Back to home page

LXR

 
 

    


0001 /*
0002  * Copyright (C) 1991, 1992 Linus Torvalds
0003  * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
0004  * Elevator latency, (C) 2000  Andrea Arcangeli <andrea@suse.de> SuSE
0005  * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
0006  * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
0007  *  -  July2000
0008  * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
0009  */
0010 
0011 /*
0012  * This handles all read/write requests to block devices
0013  */
0014 #include <linux/kernel.h>
0015 #include <linux/module.h>
0016 #include <linux/backing-dev.h>
0017 #include <linux/bio.h>
0018 #include <linux/blkdev.h>
0019 #include <linux/blk-mq.h>
0020 #include <linux/highmem.h>
0021 #include <linux/mm.h>
0022 #include <linux/kernel_stat.h>
0023 #include <linux/string.h>
0024 #include <linux/init.h>
0025 #include <linux/completion.h>
0026 #include <linux/slab.h>
0027 #include <linux/swap.h>
0028 #include <linux/writeback.h>
0029 #include <linux/task_io_accounting_ops.h>
0030 #include <linux/fault-inject.h>
0031 #include <linux/list_sort.h>
0032 #include <linux/delay.h>
0033 #include <linux/ratelimit.h>
0034 #include <linux/pm_runtime.h>
0035 #include <linux/blk-cgroup.h>
0036 
0037 #define CREATE_TRACE_POINTS
0038 #include <trace/events/block.h>
0039 
0040 #include "blk.h"
0041 #include "blk-mq.h"
0042 #include "blk-wbt.h"
0043 
0044 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
0045 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
0046 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
0047 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
0048 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
0049 
0050 DEFINE_IDA(blk_queue_ida);
0051 
0052 /*
0053  * For the allocated request tables
0054  */
0055 struct kmem_cache *request_cachep;
0056 
0057 /*
0058  * For queue allocation
0059  */
0060 struct kmem_cache *blk_requestq_cachep;
0061 
0062 /*
0063  * Controlling structure to kblockd
0064  */
0065 static struct workqueue_struct *kblockd_workqueue;
0066 
0067 static void blk_clear_congested(struct request_list *rl, int sync)
0068 {
0069 #ifdef CONFIG_CGROUP_WRITEBACK
0070     clear_wb_congested(rl->blkg->wb_congested, sync);
0071 #else
0072     /*
0073      * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
0074      * flip its congestion state for events on other blkcgs.
0075      */
0076     if (rl == &rl->q->root_rl)
0077         clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
0078 #endif
0079 }
0080 
0081 static void blk_set_congested(struct request_list *rl, int sync)
0082 {
0083 #ifdef CONFIG_CGROUP_WRITEBACK
0084     set_wb_congested(rl->blkg->wb_congested, sync);
0085 #else
0086     /* see blk_clear_congested() */
0087     if (rl == &rl->q->root_rl)
0088         set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
0089 #endif
0090 }
0091 
0092 void blk_queue_congestion_threshold(struct request_queue *q)
0093 {
0094     int nr;
0095 
0096     nr = q->nr_requests - (q->nr_requests / 8) + 1;
0097     if (nr > q->nr_requests)
0098         nr = q->nr_requests;
0099     q->nr_congestion_on = nr;
0100 
0101     nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
0102     if (nr < 1)
0103         nr = 1;
0104     q->nr_congestion_off = nr;
0105 }
0106 
0107 /**
0108  * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
0109  * @bdev:   device
0110  *
0111  * Locates the passed device's request queue and returns the address of its
0112  * backing_dev_info.  This function can only be called if @bdev is opened
0113  * and the return value is never NULL.
0114  */
0115 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
0116 {
0117     struct request_queue *q = bdev_get_queue(bdev);
0118 
0119     return &q->backing_dev_info;
0120 }
0121 EXPORT_SYMBOL(blk_get_backing_dev_info);
0122 
0123 void blk_rq_init(struct request_queue *q, struct request *rq)
0124 {
0125     memset(rq, 0, sizeof(*rq));
0126 
0127     INIT_LIST_HEAD(&rq->queuelist);
0128     INIT_LIST_HEAD(&rq->timeout_list);
0129     rq->cpu = -1;
0130     rq->q = q;
0131     rq->__sector = (sector_t) -1;
0132     INIT_HLIST_NODE(&rq->hash);
0133     RB_CLEAR_NODE(&rq->rb_node);
0134     rq->cmd = rq->__cmd;
0135     rq->cmd_len = BLK_MAX_CDB;
0136     rq->tag = -1;
0137     rq->start_time = jiffies;
0138     set_start_time_ns(rq);
0139     rq->part = NULL;
0140 }
0141 EXPORT_SYMBOL(blk_rq_init);
0142 
0143 static void req_bio_endio(struct request *rq, struct bio *bio,
0144               unsigned int nbytes, int error)
0145 {
0146     if (error)
0147         bio->bi_error = error;
0148 
0149     if (unlikely(rq->rq_flags & RQF_QUIET))
0150         bio_set_flag(bio, BIO_QUIET);
0151 
0152     bio_advance(bio, nbytes);
0153 
0154     /* don't actually finish bio if it's part of flush sequence */
0155     if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
0156         bio_endio(bio);
0157 }
0158 
0159 void blk_dump_rq_flags(struct request *rq, char *msg)
0160 {
0161     int bit;
0162 
0163     printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
0164         rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
0165         (unsigned long long) rq->cmd_flags);
0166 
0167     printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
0168            (unsigned long long)blk_rq_pos(rq),
0169            blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
0170     printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
0171            rq->bio, rq->biotail, blk_rq_bytes(rq));
0172 
0173     if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
0174         printk(KERN_INFO "  cdb: ");
0175         for (bit = 0; bit < BLK_MAX_CDB; bit++)
0176             printk("%02x ", rq->cmd[bit]);
0177         printk("\n");
0178     }
0179 }
0180 EXPORT_SYMBOL(blk_dump_rq_flags);
0181 
0182 static void blk_delay_work(struct work_struct *work)
0183 {
0184     struct request_queue *q;
0185 
0186     q = container_of(work, struct request_queue, delay_work.work);
0187     spin_lock_irq(q->queue_lock);
0188     __blk_run_queue(q);
0189     spin_unlock_irq(q->queue_lock);
0190 }
0191 
0192 /**
0193  * blk_delay_queue - restart queueing after defined interval
0194  * @q:      The &struct request_queue in question
0195  * @msecs:  Delay in msecs
0196  *
0197  * Description:
0198  *   Sometimes queueing needs to be postponed for a little while, to allow
0199  *   resources to come back. This function will make sure that queueing is
0200  *   restarted around the specified time. Queue lock must be held.
0201  */
0202 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
0203 {
0204     if (likely(!blk_queue_dead(q)))
0205         queue_delayed_work(kblockd_workqueue, &q->delay_work,
0206                    msecs_to_jiffies(msecs));
0207 }
0208 EXPORT_SYMBOL(blk_delay_queue);
0209 
0210 /**
0211  * blk_start_queue_async - asynchronously restart a previously stopped queue
0212  * @q:    The &struct request_queue in question
0213  *
0214  * Description:
0215  *   blk_start_queue_async() will clear the stop flag on the queue, and
0216  *   ensure that the request_fn for the queue is run from an async
0217  *   context.
0218  **/
0219 void blk_start_queue_async(struct request_queue *q)
0220 {
0221     queue_flag_clear(QUEUE_FLAG_STOPPED, q);
0222     blk_run_queue_async(q);
0223 }
0224 EXPORT_SYMBOL(blk_start_queue_async);
0225 
0226 /**
0227  * blk_start_queue - restart a previously stopped queue
0228  * @q:    The &struct request_queue in question
0229  *
0230  * Description:
0231  *   blk_start_queue() will clear the stop flag on the queue, and call
0232  *   the request_fn for the queue if it was in a stopped state when
0233  *   entered. Also see blk_stop_queue(). Queue lock must be held.
0234  **/
0235 void blk_start_queue(struct request_queue *q)
0236 {
0237     WARN_ON(!irqs_disabled());
0238 
0239     queue_flag_clear(QUEUE_FLAG_STOPPED, q);
0240     __blk_run_queue(q);
0241 }
0242 EXPORT_SYMBOL(blk_start_queue);
0243 
0244 /**
0245  * blk_stop_queue - stop a queue
0246  * @q:    The &struct request_queue in question
0247  *
0248  * Description:
0249  *   The Linux block layer assumes that a block driver will consume all
0250  *   entries on the request queue when the request_fn strategy is called.
0251  *   Often this will not happen, because of hardware limitations (queue
0252  *   depth settings). If a device driver gets a 'queue full' response,
0253  *   or if it simply chooses not to queue more I/O at one point, it can
0254  *   call this function to prevent the request_fn from being called until
0255  *   the driver has signalled it's ready to go again. This happens by calling
0256  *   blk_start_queue() to restart queue operations. Queue lock must be held.
0257  **/
0258 void blk_stop_queue(struct request_queue *q)
0259 {
0260     cancel_delayed_work(&q->delay_work);
0261     queue_flag_set(QUEUE_FLAG_STOPPED, q);
0262 }
0263 EXPORT_SYMBOL(blk_stop_queue);
0264 
0265 /**
0266  * blk_sync_queue - cancel any pending callbacks on a queue
0267  * @q: the queue
0268  *
0269  * Description:
0270  *     The block layer may perform asynchronous callback activity
0271  *     on a queue, such as calling the unplug function after a timeout.
0272  *     A block device may call blk_sync_queue to ensure that any
0273  *     such activity is cancelled, thus allowing it to release resources
0274  *     that the callbacks might use. The caller must already have made sure
0275  *     that its ->make_request_fn will not re-add plugging prior to calling
0276  *     this function.
0277  *
0278  *     This function does not cancel any asynchronous activity arising
0279  *     out of elevator or throttling code. That would require elevator_exit()
0280  *     and blkcg_exit_queue() to be called with queue lock initialized.
0281  *
0282  */
0283 void blk_sync_queue(struct request_queue *q)
0284 {
0285     del_timer_sync(&q->timeout);
0286 
0287     if (q->mq_ops) {
0288         struct blk_mq_hw_ctx *hctx;
0289         int i;
0290 
0291         queue_for_each_hw_ctx(q, hctx, i) {
0292             cancel_work_sync(&hctx->run_work);
0293             cancel_delayed_work_sync(&hctx->delay_work);
0294         }
0295     } else {
0296         cancel_delayed_work_sync(&q->delay_work);
0297     }
0298 }
0299 EXPORT_SYMBOL(blk_sync_queue);
0300 
0301 /**
0302  * __blk_run_queue_uncond - run a queue whether or not it has been stopped
0303  * @q:  The queue to run
0304  *
0305  * Description:
0306  *    Invoke request handling on a queue if there are any pending requests.
0307  *    May be used to restart request handling after a request has completed.
0308  *    This variant runs the queue whether or not the queue has been
0309  *    stopped. Must be called with the queue lock held and interrupts
0310  *    disabled. See also @blk_run_queue.
0311  */
0312 inline void __blk_run_queue_uncond(struct request_queue *q)
0313 {
0314     if (unlikely(blk_queue_dead(q)))
0315         return;
0316 
0317     /*
0318      * Some request_fn implementations, e.g. scsi_request_fn(), unlock
0319      * the queue lock internally. As a result multiple threads may be
0320      * running such a request function concurrently. Keep track of the
0321      * number of active request_fn invocations such that blk_drain_queue()
0322      * can wait until all these request_fn calls have finished.
0323      */
0324     q->request_fn_active++;
0325     q->request_fn(q);
0326     q->request_fn_active--;
0327 }
0328 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
0329 
0330 /**
0331  * __blk_run_queue - run a single device queue
0332  * @q:  The queue to run
0333  *
0334  * Description:
0335  *    See @blk_run_queue. This variant must be called with the queue lock
0336  *    held and interrupts disabled.
0337  */
0338 void __blk_run_queue(struct request_queue *q)
0339 {
0340     if (unlikely(blk_queue_stopped(q)))
0341         return;
0342 
0343     __blk_run_queue_uncond(q);
0344 }
0345 EXPORT_SYMBOL(__blk_run_queue);
0346 
0347 /**
0348  * blk_run_queue_async - run a single device queue in workqueue context
0349  * @q:  The queue to run
0350  *
0351  * Description:
0352  *    Tells kblockd to perform the equivalent of @blk_run_queue on behalf
0353  *    of us. The caller must hold the queue lock.
0354  */
0355 void blk_run_queue_async(struct request_queue *q)
0356 {
0357     if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
0358         mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
0359 }
0360 EXPORT_SYMBOL(blk_run_queue_async);
0361 
0362 /**
0363  * blk_run_queue - run a single device queue
0364  * @q: The queue to run
0365  *
0366  * Description:
0367  *    Invoke request handling on this queue, if it has pending work to do.
0368  *    May be used to restart queueing when a request has completed.
0369  */
0370 void blk_run_queue(struct request_queue *q)
0371 {
0372     unsigned long flags;
0373 
0374     spin_lock_irqsave(q->queue_lock, flags);
0375     __blk_run_queue(q);
0376     spin_unlock_irqrestore(q->queue_lock, flags);
0377 }
0378 EXPORT_SYMBOL(blk_run_queue);
0379 
0380 void blk_put_queue(struct request_queue *q)
0381 {
0382     kobject_put(&q->kobj);
0383 }
0384 EXPORT_SYMBOL(blk_put_queue);
0385 
0386 /**
0387  * __blk_drain_queue - drain requests from request_queue
0388  * @q: queue to drain
0389  * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
0390  *
0391  * Drain requests from @q.  If @drain_all is set, all requests are drained.
0392  * If not, only ELVPRIV requests are drained.  The caller is responsible
0393  * for ensuring that no new requests which need to be drained are queued.
0394  */
0395 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
0396     __releases(q->queue_lock)
0397     __acquires(q->queue_lock)
0398 {
0399     int i;
0400 
0401     lockdep_assert_held(q->queue_lock);
0402 
0403     while (true) {
0404         bool drain = false;
0405 
0406         /*
0407          * The caller might be trying to drain @q before its
0408          * elevator is initialized.
0409          */
0410         if (q->elevator)
0411             elv_drain_elevator(q);
0412 
0413         blkcg_drain_queue(q);
0414 
0415         /*
0416          * This function might be called on a queue which failed
0417          * driver init after queue creation or is not yet fully
0418          * active yet.  Some drivers (e.g. fd and loop) get unhappy
0419          * in such cases.  Kick queue iff dispatch queue has
0420          * something on it and @q has request_fn set.
