blob: 1f61b74867e41d3f74f61aeec539e8b00157dacf [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080029#include <linux/fault-inject.h>
Jens Axboe73c10102011-03-08 13:19:51 +010030#include <linux/list_sort.h>
Tejun Heoe3c78ca2011-10-19 14:32:38 +020031#include <linux/delay.h>
Li Zefan55782132009-06-09 13:43:05 +080032
33#define CREATE_TRACE_POINTS
34#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Jens Axboe8324aa92008-01-29 14:51:59 +010036#include "blk.h"
37
Mike Snitzerd07335e2010-11-16 12:52:38 +010038EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020039EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Li Zefan55782132009-06-09 13:43:05 +080040EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010041
Tejun Heoa73f7302011-12-14 00:33:37 +010042DEFINE_IDA(blk_queue_ida);
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/*
45 * For the allocated request tables
46 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010047static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49/*
50 * For queue allocation
51 */
Jens Axboe6728cb02008-01-31 13:03:55 +010052struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 * Controlling structure to kblockd
56 */
Jens Axboeff856ba2006-01-09 16:02:34 +010057static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Jens Axboe26b82562008-01-29 13:54:41 +010059static void drive_stat_acct(struct request *rq, int new_io)
60{
Jens Axboe28f13702008-05-07 10:15:46 +020061 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010062 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090063 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010064
Jens Axboec2553b52009-04-24 08:10:11 +020065 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010066 return;
67
Tejun Heo074a7ac2008-08-25 19:56:14 +090068 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +090069
Jerome Marchand09e099d2011-01-05 16:57:38 +010070 if (!new_io) {
71 part = rq->part;
Tejun Heo074a7ac2008-08-25 19:56:14 +090072 part_stat_inc(cpu, part, merges[rw]);
Jerome Marchand09e099d2011-01-05 16:57:38 +010073 } else {
74 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Jens Axboe6c23a962011-01-07 08:43:37 +010075 if (!hd_struct_try_get(part)) {
Jerome Marchand09e099d2011-01-05 16:57:38 +010076 /*
77 * The partition is already being removed,
78 * the request will be accounted on the disk only
79 *
80 * We take a reference on disk->part0 although that
81 * partition will never be deleted, so we can treat
82 * it as any other partition.
83 */
84 part = &rq->rq_disk->part0;
Jens Axboe6c23a962011-01-07 08:43:37 +010085 hd_struct_get(part);
Jerome Marchand09e099d2011-01-05 16:57:38 +010086 }
Tejun Heo074a7ac2008-08-25 19:56:14 +090087 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020088 part_inc_in_flight(part, rw);
Jerome Marchand09e099d2011-01-05 16:57:38 +010089 rq->part = part;
Jens Axboe26b82562008-01-29 13:54:41 +010090 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020091
Tejun Heo074a7ac2008-08-25 19:56:14 +090092 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010093}
94
Jens Axboe8324aa92008-01-29 14:51:59 +010095void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096{
97 int nr;
98
99 nr = q->nr_requests - (q->nr_requests / 8) + 1;
100 if (nr > q->nr_requests)
101 nr = q->nr_requests;
102 q->nr_congestion_on = nr;
103
104 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
105 if (nr < 1)
106 nr = 1;
107 q->nr_congestion_off = nr;
108}
109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110/**
111 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * @bdev: device
113 *
114 * Locates the passed device's request queue and returns the address of its
115 * backing_dev_info
116 *
117 * Will return NULL if the request queue cannot be located.
118 */
119struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
120{
121 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200122 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124 if (q)
125 ret = &q->backing_dev_info;
126 return ret;
127}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128EXPORT_SYMBOL(blk_get_backing_dev_info);
129
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200130void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200132 memset(rq, 0, sizeof(*rq));
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700135 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200136 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100137 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900138 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200139 INIT_HLIST_NODE(&rq->hash);
140 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200141 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800142 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100143 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100144 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900145 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700146 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100147 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200149EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
NeilBrown5bb23a62007-09-27 12:46:13 +0200151static void req_bio_endio(struct request *rq, struct bio *bio,
152 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100153{
Tejun Heo143a87f2011-01-25 12:43:52 +0100154 if (error)
155 clear_bit(BIO_UPTODATE, &bio->bi_flags);
156 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
157 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100158
Tejun Heo143a87f2011-01-25 12:43:52 +0100159 if (unlikely(nbytes > bio->bi_size)) {
160 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
161 __func__, nbytes, bio->bi_size);
162 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +0200163 }
Tejun Heo143a87f2011-01-25 12:43:52 +0100164
165 if (unlikely(rq->cmd_flags & REQ_QUIET))
166 set_bit(BIO_QUIET, &bio->bi_flags);
167
168 bio->bi_size -= nbytes;
169 bio->bi_sector += (nbytes >> 9);
170
171 if (bio_integrity(bio))
172 bio_integrity_advance(bio, nbytes);
173
174 /* don't actually finish bio if it's part of flush sequence */
175 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
176 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100177}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179void blk_dump_rq_flags(struct request *rq, char *msg)
180{
181 int bit;
182
Jens Axboe6728cb02008-01-31 13:03:55 +0100183 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200184 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
185 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Tejun Heo83096eb2009-05-07 22:24:39 +0900187 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
188 (unsigned long long)blk_rq_pos(rq),
189 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900190 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900191 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200193 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100194 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200195 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 printk("%02x ", rq->cmd[bit]);
197 printk("\n");
198 }
199}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200EXPORT_SYMBOL(blk_dump_rq_flags);
201
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500202static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200203{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500204 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200205
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500206 q = container_of(work, struct request_queue, delay_work.work);
207 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200208 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500209 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
212/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500213 * blk_delay_queue - restart queueing after defined interval
214 * @q: The &struct request_queue in question
215 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 *
217 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500218 * Sometimes queueing needs to be postponed for a little while, to allow
219 * resources to come back. This function will make sure that queueing is
220 * restarted around the specified time.
