2240 
2241         return false;
2242 }
2243 
2244 /**
2245  * blk_unprep_request - unprepare a request
2246  * @req:        the request
2247  *
2248  * This function makes a request ready for complete resubmission (or
2249  * completion).  It happens only after all error handling is complete,
2250  * so represents the appropriate moment to deallocate any resources
2251  * that were allocated to the request in the prep_rq_fn.  The queue
2252  * lock is held when calling this.
2253  */
2254 void blk_unprep_request(struct request *req)
2255 {
2256         struct request_queue *q = req->q;
2257 
2258         req->cmd_flags &= ~REQ_DONTPREP;
2259         if (q->unprep_rq_fn)
2260                 q->unprep_rq_fn(q, req);
2261 }
2262 EXPORT_SYMBOL_GPL(blk_unprep_request);
2263 
2264 /*
2265  * queue lock must be held
2266  */
2267 static void blk_finish_request(struct request *req, int error)
2268 {
2269         if (blk_rq_tagged(req))
2270                 blk_queue_end_tag(req->q, req);
2271 
2272         BUG_ON(blk_queued_rq(req));
2273 
2274         if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2275                 laptop_io_completion(&req->q->backing_dev_info);
2276 
2277         blk_delete_timer(req);
2278 
2279         if (req->cmd_flags & REQ_DONTPREP)
2280                 blk_unprep_request(req);
2281 
2282 
2283         blk_account_io_done(req);
2284 
2285         if (req->end_io)
2286                 req->end_io(req, error);
2287         else {
2288                 if (blk_bidi_rq(req))
2289                         __blk_put_request(req->next_rq->q, req->next_rq);
2290 
2291                 __blk_put_request(req->q, req);
2292         }
2293 }
2294 
2295 /**
2296  * blk_end_bidi_request - Complete a bidi request
2297  * @rq:         the request to complete
2298  * @error:      %0 for success, < %0 for error
2299  * @nr_bytes:   number of bytes to complete @rq
2300  * @bidi_bytes: number of bytes to complete @rq->next_rq
2301  *
2302  * Description:
2303  *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2304  *     Drivers that supports bidi can safely call this member for any
2305  *     type of request, bidi or uni.  In the later case @bidi_bytes is
2306  *     just ignored.
2307  *
2308  * Return:
2309  *     %false - we are done with this request
2310  *     %true  - still buffers pending for this request
2311  **/
2312 static bool blk_end_bidi_request(struct request *rq, int error,
2313                                  unsigned int nr_bytes, unsigned int bidi_bytes)
2314 {
2315         struct request_queue *q = rq->q;
2316         unsigned long flags;
2317 
2318         if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2319                 return true;
2320 
2321         spin_lock_irqsave(q->queue_lock, flags);
2322         blk_finish_request(rq, error);
2323         spin_unlock_irqrestore(q->queue_lock, flags);
2324 
2325         return false;
2326 }
2327 
2328 /**
2329  * __blk_end_bidi_request - Complete a bidi request with queue lock held
2330  * @rq:         the request to complete
2331  * @error:      %0 for success, < %0 for error
2332  * @nr_bytes:   number of bytes to complete @rq
2333  * @bidi_bytes: number of bytes to complete @rq->next_rq
2334  *
2335  * Description:
2336  *     Identical to blk_end_bidi_request() except that queue lock is
2337  *     assumed to be locked on entry and remains so on return.
2338  *
2339  * Return:
2340  *     %false - we are done with this request
2341  *     %true  - still buffers pending for this request
2342  **/
2343 bool __blk_end_bidi_request(struct request *rq, int error,
2344                                    unsigned int nr_bytes, unsigned int bidi_bytes)
2345 {
2346         if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2347                 return true;
2348 
2349         blk_finish_request(rq, error);
2350 
2351         return false;
2352 }
2353 
2354 /**
2355  * blk_end_request - Helper function for drivers to complete the request.
2356  * @rq:       the request being processed
2357  * @error:    %0 for success, < %0 for error
2358  * @nr_bytes: number of bytes to complete
2359  *
2360  * Description:
2361  *     Ends I/O on a number of bytes attached to @rq.