0421          */
0422         if (!list_empty(&q->queue_head) && q->request_fn)
0423             __blk_run_queue(q);
0424 
0425         drain |= q->nr_rqs_elvpriv;
0426         drain |= q->request_fn_active;
0427 
0428         /*
0429          * Unfortunately, requests are queued at and tracked from
0430          * multiple places and there's no single counter which can
0431          * be drained.  Check all the queues and counters.
0432          */
0433         if (drain_all) {
0434             struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
0435             drain |= !list_empty(&q->queue_head);
0436             for (i = 0; i < 2; i++) {
0437                 drain |= q->nr_rqs[i];
0438                 drain |= q->in_flight[i];
0439                 if (fq)
0440                     drain |= !list_empty(&fq->flush_queue[i]);
0441             }
0442         }
0443 
0444         if (!drain)
0445             break;
0446 
0447         spin_unlock_irq(q->queue_lock);
0448 
0449         msleep(10);
0450 
0451         spin_lock_irq(q->queue_lock);
0452     }
0453 
0454     /*
0455      * With queue marked dead, any woken up waiter will fail the
0456      * allocation path, so the wakeup chaining is lost and we're
0457      * left with hung waiters. We need to wake up those waiters.
0458      */
0459     if (q->request_fn) {
0460         struct request_list *rl;
0461 
0462         blk_queue_for_each_rl(rl, q)
0463             for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
0464                 wake_up_all(&rl->wait[i]);
0465     }
0466 }
0467 
0468 /**
0469  * blk_queue_bypass_start - enter queue bypass mode
0470  * @q: queue of interest
0471  *
0472  * In bypass mode, only the dispatch FIFO queue of @q is used.  This
0473  * function makes @q enter bypass mode and drains all requests which were
0474  * throttled or issued before.  On return, it's guaranteed that no request
0475  * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
0476  * inside queue or RCU read lock.
0477  */
0478 void blk_queue_bypass_start(struct request_queue *q)
0479 {
0480     spin_lock_irq(q->queue_lock);
0481     q->bypass_depth++;
0482     queue_flag_set(QUEUE_FLAG_BYPASS, q);
0483     spin_unlock_irq(q->queue_lock);
0484 
0485     /*
0486      * Queues start drained.  Skip actual draining till init is
0487      * complete.  This avoids lenghty delays during queue init which
0488      * can happen many times during boot.
0489      */
0490     if (blk_queue_init_done(q)) {
0491         spin_lock_irq(q->queue_lock);
0492         __blk_drain_queue(q, false);
0493         spin_unlock_irq(q->queue_lock);
0494 
0495         /* ensure blk_queue_bypass() is %true inside RCU read lock */
0496         synchronize_rcu();
0497     }
0498 }
0499 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
0500 
0501 /**
0502  * blk_queue_bypass_end - leave queue bypass mode
0503  * @q: queue of interest
0504  *
0505  * Leave bypass mode and restore the normal queueing behavior.
0506  */
0507 void blk_queue_bypass_end(struct request_queue *q)
0508 {
0509     spin_lock_irq(q->queue_lock);
0510     if (!--q->bypass_depth)
0511         queue_flag_clear(QUEUE_FLAG_BYPASS, q);
0512     WARN_ON_ONCE(q->bypass_depth < 0);
0513     spin_unlock_irq(q->queue_lock);
0514 }
0515 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
0516 
0517 void blk_set_queue_dying(struct request_queue *q)
0518 {
0519     spin_lock_irq(q->queue_lock);
0520     queue_flag_set(QUEUE_FLAG_DYING, q);
0521     spin_unlock_irq(q->queue_lock);
0522 
0523     if (q->mq_ops)
0524         blk_mq_wake_waiters(q);
0525     else {
0526         struct request_list *rl;
0527 
0528         blk_queue_for_each_rl(rl, q) {
0529             if (rl->rq_pool) {
0530                 wake_up(&rl->wait[BLK_RW_SYNC]);
0531                 wake_up(&rl->wait[BLK_RW_ASYNC]);
0532             }
0533         }
0534     }
0535 }
0536 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
0537 
0538 /**
0539  * blk_cleanup_queue - shutdown a request queue
0540  * @q: request queue to shutdown
0541  *
0542  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
0543  * put it.  All future requests will be failed immediately with -ENODEV.
0544  */
0545 void blk_cleanup_queue(struct request_queue *q)
0546 {
0547     spinlock_t *lock = q->queue_lock;
0548 
0549     /* mark @q DYING, no new request or merges will be allowed afterwards */
0550     mutex_lock(&q->sysfs_lock);
0551     blk_set_queue_dying(q);
0552     spin_lock_irq(lock);
0553 
0554     /*
0555      * A dying queue is permanently in bypass mode till released.  Note
0556      * that, unlike blk_queue_bypass_start(), we aren't performing
0557      * synchronize_rcu() after entering bypass mode to avoid the delay
0558      * as some drivers create and destroy a lot of queues while
0559      * probing.  This is still safe because blk_release_queue() will be
0560      * called only after the queue refcnt drops to zero and nothing,
0561      * RCU or not, would be traversing the queue by then.
0562      */
0563     q->bypass_depth++;
0564     queue_flag_set(QUEUE_FLAG_BYPASS, q);
0565 
0566     queue_flag_set(QUEUE_FLAG_NOMERGES, q);
0567     queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
0568     queue_flag_set(QUEUE_FLAG_DYING, q);
0569     spin_unlock_irq(lock);
0570     mutex_unlock(&q->sysfs_lock);
0571 
0572     /*
0573      * Drain all requests queued before DYING marking. Set DEAD flag to
0574      * prevent that q->request_fn() gets invoked after draining finished.
0575      */
0576     blk_freeze_queue(q);
0577     spin_lock_irq(lock);
0578     if (!q->mq_ops)
0579         __blk_drain_queue(q, true);
0580     queue_flag_set(QUEUE_FLAG_DEAD, q);
0581     spin_unlock_irq(lock);
0582 
0583     /* for synchronous bio-based driver finish in-flight integrity i/o */
0584     blk_flush_integrity();
0585 
0586     /* @q won't process any more request, flush async actions */
0587     del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
0588     blk_sync_queue(q);
0589 
0590     if (q->mq_ops)
0591         blk_mq_free_queue(q);
0592     percpu_ref_exit(&q->q_usage_counter);
0593 
0594     spin_lock_irq(lock);
0595     if (q->queue_lock != &q->__queue_lock)
0596         q->queue_lock = &q->__queue_lock;
0597     spin_unlock_irq(lock);
0598 
0599     bdi_unregister(&q->backing_dev_info);
0600 
0601     /* @q is and will stay empty, shutdown and put */
0602     blk_put_queue(q);
0603 }
0604 EXPORT_SYMBOL(blk_cleanup_queue);
0605 
0606 /* Allocate memory local to the request queue */
0607 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
0608 {
0609     int nid = (int)(long)data;
0610     return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
0611 }
0612 
0613 static void free_request_struct(void *element, void *unused)
0614 {
0615     kmem_cache_free(request_cachep, element);
0616 }
0617 
0618 int blk_init_rl(struct request_list *rl, struct request_queue *q,
0619         gfp_t gfp_mask)
0620 {
0621     if (unlikely(rl->rq_pool))
0622         return 0;
0623 
0624     rl->q = q;
0625     rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
0626     rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
0627     init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
0628     init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
0629 
0630     rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
0631                       free_request_struct,
0632                       (void *)(long)q->node, gfp_mask,
0633                       q->node);
0634     if (!rl->rq_pool)
0635         return -ENOMEM;
0636 
0637     return 0;
0638 }
0639 
0640 void blk_exit_rl(struct request_list *rl)
0641 {
0642     if (rl->rq_pool)
0643         mempool_destroy(rl->rq_pool);
0644 }
0645 
0646 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
0647 {
0648     return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
0649 }
0650 EXPORT_SYMBOL(blk_alloc_queue);
0651 
0652 int blk_queue_enter(struct request_queue *q, bool nowait)
0653 {
0654     while (true) {
0655         int ret;
0656 
0657         if (percpu_ref_tryget_live(&q->q_usage_counter))
0658             return 0;
0659 
0660         if (nowait)
0661             return -EBUSY;
0662 
0663         ret = wait_event_interruptible(q->mq_freeze_wq,
0664                 !atomic_read(&q->mq_freeze_depth) ||
0665                 blk_queue_dying(q));
0666         if (blk_queue_dying(q))
0667             return -ENODEV;
0668         if (ret)
0669             return ret;
0670     }
0671 }
0672 
0673 void blk_queue_exit(struct request_queue *q)
0674 {
0675     percpu_ref_put(&q->q_usage_counter);
0676 }
0677 
0678 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
0679 {
0680     struct request_queue *q =
0681         container_of(ref, struct request_queue, q_usage_counter);
0682 
0683     wake_up_all(&q->mq_freeze_wq);
0684 }
0685 
0686 static void blk_rq_timed_out_timer(unsigned long data)
0687 {
0688     struct request_queue *q = (struct request_queue *)data;
0689 
0690     kblockd_schedule_work(&q->timeout_work);
0691 }
0692 
0693 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
0694 {
0695     struct request_queue *q;
0696     int err;
0697 
0698     q = kmem_cache_alloc_node(blk_requestq_cachep,
0699                 gfp_mask | __GFP_ZERO, node_id);
0700     if (!q)
0701         return NULL;
0702 
0703     q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
0704     if (q->id < 0)
0705         goto fail_q;
0706 
0707     q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
0708     if (!q->bio_split)
0709         goto fail_id;
0710 
0711     q->backing_dev_info.ra_pages =
0712             (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
0713     q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
0714     q->backing_dev_info.name = "block";
0715     q->node = node_id;
0716 
0717     err = bdi_init(&q->backing_dev_info);
0718     if (err)
0719         goto fail_split;
0720 
0721     setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
0722             laptop_mode_timer_fn, (unsigned long) q);
0723     setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
0724     INIT_LIST_HEAD(&q->queue_head);
0725     INIT_LIST_HEAD(&q->timeout_list);
0726     INIT_LIST_HEAD(&q->icq_list);
0727 #ifdef CONFIG_BLK_CGROUP
0728     INIT_LIST_HEAD(&q->blkg_list);
0729 #endif
0730     INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
0731 
0732     kobject_init(&q->kobj, &blk_queue_ktype);
0733 
0734     mutex_init(&q->sysfs_lock);
0735     spin_lock_init(&q->__queue_lock);
0736 
0737     /*
0738      * By default initialize queue_lock to internal lock and driver can
0739      * override it later if need be.
0740      */
0741     q->queue_lock = &q->__queue_lock;
0742 
0743     /*
0744      * A queue starts its life with bypass turned on to avoid
0745      * unnecessary bypass on/off overhead and nasty surprises during
0746      * init.  The initial bypass will be finished when the queue is
0747      * registered by blk_register_queue().
0748      */
0749     q->bypass_depth = 1;
0750     __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
0751 
0752     init_waitqueue_head(&q->mq_freeze_wq);
0753 
0754     /*
0755      * Init percpu_ref in atomic mode so that it's faster to shutdown.
0756      * See blk_register_queue() for details.
0757      */
0758     if (percpu_ref_init(&q->q_usage_counter,
0759                 blk_queue_usage_counter_release,
0760                 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
0761         goto fail_bdi;
0762 
0763     if (blkcg_init_queue(q))
0764         goto fail_ref;
0765 
0766     return q;
0767 
0768 fail_ref:
0769     percpu_ref_exit(&q->q_usage_counter);
0770 fail_bdi:
0771     bdi_destroy(&q->backing_dev_info);
0772 fail_split:
0773     bioset_free(q->bio_split);
0774 fail_id:
0775     ida_simple_remove(&blk_queue_ida, q->id);
0776 fail_q:
0777     kmem_cache_free(blk_requestq_cachep, q);
0778     return NULL;
0779 }
0780 EXPORT_SYMBOL(blk_alloc_queue_node);
0781 
0782 /**
0783  * blk_init_queue  - prepare a request queue for use with a block device
0784  * @rfn:  The function to be called to process requests that have been
0785  *        placed on the queue.
0786  * @lock: Request queue spin lock
0787  *
0788  * Description:
0789  *    If a block device wishes to use the standard request handling procedures,
0790  *    which sorts requests and coalesces adjacent requests, then it must
0791  *    call blk_init_queue().  The function @rfn will be called when there
0792  *    are requests on the queue that need to be processed.  If the device
0793  *    supports plugging, then @rfn may not be called immediately when requests
0794  *    are available on the queue, but may be called at some time later instead.
0795  *    Plugged queues are generally unplugged when a buffer belonging to one
0796  *    of the requests on the queue is needed, or due to memory pressure.
0797  *
0798  *    @rfn is not required, or even expected, to remove all requests off the
0799  *    queue, but only as many as it can handle at a time.  If it does leave
0800  *    requests on the queue, it is responsible for arranging that the requests
0801  *    get dealt with eventually.
0802  *
0803  *    The queue spin lock must be held while manipulating the requests on the
0804  *    request queue; this lock will be taken also from interrupt context, so irq
0805  *    disabling is needed for it.
0806  *
0807  *    Function returns a pointer to the initialized request queue, or %NULL if
0808  *    it didn't succeed.
0809  *
0810  * Note:
0811  *    blk_init_queue() must be paired with a blk_cleanup_queue() call
0812  *    when the block device is deactivated (such as at module unload).