221 */
222void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223{
Jens Axboe4521cc42011-04-18 11:36:39 +0200224 queue_delayed_work(kblockd_workqueue, &q->delay_work,
225 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500227EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500228
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229/**
230 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200231 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 *
233 * Description:
234 * blk_start_queue() will clear the stop flag on the queue, and call
235 * the request_fn for the queue if it was in a stopped state when
236 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200238void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200240 WARN_ON(!irqs_disabled());
241
Nick Piggin75ad23b2008-04-29 14:48:33 +0200242 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200243 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245EXPORT_SYMBOL(blk_start_queue);
246
247/**
248 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200249 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 *
251 * Description:
252 * The Linux block layer assumes that a block driver will consume all
253 * entries on the request queue when the request_fn strategy is called.
254 * Often this will not happen, because of hardware limitations (queue
255 * depth settings). If a device driver gets a 'queue full' response,
256 * or if it simply chooses not to queue more I/O at one point, it can
257 * call this function to prevent the request_fn from being called until
258 * the driver has signalled it's ready to go again. This happens by calling
259 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200261void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
Jens Axboead3d9d72011-03-25 16:58:59 +0100263 __cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200264 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
266EXPORT_SYMBOL(blk_stop_queue);
267
268/**
269 * blk_sync_queue - cancel any pending callbacks on a queue
270 * @q: the queue
271 *
272 * Description:
273 * The block layer may perform asynchronous callback activity
274 * on a queue, such as calling the unplug function after a timeout.
275 * A block device may call blk_sync_queue to ensure that any
276 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200277 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 * that its ->make_request_fn will not re-add plugging prior to calling
279 * this function.
280 *
Vivek Goyalda527772011-03-02 19:05:33 -0500281 * This function does not cancel any asynchronous activity arising
282 * out of elevator or throttling code. That would require elevaotor_exit()
283 * and blk_throtl_exit() to be called with queue lock initialized.
284 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 */
286void blk_sync_queue(struct request_queue *q)
287{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100288 del_timer_sync(&q->timeout);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500289 cancel_delayed_work_sync(&q->delay_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290}
291EXPORT_SYMBOL(blk_sync_queue);
292
293/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200294 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200296 *
297 * Description:
298 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200299 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200301void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Tejun Heoa538cd02009-04-23 11:05:17 +0900303 if (unlikely(blk_queue_stopped(q)))
304 return;
305
Jens Axboec21e6be2011-04-19 13:32:46 +0200306 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200307}
308EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200309
Nick Piggin75ad23b2008-04-29 14:48:33 +0200310/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200311 * blk_run_queue_async - run a single device queue in workqueue context
312 * @q: The queue to run
313 *
314 * Description:
315 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
316 * of us.
317 */
318void blk_run_queue_async(struct request_queue *q)
319{
Shaohua Li3ec717b2011-05-18 11:22:43 +0200320 if (likely(!blk_queue_stopped(q))) {
321 __cancel_delayed_work(&q->delay_work);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200322 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Shaohua Li3ec717b2011-05-18 11:22:43 +0200323 }
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200324}
Jens Axboec21e6be2011-04-19 13:32:46 +0200325EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200326
327/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200328 * blk_run_queue - run a single device queue
329 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200330 *
331 * Description:
332 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900333 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200334 */
335void blk_run_queue(struct request_queue *q)
336{
337 unsigned long flags;
338
339 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200340 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 spin_unlock_irqrestore(q->queue_lock, flags);
342}
343EXPORT_SYMBOL(blk_run_queue);
344
Jens Axboe165125e2007-07-24 09:28:11 +0200345void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500346{
347 kobject_put(&q->kobj);
348}
Jens Axboed86e0e82011-05-27 07:44:43 +0200349EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500350
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200351/**
352 * blk_drain_queue - drain requests from request_queue
353 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200354 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200355 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200356 * Drain requests from @q. If @drain_all is set, all requests are drained.
357 * If not, only ELVPRIV requests are drained. The caller is responsible
358 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200359 */
Tejun Heoc9a929d2011-10-19 14:42:16 +0200360void blk_drain_queue(struct request_queue *q, bool drain_all)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200361{
362 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100363 bool drain = false;
364 int i;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200365
366 spin_lock_irq(q->queue_lock);
367
368 elv_drain_elevator(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200369 if (drain_all)
370 blk_throtl_drain(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200371
Tejun Heo4eabc942011-12-15 20:03:04 +0100372 /*
373 * This function might be called on a queue which failed
374 * driver init after queue creation. Some drivers
375 * (e.g. fd) get unhappy in such cases. Kick queue iff
376 * dispatch queue has something on it.
377 */
378 if (!list_empty(&q->queue_head))
379 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200380
Tejun Heo481a7d62011-12-14 00:33:37 +0100381 drain |= q->rq.elvpriv;
382
383 /*
384 * Unfortunately, requests are queued at and tracked from
385 * multiple places and there's no single counter which can
386 * be drained. Check all the queues and counters.
387 */
388 if (drain_all) {
389 drain |= !list_empty(&q->queue_head);
390 for (i = 0; i < 2; i++) {
391 drain |= q->rq.count[i];
392 drain |= q->in_flight[i];
393 drain |= !list_empty(&q->flush_queue[i]);
394 }
395 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200396
397 spin_unlock_irq(q->queue_lock);
398
Tejun Heo481a7d62011-12-14 00:33:37 +0100399 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200400 break;
401 msleep(10);
402 }
403}
404
Tejun Heoc9a929d2011-10-19 14:42:16 +0200405/**
406 * blk_cleanup_queue - shutdown a request queue
407 * @q: request queue to shutdown
408 *
409 * Mark @q DEAD, drain all pending requests, destroy and put it. All
410 * future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500411 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100412void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500413{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200414 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700415
Tejun Heoc9a929d2011-10-19 14:42:16 +0200416 /* mark @q DEAD, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500417 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200418 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200419
420 spin_lock_irq(lock);
421 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
422 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
423 queue_flag_set(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500424
Hannes Reinecke777eb1b2011-09-28 08:07:01 -0600425 if (q->queue_lock != &q->__queue_lock)
426 q->queue_lock = &q->__queue_lock;
Vivek Goyalda527772011-03-02 19:05:33 -0500427
Tejun Heoc9a929d2011-10-19 14:42:16 +0200428 spin_unlock_irq(lock);
429 mutex_unlock(&q->sysfs_lock);
430
Tejun Heo6dd9ad72011-11-03 18:52:11 +0100431 /*
432 * Drain all requests queued before DEAD marking. The caller might
433 * be trying to tear down @q before its elevator is initialized, in
434 * which case we don't want to call into draining.