2362  *     If @rq has leftover, sets it up for the next range of segments.
2363  *
2364  * Return:
2365  *     %false - we are done with this request
2366  *     %true  - still buffers pending for this request
2367  **/
2368 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2369 {
2370         return blk_end_bidi_request(rq, error, nr_bytes, 0);
2371 }
2372 EXPORT_SYMBOL(blk_end_request);
2373 
2374 /**
2375  * blk_end_request_all - Helper function for drives to finish the request.
2376  * @rq: the request to finish
2377  * @error: %0 for success, < %0 for error
2378  *
2379  * Description:
2380  *     Completely finish @rq.
2381  */
2382 void blk_end_request_all(struct request *rq, int error)
2383 {
2384         bool pending;
2385         unsigned int bidi_bytes = 0;
2386 
2387         if (unlikely(blk_bidi_rq(rq)))
2388                 bidi_bytes = blk_rq_bytes(rq->next_rq);
2389 
2390         pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2391         BUG_ON(pending);
2392 }
2393 EXPORT_SYMBOL(blk_end_request_all);
2394 
2395 /**
2396  * blk_end_request_cur - Helper function to finish the current request chunk.
2397  * @rq: the request to finish the current chunk for
2398  * @error: %0 for success, < %0 for error
2399  *
2400  * Description:
2401  *     Complete the current consecutively mapped chunk from @rq.
2402  *
2403  * Return:
2404  *     %false - we are done with this request
2405  *     %true  - still buffers pending for this request
2406  */
2407 bool blk_end_request_cur(struct request *rq, int error)
2408 {
2409         return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2410 }
2411 EXPORT_SYMBOL(blk_end_request_cur);
2412 
2413 /**
2414  * blk_end_request_err - Finish a request till the next failure boundary.
2415  * @rq: the request to finish till the next failure boundary for
2416  * @error: must be negative errno
2417  *
2418  * Description:
2419  *     Complete @rq till the next failure boundary.
2420  *
2421  * Return:
2422  *     %false - we are done with this request
2423  *     %true  - still buffers pending for this request
2424  */
2425 bool blk_end_request_err(struct request *rq, int error)
2426 {
2427         WARN_ON(error >= 0);
2428         return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2429 }
2430 EXPORT_SYMBOL_GPL(blk_end_request_err);
2431 
2432 /**
2433  * __blk_end_request - Helper function for drivers to complete the request.
2434  * @rq:       the request being processed
2435  * @error:    %0 for success, < %0 for error
2436  * @nr_bytes: number of bytes to complete
2437  *
2438  * Description:
2439  *     Must be called with queue lock held unlike blk_end_request().
2440  *
2441  * Return:
2442  *     %false - we are done with this request
2443  *     %true  - still buffers pending for this request
2444  **/
2445 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2446 {
2447         return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2448 }
2449 EXPORT_SYMBOL(__blk_end_request);
2450 
2451 /**
2452  * __blk_end_request_all - Helper function for drives to finish the request.
2453  * @rq: the request to finish
2454  * @error: %0 for success, < %0 for error
2455  *
2456  * Description:
2457  *     Completely finish @rq.  Must be called with queue lock held.
2458  */
2459 void __blk_end_request_all(struct request *rq, int error)
2460 {
2461         bool pending;
2462         unsigned int bidi_bytes = 0;
2463 
2464         if (unlikely(blk_bidi_rq(rq)))
2465                 bidi_bytes = blk_rq_bytes(rq->next_rq);
2466 
2467         pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2468         BUG_ON(pending);
2469 }
2470 EXPORT_SYMBOL(__blk_end_request_all);
2471 
2472 /**
2473  * __blk_end_request_cur - Helper function to finish the current request chunk.
2474  * @rq: the request to finish the current chunk for
2475  * @error: %0 for success, < %0 for error
2476  *
2477  * Description:
2478  *     Complete the current consecutively mapped chunk from @rq.  Must
2479  *     be called with queue lock held.