0813  **/
0814 
0815 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
0816 {
0817     return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
0818 }
0819 EXPORT_SYMBOL(blk_init_queue);
0820 
0821 struct request_queue *
0822 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
0823 {
0824     struct request_queue *uninit_q, *q;
0825 
0826     uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
0827     if (!uninit_q)
0828         return NULL;
0829 
0830     q = blk_init_allocated_queue(uninit_q, rfn, lock);
0831     if (!q)
0832         blk_cleanup_queue(uninit_q);
0833 
0834     return q;
0835 }
0836 EXPORT_SYMBOL(blk_init_queue_node);
0837 
0838 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
0839 
0840 struct request_queue *
0841 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
0842              spinlock_t *lock)
0843 {
0844     if (!q)
0845         return NULL;
0846 
0847     q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
0848     if (!q->fq)
0849         return NULL;
0850 
0851     if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
0852         goto fail;
0853 
0854     INIT_WORK(&q->timeout_work, blk_timeout_work);
0855     q->request_fn       = rfn;
0856     q->prep_rq_fn       = NULL;
0857     q->unprep_rq_fn     = NULL;
0858     q->queue_flags      |= QUEUE_FLAG_DEFAULT;
0859 
0860     /* Override internal queue lock with supplied lock pointer */
0861     if (lock)
0862         q->queue_lock       = lock;
0863 
0864     /*
0865      * This also sets hw/phys segments, boundary and size
0866      */
0867     blk_queue_make_request(q, blk_queue_bio);
0868 
0869     q->sg_reserved_size = INT_MAX;
0870 
0871     /* Protect q->elevator from elevator_change */
0872     mutex_lock(&q->sysfs_lock);
0873 
0874     /* init elevator */
0875     if (elevator_init(q, NULL)) {
0876         mutex_unlock(&q->sysfs_lock);
0877         goto fail;
0878     }
0879 
0880     mutex_unlock(&q->sysfs_lock);
0881 
0882     return q;
0883 
0884 fail:
0885     blk_free_flush_queue(q->fq);
0886     wbt_exit(q);
0887     return NULL;
0888 }
0889 EXPORT_SYMBOL(blk_init_allocated_queue);
0890 
0891 bool blk_get_queue(struct request_queue *q)
0892 {
0893     if (likely(!blk_queue_dying(q))) {
0894         __blk_get_queue(q);
0895         return true;
0896     }
0897 
0898     return false;
0899 }
0900 EXPORT_SYMBOL(blk_get_queue);
0901 
0902 static inline void blk_free_request(struct request_list *rl, struct request *rq)
0903 {
0904     if (rq->rq_flags & RQF_ELVPRIV) {
0905         elv_put_request(rl->q, rq);
0906         if (rq->elv.icq)
0907             put_io_context(rq->elv.icq->ioc);
0908     }
0909 
0910     mempool_free(rq, rl->rq_pool);
0911 }
0912 
0913 /*
0914  * ioc_batching returns true if the ioc is a valid batching request and
0915  * should be given priority access to a request.
0916  */
0917 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
0918 {
0919     if (!ioc)
0920         return 0;
0921 
0922     /*
0923      * Make sure the process is able to allocate at least 1 request
0924      * even if the batch times out, otherwise we could theoretically
0925      * lose wakeups.
0926      */
0927     return ioc->nr_batch_requests == q->nr_batching ||
0928         (ioc->nr_batch_requests > 0
0929         && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
0930 }
0931 
0932 /*
0933  * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
0934  * will cause the process to be a "batcher" on all queues in the system. This
0935  * is the behaviour we want though - once it gets a wakeup it should be given
0936  * a nice run.
0937  */
0938 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
0939 {
0940     if (!ioc || ioc_batching(q, ioc))
0941         return;
0942 
0943     ioc->nr_batch_requests = q->nr_batching;
0944     ioc->last_waited = jiffies;
0945 }
0946 
0947 static void __freed_request(struct request_list *rl, int sync)
0948 {
0949     struct request_queue *q = rl->q;
0950 
0951     if (rl->count[sync] < queue_congestion_off_threshold(q))
0952         blk_clear_congested(rl, sync);
0953 
0954     if (rl->count[sync] + 1 <= q->nr_requests) {
0955         if (waitqueue_active(&rl->wait[sync]))
0956             wake_up(&rl->wait[sync]);
0957 
0958         blk_clear_rl_full(rl, sync);
0959     }
0960 }
0961 
0962 /*
0963  * A request has just been released.  Account for it, update the full and
0964  * congestion status, wake up any waiters.   Called under q->queue_lock.
0965  */
0966 static void freed_request(struct request_list *rl, bool sync,
0967         req_flags_t rq_flags)
0968 {
0969     struct request_queue *q = rl->q;
0970 
0971     q->nr_rqs[sync]--;
0972     rl->count[sync]--;
0973     if (rq_flags & RQF_ELVPRIV)
0974         q->nr_rqs_elvpriv--;
0975 
0976     __freed_request(rl, sync);
0977 
0978     if (unlikely(rl->starved[sync ^ 1]))
0979         __freed_request(rl, sync ^ 1);
0980 }
0981 
0982 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
0983 {
0984     struct request_list *rl;
0985     int on_thresh, off_thresh;
0986 
0987     spin_lock_irq(q->queue_lock);
0988     q->nr_requests = nr;
0989     blk_queue_congestion_threshold(q);
0990     on_thresh = queue_congestion_on_threshold(q);
0991     off_thresh = queue_congestion_off_threshold(q);
0992 
0993     blk_queue_for_each_rl(rl, q) {
0994         if (rl->count[BLK_RW_SYNC] >= on_thresh)
0995             blk_set_congested(rl, BLK_RW_SYNC);
0996         else if (rl->count[BLK_RW_SYNC] < off_thresh)
0997             blk_clear_congested(rl, BLK_RW_SYNC);
0998 
0999         if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1000             blk_set_congested(rl, BLK_RW_ASYNC);
1001         else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1002             blk_clear_congested(rl, BLK_RW_ASYNC);
1003 
1004         if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1005             blk_set_rl_full(rl, BLK_RW_SYNC);
1006         } else {
1007             blk_clear_rl_full(rl, BLK_RW_SYNC);
1008             wake_up(&rl->wait[BLK_RW_SYNC]);
1009         }
1010 
1011         if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1012             blk_set_rl_full(rl, BLK_RW_ASYNC);
1013         } else {
1014             blk_clear_rl_full(rl, BLK_RW_ASYNC);
1015             wake_up(&rl->wait[BLK_RW_ASYNC]);
1016         }
1017     }
1018 
1019     spin_unlock_irq(q->queue_lock);
1020     return 0;
1021 }
1022 
1023 /*
1024  * Determine if elevator data should be initialized when allocating the
1025  * request associated with @bio.
1026  */
1027 static bool blk_rq_should_init_elevator(struct bio *bio)
1028 {
1029     if (!bio)
1030         return true;
1031 
1032     /*
1033      * Flush requests do not use the elevator so skip initialization.
1034      * This allows a request to share the flush and elevator data.
1035      */
1036     if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA))
1037         return false;
1038 
1039     return true;
1040 }
1041 
1042 /**
1043  * rq_ioc - determine io_context for request allocation
1044  * @bio: request being allocated is for this bio (can be %NULL)
1045  *
1046  * Determine io_context to use for request allocation for @bio.  May return
1047  * %NULL if %current->io_context doesn't exist.
1048  */
1049 static struct io_context *rq_ioc(struct bio *bio)
1050 {
1051 #ifdef CONFIG_BLK_CGROUP
1052     if (bio && bio->bi_ioc)
1053         return bio->bi_ioc;
1054 #endif
1055     return current->io_context;
1056 }
1057 
1058 /**
1059  * __get_request - get a free request
1060  * @rl: request list to allocate from
1061  * @op: operation and flags
1062  * @bio: bio to allocate request for (can be %NULL)
1063  * @gfp_mask: allocation mask
1064  *
1065  * Get a free request from @q.  This function may fail under memory
1066  * pressure or if @q is dead.
1067  *
1068  * Must be called with @q->queue_lock held and,
1069  * Returns ERR_PTR on failure, with @q->queue_lock held.
1070  * Returns request pointer on success, with @q->queue_lock *not held*.
1071  */
1072 static struct request *__get_request(struct request_list *rl, unsigned int op,
1073         struct bio *bio, gfp_t gfp_mask)
1074 {
1075     struct request_queue *q = rl->q;
1076     struct request *rq;
1077     struct elevator_type *et = q->elevator->type;
1078     struct io_context *ioc = rq_ioc(bio);
1079     struct io_cq *icq = NULL;
1080     const bool is_sync = op_is_sync(op);
1081     int may_queue;
1082     req_flags_t rq_flags = RQF_ALLOCED;
1083 
1084     if (unlikely(blk_queue_dying(q)))
1085         return ERR_PTR(-ENODEV);
1086 
1087     may_queue = elv_may_queue(q, op);
1088     if (may_queue == ELV_MQUEUE_NO)
1089         goto rq_starved;
1090 
1091     if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1092         if (rl->count[is_sync]+1 >= q->nr_requests) {
1093             /*
1094              * The queue will fill after this allocation, so set
1095              * it as full, and mark this process as "batching".
1096              * This process will be allowed to complete a batch of
1097              * requests, others will be blocked.
1098              */
1099             if (!blk_rl_full(rl, is_sync)) {
1100                 ioc_set_batching(q, ioc);
1101                 blk_set_rl_full(rl, is_sync);
1102             } else {
1103                 if (may_queue != ELV_MQUEUE_MUST
1104                         && !ioc_batching(q, ioc)) {
1105                     /*
1106                      * The queue is full and the allocating
1107                      * process is not a "batcher", and not
1108                      * exempted by the IO scheduler
1109                      */
1110                     return ERR_PTR(-ENOMEM);
1111                 }
1112             }
1113         }
1114         blk_set_congested(rl, is_sync);
1115     }
1116 
1117     /*
1118      * Only allow batching queuers to allocate up to 50% over the defined
1119      * limit of requests, otherwise we could have thousands of requests
1120      * allocated with any setting of ->nr_requests
1121      */
1122     if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1123         return ERR_PTR(-ENOMEM);
1124 
1125     q->nr_rqs[is_sync]++;
1126     rl->count[is_sync]++;
1127     rl->starved[is_sync] = 0;
1128 
1129     /*
1130      * Decide whether the new request will be managed by elevator.  If
1131      * so, mark @rq_flags and increment elvpriv.  Non-zero elvpriv will
1132      * prevent the current elevator from being destroyed until the new
1133      * request is freed.  This guarantees icq's won't be destroyed and
1134      * makes creating new ones safe.
1135      *
1136      * Also, lookup icq while holding queue_lock.  If it doesn't exist,
1137      * it will be created after releasing queue_lock.
1138      */
1139     if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1140         rq_flags |= RQF_ELVPRIV;
1141         q->nr_rqs_elvpriv++;
1142         if (et->icq_cache && ioc)
1143             icq = ioc_lookup_icq(ioc, q);
1144     }
1145 
1146     if (blk_queue_io_stat(q))
1147         rq_flags |= RQF_IO_STAT;
1148     spin_unlock_irq(q->queue_lock);
1149 
1150     /* allocate and init request */
1151     rq = mempool_alloc(rl->rq_pool, gfp_mask);
1152     if (!rq)
1153         goto fail_alloc;
1154 
1155     blk_rq_init(q, rq);
1156     blk_rq_set_rl(rq, rl);
1157     blk_rq_set_prio(rq, ioc);
1158     rq->cmd_flags = op;
1159     rq->rq_flags = rq_flags;
1160 
1161     /* init elvpriv */
1162     if (rq_flags & RQF_ELVPRIV) {
1163         if (unlikely(et->icq_cache && !icq)) {
1164             if (ioc)
1165                 icq = ioc_create_icq(ioc, q, gfp_mask);
1166             if (!icq)
1167                 goto fail_elvpriv;
1168         }
1169 
1170         rq->elv.icq = icq;
1171         if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1172             goto fail_elvpriv;
1173 
1174         /* @rq->elv.icq holds io_context until @rq is freed */
1175         if (icq)
1176             get_io_context(icq->ioc);
1177     }
1178 out:
1179     /*
1180      * ioc may be NULL here, and ioc_batching will be false. That's
1181      * OK, if the queue is under the request limit then requests need
1182      * not count toward the nr_batch_requests limit. There will always
1183      * be some limit enforced by BLK_BATCH_TIME.
1184      */
1185     if (ioc_batching(q, ioc))
1186         ioc->nr_batch_requests--;
1187 
1188     trace_block_getrq(q, bio, op);
1189     return rq;
1190 
1191 fail_elvpriv:
1192     /*
1193      * elvpriv init failed.  ioc, icq and elvpriv aren't mempool backed
1194      * and may fail indefinitely under memory pressure and thus
1195      * shouldn't stall IO.  Treat this request as !elvpriv.  This will
1196      * disturb iosched and blkcg but weird is bettern than dead.
1197      */
1198     printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1199                __func__, dev_name(q->backing_dev_info.dev));
1200 
1201     rq->rq_flags &= ~RQF_ELVPRIV;
1202     rq->elv.icq = NULL;
1203 
1204     spin_lock_irq(q->queue_lock);
1205     q->nr_rqs_elvpriv--;
1206     spin_unlock_irq(q->queue_lock);
1207     goto out;
1208 
1209 fail_alloc:
1210     /*
1211      * Allocation failed presumably due to memory. Undo anything we
1212      * might have messed up.
1213      *
1214      * Allocating task should really be put onto the front of the wait
1215      * queue, but this is pretty rare.
1216      */
1217     spin_lock_irq(q->queue_lock);
1218     freed_request(rl, is_sync, rq_flags);
1219 
1220     /*
1221      * in the very unlikely event that allocation failed and no
1222      * requests for this direction was pending, mark us starved so that
1223      * freeing of a request in the other direction will notice
1224      * us. another possible fix would be to split the rq mempool into
1225      * READ and WRITE
1226      */
1227 rq_starved:
1228     if (unlikely(rl->count[is_sync] == 0))
1229         rl->starved[is_sync] = 1;
1230     return ERR_PTR(-ENOMEM);
1231 }
1232 
1233 /**
1234  * get_request - get a free request
1235  * @q: request_queue to allocate request from
1236  * @op: operation and flags
1237  * @bio: bio to allocate request for (can be %NULL)
1238  * @gfp_mask: allocation mask
1239  *
1240  * Get a free request from @q.  If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1241  * this function keeps retrying under memory pressure and fails iff @q is dead.
1242  *
1243  * Must be called with @q->queue_lock held and,
1244  * Returns ERR_PTR on failure, with @q->queue_lock held.
1245  * Returns request pointer on success, with @q->queue_lock *not held*.