435 */
436 if (q->elevator)
437 blk_drain_queue(q, true);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200438
439 /* @q won't process any more request, flush async actions */
440 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
441 blk_sync_queue(q);
442
443 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500444 blk_put_queue(q);
445}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446EXPORT_SYMBOL(blk_cleanup_queue);
447
Jens Axboe165125e2007-07-24 09:28:11 +0200448static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449{
450 struct request_list *rl = &q->rq;
451
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400452 if (unlikely(rl->rq_pool))
453 return 0;
454
Jens Axboe1faa16d2009-04-06 14:48:01 +0200455 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
456 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200457 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200458 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
459 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
Christoph Lameter19460892005-06-23 00:08:19 -0700461 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
462 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
464 if (!rl->rq_pool)
465 return -ENOMEM;
466
467 return 0;
468}
469
Jens Axboe165125e2007-07-24 09:28:11 +0200470struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
Christoph Lameter19460892005-06-23 00:08:19 -0700472 return blk_alloc_queue_node(gfp_mask, -1);
473}
474EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Jens Axboe165125e2007-07-24 09:28:11 +0200476struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700477{
Jens Axboe165125e2007-07-24 09:28:11 +0200478 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700479 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700480
Jens Axboe8324aa92008-01-29 14:51:59 +0100481 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700482 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 if (!q)
484 return NULL;
485
Dan Carpenter00380a42012-03-23 09:58:54 +0100486 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100487 if (q->id < 0)
488 goto fail_q;
489
Jens Axboe0989a022009-06-12 14:42:56 +0200490 q->backing_dev_info.ra_pages =
491 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
492 q->backing_dev_info.state = 0;
493 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200494 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100495 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200496
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700497 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100498 if (err)
499 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700500
Tejun Heoa73f7302011-12-14 00:33:37 +0100501 if (blk_throtl_init(q))
502 goto fail_id;
Vivek Goyale43473b2010-09-15 17:06:35 -0400503
Matthew Garrett31373d02010-04-06 14:25:14 +0200504 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
505 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700506 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
507 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100508 INIT_LIST_HEAD(&q->icq_list);
Tejun Heoae1b1532011-01-25 12:43:54 +0100509 INIT_LIST_HEAD(&q->flush_queue[0]);
510 INIT_LIST_HEAD(&q->flush_queue[1]);
511 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500512 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500513
Jens Axboe8324aa92008-01-29 14:51:59 +0100514 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Al Viro483f4af2006-03-18 18:34:37 -0500516 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700517 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500518
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500519 /*
520 * By default initialize queue_lock to internal lock and driver can
521 * override it later if need be.
522 */
523 q->queue_lock = &q->__queue_lock;
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100526
527fail_id:
528 ida_simple_remove(&blk_queue_ida, q->id);
529fail_q:
530 kmem_cache_free(blk_requestq_cachep, q);
531 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532}
Christoph Lameter19460892005-06-23 00:08:19 -0700533EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535/**
536 * blk_init_queue - prepare a request queue for use with a block device
537 * @rfn: The function to be called to process requests that have been
538 * placed on the queue.
539 * @lock: Request queue spin lock
540 *
541 * Description:
542 * If a block device wishes to use the standard request handling procedures,
543 * which sorts requests and coalesces adjacent requests, then it must
544 * call blk_init_queue(). The function @rfn will be called when there
545 * are requests on the queue that need to be processed. If the device
546 * supports plugging, then @rfn may not be called immediately when requests
547 * are available on the queue, but may be called at some time later instead.
548 * Plugged queues are generally unplugged when a buffer belonging to one
549 * of the requests on the queue is needed, or due to memory pressure.
550 *
551 * @rfn is not required, or even expected, to remove all requests off the
552 * queue, but only as many as it can handle at a time. If it does leave
553 * requests on the queue, it is responsible for arranging that the requests
554 * get dealt with eventually.
555 *
556 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200557 * request queue; this lock will be taken also from interrupt context, so irq
558 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200560 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 * it didn't succeed.
562 *
563 * Note:
564 * blk_init_queue() must be paired with a blk_cleanup_queue() call
565 * when the block device is deactivated (such as at module unload).