2480  *
2481  * Return:
2482  *     %false - we are done with this request
2483  *     %true  - still buffers pending for this request
2484  */
2485 bool __blk_end_request_cur(struct request *rq, int error)
2486 {
2487         return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2488 }
2489 EXPORT_SYMBOL(__blk_end_request_cur);
2490 
2491 /**
2492  * __blk_end_request_err - Finish a request till the next failure boundary.
2493  * @rq: the request to finish till the next failure boundary for
2494  * @error: must be negative errno
2495  *
2496  * Description:
2497  *     Complete @rq till the next failure boundary.  Must be called
2498  *     with queue lock held.
2499  *
2500  * Return:
2501  *     %false - we are done with this request
2502  *     %true  - still buffers pending for this request
2503  */
2504 bool __blk_end_request_err(struct request *rq, int error)
2505 {
2506         WARN_ON(error >= 0);
2507         return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2508 }
2509 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2510 
2511 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2512                      struct bio *bio)
2513 {
2514         /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2515         rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2516 
2517         if (bio_has_data(bio)) {
2518                 rq->nr_phys_segments = bio_phys_segments(q, bio);
2519                 rq->buffer = bio_data(bio);
2520         }
2521         rq->__data_len = bio->bi_size;
2522         rq->bio = rq->biotail = bio;
2523 
2524         if (bio->bi_bdev)
2525                 rq->rq_disk = bio->bi_bdev->bd_disk;
2526 }
2527 
2528 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2529 /**
2530  * rq_flush_dcache_pages - Helper function to flush all pages in a request
2531  * @rq: the request to be flushed
2532  *
2533  * Description:
2534  *     Flush all pages in @rq.
2535  */
2536 void rq_flush_dcache_pages(struct request *rq)
2537 {
2538         struct req_iterator iter;
2539         struct bio_vec *bvec;
2540 
2541         rq_for_each_segment(bvec, rq, iter)
2542                 flush_dcache_page(bvec->bv_page);
2543 }
2544 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2545 #endif
2546 
2547 /**
2548  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2549  * @q : the queue of the device being checked
2550  *
2551  * Description:
2552  *    Check if underlying low-level drivers of a device are busy.
2553  *    If the drivers want to export their busy state, they must set own
2554  *    exporting function using blk_queue_lld_busy() first.
2555  *
2556  *    Basically, this function is used only by request stacking drivers
2557  *    to stop dispatching requests to underlying devices when underlying
2558  *    devices are busy.  This behavior helps more I/O merging on the queue
2559  *    of the request stacking driver and prevents I/O throughput regression
2560  *    on burst I/O load.
2561  *
2562  * Return:
2563  *    0 - Not busy (The request stacking driver should dispatch request)
2564  *    1 - Busy (The request stacking driver should stop dispatching request)
2565  */
2566 int blk_lld_busy(struct request_queue *q)
2567 {
2568         if (q->lld_busy_fn)
2569                 return q->lld_busy_fn(q);
2570 
2571         return 0;
2572 }
2573 EXPORT_SYMBOL_GPL(blk_lld_busy);
2574 
2575 /**
2576  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2577  * @rq: the clone request to be cleaned up
2578  *
2579  * Description:
2580  *     Free all bios in @rq for a cloned request.
2581  */
2582 void blk_rq_unprep_clone(struct request *rq)
2583 {
2584         struct bio *bio;
2585 
2586         while ((bio = rq->bio) != NULL) {
2587                 rq->bio = bio->bi_next;
2588 
2589                 bio_put(bio);
2590         }
2591 }
2592 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2593 
2594 /*
2595  * Copy attributes of the original request to the clone request.
2596  * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2597  */
2598 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2599 {
2600         dst->cpu = src->cpu;
2601         dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2602         dst->cmd_type = src->cmd_type;
2603         dst->__sector = blk_rq_pos(src);
2604         dst->__data_len = blk_rq_bytes(src);
2605         dst->nr_phys_segments = src->nr_phys_segments;
2606         dst->ioprio = src->ioprio;
2607         dst->extra_len = src->extra_len;
2608 }
2609 
2610 /**
2611  * blk_rq_prep_clone - Helper function to setup clone request
2612  * @rq: the request to be setup
2613  * @rq_src: original request to be cloned
2614  * @bs: bio_set that bios for clone are allocated from
2615  * @gfp_mask: memory allocation mask for bio
2616  * @bio_ctr: setup function to be called for each clone bio.