1246  */
1247 static struct request *get_request(struct request_queue *q, unsigned int op,
1248         struct bio *bio, gfp_t gfp_mask)
1249 {
1250     const bool is_sync = op_is_sync(op);
1251     DEFINE_WAIT(wait);
1252     struct request_list *rl;
1253     struct request *rq;
1254 
1255     rl = blk_get_rl(q, bio);    /* transferred to @rq on success */
1256 retry:
1257     rq = __get_request(rl, op, bio, gfp_mask);
1258     if (!IS_ERR(rq))
1259         return rq;
1260 
1261     if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1262         blk_put_rl(rl);
1263         return rq;
1264     }
1265 
1266     /* wait on @rl and retry */
1267     prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1268                   TASK_UNINTERRUPTIBLE);
1269 
1270     trace_block_sleeprq(q, bio, op);
1271 
1272     spin_unlock_irq(q->queue_lock);
1273     io_schedule();
1274 
1275     /*
1276      * After sleeping, we become a "batching" process and will be able
1277      * to allocate at least one request, and up to a big batch of them
1278      * for a small period time.  See ioc_batching, ioc_set_batching
1279      */
1280     ioc_set_batching(q, current->io_context);
1281 
1282     spin_lock_irq(q->queue_lock);
1283     finish_wait(&rl->wait[is_sync], &wait);
1284 
1285     goto retry;
1286 }
1287 
1288 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1289         gfp_t gfp_mask)
1290 {
1291     struct request *rq;
1292 
1293     BUG_ON(rw != READ && rw != WRITE);
1294 
1295     /* create ioc upfront */
1296     create_io_context(gfp_mask, q->node);
1297 
1298     spin_lock_irq(q->queue_lock);
1299     rq = get_request(q, rw, NULL, gfp_mask);
1300     if (IS_ERR(rq)) {
1301         spin_unlock_irq(q->queue_lock);
1302         return rq;
1303     }
1304 
1305     /* q->queue_lock is unlocked at this point */
1306     rq->__data_len = 0;
1307     rq->__sector = (sector_t) -1;
1308     rq->bio = rq->biotail = NULL;
1309     return rq;
1310 }
1311 
1312 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1313 {
1314     if (q->mq_ops)
1315         return blk_mq_alloc_request(q, rw,
1316             (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1317                 0 : BLK_MQ_REQ_NOWAIT);
1318     else
1319         return blk_old_get_request(q, rw, gfp_mask);
1320 }
1321 EXPORT_SYMBOL(blk_get_request);
1322 
1323 /**
1324  * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1325  * @rq:     request to be initialized
1326  *
1327  */
1328 void blk_rq_set_block_pc(struct request *rq)
1329 {
1330     rq->cmd_type = REQ_TYPE_BLOCK_PC;
1331     memset(rq->__cmd, 0, sizeof(rq->__cmd));
1332 }
1333 EXPORT_SYMBOL(blk_rq_set_block_pc);
1334 
1335 /**
1336  * blk_requeue_request - put a request back on queue
1337  * @q:      request queue where request should be inserted
1338  * @rq:     request to be inserted
1339  *
1340  * Description:
1341  *    Drivers often keep queueing requests until the hardware cannot accept
1342  *    more, when that condition happens we need to put the request back
1343  *    on the queue. Must be called with queue lock held.
1344  */
1345 void blk_requeue_request(struct request_queue *q, struct request *rq)
1346 {
1347     blk_delete_timer(rq);
1348     blk_clear_rq_complete(rq);
1349     trace_block_rq_requeue(q, rq);
1350     wbt_requeue(q->rq_wb, &rq->issue_stat);
1351 
1352     if (rq->rq_flags & RQF_QUEUED)
1353         blk_queue_end_tag(q, rq);
1354 
1355     BUG_ON(blk_queued_rq(rq));
1356 
1357     elv_requeue_request(q, rq);
1358 }
1359 EXPORT_SYMBOL(blk_requeue_request);
1360 
1361 static void add_acct_request(struct request_queue *q, struct request *rq,
1362                  int where)
1363 {
1364     blk_account_io_start(rq, true);
1365     __elv_add_request(q, rq, where);
1366 }
1367 
1368 static void part_round_stats_single(int cpu, struct hd_struct *part,
1369                     unsigned long now)
1370 {
1371     int inflight;
1372 
1373     if (now == part->stamp)
1374         return;
1375 
1376     inflight = part_in_flight(part);
1377     if (inflight) {
1378         __part_stat_add(cpu, part, time_in_queue,
1379                 inflight * (now - part->stamp));
1380         __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1381     }
1382     part->stamp = now;
1383 }
1384 
1385 /**
1386  * part_round_stats() - Round off the performance stats on a struct disk_stats.
1387  * @cpu: cpu number for stats access
1388  * @part: target partition
1389  *
1390  * The average IO queue length and utilisation statistics are maintained
1391  * by observing the current state of the queue length and the amount of
1392  * time it has been in this state for.
1393  *
1394  * Normally, that accounting is done on IO completion, but that can result
1395  * in more than a second's worth of IO being accounted for within any one
1396  * second, leading to >100% utilisation.  To deal with that, we call this
1397  * function to do a round-off before returning the results when reading
1398  * /proc/diskstats.  This accounts immediately for all queue usage up to
1399  * the current jiffies and restarts the counters again.
1400  */
1401 void part_round_stats(int cpu, struct hd_struct *part)
1402 {
1403     unsigned long now = jiffies;
1404 
1405     if (part->partno)
1406         part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1407     part_round_stats_single(cpu, part, now);
1408 }
1409 EXPORT_SYMBOL_GPL(part_round_stats);
1410 
1411 #ifdef CONFIG_PM
1412 static void blk_pm_put_request(struct request *rq)
1413 {
1414     if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1415         pm_runtime_mark_last_busy(rq->q->dev);
1416 }
1417 #else
1418 static inline void blk_pm_put_request(struct request *rq) {}
1419 #endif
1420 
1421 /*
1422  * queue lock must be held
1423  */
1424 void __blk_put_request(struct request_queue *q, struct request *req)
1425 {
1426     req_flags_t rq_flags = req->rq_flags;
1427 
1428     if (unlikely(!q))
1429         return;
1430 
1431     if (q->mq_ops) {
1432         blk_mq_free_request(req);
1433         return;
1434     }
1435 
1436     blk_pm_put_request(req);
1437 
1438     elv_completed_request(q, req);
1439 
1440     /* this is a bio leak */
1441     WARN_ON(req->bio != NULL);
1442 
1443     wbt_done(q->rq_wb, &req->issue_stat);
1444 
1445     /*
1446      * Request may not have originated from ll_rw_blk. if not,
1447      * it didn't come out of our reserved rq pools
1448      */
1449     if (rq_flags & RQF_ALLOCED) {
1450         struct request_list *rl = blk_rq_rl(req);
1451         bool sync = op_is_sync(req->cmd_flags);
1452 
1453         BUG_ON(!list_empty(&req->queuelist));
1454         BUG_ON(ELV_ON_HASH(req));
1455 
1456         blk_free_request(rl, req);
1457         freed_request(rl, sync, rq_flags);
1458         blk_put_rl(rl);
1459     }
1460 }
1461 EXPORT_SYMBOL_GPL(__blk_put_request);
1462 
1463 void blk_put_request(struct request *req)
1464 {
1465     struct request_queue *q = req->q;
1466 
1467     if (q->mq_ops)
1468         blk_mq_free_request(req);
1469     else {
1470         unsigned long flags;
1471 
1472         spin_lock_irqsave(q->queue_lock, flags);
1473         __blk_put_request(q, req);
1474         spin_unlock_irqrestore(q->queue_lock, flags);
1475     }
1476 }
1477 EXPORT_SYMBOL(blk_put_request);
1478 
1479 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1480                 struct bio *bio)
1481 {
1482     const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1483 
1484     if (!ll_back_merge_fn(q, req, bio))
1485         return false;
1486 
1487     trace_block_bio_backmerge(q, req, bio);
1488 
1489     if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1490         blk_rq_set_mixed_merge(req);
1491 
1492     req->biotail->bi_next = bio;
1493     req->biotail = bio;
1494     req->__data_len += bio->bi_iter.bi_size;
1495     req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1496 
1497     blk_account_io_start(req, false);
1498     return true;
1499 }
1500 
1501 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1502                  struct bio *bio)
1503 {
1504     const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1505 
1506     if (!ll_front_merge_fn(q, req, bio))
1507         return false;
1508 
1509     trace_block_bio_frontmerge(q, req, bio);
1510 
1511     if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1512         blk_rq_set_mixed_merge(req);
1513 
1514     bio->bi_next = req->bio;
1515     req->bio = bio;
1516 
1517     req->__sector = bio->bi_iter.bi_sector;
1518     req->__data_len += bio->bi_iter.bi_size;
1519     req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1520 
1521     blk_account_io_start(req, false);
1522     return true;
1523 }
1524 
1525 /**
1526  * blk_attempt_plug_merge - try to merge with %current's plugged list
1527  * @q: request_queue new bio is being queued at
1528  * @bio: new bio being queued
1529  * @request_count: out parameter for number of traversed plugged requests
1530  * @same_queue_rq: pointer to &struct request that gets filled in when
1531  * another request associated with @q is found on the plug list
1532  * (optional, may be %NULL)
1533  *
1534  * Determine whether @bio being queued on @q can be merged with a request
1535  * on %current's plugged list.  Returns %true if merge was successful,
1536  * otherwise %false.
1537  *
1538  * Plugging coalesces IOs from the same issuer for the same purpose without
1539  * going through @q->queue_lock.  As such it's more of an issuing mechanism
1540  * than scheduling, and the request, while may have elvpriv data, is not
1541  * added on the elevator at this point.  In addition, we don't have
1542  * reliable access to the elevator outside queue lock.  Only check basic
1543  * merging parameters without querying the elevator.
1544  *
1545  * Caller must ensure !blk_queue_nomerges(q) beforehand.
1546  */
1547 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1548                 unsigned int *request_count,
1549                 struct request **same_queue_rq)
1550 {
1551     struct blk_plug *plug;
1552     struct request *rq;
1553     bool ret = false;
1554     struct list_head *plug_list;
1555 
1556     plug = current->plug;
1557     if (!plug)
1558         goto out;
1559     *request_count = 0;
1560 
1561     if (q->mq_ops)
1562         plug_list = &plug->mq_list;
1563     else
1564         plug_list = &plug->list;
1565 
1566     list_for_each_entry_reverse(rq, plug_list, queuelist) {
1567         int el_ret;
1568 
1569         if (rq->q == q) {
1570             (*request_count)++;
1571             /*
1572              * Only blk-mq multiple hardware queues case checks the
1573              * rq in the same queue, there should be only one such
1574              * rq in a queue
1575              **/
1576             if (same_queue_rq)
1577                 *same_queue_rq = rq;
1578         }
1579 
1580         if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1581             continue;
1582 
1583         el_ret = blk_try_merge(rq, bio);
1584         if (el_ret == ELEVATOR_BACK_MERGE) {
1585             ret = bio_attempt_back_merge(q, rq, bio);
1586             if (ret)
1587                 break;
1588         } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1589             ret = bio_attempt_front_merge(q, rq, bio);
1590             if (ret)
1591                 break;
1592         }
1593     }
1594 out:
1595     return ret;
1596 }
1597 
1598 unsigned int blk_plug_queued_count(struct request_queue *q)
1599 {
1600     struct blk_plug *plug;
1601     struct request *rq;
1602     struct list_head *plug_list;
1603     unsigned int ret = 0;
1604 
1605     plug = current->plug;
1606     if (!plug)
1607         goto out;
1608 
1609     if (q->mq_ops)
1610         plug_list = &plug->mq_list;
1611     else
1612         plug_list = &plug->list;
1613 
1614     list_for_each_entry(rq, plug_list, queuelist) {
1615         if (rq->q == q)
1616             ret++;
1617     }
1618 out:
1619     return ret;
1620 }
1621 
1622 void init_request_from_bio(struct request *req, struct bio *bio)
1623 {
1624     req->cmd_type = REQ_TYPE_FS;
1625     if (bio->bi_opf & REQ_RAHEAD)
1626         req->cmd_flags |= REQ_FAILFAST_MASK;
1627 
1628     req->errors = 0;
1629     req->__sector = bio->bi_iter.bi_sector;
1630     if (ioprio_valid(bio_prio(bio)))
1631         req->ioprio = bio_prio(bio);
1632     blk_rq_bio_prep(req->q, req, bio);
1633 }
1634 
1635 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1636 {
1637     struct blk_plug *plug;
1638     int el_ret, where = ELEVATOR_INSERT_SORT;
1639     struct request *req;
1640     unsigned int request_count = 0;
1641     unsigned int wb_acct;
1642 
1643     /*
1644      * low level driver can indicate that it wants pages above a
1645      * certain limit bounced to low memory (ie for highmem, or even
1646      * ISA dma in theory)
1647      */
1648     blk_queue_bounce(q, &bio);
1649 
1650     blk_queue_split(q, &bio, q->bio_split);
1651 
1652     if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1653         bio->bi_error = -EIO;
1654         bio_endio(bio);
1655         return BLK_QC_T_NONE;
1656     }
1657 
1658     if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) {
1659         spin_lock_irq(q->queue_lock);
1660         where = ELEVATOR_INSERT_FLUSH;
1661         goto get_rq;
1662     }
1663 
1664     /*
1665      * Check if we can merge with the plugged list before grabbing
1666      * any locks.
1667      */
1668     if (!blk_queue_nomerges(q)) {
1669         if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1670             return BLK_QC_T_NONE;
1671     } else
1672         request_count = blk_plug_queued_count(q);
1673 
1674     spin_lock_irq(q->queue_lock);
1675 
1676     el_ret = elv_merge(q, &req, bio);
1677     if (el_ret == ELEVATOR_BACK_MERGE) {
1678         if (bio_attempt_back_merge(q, req, bio)) {
1679             elv_bio_merged(q, req, bio);
1680             if (!attempt_back_merge(q, req))
1681                 elv_merged_request(q, req, el_ret);
1682             goto out_unlock;
1683         }
1684     } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1685         if (bio_attempt_front_merge(q, req, bio)) {
1686             elv_bio_merged(q, req, bio);
1687             if (!attempt_front_merge(q, req))
1688                 elv_merged_request(q, req, el_ret);
1689             goto out_unlock;
1690         }
1691     }
1692 
1693 get_rq:
1694     wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
1695 
1696     /*
1697      * Grab a free request. This is might sleep but can not fail.
1698      * Returns with the queue unlocked.
1699      */
1700     req = get_request(q, bio->bi_opf, bio, GFP_NOIO);
1701     if (IS_ERR(req)) {
1702         __wbt_done(q->rq_wb, wb_acct);
1703         bio->bi_error = PTR_ERR(req);
1704         bio_endio(bio);
1705         goto out_unlock;
1706     }
1707 
1708     wbt_track(&req->issue_stat, wb_acct);
1709 
1710     /*
1711      * After dropping the lock and possibly sleeping here, our request
1712      * may now be mergeable after it had proven unmergeable (above).