566 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700567
Jens Axboe165125e2007-07-24 09:28:11 +0200568struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
Christoph Lameter19460892005-06-23 00:08:19 -0700570 return blk_init_queue_node(rfn, lock, -1);
571}
572EXPORT_SYMBOL(blk_init_queue);
573
Jens Axboe165125e2007-07-24 09:28:11 +0200574struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700575blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
576{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600577 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600579 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
580 if (!uninit_q)
581 return NULL;
582
Mike Snitzer51514122011-11-23 10:59:13 +0100583 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600584 if (!q)
585 blk_cleanup_queue(uninit_q);
586
587 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200588}
589EXPORT_SYMBOL(blk_init_queue_node);
590
591struct request_queue *
592blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
593 spinlock_t *lock)
594{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 if (!q)
596 return NULL;
597
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600598 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500599 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
601 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900603 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100604 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500605
606 /* Override internal queue lock with supplied lock pointer */
607 if (lock)
608 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Jens Axboef3b144a2009-03-06 08:48:33 +0100610 /*
611 * This also sets hw/phys segments, boundary and size
612 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200613 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Alan Stern44ec9542007-02-20 11:01:57 -0500615 q->sg_reserved_size = INT_MAX;
616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 /*
618 * all done
619 */
620 if (!elevator_init(q, NULL)) {
621 blk_queue_congestion_threshold(q);
622 return q;
623 }
624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 return NULL;
626}
Mike Snitzer51514122011-11-23 10:59:13 +0100627EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Tejun Heo09ac46c2011-12-14 00:33:38 +0100629bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Tejun Heo34f60552011-12-14 00:33:37 +0100631 if (likely(!blk_queue_dead(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100632 __blk_get_queue(q);
633 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
635
Tejun Heo09ac46c2011-12-14 00:33:38 +0100636 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
Jens Axboed86e0e82011-05-27 07:44:43 +0200638EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Jens Axboe165125e2007-07-24 09:28:11 +0200640static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100642 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200643 elv_put_request(q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100644 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100645 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100646 }
647
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 mempool_free(rq, q->rq.rq_pool);
649}
650
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200651static struct request *
Tejun Heof1f8cc92011-12-14 00:33:42 +0100652blk_alloc_request(struct request_queue *q, struct io_cq *icq,
653 unsigned int flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
655 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
656
657 if (!rq)
658 return NULL;
659
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200660 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200661
Jerome Marchand42dad762009-04-22 14:01:49 +0200662 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Tejun Heof1f8cc92011-12-14 00:33:42 +0100664 if (flags & REQ_ELVPRIV) {
665 rq->elv.icq = icq;
666 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
667 mempool_free(rq, q->rq.rq_pool);
668 return NULL;
669 }
670 /* @rq->elv.icq holds on to io_context until @rq is freed */
671 if (icq)
672 get_io_context(icq->ioc);
Tejun Heocb98fc82005-10-28 08:29:39 +0200673 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Tejun Heocb98fc82005-10-28 08:29:39 +0200675 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
678/*
679 * ioc_batching returns true if the ioc is a valid batching request and
680 * should be given priority access to a request.
681 */
Jens Axboe165125e2007-07-24 09:28:11 +0200682static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
684 if (!ioc)
685 return 0;
686
687 /*
688 * Make sure the process is able to allocate at least 1 request
689 * even if the batch times out, otherwise we could theoretically
690 * lose wakeups.
691 */
692 return ioc->nr_batch_requests == q->nr_batching ||
693 (ioc->nr_batch_requests > 0
694 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
695}
696
697/*
698 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
699 * will cause the process to be a "batcher" on all queues in the system. This
700 * is the behaviour we want though - once it gets a wakeup it should be given
701 * a nice run.
702 */
Jens Axboe165125e2007-07-24 09:28:11 +0200703static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704{
705 if (!ioc || ioc_batching(q, ioc))
706 return;
707
708 ioc->nr_batch_requests = q->nr_batching;
709 ioc->last_waited = jiffies;
710}
711
Jens Axboe1faa16d2009-04-06 14:48:01 +0200712static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
714 struct request_list *rl = &q->rq;
715
Jens Axboe1faa16d2009-04-06 14:48:01 +0200716 if (rl->count[sync] < queue_congestion_off_threshold(q))
717 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Jens Axboe1faa16d2009-04-06 14:48:01 +0200719 if (rl->count[sync] + 1 <= q->nr_requests) {
720 if (waitqueue_active(&rl->wait[sync]))
721 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Jens Axboe1faa16d2009-04-06 14:48:01 +0200723 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 }
725}
726
727/*
728 * A request has just been released. Account for it, update the full and
729 * congestion status, wake up any waiters. Called under q->queue_lock.
730 */
Tejun Heo75eb6c32011-10-19 14:31:22 +0200731static void freed_request(struct request_queue *q, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
733 struct request_list *rl = &q->rq;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200734 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
Jens Axboe1faa16d2009-04-06 14:48:01 +0200736 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200737 if (flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200738 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Jens Axboe1faa16d2009-04-06 14:48:01 +0200740 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Jens Axboe1faa16d2009-04-06 14:48:01 +0200742 if (unlikely(rl->starved[sync ^ 1]))
743 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744}
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100747 * Determine if elevator data should be initialized when allocating the
748 * request associated with @bio.
749 */
750static bool blk_rq_should_init_elevator(struct bio *bio)
751{
752 if (!bio)
753 return true;
754
755 /*
756 * Flush requests do not use the elevator so skip initialization.
757 * This allows a request to share the flush and elevator data.
758 */
759 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
760 return false;
761
762 return true;
763}
764
Tejun Heoda8303c2011-10-19 14:33:05 +0200765/**
766 * get_request - get a free request
767 * @q: request_queue to allocate request from
768 * @rw_flags: RW and SYNC flags
769 * @bio: bio to allocate request for (can be %NULL)
770 * @gfp_mask: allocation mask
771 *
772 * Get a free request from @q. This function may fail under memory
773 * pressure or if @q is dead.
774 *
775 * Must be callled with @q->queue_lock held and,
776 * Returns %NULL on failure, with @q->queue_lock held.
777 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
Jens Axboe165125e2007-07-24 09:28:11 +0200779static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100780 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781{
782 struct request *rq = NULL;
783 struct request_list *rl = &q->rq;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100784 struct elevator_type *et;
Tejun Heof2dbd762011-12-14 00:33:40 +0100785 struct io_context *ioc;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100786 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200787 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heof2dbd762011-12-14 00:33:40 +0100788 bool retried = false;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200789 int may_queue;
Tejun Heof2dbd762011-12-14 00:33:40 +0100790retry:
Tejun Heof1f8cc92011-12-14 00:33:42 +0100791 et = q->elevator->type;
Tejun Heof2dbd762011-12-14 00:33:40 +0100792 ioc = current->io_context;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
Tejun Heo34f60552011-12-14 00:33:37 +0100794 if (unlikely(blk_queue_dead(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200795 return NULL;
796
Jens Axboe7749a8d2006-12-13 13:02:26 +0100797 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100798 if (may_queue == ELV_MQUEUE_NO)
799 goto rq_starved;
800
Jens Axboe1faa16d2009-04-06 14:48:01 +0200801 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
802 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100803 /*
804 * We want ioc to record batching state. If it's
805 * not already there, creating a new one requires
806 * dropping queue_lock, which in turn requires
807 * retesting conditions to avoid queue hang.