2617  *           Returns %0 for success, non %0 for failure.
2618  * @data: private data to be passed to @bio_ctr
2619  *
2620  * Description:
2621  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2622  *     The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2623  *     are not copied, and copying such parts is the caller's responsibility.
2624  *     Also, pages which the original bios are pointing to are not copied
2625  *     and the cloned bios just point same pages.
2626  *     So cloned bios must be completed before original bios, which means
2627  *     the caller must complete @rq before @rq_src.
2628  */
2629 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2630                       struct bio_set *bs, gfp_t gfp_mask,
2631                       int (*bio_ctr)(struct bio *, struct bio *, void *),
2632                       void *data)
2633 {
2634         struct bio *bio, *bio_src;
2635 
2636         if (!bs)
2637                 bs = fs_bio_set;
2638 
2639         blk_rq_init(NULL, rq);
2640 
2641         __rq_for_each_bio(bio_src, rq_src) {
2642                 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2643                 if (!bio)
2644                         goto free_and_out;
2645 
2646                 __bio_clone(bio, bio_src);
2647 
2648                 if (bio_integrity(bio_src) &&
2649                     bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2650                         goto free_and_out;
2651 
2652                 if (bio_ctr && bio_ctr(bio, bio_src, data))
2653                         goto free_and_out;
2654 
2655                 if (rq->bio) {
2656                         rq->biotail->bi_next = bio;
2657                         rq->biotail = bio;
2658                 } else
2659                         rq->bio = rq->biotail = bio;
2660         }
2661 
2662         __blk_rq_prep_clone(rq, rq_src);
2663 
2664         return 0;
2665 
2666 free_and_out:
2667         if (bio)
2668                 bio_free(bio, bs);
2669         blk_rq_unprep_clone(rq);
2670 
2671         return -ENOMEM;
2672 }
2673 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2674 
2675 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2676 {
2677         return queue_work(kblockd_workqueue, work);
2678 }
2679 EXPORT_SYMBOL(kblockd_schedule_work);
2680 
2681 int kblockd_schedule_delayed_work(struct request_queue *q,
2682                         struct delayed_work *dwork, unsigned long delay)
2683 {
2684         return queue_delayed_work(kblockd_workqueue, dwork, delay);
2685 }
2686 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2687 
2688 #define PLUG_MAGIC      0x91827364
2689 
2690 /**
2691  * blk_start_plug - initialize blk_plug and track it inside the task_struct
2692  * @plug:       The &struct blk_plug that needs to be initialized
2693  *
2694  * Description:
2695  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
2696  *   pending I/O should the task end up blocking between blk_start_plug() and
2697  *   blk_finish_plug(). This is important from a performance perspective, but
2698  *   also ensures that we don't deadlock. For instance, if the task is blocking
2699  *   for a memory allocation, memory reclaim could end up wanting to free a
2700  *   page belonging to that request that is currently residing in our private
2701  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
2702  *   this kind of deadlock.
2703  */
2704 void blk_start_plug(struct blk_plug *plug)
2705 {
2706         struct task_struct *tsk = current;
2707 
2708         plug->magic = PLUG_MAGIC;
2709         INIT_LIST_HEAD(&plug->list);
2710         INIT_LIST_HEAD(&plug->cb_list);
2711         plug->should_sort = 0;
2712 
2713         /*
2714          * If this is a nested plug, don't actually assign it. It will be
2715          * flushed on its own.
2716          */
2717         if (!tsk->plug) {
2718                 /*
2719                  * Store ordering should not be needed here, since a potential
2720                  * preempt will imply a full memory barrier
2721                  */
2722                 tsk->plug = plug;
2723         }
2724 }
2725 EXPORT_SYMBOL(blk_start_plug);
2726 
2727 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2728 {
2729         struct request *rqa = container_of(a, struct request, queuelist);
2730         struct request *rqb = container_of(b, struct request, queuelist);
2731 
2732         return !(rqa->q <= rqb->q);
2733 }
2734 
2735 /*
2736  * If 'from_schedule' is true, then postpone the dispatch of requests
2737  * until a safe kblockd context. We due this to avoid accidental big
2738  * additional stack usage in driver dispatch, in places where the originally
2739  * plugger did not intend it.