1713      * We don't worry about that case for efficiency. It won't happen
1714      * often, and the elevators are able to handle it.
1715      */
1716     init_request_from_bio(req, bio);
1717 
1718     if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1719         req->cpu = raw_smp_processor_id();
1720 
1721     plug = current->plug;
1722     if (plug) {
1723         /*
1724          * If this is the first request added after a plug, fire
1725          * of a plug trace.
1726          *
1727          * @request_count may become stale because of schedule
1728          * out, so check plug list again.
1729          */
1730         if (!request_count || list_empty(&plug->list))
1731             trace_block_plug(q);
1732         else {
1733             struct request *last = list_entry_rq(plug->list.prev);
1734             if (request_count >= BLK_MAX_REQUEST_COUNT ||
1735                 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
1736                 blk_flush_plug_list(plug, false);
1737                 trace_block_plug(q);
1738             }
1739         }
1740         list_add_tail(&req->queuelist, &plug->list);
1741         blk_account_io_start(req, true);
1742     } else {
1743         spin_lock_irq(q->queue_lock);
1744         add_acct_request(q, req, where);
1745         __blk_run_queue(q);
1746 out_unlock:
1747         spin_unlock_irq(q->queue_lock);
1748     }
1749 
1750     return BLK_QC_T_NONE;
1751 }
1752 
1753 /*
1754  * If bio->bi_dev is a partition, remap the location
1755  */
1756 static inline void blk_partition_remap(struct bio *bio)
1757 {
1758     struct block_device *bdev = bio->bi_bdev;
1759 
1760     /*
1761      * Zone reset does not include bi_size so bio_sectors() is always 0.
1762      * Include a test for the reset op code and perform the remap if needed.
1763      */
1764     if (bdev != bdev->bd_contains &&
1765         (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET)) {
1766         struct hd_struct *p = bdev->bd_part;
1767 
1768         bio->bi_iter.bi_sector += p->start_sect;
1769         bio->bi_bdev = bdev->bd_contains;
1770 
1771         trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1772                       bdev->bd_dev,
1773                       bio->bi_iter.bi_sector - p->start_sect);
1774     }
1775 }
1776 
1777 static void handle_bad_sector(struct bio *bio)
1778 {
1779     char b[BDEVNAME_SIZE];
1780 
1781     printk(KERN_INFO "attempt to access beyond end of device\n");
1782     printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1783             bdevname(bio->bi_bdev, b),
1784             bio->bi_opf,
1785             (unsigned long long)bio_end_sector(bio),
1786             (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1787 }
1788 
1789 #ifdef CONFIG_FAIL_MAKE_REQUEST
1790 
1791 static DECLARE_FAULT_ATTR(fail_make_request);
1792 
1793 static int __init setup_fail_make_request(char *str)
1794 {
1795     return setup_fault_attr(&fail_make_request, str);
1796 }
1797 __setup("fail_make_request=", setup_fail_make_request);
1798 
1799 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1800 {
1801     return part->make_it_fail && should_fail(&fail_make_request, bytes);
1802 }
1803 
1804 static int __init fail_make_request_debugfs(void)
1805 {
1806     struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1807                         NULL, &fail_make_request);
1808 
1809     return PTR_ERR_OR_ZERO(dir);
1810 }
1811 
1812 late_initcall(fail_make_request_debugfs);
1813 
1814 #else /* CONFIG_FAIL_MAKE_REQUEST */
1815 
1816 static inline bool should_fail_request(struct hd_struct *part,
1817                     unsigned int bytes)
1818 {
1819     return false;
1820 }
1821 
1822 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1823 
1824 /*
1825  * Check whether this bio extends beyond the end of the device.
1826  */
1827 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1828 {
1829     sector_t maxsector;
1830 
1831     if (!nr_sectors)
1832         return 0;
1833 
1834     /* Test device or partition size, when known. */
1835     maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1836     if (maxsector) {
1837         sector_t sector = bio->bi_iter.bi_sector;
1838 
1839         if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1840             /*
1841              * This may well happen - the kernel calls bread()
1842              * without checking the size of the device, e.g., when
1843              * mounting a device.
1844              */
1845             handle_bad_sector(bio);
1846             return 1;
1847         }
1848     }
1849 
1850     return 0;
1851 }
1852 
1853 static noinline_for_stack bool
1854 generic_make_request_checks(struct bio *bio)
1855 {
1856     struct request_queue *q;
1857     int nr_sectors = bio_sectors(bio);
1858     int err = -EIO;
1859     char b[BDEVNAME_SIZE];
1860     struct hd_struct *part;
1861 
1862     might_sleep();
1863 
1864     if (bio_check_eod(bio, nr_sectors))
1865         goto end_io;
1866 
1867     q = bdev_get_queue(bio->bi_bdev);
1868     if (unlikely(!q)) {
1869         printk(KERN_ERR
1870                "generic_make_request: Trying to access "
1871             "nonexistent block-device %s (%Lu)\n",
1872             bdevname(bio->bi_bdev, b),
1873             (long long) bio->bi_iter.bi_sector);
1874         goto end_io;
1875     }
1876 
1877     part = bio->bi_bdev->bd_part;
1878     if (should_fail_request(part, bio->bi_iter.bi_size) ||
1879         should_fail_request(&part_to_disk(part)->part0,
1880                 bio->bi_iter.bi_size))
1881         goto end_io;
1882 
1883     /*
1884      * If this device has partitions, remap block n
1885      * of partition p to block n+start(p) of the disk.
1886      */
1887     blk_partition_remap(bio);
1888 
1889     if (bio_check_eod(bio, nr_sectors))
1890         goto end_io;
1891 
1892     /*
1893      * Filter flush bio's early so that make_request based
1894      * drivers without flush support don't have to worry
1895      * about them.
1896      */
1897     if ((bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) &&
1898         !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1899         bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
1900         if (!nr_sectors) {
1901             err = 0;
1902             goto end_io;
1903         }
1904     }
1905 
1906     switch (bio_op(bio)) {
1907     case REQ_OP_DISCARD:
1908         if (!blk_queue_discard(q))
1909             goto not_supported;
1910         break;
1911     case REQ_OP_SECURE_ERASE:
1912         if (!blk_queue_secure_erase(q))
1913             goto not_supported;
1914         break;
1915     case REQ_OP_WRITE_SAME:
1916         if (!bdev_write_same(bio->bi_bdev))
1917             goto not_supported;
1918         break;
1919     case REQ_OP_ZONE_REPORT:
1920     case REQ_OP_ZONE_RESET:
1921         if (!bdev_is_zoned(bio->bi_bdev))
1922             goto not_supported;
1923         break;
1924     case REQ_OP_WRITE_ZEROES:
1925         if (!bdev_write_zeroes_sectors(bio->bi_bdev))
1926             goto not_supported;
1927         break;
1928     default:
1929         break;
1930     }
1931 
1932     /*
1933      * Various block parts want %current->io_context and lazy ioc
1934      * allocation ends up trading a lot of pain for a small amount of
1935      * memory.  Just allocate it upfront.  This may fail and block
1936      * layer knows how to live with it.
1937      */
1938     create_io_context(GFP_ATOMIC, q->node);
1939 
1940     if (!blkcg_bio_issue_check(q, bio))
1941         return false;
1942 
1943     trace_block_bio_queue(q, bio);
1944     return true;
1945 
1946 not_supported:
1947     err = -EOPNOTSUPP;
1948 end_io:
1949     bio->bi_error = err;
1950     bio_endio(bio);
1951     return false;
1952 }
1953 
1954 /**
1955  * generic_make_request - hand a buffer to its device driver for I/O
1956  * @bio:  The bio describing the location in memory and on the device.
1957  *
1958  * generic_make_request() is used to make I/O requests of block
1959  * devices. It is passed a &struct bio, which describes the I/O that needs
1960  * to be done.
1961  *
1962  * generic_make_request() does not return any status.  The
1963  * success/failure status of the request, along with notification of
1964  * completion, is delivered asynchronously through the bio->bi_end_io
1965  * function described (one day) else where.
1966  *
1967  * The caller of generic_make_request must make sure that bi_io_vec
1968  * are set to describe the memory buffer, and that bi_dev and bi_sector are
1969  * set to describe the device address, and the
1970  * bi_end_io and optionally bi_private are set to describe how
1971  * completion notification should be signaled.
1972  *
1973  * generic_make_request and the drivers it calls may use bi_next if this
1974  * bio happens to be merged with someone else, and may resubmit the bio to
1975  * a lower device by calling into generic_make_request recursively, which
1976  * means the bio should NOT be touched after the call to ->make_request_fn.
1977  */
1978 blk_qc_t generic_make_request(struct bio *bio)
1979 {
1980     struct bio_list bio_list_on_stack;
1981     blk_qc_t ret = BLK_QC_T_NONE;
1982 
1983     if (!generic_make_request_checks(bio))
1984         goto out;
1985 
1986     /*
1987      * We only want one ->make_request_fn to be active at a time, else
1988      * stack usage with stacked devices could be a problem.  So use
1989      * current->bio_list to keep a list of requests submited by a
1990      * make_request_fn function.  current->bio_list is also used as a
1991      * flag to say if generic_make_request is currently active in this
1992      * task or not.  If it is NULL, then no make_request is active.  If
1993      * it is non-NULL, then a make_request is active, and new requests
1994      * should be added at the tail
1995      */
1996     if (current->bio_list) {
1997         bio_list_add(current->bio_list, bio);
1998         goto out;
1999     }
2000 
2001     /* following loop may be a bit non-obvious, and so deserves some
2002      * explanation.
2003      * Before entering the loop, bio->bi_next is NULL (as all callers
2004      * ensure that) so we have a list with a single bio.
2005      * We pretend that we have just taken it off a longer list, so
2006      * we assign bio_list to a pointer to the bio_list_on_stack,
2007      * thus initialising the bio_list of new bios to be
2008      * added.  ->make_request() may indeed add some more bios
2009      * through a recursive call to generic_make_request.  If it
2010      * did, we find a non-NULL value in bio_list and re-enter the loop
2011      * from the top.  In this case we really did just take the bio
2012      * of the top of the list (no pretending) and so remove it from
2013      * bio_list, and call into ->make_request() again.
2014      */
2015     BUG_ON(bio->bi_next);
2016     bio_list_init(&bio_list_on_stack);
2017     current->bio_list = &bio_list_on_stack;
2018     do {
2019         struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2020 
2021         if (likely(blk_queue_enter(q, false) == 0)) {
2022             ret = q->make_request_fn(q, bio);
2023 
2024             blk_queue_exit(q);
2025 
2026             bio = bio_list_pop(current->bio_list);
2027         } else {
2028             struct bio *bio_next = bio_list_pop(current->bio_list);
2029 
2030             bio_io_error(bio);
2031             bio = bio_next;
2032         }
2033     } while (bio);
2034     current->bio_list = NULL; /* deactivate */
2035 
2036 out:
2037     return ret;
2038 }
2039 EXPORT_SYMBOL(generic_make_request);
2040 
2041 /**
2042  * submit_bio - submit a bio to the block device layer for I/O
2043  * @bio: The &struct bio which describes the I/O
2044  *
2045  * submit_bio() is very similar in purpose to generic_make_request(), and
2046  * uses that function to do most of the work. Both are fairly rough
2047  * interfaces; @bio must be presetup and ready for I/O.
2048  *
2049  */
2050 blk_qc_t submit_bio(struct bio *bio)
2051 {
2052     /*
2053      * If it's a regular read/write or a barrier with data attached,
2054      * go through the normal accounting stuff before submission.
2055      */
2056     if (bio_has_data(bio)) {
2057         unsigned int count;
2058 
2059         if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2060             count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2061         else
2062             count = bio_sectors(bio);
2063 
2064         if (op_is_write(bio_op(bio))) {
2065             count_vm_events(PGPGOUT, count);
2066         } else {
2067             task_io_account_read(bio->bi_iter.bi_size);
2068             count_vm_events(PGPGIN, count);
2069         }
2070 
2071         if (unlikely(block_dump)) {
2072             char b[BDEVNAME_SIZE];
2073             printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2074             current->comm, task_pid_nr(current),
2075                 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2076                 (unsigned long long)bio->bi_iter.bi_sector,
2077                 bdevname(bio->bi_bdev, b),
2078                 count);
2079         }
2080     }
2081 
2082     return generic_make_request(bio);
2083 }
2084 EXPORT_SYMBOL(submit_bio);
2085 
2086 /**
2087  * blk_cloned_rq_check_limits - Helper function to check a cloned request
2088  *                              for new the queue limits
2089  * @q:  the queue
2090  * @rq: the request being checked
2091  *
2092  * Description:
2093  *    @rq may have been made based on weaker limitations of upper-level queues
2094  *    in request stacking drivers, and it may violate the limitation of @q.
2095  *    Since the block layer and the underlying device driver trust @rq
2096  *    after it is inserted to @q, it should be checked against @q before
2097  *    the insertion using this generic function.
2098  *
2099  *    Request stacking drivers like request-based dm may change the queue
2100  *    limits when retrying requests on other queues. Those requests need
2101  *    to be checked against the new queue limits again during dispatch.
2102  */
2103 static int blk_cloned_rq_check_limits(struct request_queue *q,
2104                       struct request *rq)
2105 {
2106     if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2107         printk(KERN_ERR "%s: over max size limit.\n", __func__);
2108         return -EIO;
2109     }
2110 
2111     /*
2112      * queue's settings related to segment counting like q->bounce_pfn
2113      * may differ from that of other stacking queues.
2114      * Recalculate it to check the request correctly on this queue's
2115      * limitation.