808 */
809 if (!ioc && !retried) {
810 spin_unlock_irq(q->queue_lock);
811 create_io_context(current, gfp_mask, q->node);
812 spin_lock_irq(q->queue_lock);
813 retried = true;
814 goto retry;
815 }
816
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100817 /*
818 * The queue will fill after this allocation, so set
819 * it as full, and mark this process as "batching".
820 * This process will be allowed to complete a batch of
821 * requests, others will be blocked.
822 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200823 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100824 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200825 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100826 } else {
827 if (may_queue != ELV_MQUEUE_MUST
828 && !ioc_batching(q, ioc)) {
829 /*
830 * The queue is full and the allocating
831 * process is not a "batcher", and not
832 * exempted by the IO scheduler
833 */
834 goto out;
835 }
836 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200838 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 }
840
Jens Axboe082cf692005-06-28 16:35:11 +0200841 /*
842 * Only allow batching queuers to allocate up to 50% over the defined
843 * limit of requests, otherwise we could have thousands of requests
844 * allocated with any setting of ->nr_requests
845 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200846 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200847 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100848
Jens Axboe1faa16d2009-04-06 14:48:01 +0200849 rl->count[is_sync]++;
850 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200851
Tejun Heof1f8cc92011-12-14 00:33:42 +0100852 /*
853 * Decide whether the new request will be managed by elevator. If
854 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
855 * prevent the current elevator from being destroyed until the new
856 * request is freed. This guarantees icq's won't be destroyed and
857 * makes creating new ones safe.
858 *
859 * Also, lookup icq while holding queue_lock. If it doesn't exist,
860 * it will be created after releasing queue_lock.
861 */
Tejun Heo75eb6c32011-10-19 14:31:22 +0200862 if (blk_rq_should_init_elevator(bio) &&
863 !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
864 rw_flags |= REQ_ELVPRIV;
865 rl->elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100866 if (et->icq_cache && ioc)
867 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100868 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200869
Jens Axboef253b862010-10-24 22:06:02 +0200870 if (blk_queue_io_stat(q))
871 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 spin_unlock_irq(q->queue_lock);
873
Tejun Heof1f8cc92011-12-14 00:33:42 +0100874 /* create icq if missing */
Shaohua Li05c30b92012-01-19 09:20:10 +0100875 if ((rw_flags & REQ_ELVPRIV) && unlikely(et->icq_cache && !icq)) {
Tejun Heof1f8cc92011-12-14 00:33:42 +0100876 icq = ioc_create_icq(q, gfp_mask);
Shaohua Li05c30b92012-01-19 09:20:10 +0100877 if (!icq)
878 goto fail_icq;
879 }
Tejun Heof1f8cc92011-12-14 00:33:42 +0100880
Shaohua Li05c30b92012-01-19 09:20:10 +0100881 rq = blk_alloc_request(q, icq, rw_flags, gfp_mask);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100882
Shaohua Li05c30b92012-01-19 09:20:10 +0100883fail_icq:
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100884 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 /*
886 * Allocation failed presumably due to memory. Undo anything
887 * we might have messed up.
888 *
889 * Allocating task should really be put onto the front of the
890 * wait queue, but this is pretty rare.
891 */
892 spin_lock_irq(q->queue_lock);
Tejun Heo75eb6c32011-10-19 14:31:22 +0200893 freed_request(q, rw_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
895 /*
896 * in the very unlikely event that allocation failed and no
897 * requests for this direction was pending, mark us starved
898 * so that freeing of a request in the other direction will
899 * notice us. another possible fix would be to split the
900 * rq mempool into READ and WRITE
901 */
902rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200903 if (unlikely(rl->count[is_sync] == 0))
904 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 goto out;
907 }
908
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100909 /*
910 * ioc may be NULL here, and ioc_batching will be false. That's
911 * OK, if the queue is under the request limit then requests need
912 * not count toward the nr_batch_requests limit. There will always
913 * be some limit enforced by BLK_BATCH_TIME.
914 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 if (ioc_batching(q, ioc))
916 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100917
Jens Axboe1faa16d2009-04-06 14:48:01 +0200918 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 return rq;
921}
922
Tejun Heoda8303c2011-10-19 14:33:05 +0200923/**
924 * get_request_wait - get a free request with retry
925 * @q: request_queue to allocate request from
926 * @rw_flags: RW and SYNC flags
927 * @bio: bio to allocate request for (can be %NULL)
Nick Piggind6344532005-06-28 20:45:14 -0700928 *
Tejun Heoda8303c2011-10-19 14:33:05 +0200929 * Get a free request from @q. This function keeps retrying under memory
930 * pressure and fails iff @q is dead.
931 *
932 * Must be callled with @q->queue_lock held and,
933 * Returns %NULL on failure, with @q->queue_lock held.
934 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 */
Jens Axboe165125e2007-07-24 09:28:11 +0200936static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200937 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200939 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 struct request *rq;
941
Jens Axboe7749a8d2006-12-13 13:02:26 +0100942 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700943 while (!rq) {
944 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 struct request_list *rl = &q->rq;
946
Tejun Heo34f60552011-12-14 00:33:37 +0100947 if (unlikely(blk_queue_dead(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200948 return NULL;
949
Jens Axboe1faa16d2009-04-06 14:48:01 +0200950 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 TASK_UNINTERRUPTIBLE);
952
Jens Axboe1faa16d2009-04-06 14:48:01 +0200953 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200955 spin_unlock_irq(q->queue_lock);
956 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200958 /*
959 * After sleeping, we become a "batching" process and
960 * will be able to allocate at least one request, and
961 * up to a big batch of them for a small period time.
