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

Archive Download this file



interactive