2116      */
2117     blk_recalc_rq_segments(rq);
2118     if (rq->nr_phys_segments > queue_max_segments(q)) {
2119         printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2120         return -EIO;
2121     }
2122 
2123     return 0;
2124 }
2125 
2126 /**
2127  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2128  * @q:  the queue to submit the request
2129  * @rq: the request being queued
2130  */
2131 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2132 {
2133     unsigned long flags;
2134     int where = ELEVATOR_INSERT_BACK;
2135 
2136     if (blk_cloned_rq_check_limits(q, rq))
2137         return -EIO;
2138 
2139     if (rq->rq_disk &&
2140         should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2141         return -EIO;
2142 
2143     if (q->mq_ops) {
2144         if (blk_queue_io_stat(q))
2145             blk_account_io_start(rq, true);
2146         blk_mq_insert_request(rq, false, true, false);
2147         return 0;
2148     }
2149 
2150     spin_lock_irqsave(q->queue_lock, flags);
2151     if (unlikely(blk_queue_dying(q))) {
2152         spin_unlock_irqrestore(q->queue_lock, flags);
2153         return -ENODEV;
2154     }
2155 
2156     /*
2157      * Submitting request must be dequeued before calling this function
2158      * because it will be linked to another request_queue
2159      */
2160     BUG_ON(blk_queued_rq(rq));
2161 
2162     if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
2163         where = ELEVATOR_INSERT_FLUSH;
2164 
2165     add_acct_request(q, rq, where);
2166     if (where == ELEVATOR_INSERT_FLUSH)
2167         __blk_run_queue(q);
2168     spin_unlock_irqrestore(q->queue_lock, flags);
2169 
2170     return 0;
2171 }
2172 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2173 
2174 /**
2175  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2176  * @rq: request to examine
2177  *
2178  * Description:
2179  *     A request could be merge of IOs which require different failure
2180  *     handling.  This function determines the number of bytes which
2181  *     can be failed from the beginning of the request without
2182  *     crossing into area which need to be retried further.
2183  *
2184  * Return:
2185  *     The number of bytes to fail.
2186  *
2187  * Context:
2188  *     queue_lock must be held.
2189  */
2190 unsigned int blk_rq_err_bytes(const struct request *rq)
2191 {
2192     unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2193     unsigned int bytes = 0;
2194     struct bio *bio;
2195 
2196     if (!(rq->rq_flags & RQF_MIXED_MERGE))
2197         return blk_rq_bytes(rq);
2198 
2199     /*
2200      * Currently the only 'mixing' which can happen is between
2201      * different fastfail types.  We can safely fail portions
2202      * which have all the failfast bits that the first one has -
2203      * the ones which are at least as eager to fail as the first
2204      * one.
2205      */
2206     for (bio = rq->bio; bio; bio = bio->bi_next) {
2207         if ((bio->bi_opf & ff) != ff)
2208             break;
2209         bytes += bio->bi_iter.bi_size;
2210     }
2211 
2212     /* this could lead to infinite loop */
2213     BUG_ON(blk_rq_bytes(rq) && !bytes);
2214     return bytes;
2215 }
2216 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2217 
2218 void blk_account_io_completion(struct request *req, unsigned int bytes)
2219 {
2220     if (blk_do_io_stat(req)) {
2221         const int rw = rq_data_dir(req);
2222         struct hd_struct *part;
2223         int cpu;
2224 
2225         cpu = part_stat_lock();
2226         part = req->part;
2227         part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2228         part_stat_unlock();
2229     }
2230 }
2231 
2232 void blk_account_io_done(struct request *req)
2233 {
2234     /*
2235      * Account IO completion.  flush_rq isn't accounted as a
2236      * normal IO on queueing nor completion.  Accounting the
2237      * containing request is enough.
2238      */
2239     if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2240         unsigned long duration = jiffies - req->start_time;
2241         const int rw = rq_data_dir(req);
2242         struct hd_struct *part;
2243         int cpu;
2244 
2245         cpu = part_stat_lock();
2246         part = req->part;
2247 
2248         part_stat_inc(cpu, part, ios[rw]);
2249         part_stat_add(cpu, part, ticks[rw], duration);
2250         part_round_stats(cpu, part);
2251         part_dec_in_flight(part, rw);
2252 
2253         hd_struct_put(part);
2254         part_stat_unlock();
2255     }
2256 }
2257 
2258 #ifdef CONFIG_PM
2259 /*
2260  * Don't process normal requests when queue is suspended
2261  * or in the process of suspending/resuming
2262  */
2263 static struct request *blk_pm_peek_request(struct request_queue *q,
2264                        struct request *rq)
2265 {
2266     if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2267         (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM))))
2268         return NULL;
2269     else
2270         return rq;
2271 }
2272 #else
2273 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2274                           struct request *rq)
2275 {
2276     return rq;
2277 }
2278 #endif
2279 
2280 void blk_account_io_start(struct request *rq, bool new_io)
2281 {
2282     struct hd_struct *part;
2283     int rw = rq_data_dir(rq);
2284     int cpu;
2285 
2286     if (!blk_do_io_stat(rq))
2287         return;
2288 
2289     cpu = part_stat_lock();
2290 
2291     if (!new_io) {
2292         part = rq->part;
2293         part_stat_inc(cpu, part, merges[rw]);
2294     } else {
2295         part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2296         if (!hd_struct_try_get(part)) {
2297             /*
2298              * The partition is already being removed,
2299              * the request will be accounted on the disk only
2300              *
2301              * We take a reference on disk->part0 although that
2302              * partition will never be deleted, so we can treat
2303              * it as any other partition.
2304              */
2305             part = &rq->rq_disk->part0;
2306             hd_struct_get(part);
2307         }
2308         part_round_stats(cpu, part);
2309         part_inc_in_flight(part, rw);
2310         rq->part = part;
2311     }
2312 
2313     part_stat_unlock();
2314 }
2315 
2316 /**
2317  * blk_peek_request - peek at the top of a request queue
2318  * @q: request queue to peek at
2319  *
2320  * Description:
2321  *     Return the request at the top of @q.  The returned request
2322  *     should be started using blk_start_request() before LLD starts
2323  *     processing it.
2324  *
2325  * Return:
2326  *     Pointer to the request at the top of @q if available.  Null
2327  *     otherwise.
2328  *
2329  * Context:
2330  *     queue_lock must be held.
2331  */
2332 struct request *blk_peek_request(struct request_queue *q)
2333 {
2334     struct request *rq;
2335     int ret;
2336 
2337     while ((rq = __elv_next_request(q)) != NULL) {
2338 
2339         rq = blk_pm_peek_request(q, rq);
2340         if (!rq)
2341             break;
2342 
2343         if (!(rq->rq_flags & RQF_STARTED)) {
2344             /*
2345              * This is the first time the device driver
2346              * sees this request (possibly after
2347              * requeueing).  Notify IO scheduler.
2348              */
2349             if (rq->rq_flags & RQF_SORTED)
2350                 elv_activate_rq(q, rq);
2351 
2352             /*
2353              * just mark as started even if we don't start
2354              * it, a request that has been delayed should
2355              * not be passed by new incoming requests
2356              */
2357             rq->rq_flags |= RQF_STARTED;
2358             trace_block_rq_issue(q, rq);
2359         }
2360 
2361         if (!q->boundary_rq || q->boundary_rq == rq) {
2362             q->end_sector = rq_end_sector(rq);
2363             q->boundary_rq = NULL;
2364         }
2365 
2366         if (rq->rq_flags & RQF_DONTPREP)
2367             break;
2368 
2369         if (q->dma_drain_size && blk_rq_bytes(rq)) {
2370             /*
2371              * make sure space for the drain appears we
2372              * know we can do this because max_hw_segments
2373              * has been adjusted to be one fewer than the
2374              * device can handle
2375              */
2376             rq->nr_phys_segments++;
2377         }
2378 
2379         if (!q->prep_rq_fn)
2380             break;
2381 
2382         ret = q->prep_rq_fn(q, rq);
2383         if (ret == BLKPREP_OK) {
2384             break;
2385         } else if (ret == BLKPREP_DEFER) {
2386             /*
2387              * the request may have been (partially) prepped.
2388              * we need to keep this request in the front to
2389              * avoid resource deadlock.  RQF_STARTED will
2390              * prevent other fs requests from passing this one.
2391              */
2392             if (q->dma_drain_size && blk_rq_bytes(rq) &&
2393                 !(rq->rq_flags & RQF_DONTPREP)) {
2394                 /*
2395                  * remove the space for the drain we added
2396                  * so that we don't add it again
2397                  */
2398                 --rq->nr_phys_segments;
2399             }
2400 
2401             rq = NULL;
2402             break;
2403         } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2404             int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2405 
2406             rq->rq_flags |= RQF_QUIET;
2407             /*
2408              * Mark this request as started so we don't trigger
2409              * any debug logic in the end I/O path.
2410              */
2411             blk_start_request(rq);
2412             __blk_end_request_all(rq, err);
2413         } else {
2414             printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2415             break;
2416         }
2417     }
2418 
2419     return rq;
2420 }
2421 EXPORT_SYMBOL(blk_peek_request);
2422 
2423 void blk_dequeue_request(struct request *rq)
2424 {
2425     struct request_queue *q = rq->q;
2426 
2427     BUG_ON(list_empty(&rq->queuelist));
2428     BUG_ON(ELV_ON_HASH(rq));
2429 
2430     list_del_init(&rq->queuelist);
2431 
2432     /*
2433      * the time frame between a request being removed from the lists
2434      * and to it is freed is accounted as io that is in progress at
2435      * the driver side.
2436      */
2437     if (blk_account_rq(rq)) {
2438         q->in_flight[rq_is_sync(rq)]++;
2439         set_io_start_time_ns(rq);
2440     }
2441 }
2442 
2443 /**
2444  * blk_start_request - start request processing on the driver
2445  * @req: request to dequeue
2446  *
2447  * Description:
2448  *     Dequeue @req and start timeout timer on it.  This hands off the
2449  *     request to the driver.
2450  *
2451  *     Block internal functions which don't want to start timer should
2452  *     call blk_dequeue_request().
2453  *
2454  * Context:
2455  *     queue_lock must be held.
2456  */
2457 void blk_start_request(struct request *req)
2458 {
2459     blk_dequeue_request(req);
2460 
2461     if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2462         blk_stat_set_issue_time(&req->issue_stat);
2463         req->rq_flags |= RQF_STATS;
2464         wbt_issue(req->q->rq_wb, &req->issue_stat);
2465     }
2466 
2467     /*
2468      * We are now handing the request to the hardware, initialize
2469      * resid_len to full count and add the timeout handler.
2470      */
2471     req->resid_len = blk_rq_bytes(req);
2472     if (unlikely(blk_bidi_rq(req)))
2473         req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2474 
2475     BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2476     blk_add_timer(req);
2477 }
2478 EXPORT_SYMBOL(blk_start_request);
2479 
2480 /**
2481  * blk_fetch_request - fetch a request from a request queue
2482  * @q: request queue to fetch a request from
2483  *
2484  * Description:
2485  *     Return the request at the top of @q.  The request is started on
2486  *     return and LLD can start processing it immediately.
2487  *
2488  * Return:
2489  *     Pointer to the request at the top of @q if available.  Null
2490  *     otherwise.
2491  *
2492  * Context:
2493  *     queue_lock must be held.
2494  */
2495 struct request *blk_fetch_request(struct request_queue *q)
2496 {
2497     struct request *rq;
2498 
2499     rq = blk_peek_request(q);
2500     if (rq)
2501         blk_start_request(rq);
2502     return rq;
2503 }
2504 EXPORT_SYMBOL(blk_fetch_request);
2505 
2506 /**
2507  * blk_update_request - Special helper function for request stacking drivers
2508  * @req:      the request being processed
2509  * @error:    %0 for success, < %0 for error
2510  * @nr_bytes: number of bytes to complete @req
2511  *
2512  * Description:
2513  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
2514  *     the request structure even if @req doesn't have leftover.
2515  *     If @req has leftover, sets it up for the next range of segments.
2516  *
2517  *     This special helper function is only for request stacking drivers
2518  *     (e.g. request-based dm) so that they can handle partial completion.
2519  *     Actual device drivers should use blk_end_request instead.
2520  *
2521  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2522  *     %false return from this function.
2523  *
2524  * Return:
2525  *     %false - this request doesn't have any more data
2526  *     %true  - this request has more data
2527  **/
2528 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2529 {
2530     int total_bytes;
2531 
2532     trace_block_rq_complete(req->q, req, nr_bytes);
2533 
2534     if (!req->bio)
2535         return false;
2536 
2537     /*
2538      * For fs requests, rq is just carrier of independent bio's
2539      * and each partial completion should be handled separately.
2540      * Reset per-request error on each partial completion.
2541      *
2542      * TODO: tj: This is too subtle.  It would be better to let
2543      * low level drivers do what they see fit.
2544      */
2545     if (req->cmd_type == REQ_TYPE_FS)
2546         req->errors = 0;
2547 
2548     if (error && req->cmd_type == REQ_TYPE_FS &&
2549         !(req->rq_flags & RQF_QUIET)) {
2550         char *error_type;
2551 
2552         switch (error) {
2553         case -ENOLINK:
2554             error_type = "recoverable transport";
2555             break;
2556         case -EREMOTEIO:
2557             error_type = "critical target";
2558             break;
2559         case -EBADE:
2560             error_type = "critical nexus";
2561             break;
2562         case -ETIMEDOUT:
2563             error_type = "timeout";
2564             break;
2565         case -ENOSPC:
2566             error_type = "critical space allocation";
2567             break;
2568         case -ENODATA:
2569             error_type = "critical medium";
2570             break;
2571         case -EIO:
2572         default:
2573             error_type = "I/O";
2574             break;
2575         }
2576         printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2577                    __func__, error_type, req->rq_disk ?
2578                    req->rq_disk->disk_name : "?",
2579                    (unsigned long long)blk_rq_pos(req));
2580 
2581     }
2582 
2583     blk_account_io_completion(req, nr_bytes);
2584 
2585     total_bytes = 0;
2586     while (req->bio) {
2587         struct bio *bio = req->bio;
2588         unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2589 
2590         if (bio_bytes == bio->bi_iter.bi_size)
2591             req->bio = bio->bi_next;
2592 
2593         req_bio_endio(req, bio, bio_bytes, error);
2594 
2595         total_bytes += bio_bytes;
2596         nr_bytes -= bio_bytes;
2597 
2598         if (!nr_bytes)
2599             break;
2600     }
2601 
2602     /*
2603      * completely done
2604      */
2605     if (!req->bio) {
2606         /*
2607          * Reset counters so that the request stacking driver
2608          * can find how many bytes remain in the request
2609          * later.
2610          */
2611         req->__data_len = 0;
2612         return false;
2613     }
2614 
2615     WARN_ON_ONCE(req->rq_flags & RQF_SPECIAL_PAYLOAD);
2616 
2617     req->__data_len -= total_bytes;
2618 
2619     /* update sector only for requests with clear definition of sector */
2620     if (req->cmd_type == REQ_TYPE_FS)
2621         req->__sector += total_bytes >> 9;
2622 
2623     /* mixed attributes always follow the first bio */
2624     if (req->rq_flags & RQF_MIXED_MERGE) {
2625         req->cmd_flags &= ~REQ_FAILFAST_MASK;
2626         req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2627     }
2628 
2629     /*
2630      * If total number of sectors is less than the first segment
2631      * size, something has gone terribly wrong.