962 * See ioc_batching, ioc_set_batching
963 */
Tejun Heof2dbd762011-12-14 00:33:40 +0100964 create_io_context(current, GFP_NOIO, q->node);
965 ioc_set_batching(q, current->io_context);
Jens Axboe2056a782006-03-23 20:00:26 +0100966
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200967 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200968 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200969
970 rq = get_request(q, rw_flags, bio, GFP_NOIO);
971 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 return rq;
974}
975
Jens Axboe165125e2007-07-24 09:28:11 +0200976struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977{
978 struct request *rq;
979
980 BUG_ON(rw != READ && rw != WRITE);
981
Nick Piggind6344532005-06-28 20:45:14 -0700982 spin_lock_irq(q->queue_lock);
Tejun Heoda8303c2011-10-19 14:33:05 +0200983 if (gfp_mask & __GFP_WAIT)
Jens Axboe22e2c502005-06-27 10:55:12 +0200984 rq = get_request_wait(q, rw, NULL);
Tejun Heoda8303c2011-10-19 14:33:05 +0200985 else
Jens Axboe22e2c502005-06-27 10:55:12 +0200986 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +0200987 if (!rq)
988 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -0700989 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
991 return rq;
992}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993EXPORT_SYMBOL(blk_get_request);
994
995/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300996 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700997 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300998 * @bio: The bio describing the memory mappings that will be submitted for IO.
999 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001000 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001001 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001002 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1003 * type commands. Where the struct request needs to be farther initialized by
1004 * the caller. It is passed a &struct bio, which describes the memory info of
1005 * the I/O transfer.
1006 *
1007 * The caller of blk_make_request must make sure that bi_io_vec
1008 * are set to describe the memory buffers. That bio_data_dir() will return
1009 * the needed direction of the request. (And all bio's in the passed bio-chain
1010 * are properly set accordingly)
1011 *
1012 * If called under none-sleepable conditions, mapped bio buffers must not
1013 * need bouncing, by calling the appropriate masked or flagged allocator,
1014 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1015 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001016 *
1017 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1018 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1019 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1020 * completion of a bio that hasn't been submitted yet, thus resulting in a
1021 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1022 * of bio_alloc(), as that avoids the mempool deadlock.
1023 * If possible a big IO should be split into smaller parts when allocation
1024 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001025 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001026struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1027 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001028{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001029 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1030
1031 if (unlikely(!rq))
1032 return ERR_PTR(-ENOMEM);
1033
1034 for_each_bio(bio) {
1035 struct bio *bounce_bio = bio;
1036 int ret;
1037
1038 blk_queue_bounce(q, &bounce_bio);
1039 ret = blk_rq_append_bio(q, rq, bounce_bio);
1040 if (unlikely(ret)) {
1041 blk_put_request(rq);
1042 return ERR_PTR(ret);
1043 }
1044 }
1045
1046 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001047}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001048EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001049
1050/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 * blk_requeue_request - put a request back on queue
1052 * @q: request queue where request should be inserted
1053 * @rq: request to be inserted
1054 *
1055 * Description:
1056 * Drivers often keep queueing requests until the hardware cannot accept
1057 * more, when that condition happens we need to put the request back
1058 * on the queue. Must be called with queue lock held.
1059 */
Jens Axboe165125e2007-07-24 09:28:11 +02001060void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001062 blk_delete_timer(rq);
1063 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001064 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001065
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 if (blk_rq_tagged(rq))
1067 blk_queue_end_tag(q, rq);
1068
James Bottomleyba396a62009-05-27 14:17:08 +02001069 BUG_ON(blk_queued_rq(rq));
1070
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 elv_requeue_request(q, rq);
1072}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073EXPORT_SYMBOL(blk_requeue_request);
1074
Jens Axboe73c10102011-03-08 13:19:51 +01001075static void add_acct_request(struct request_queue *q, struct request *rq,
1076 int where)
1077{
1078 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001079 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001080}
1081
Tejun Heo074a7ac2008-08-25 19:56:14 +09001082static void part_round_stats_single(int cpu, struct hd_struct *part,
1083 unsigned long now)
1084{
1085 if (now == part->stamp)
1086 return;
1087
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001088 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001089 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001090 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001091 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1092 }
1093 part->stamp = now;
1094}
1095
1096/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001097 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1098 * @cpu: cpu number for stats access
1099 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 *
1101 * The average IO queue length and utilisation statistics are maintained
1102 * by observing the current state of the queue length and the amount of
1103 * time it has been in this state for.
1104 *
1105 * Normally, that accounting is done on IO completion, but that can result
1106 * in more than a second's worth of IO being accounted for within any one
1107 * second, leading to >100% utilisation. To deal with that, we call this
1108 * function to do a round-off before returning the results when reading
1109 * /proc/diskstats. This accounts immediately for all queue usage up to
1110 * the current jiffies and restarts the counters again.
1111 */
Tejun Heoc9959052008-08-25 19:47:21 +09001112void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001113{
1114 unsigned long now = jiffies;
1115
Tejun Heo074a7ac2008-08-25 19:56:14 +09001116 if (part->partno)
1117 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1118 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001119}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001120EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001121
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122/*
1123 * queue lock must be held
1124 */
Jens Axboe165125e2007-07-24 09:28:11 +02001125void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 if (unlikely(!q))
1128 return;
1129 if (unlikely(--req->ref_count))
1130 return;
1131
Tejun Heo8922e162005-10-20 16:23:44 +02001132 elv_completed_request(q, req);
1133
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001134 /* this is a bio leak */
1135 WARN_ON(req->bio != NULL);
1136
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 /*
1138 * Request may not have originated from ll_rw_blk. if not,
1139 * it didn't come out of our reserved rq pools
1140 */
Jens Axboe49171e52006-08-10 08:59:11 +02001141 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001142 unsigned int flags = req->cmd_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001145 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
1147 blk_free_request(q, req);
Tejun Heo75eb6c32011-10-19 14:31:22 +02001148 freed_request(q, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 }
1150}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001151EXPORT_SYMBOL_GPL(__blk_put_request);
1152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153void blk_put_request(struct request *req)
1154{
Tejun Heo8922e162005-10-20 16:23:44 +02001155 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001156 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001158 spin_lock_irqsave(q->queue_lock, flags);
1159 __blk_put_request(q, req);
1160 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162EXPORT_SYMBOL(blk_put_request);
1163
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001164/**
1165 * blk_add_request_payload - add a payload to a request
1166 * @rq: request to update
1167 * @page: page backing the payload
1168 * @len: length of the payload.