2740  */
2741 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2742                             bool from_schedule)
2743         __releases(q->queue_lock)
2744 {
2745         trace_block_unplug(q, depth, !from_schedule);
2746 
2747         /*
2748          * If we are punting this to kblockd, then we can safely drop
2749          * the queue_lock before waking kblockd (which needs to take
2750          * this lock).
2751          */
2752         if (from_schedule) {
2753                 spin_unlock(q->queue_lock);
2754                 blk_run_queue_async(q);
2755         } else {
2756                 __blk_run_queue(q);
2757                 spin_unlock(q->queue_lock);
2758         }
2759 
2760 }
2761 
2762 static void flush_plug_callbacks(struct blk_plug *plug)
2763 {
2764         LIST_HEAD(callbacks);
2765 
2766         if (list_empty(&plug->cb_list))
2767                 return;
2768 
2769         list_splice_init(&plug->cb_list, &callbacks);
2770 
2771         while (!list_empty(&callbacks)) {
2772                 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2773                                                           struct blk_plug_cb,
2774                                                           list);
2775                 list_del(&cb->list);
2776                 cb->callback(cb);
2777         }
2778 }
2779 
2780 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2781 {
2782         struct request_queue *q;
2783         unsigned long flags;
2784         struct request *rq;
2785         LIST_HEAD(list);
2786         unsigned int depth;
2787 
2788         BUG_ON(plug->magic != PLUG_MAGIC);
2789 
2790         flush_plug_callbacks(plug);
2791         if (list_empty(&plug->list))
2792                 return;
2793 
2794         list_splice_init(&plug->list, &list);
2795 
2796         if (plug->should_sort) {
2797                 list_sort(NULL, &list, plug_rq_cmp);
2798                 plug->should_sort = 0;
2799         }
2800 
2801         q = NULL;
2802         depth = 0;
2803 
2804         /*
2805          * Save and disable interrupts here, to avoid doing it for every
2806          * queue lock we have to take.
2807          */
2808         local_irq_save(flags);
2809         while (!list_empty(&list)) {
2810                 rq = list_entry_rq(list.next);
2811                 list_del_init(&rq->queuelist);
2812                 BUG_ON(!rq->q);
2813                 if (rq->q != q) {
2814                         /*
2815                          * This drops the queue lock
2816                          */
2817                         if (q)
2818                                 queue_unplugged(q, depth, from_schedule);
2819                         q = rq->q;
2820                         depth = 0;
2821                         spin_lock(q->queue_lock);
2822                 }
2823                 /*
2824                  * rq is already accounted, so use raw insert
2825                  */
2826                 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2827                         __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2828                 else
2829                         __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2830 
2831                 depth++;
2832         }
2833 
2834         /*
2835          * This drops the queue lock
2836          */
2837         if (q)
2838                 queue_unplugged(q, depth, from_schedule);
2839 
2840         local_irq_restore(flags);
2841 }
2842 
2843 void blk_finish_plug(struct blk_plug *plug)
2844 {
2845         blk_flush_plug_list(plug, false);
2846 
2847         if (plug == current->plug)
2848                 current->plug = NULL;
2849 }
2850 EXPORT_SYMBOL(blk_finish_plug);
2851 
2852 int __init blk_dev_init(void)
2853 {
2854         BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2855                         sizeof(((struct request *)0)->cmd_flags));
2856 
2857         /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2858         kblockd_workqueue = alloc_workqueue("kblockd",
2859                                             WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2860         if (!kblockd_workqueue)
2861                 panic("Failed to create kblockd\n");
2862 
2863         request_cachep = kmem_cache_create("blkdev_requests",
2864                         sizeof(struct request), 0, SLAB_PANIC, NULL);
2865 
2866         blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2867                         sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2868 
2869         return 0;
2870 }
2871 

~ [ 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