2632      */
2633     if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2634         blk_dump_rq_flags(req, "request botched");
2635         req->__data_len = blk_rq_cur_bytes(req);
2636     }
2637 
2638     /* recalculate the number of segments */
2639     blk_recalc_rq_segments(req);
2640 
2641     return true;
2642 }
2643 EXPORT_SYMBOL_GPL(blk_update_request);
2644 
2645 static bool blk_update_bidi_request(struct request *rq, int error,
2646                     unsigned int nr_bytes,
2647                     unsigned int bidi_bytes)
2648 {
2649     if (blk_update_request(rq, error, nr_bytes))
2650         return true;
2651 
2652     /* Bidi request must be completed as a whole */
2653     if (unlikely(blk_bidi_rq(rq)) &&
2654         blk_update_request(rq->next_rq, error, bidi_bytes))
2655         return true;
2656 
2657     if (blk_queue_add_random(rq->q))
2658         add_disk_randomness(rq->rq_disk);
2659 
2660     return false;
2661 }
2662 
2663 /**
2664  * blk_unprep_request - unprepare a request
2665  * @req:    the request
2666  *
2667  * This function makes a request ready for complete resubmission (or
2668  * completion).  It happens only after all error handling is complete,
2669  * so represents the appropriate moment to deallocate any resources
2670  * that were allocated to the request in the prep_rq_fn.  The queue
2671  * lock is held when calling this.
2672  */
2673 void blk_unprep_request(struct request *req)
2674 {
2675     struct request_queue *q = req->q;
2676 
2677     req->rq_flags &= ~RQF_DONTPREP;
2678     if (q->unprep_rq_fn)
2679         q->unprep_rq_fn(q, req);
2680 }
2681 EXPORT_SYMBOL_GPL(blk_unprep_request);
2682 
2683 /*
2684  * queue lock must be held
2685  */
2686 void blk_finish_request(struct request *req, int error)
2687 {
2688     struct request_queue *q = req->q;
2689 
2690     if (req->rq_flags & RQF_STATS)
2691         blk_stat_add(&q->rq_stats[rq_data_dir(req)], req);
2692 
2693     if (req->rq_flags & RQF_QUEUED)
2694         blk_queue_end_tag(q, req);
2695 
2696     BUG_ON(blk_queued_rq(req));
2697 
2698     if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2699         laptop_io_completion(&req->q->backing_dev_info);
2700 
2701     blk_delete_timer(req);
2702 
2703     if (req->rq_flags & RQF_DONTPREP)
2704         blk_unprep_request(req);
2705 
2706     blk_account_io_done(req);
2707 
2708     if (req->end_io) {
2709         wbt_done(req->q->rq_wb, &req->issue_stat);
2710         req->end_io(req, error);
2711     } else {
2712         if (blk_bidi_rq(req))
2713             __blk_put_request(req->next_rq->q, req->next_rq);
2714 
2715         __blk_put_request(q, req);
2716     }
2717 }
2718 EXPORT_SYMBOL(blk_finish_request);
2719 
2720 /**
2721  * blk_end_bidi_request - Complete a bidi request
2722  * @rq:         the request to complete
2723  * @error:      %0 for success, < %0 for error
2724  * @nr_bytes:   number of bytes to complete @rq
2725  * @bidi_bytes: number of bytes to complete @rq->next_rq
2726  *
2727  * Description:
2728  *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2729  *     Drivers that supports bidi can safely call this member for any
2730  *     type of request, bidi or uni.  In the later case @bidi_bytes is
2731  *     just ignored.
2732  *
2733  * Return:
2734  *     %false - we are done with this request
2735  *     %true  - still buffers pending for this request
2736  **/
2737 static bool blk_end_bidi_request(struct request *rq, int error,
2738                  unsigned int nr_bytes, unsigned int bidi_bytes)
2739 {
2740     struct request_queue *q = rq->q;
2741     unsigned long flags;
2742 
2743     if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2744         return true;
2745 
2746     spin_lock_irqsave(q->queue_lock, flags);
2747     blk_finish_request(rq, error);
2748     spin_unlock_irqrestore(q->queue_lock, flags);
2749 
2750     return false;
2751 }
2752 
2753 /**
2754  * __blk_end_bidi_request - Complete a bidi request with queue lock held
2755  * @rq:         the request to complete
2756  * @error:      %0 for success, < %0 for error
2757  * @nr_bytes:   number of bytes to complete @rq
2758  * @bidi_bytes: number of bytes to complete @rq->next_rq
2759  *
2760  * Description:
2761  *     Identical to blk_end_bidi_request() except that queue lock is
2762  *     assumed to be locked on entry and remains so on return.
2763  *
2764  * Return:
2765  *     %false - we are done with this request
2766  *     %true  - still buffers pending for this request
2767  **/
2768 bool __blk_end_bidi_request(struct request *rq, int error,
2769                    unsigned int nr_bytes, unsigned int bidi_bytes)
2770 {
2771     if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2772         return true;
2773 
2774     blk_finish_request(rq, error);
2775 
2776     return false;
2777 }
2778 
2779 /**
2780  * blk_end_request - Helper function for drivers to complete the request.
2781  * @rq:       the request being processed
2782  * @error:    %0 for success, < %0 for error
2783  * @nr_bytes: number of bytes to complete
2784  *
2785  * Description:
2786  *     Ends I/O on a number of bytes attached to @rq.
2787  *     If @rq has leftover, sets it up for the next range of segments.
2788  *
2789  * Return:
2790  *     %false - we are done with this request
2791  *     %true  - still buffers pending for this request
2792  **/
2793 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2794 {
2795     return blk_end_bidi_request(rq, error, nr_bytes, 0);
2796 }
2797 EXPORT_SYMBOL(blk_end_request);
2798 
2799 /**
2800  * blk_end_request_all - Helper function for drives to finish the request.
2801  * @rq: the request to finish
2802  * @error: %0 for success, < %0 for error
2803  *
2804  * Description:
2805  *     Completely finish @rq.
2806  */
2807 void blk_end_request_all(struct request *rq, int error)
2808 {
2809     bool pending;
2810     unsigned int bidi_bytes = 0;
2811 
2812     if (unlikely(blk_bidi_rq(rq)))
2813         bidi_bytes = blk_rq_bytes(rq->next_rq);
2814 
2815     pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2816     BUG_ON(pending);
2817 }
2818 EXPORT_SYMBOL(blk_end_request_all);
2819 
2820 /**
2821  * blk_end_request_cur - Helper function to finish the current request chunk.
2822  * @rq: the request to finish the current chunk for
2823  * @error: %0 for success, < %0 for error
2824  *
2825  * Description:
2826  *     Complete the current consecutively mapped chunk from @rq.
2827  *
2828  * Return:
2829  *     %false - we are done with this request
2830  *     %true  - still buffers pending for this request
2831  */
2832 bool blk_end_request_cur(struct request *rq, int error)
2833 {
2834     return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2835 }
2836 EXPORT_SYMBOL(blk_end_request_cur);
2837 
2838 /**
2839  * blk_end_request_err - Finish a request till the next failure boundary.
2840  * @rq: the request to finish till the next failure boundary for
2841  * @error: must be negative errno
2842  *
2843  * Description:
2844  *     Complete @rq till the next failure boundary.
2845  *
2846  * Return:
2847  *     %false - we are done with this request
2848  *     %true  - still buffers pending for this request
2849  */
2850 bool blk_end_request_err(struct request *rq, int error)
2851 {
2852     WARN_ON(error >= 0);
2853     return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2854 }
2855 EXPORT_SYMBOL_GPL(blk_end_request_err);
2856 
2857 /**
2858  * __blk_end_request - Helper function for drivers to complete the request.
2859  * @rq:       the request being processed
2860  * @error:    %0 for success, < %0 for error
2861  * @nr_bytes: number of bytes to complete
2862  *
2863  * Description:
2864  *     Must be called with queue lock held unlike blk_end_request().
2865  *
2866  * Return:
2867  *     %false - we are done with this request
2868  *     %true  - still buffers pending for this request
2869  **/
2870 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2871 {
2872     return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2873 }
2874 EXPORT_SYMBOL(__blk_end_request);
2875 
2876 /**
2877  * __blk_end_request_all - Helper function for drives to finish the request.
2878  * @rq: the request to finish
2879  * @error: %0 for success, < %0 for error
2880  *
2881  * Description:
2882  *     Completely finish @rq.  Must be called with queue lock held.
2883  */
2884 void __blk_end_request_all(struct request *rq, int error)
2885 {
2886     bool pending;
2887     unsigned int bidi_bytes = 0;
2888 
2889     if (unlikely(blk_bidi_rq(rq)))
2890         bidi_bytes = blk_rq_bytes(rq->next_rq);
2891 
2892     pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2893     BUG_ON(pending);
2894 }
2895 EXPORT_SYMBOL(__blk_end_request_all);
2896 
2897 /**
2898  * __blk_end_request_cur - Helper function to finish the current request chunk.
2899  * @rq: the request to finish the current chunk for
2900  * @error: %0 for success, < %0 for error
2901  *
2902  * Description:
2903  *     Complete the current consecutively mapped chunk from @rq.  Must
2904  *     be called with queue lock held.
2905  *
2906  * Return:
2907  *     %false - we are done with this request
2908  *     %true  - still buffers pending for this request
2909  */
2910 bool __blk_end_request_cur(struct request *rq, int error)
2911 {
2912     return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2913 }
2914 EXPORT_SYMBOL(__blk_end_request_cur);
2915 
2916 /**
2917  * __blk_end_request_err - Finish a request till the next failure boundary.
2918  * @rq: the request to finish till the next failure boundary for
2919  * @error: must be negative errno
2920  *
2921  * Description:
2922  *     Complete @rq till the next failure boundary.  Must be called
2923  *     with queue lock held.
2924  *
2925  * Return:
2926  *     %false - we are done with this request
2927  *     %true  - still buffers pending for this request
2928  */
2929 bool __blk_end_request_err(struct request *rq, int error)
2930 {
2931     WARN_ON(error >= 0);
2932     return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2933 }
2934 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2935 
2936 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2937              struct bio *bio)
2938 {
2939     if (bio_has_data(bio))
2940         rq->nr_phys_segments = bio_phys_segments(q, bio);
2941 
2942     rq->__data_len = bio->bi_iter.bi_size;
2943     rq->bio = rq->biotail = bio;
2944 
2945     if (bio->bi_bdev)
2946         rq->rq_disk = bio->bi_bdev->bd_disk;
2947 }
2948 
2949 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2950 /**
2951  * rq_flush_dcache_pages - Helper function to flush all pages in a request
2952  * @rq: the request to be flushed
2953  *
2954  * Description:
2955  *     Flush all pages in @rq.
2956  */
2957 void rq_flush_dcache_pages(struct request *rq)
2958 {
2959     struct req_iterator iter;
2960     struct bio_vec bvec;
2961 
2962     rq_for_each_segment(bvec, rq, iter)
2963         flush_dcache_page(bvec.bv_page);
2964 }
2965 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2966 #endif
2967 
2968 /**
2969  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2970  * @q : the queue of the device being checked
2971  *
2972  * Description:
2973  *    Check if underlying low-level drivers of a device are busy.
2974  *    If the drivers want to export their busy state, they must set own
2975  *    exporting function using blk_queue_lld_busy() first.
2976  *
2977  *    Basically, this function is used only by request stacking drivers
2978  *    to stop dispatching requests to underlying devices when underlying
2979  *    devices are busy.  This behavior helps more I/O merging on the queue
2980  *    of the request stacking driver and prevents I/O throughput regression
2981  *    on burst I/O load.
2982  *
2983  * Return:
2984  *    0 - Not busy (The request stacking driver should dispatch request)
2985  *    1 - Busy (The request stacking driver should stop dispatching request)
2986  */
2987 int blk_lld_busy(struct request_queue *q)
2988 {
2989     if (q->lld_busy_fn)
2990         return q->lld_busy_fn(q);
2991 
2992     return 0;
2993 }
2994 EXPORT_SYMBOL_GPL(blk_lld_busy);
2995 
2996 /**
2997  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2998  * @rq: the clone request to be cleaned up
2999  *
3000  * Description:
3001  *     Free all bios in @rq for a cloned request.
3002  */
3003 void blk_rq_unprep_clone(struct request *rq)
3004 {
3005     struct bio *bio;
3006 
3007     while ((bio = rq->bio) != NULL) {
3008         rq->bio = bio->bi_next;
3009 
3010         bio_put(bio);
3011     }
3012 }
3013 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3014 
3015 /*
3016  * Copy attributes of the original request to the clone request.
3017  * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3018  */
3019 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3020 {
3021     dst->cpu = src->cpu;
3022     dst->cmd_flags = src->cmd_flags | REQ_NOMERGE;
3023     dst->cmd_type = src->cmd_type;
3024     dst->__sector = blk_rq_pos(src);
3025     dst->__data_len = blk_rq_bytes(src);
3026     dst->nr_phys_segments = src->nr_phys_segments;
3027     dst->ioprio = src->ioprio;
3028     dst->extra_len = src->extra_len;
3029 }
3030 
3031 /**
3032  * blk_rq_prep_clone - Helper function to setup clone request
3033  * @rq: the request to be setup
3034  * @rq_src: original request to be cloned
3035  * @bs: bio_set that bios for clone are allocated from
3036  * @gfp_mask: memory allocation mask for bio
3037  * @bio_ctr: setup function to be called for each clone bio.
3038  *           Returns %0 for success, non %0 for failure.
3039  * @data: private data to be passed to @bio_ctr
3040  *
3041  * Description:
3042  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3043  *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3044  *     are not copied, and copying such parts is the caller's responsibility.
3045  *     Also, pages which the original bios are pointing to are not copied
3046  *     and the cloned bios just point same pages.
3047  *     So cloned bios must be completed before original bios, which means
3048  *     the caller must complete @rq before @rq_src.