1169 *
1170 * This allows to later add a payload to an already submitted request by
1171 * a block driver. The driver needs to take care of freeing the payload
1172 * itself.
1173 *
1174 * Note that this is a quite horrible hack and nothing but handling of
1175 * discard requests should ever use it.
1176 */
1177void blk_add_request_payload(struct request *rq, struct page *page,
1178 unsigned int len)
1179{
1180 struct bio *bio = rq->bio;
1181
1182 bio->bi_io_vec->bv_page = page;
1183 bio->bi_io_vec->bv_offset = 0;
1184 bio->bi_io_vec->bv_len = len;
1185
1186 bio->bi_size = len;
1187 bio->bi_vcnt = 1;
1188 bio->bi_phys_segments = 1;
1189
1190 rq->__data_len = rq->resid_len = len;
1191 rq->nr_phys_segments = 1;
1192 rq->buffer = bio_data(bio);
1193}
1194EXPORT_SYMBOL_GPL(blk_add_request_payload);
1195
Jens Axboe73c10102011-03-08 13:19:51 +01001196static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1197 struct bio *bio)
1198{
1199 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1200
Jens Axboe73c10102011-03-08 13:19:51 +01001201 if (!ll_back_merge_fn(q, req, bio))
1202 return false;
1203
1204 trace_block_bio_backmerge(q, bio);
1205
1206 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1207 blk_rq_set_mixed_merge(req);
1208
1209 req->biotail->bi_next = bio;
1210 req->biotail = bio;
1211 req->__data_len += bio->bi_size;
1212 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1213
1214 drive_stat_acct(req, 0);
1215 return true;
1216}
1217
1218static bool bio_attempt_front_merge(struct request_queue *q,
1219 struct request *req, struct bio *bio)
1220{
1221 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001222
Jens Axboe73c10102011-03-08 13:19:51 +01001223 if (!ll_front_merge_fn(q, req, bio))
1224 return false;
1225
1226 trace_block_bio_frontmerge(q, bio);
1227
1228 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1229 blk_rq_set_mixed_merge(req);
1230
Jens Axboe73c10102011-03-08 13:19:51 +01001231 bio->bi_next = req->bio;
1232 req->bio = bio;
1233
1234 /*
1235 * may not be valid. if the low level driver said
1236 * it didn't need a bounce buffer then it better
1237 * not touch req->buffer either...
1238 */
1239 req->buffer = bio_data(bio);
1240 req->__sector = bio->bi_sector;
1241 req->__data_len += bio->bi_size;
1242 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1243
1244 drive_stat_acct(req, 0);
1245 return true;
1246}
1247
Tejun Heobd87b582011-10-19 14:33:08 +02001248/**
1249 * attempt_plug_merge - try to merge with %current's plugged list
1250 * @q: request_queue new bio is being queued at
1251 * @bio: new bio being queued
1252 * @request_count: out parameter for number of traversed plugged requests
1253 *
1254 * Determine whether @bio being queued on @q can be merged with a request
1255 * on %current's plugged list. Returns %true if merge was successful,
1256 * otherwise %false.
1257 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001258 * Plugging coalesces IOs from the same issuer for the same purpose without
1259 * going through @q->queue_lock. As such it's more of an issuing mechanism
1260 * than scheduling, and the request, while may have elvpriv data, is not
1261 * added on the elevator at this point. In addition, we don't have
1262 * reliable access to the elevator outside queue lock. Only check basic
1263 * merging parameters without querying the elevator.
Jens Axboe73c10102011-03-08 13:19:51 +01001264 */
Tejun Heobd87b582011-10-19 14:33:08 +02001265static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1266 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001267{
1268 struct blk_plug *plug;
1269 struct request *rq;
1270 bool ret = false;
1271
Tejun Heobd87b582011-10-19 14:33:08 +02001272 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001273 if (!plug)
1274 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001275 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001276
1277 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1278 int el_ret;
1279
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001280 if (rq->q == q)
1281 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001282
Tejun Heo07c2bd32012-02-08 09:19:42 +01001283 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001284 continue;
1285
Tejun Heo050c8ea2012-02-08 09:19:38 +01001286 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001287 if (el_ret == ELEVATOR_BACK_MERGE) {
1288 ret = bio_attempt_back_merge(q, rq, bio);
1289 if (ret)
1290 break;
1291 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1292 ret = bio_attempt_front_merge(q, rq, bio);
1293 if (ret)
1294 break;
1295 }
1296 }
1297out:
1298 return ret;
1299}
1300
Jens Axboe86db1e22008-01-29 14:53:40 +01001301void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001302{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001303 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001304
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001305 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1306 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001307 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001308
Tejun Heo52d9e672006-01-06 09:49:58 +01001309 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001310 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001311 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001312 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001313}
1314
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001315void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001317 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001318 struct blk_plug *plug;
1319 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1320 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001321 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 /*
1324 * low level driver can indicate that it wants pages above a
1325 * certain limit bounced to low memory (ie for highmem, or even
1326 * ISA dma in theory)
1327 */
1328 blk_queue_bounce(q, &bio);
1329
Tejun Heo4fed9472010-09-03 11:56:17 +02001330 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001331 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001332 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001333 goto get_rq;
1334 }
1335
Jens Axboe73c10102011-03-08 13:19:51 +01001336 /*
1337 * Check if we can merge with the plugged list before grabbing
1338 * any locks.
