Root/block/blk-core.c

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

Archive Download this file



interactive