3049  */
3050 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3051               struct bio_set *bs, gfp_t gfp_mask,
3052               int (*bio_ctr)(struct bio *, struct bio *, void *),
3053               void *data)
3054 {
3055     struct bio *bio, *bio_src;
3056 
3057     if (!bs)
3058         bs = fs_bio_set;
3059 
3060     __rq_for_each_bio(bio_src, rq_src) {
3061         bio = bio_clone_fast(bio_src, gfp_mask, bs);
3062         if (!bio)
3063             goto free_and_out;
3064 
3065         if (bio_ctr && bio_ctr(bio, bio_src, data))
3066             goto free_and_out;
3067 
3068         if (rq->bio) {
3069             rq->biotail->bi_next = bio;
3070             rq->biotail = bio;
3071         } else
3072             rq->bio = rq->biotail = bio;
3073     }
3074 
3075     __blk_rq_prep_clone(rq, rq_src);
3076 
3077     return 0;
3078 
3079 free_and_out:
3080     if (bio)
3081         bio_put(bio);
3082     blk_rq_unprep_clone(rq);
3083 
3084     return -ENOMEM;
3085 }
3086 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3087 
3088 int kblockd_schedule_work(struct work_struct *work)
3089 {
3090     return queue_work(kblockd_workqueue, work);
3091 }
3092 EXPORT_SYMBOL(kblockd_schedule_work);
3093 
3094 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3095 {
3096     return queue_work_on(cpu, kblockd_workqueue, work);
3097 }
3098 EXPORT_SYMBOL(kblockd_schedule_work_on);
3099 
3100 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3101                   unsigned long delay)
3102 {
3103     return queue_delayed_work(kblockd_workqueue, dwork, delay);
3104 }
3105 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3106 
3107 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3108                      unsigned long delay)
3109 {
3110     return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3111 }
3112 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3113 
3114 /**
3115  * blk_start_plug - initialize blk_plug and track it inside the task_struct
3116  * @plug:   The &struct blk_plug that needs to be initialized
3117  *
3118  * Description:
3119  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
3120  *   pending I/O should the task end up blocking between blk_start_plug() and
3121  *   blk_finish_plug(). This is important from a performance perspective, but
3122  *   also ensures that we don't deadlock. For instance, if the task is blocking
3123  *   for a memory allocation, memory reclaim could end up wanting to free a
3124  *   page belonging to that request that is currently residing in our private
3125  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
3126  *   this kind of deadlock.
3127  */
3128 void blk_start_plug(struct blk_plug *plug)
3129 {
3130     struct task_struct *tsk = current;
3131 
3132     /*
3133      * If this is a nested plug, don't actually assign it.
3134      */
3135     if (tsk->plug)
3136         return;
3137 
3138     INIT_LIST_HEAD(&plug->list);
3139     INIT_LIST_HEAD(&plug->mq_list);
3140     INIT_LIST_HEAD(&plug->cb_list);
3141     /*
3142      * Store ordering should not be needed here, since a potential
3143      * preempt will imply a full memory barrier
3144      */
3145     tsk->plug = plug;
3146 }
3147 EXPORT_SYMBOL(blk_start_plug);
3148 
3149 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3150 {
3151     struct request *rqa = container_of(a, struct request, queuelist);
3152     struct request *rqb = container_of(b, struct request, queuelist);
3153 
3154     return !(rqa->q < rqb->q ||
3155         (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3156 }
3157 
3158 /*
3159  * If 'from_schedule' is true, then postpone the dispatch of requests
3160  * until a safe kblockd context. We due this to avoid accidental big
3161  * additional stack usage in driver dispatch, in places where the originally
3162  * plugger did not intend it.
3163  */
3164 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3165                 bool from_schedule)
3166     __releases(q->queue_lock)
3167 {
3168     trace_block_unplug(q, depth, !from_schedule);
3169 
3170     if (from_schedule)
3171         blk_run_queue_async(q);
3172     else
3173         __blk_run_queue(q);
3174     spin_unlock(q->queue_lock);
3175 }
3176 
3177 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3178 {
3179     LIST_HEAD(callbacks);
3180 
3181     while (!list_empty(&plug->cb_list)) {
3182         list_splice_init(&plug->cb_list, &callbacks);
3183 
3184         while (!list_empty(&callbacks)) {
3185             struct blk_plug_cb *cb = list_first_entry(&callbacks,
3186                               struct blk_plug_cb,
3187                               list);
3188             list_del(&cb->list);
3189             cb->callback(cb, from_schedule);
3190         }
3191     }
3192 }
3193 
3194 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3195                       int size)
3196 {
3197     struct blk_plug *plug = current->plug;
3198     struct blk_plug_cb *cb;
3199 
3200     if (!plug)
3201         return NULL;
3202 
3203     list_for_each_entry(cb, &plug->cb_list, list)
3204         if (cb->callback == unplug && cb->data == data)
3205             return cb;
3206 
3207     /* Not currently on the callback list */
3208     BUG_ON(size < sizeof(*cb));
3209     cb = kzalloc(size, GFP_ATOMIC);
3210     if (cb) {
3211         cb->data = data;
3212         cb->callback = unplug;
3213         list_add(&cb->list, &plug->cb_list);
3214     }
3215     return cb;
3216 }
3217 EXPORT_SYMBOL(blk_check_plugged);
3218 
3219 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3220 {
3221     struct request_queue *q;
3222     unsigned long flags;
3223     struct request *rq;
3224     LIST_HEAD(list);
3225     unsigned int depth;
3226 
3227     flush_plug_callbacks(plug, from_schedule);
3228 
3229     if (!list_empty(&plug->mq_list))
3230         blk_mq_flush_plug_list(plug, from_schedule);
3231 
3232     if (list_empty(&plug->list))
3233         return;
3234 
3235     list_splice_init(&plug->list, &list);
3236 
3237     list_sort(NULL, &list, plug_rq_cmp);
3238 
3239     q = NULL;
3240     depth = 0;
3241 
3242     /*
3243      * Save and disable interrupts here, to avoid doing it for every
3244      * queue lock we have to take.
3245      */
3246     local_irq_save(flags);
3247     while (!list_empty(&list)) {
3248         rq = list_entry_rq(list.next);
3249         list_del_init(&rq->queuelist);
3250         BUG_ON(!rq->q);
3251         if (rq->q != q) {
3252             /*
3253              * This drops the queue lock
3254              */
3255             if (q)
3256                 queue_unplugged(q, depth, from_schedule);
3257             q = rq->q;
3258             depth = 0;
3259             spin_lock(q->queue_lock);
3260         }
3261 
3262         /*
3263          * Short-circuit if @q is dead
3264          */
3265         if (unlikely(blk_queue_dying(q))) {
3266             __blk_end_request_all(rq, -ENODEV);
3267             continue;
3268         }
3269 
3270         /*
3271          * rq is already accounted, so use raw insert
3272          */
3273         if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
3274             __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3275         else
3276             __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3277 
3278         depth++;
3279     }
3280 
3281     /*
3282      * This drops the queue lock
3283      */
3284     if (q)
3285         queue_unplugged(q, depth, from_schedule);
3286 
3287     local_irq_restore(flags);
3288 }
3289 
3290 void blk_finish_plug(struct blk_plug *plug)
3291 {
3292     if (plug != current->plug)
3293         return;
3294     blk_flush_plug_list(plug, false);
3295 
3296     current->plug = NULL;
3297 }
3298 EXPORT_SYMBOL(blk_finish_plug);
3299 
3300 #ifdef CONFIG_PM
3301 /**
3302  * blk_pm_runtime_init - Block layer runtime PM initialization routine
3303  * @q: the queue of the device
3304  * @dev: the device the queue belongs to
3305  *
3306  * Description:
3307  *    Initialize runtime-PM-related fields for @q and start auto suspend for
3308  *    @dev. Drivers that want to take advantage of request-based runtime PM
3309  *    should call this function after @dev has been initialized, and its
3310  *    request queue @q has been allocated, and runtime PM for it can not happen
3311  *    yet(either due to disabled/forbidden or its usage_count > 0). In most
3312  *    cases, driver should call this function before any I/O has taken place.
3313  *
3314  *    This function takes care of setting up using auto suspend for the device,
3315  *    the autosuspend delay is set to -1 to make runtime suspend impossible
3316  *    until an updated value is either set by user or by driver. Drivers do
3317  *    not need to touch other autosuspend settings.
3318  *
3319  *    The block layer runtime PM is request based, so only works for drivers
3320  *    that use request as their IO unit instead of those directly use bio's.
3321  */
3322 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3323 {
3324     q->dev = dev;
3325     q->rpm_status = RPM_ACTIVE;
3326     pm_runtime_set_autosuspend_delay(q->dev, -1);
3327     pm_runtime_use_autosuspend(q->dev);
3328 }
3329 EXPORT_SYMBOL(blk_pm_runtime_init);
3330 
3331 /**
3332  * blk_pre_runtime_suspend - Pre runtime suspend check
3333  * @q: the queue of the device
3334  *
3335  * Description:
3336  *    This function will check if runtime suspend is allowed for the device
3337  *    by examining if there are any requests pending in the queue. If there
3338  *    are requests pending, the device can not be runtime suspended; otherwise,
3339  *    the queue's status will be updated to SUSPENDING and the driver can
3340  *    proceed to suspend the device.
3341  *
3342  *    For the not allowed case, we mark last busy for the device so that
3343  *    runtime PM core will try to autosuspend it some time later.
3344  *
3345  *    This function should be called near the start of the device's
3346  *    runtime_suspend callback.
3347  *
3348  * Return:
3349  *    0     - OK to runtime suspend the device
3350  *    -EBUSY    - Device should not be runtime suspended
3351  */
3352 int blk_pre_runtime_suspend(struct request_queue *q)
3353 {
3354     int ret = 0;
3355 
3356     if (!q->dev)
3357         return ret;
3358 
3359     spin_lock_irq(q->queue_lock);
3360     if (q->nr_pending) {
3361         ret = -EBUSY;
3362         pm_runtime_mark_last_busy(q->dev);
3363     } else {
3364         q->rpm_status = RPM_SUSPENDING;
3365     }
3366     spin_unlock_irq(q->queue_lock);
3367     return ret;
3368 }
3369 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3370 
3371 /**
3372  * blk_post_runtime_suspend - Post runtime suspend processing
3373  * @q: the queue of the device
3374  * @err: return value of the device's runtime_suspend function
3375  *
3376  * Description:
3377  *    Update the queue's runtime status according to the return value of the
3378  *    device's runtime suspend function and mark last busy for the device so
3379  *    that PM core will try to auto suspend the device at a later time.
3380  *
3381  *    This function should be called near the end of the device's
3382  *    runtime_suspend callback.
3383  */
3384 void blk_post_runtime_suspend(struct request_queue *q, int err)
3385 {
3386     if (!q->dev)
3387         return;
3388 
3389     spin_lock_irq(q->queue_lock);
3390     if (!err) {
3391         q->rpm_status = RPM_SUSPENDED;
3392     } else {
3393         q->rpm_status = RPM_ACTIVE;
3394         pm_runtime_mark_last_busy(q->dev);
3395     }
3396     spin_unlock_irq(q->queue_lock);
3397 }
3398 EXPORT_SYMBOL(blk_post_runtime_suspend);
3399 
3400 /**
3401  * blk_pre_runtime_resume - Pre runtime resume processing
3402  * @q: the queue of the device
3403  *
3404  * Description:
3405  *    Update the queue's runtime status to RESUMING in preparation for the
3406  *    runtime resume of the device.
3407  *
3408  *    This function should be called near the start of the device's
3409  *    runtime_resume callback.
3410  */
3411 void blk_pre_runtime_resume(struct request_queue *q)
3412 {
3413     if (!q->dev)
3414         return;
3415 
3416     spin_lock_irq(q->queue_lock);
3417     q->rpm_status = RPM_RESUMING;
3418     spin_unlock_irq(q->queue_lock);
3419 }
3420 EXPORT_SYMBOL(blk_pre_runtime_resume);
3421 
3422 /**
3423  * blk_post_runtime_resume - Post runtime resume processing
3424  * @q: the queue of the device
3425  * @err: return value of the device's runtime_resume function
3426  *
3427  * Description:
3428  *    Update the queue's runtime status according to the return value of the
3429  *    device's runtime_resume function. If it is successfully resumed, process
3430  *    the requests that are queued into the device's queue when it is resuming
3431  *    and then mark last busy and initiate autosuspend for it.
3432  *
3433  *    This function should be called near the end of the device's
3434  *    runtime_resume callback.
3435  */
3436 void blk_post_runtime_resume(struct request_queue *q, int err)
3437 {
3438     if (!q->dev)
3439         return;
3440 
3441     spin_lock_irq(q->queue_lock);
3442     if (!err) {
3443         q->rpm_status = RPM_ACTIVE;
3444         __blk_run_queue(q);
3445         pm_runtime_mark_last_busy(q->dev);
3446         pm_request_autosuspend(q->dev);
3447     } else {
3448         q->rpm_status = RPM_SUSPENDED;
3449     }
3450     spin_unlock_irq(q->queue_lock);
3451 }
3452 EXPORT_SYMBOL(blk_post_runtime_resume);
3453 
3454 /**
3455  * blk_set_runtime_active - Force runtime status of the queue to be active
3456  * @q: the queue of the device
3457  *
3458  * If the device is left runtime suspended during system suspend the resume
3459  * hook typically resumes the device and corrects runtime status
3460  * accordingly. However, that does not affect the queue runtime PM status
3461  * which is still "suspended". This prevents processing requests from the
3462  * queue.
3463  *
3464  * This function can be used in driver's resume hook to correct queue
3465  * runtime PM status and re-enable peeking requests from the queue. It
3466  * should be called before first request is added to the queue.
3467  */
3468 void blk_set_runtime_active(struct request_queue *q)
3469 {
3470     spin_lock_irq(q->queue_lock);
3471     q->rpm_status = RPM_ACTIVE;
3472     pm_runtime_mark_last_busy(q->dev);
3473     pm_request_autosuspend(q->dev);
3474     spin_unlock_irq(q->queue_lock);
3475 }
3476 EXPORT_SYMBOL(blk_set_runtime_active);
3477 #endif
3478 
3479 int __init blk_dev_init(void)
3480 {
3481     BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3482     BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3483             FIELD_SIZEOF(struct request, cmd_flags));
3484     BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3485             FIELD_SIZEOF(struct bio, bi_opf));
3486 
3487     /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3488     kblockd_workqueue = alloc_workqueue("kblockd",
3489                         WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3490     if (!kblockd_workqueue)
3491         panic("Failed to create kblockd\n");
3492 
3493     request_cachep = kmem_cache_create("blkdev_requests",
3494             sizeof(struct request), 0, SLAB_PANIC, NULL);
3495 
3496     blk_requestq_cachep = kmem_cache_create("request_queue",
3497             sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3498 
3499     return 0;
3500 }