1339 */
Tejun Heobd87b582011-10-19 14:33:08 +02001340 if (attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001341 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001342
1343 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
1345 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001346 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001347 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001348 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001349 if (!attempt_back_merge(q, req))
1350 elv_merged_request(q, req, el_ret);
1351 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001352 }
Jens Axboe73c10102011-03-08 13:19:51 +01001353 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001354 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001355 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001356 if (!attempt_front_merge(q, req))
1357 elv_merged_request(q, req, el_ret);
1358 goto out_unlock;
1359 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 }
1361
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001363 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001364 * This sync check and mask will be re-done in init_request_from_bio(),
1365 * but we need to set it earlier to expose the sync flag to the
1366 * rq allocator and io schedulers.
1367 */
1368 rw_flags = bio_data_dir(bio);
1369 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001370 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001371
1372 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001373 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001374 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001375 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001376 req = get_request_wait(q, rw_flags, bio);
Tejun Heoda8303c2011-10-19 14:33:05 +02001377 if (unlikely(!req)) {
1378 bio_endio(bio, -ENODEV); /* @q is dead */
1379 goto out_unlock;
1380 }
Nick Piggind6344532005-06-28 20:45:14 -07001381
Nick Piggin450991b2005-06-28 20:45:13 -07001382 /*
1383 * After dropping the lock and possibly sleeping here, our request
1384 * may now be mergeable after it had proven unmergeable (above).
1385 * We don't worry about that case for efficiency. It won't happen
1386 * often, and the elevators are able to handle it.
1387 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001388 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Tao Ma9562ad92011-10-24 16:11:30 +02001390 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001391 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001392
Jens Axboe73c10102011-03-08 13:19:51 +01001393 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001394 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001395 /*
1396 * If this is the first request added after a plug, fire
1397 * of a plug trace. If others have been added before, check
1398 * if we have multiple devices in this plug. If so, make a
1399 * note to sort the list before dispatch.
1400 */
1401 if (list_empty(&plug->list))
1402 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001403 else {
1404 if (!plug->should_sort) {
1405 struct request *__rq;
Jens Axboe73c10102011-03-08 13:19:51 +01001406
Shaohua Li3540d5e2011-11-16 09:21:50 +01001407 __rq = list_entry_rq(plug->list.prev);
1408 if (__rq->q != q)
1409 plug->should_sort = 1;
1410 }
Shaohua Li019ceb72011-11-16 09:21:50 +01001411 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001412 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001413 trace_block_plug(q);
1414 }
Jens Axboe73c10102011-03-08 13:19:51 +01001415 }
Shaohua Lia6327162011-08-24 16:04:32 +02001416 list_add_tail(&req->queuelist, &plug->list);
1417 drive_stat_acct(req, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001418 } else {
1419 spin_lock_irq(q->queue_lock);
1420 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001421 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001422out_unlock:
1423 spin_unlock_irq(q->queue_lock);
1424 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425}
Jens Axboec20e8de2011-09-12 12:03:37 +02001426EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428/*
1429 * If bio->bi_dev is a partition, remap the location
1430 */
1431static inline void blk_partition_remap(struct bio *bio)
1432{
1433 struct block_device *bdev = bio->bi_bdev;
1434
Jens Axboebf2de6f2007-09-27 13:01:25 +02001435 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001437
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 bio->bi_sector += p->start_sect;
1439 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001440
Mike Snitzerd07335e2010-11-16 12:52:38 +01001441 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1442 bdev->bd_dev,
1443 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 }
1445}
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447static void handle_bad_sector(struct bio *bio)
1448{
1449 char b[BDEVNAME_SIZE];
1450
1451 printk(KERN_INFO "attempt to access beyond end of device\n");
1452 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1453 bdevname(bio->bi_bdev, b),
1454 bio->bi_rw,
1455 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001456 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
1458 set_bit(BIO_EOF, &bio->bi_flags);
1459}
1460
Akinobu Mitac17bb492006-12-08 02:39:46 -08001461#ifdef CONFIG_FAIL_MAKE_REQUEST
1462
1463static DECLARE_FAULT_ATTR(fail_make_request);
1464
1465static int __init setup_fail_make_request(char *str)
1466{
1467 return setup_fault_attr(&fail_make_request, str);
1468}
1469__setup("fail_make_request=", setup_fail_make_request);
1470
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001471static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001472{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001473 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001474}
1475
1476static int __init fail_make_request_debugfs(void)
1477{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001478 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1479 NULL, &fail_make_request);
1480
1481 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001482}
1483
1484late_initcall(fail_make_request_debugfs);
1485
1486#else /* CONFIG_FAIL_MAKE_REQUEST */
1487
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001488static inline bool should_fail_request(struct hd_struct *part,
1489 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001490{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001491 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001492}
1493
1494#endif /* CONFIG_FAIL_MAKE_REQUEST */
1495
Jens Axboec07e2b42007-07-18 13:27:58 +02001496/*
1497 * Check whether this bio extends beyond the end of the device.
1498 */
1499static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1500{
1501 sector_t maxsector;
1502
1503 if (!nr_sectors)
1504 return 0;
1505
1506 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001507 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001508 if (maxsector) {
1509 sector_t sector = bio->bi_sector;
1510
1511 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1512 /*
1513 * This may well happen - the kernel calls bread()
1514 * without checking the size of the device, e.g., when
1515 * mounting a device.
1516 */
1517 handle_bad_sector(bio);
1518 return 1;
1519 }
1520 }
1521
1522 return 0;
1523}
1524
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001525static noinline_for_stack bool
1526generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527{
Jens Axboe165125e2007-07-24 09:28:11 +02001528 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001529 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001530 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001531 char b[BDEVNAME_SIZE];
1532 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
1534 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
Jens Axboec07e2b42007-07-18 13:27:58 +02001536 if (bio_check_eod(bio, nr_sectors))
1537 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001539 q = bdev_get_queue(bio->bi_bdev);
1540 if (unlikely(!q)) {
1541 printk(KERN_ERR
1542 "generic_make_request: Trying to access "
1543 "nonexistent block-device %s (%Lu)\n",
1544 bdevname(bio->bi_bdev, b),
1545 (long long) bio->bi_sector);
1546 goto end_io;
1547 }
1548
1549 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1550 nr_sectors > queue_max_hw_sectors(q))) {
1551 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1552 bdevname(bio->bi_bdev, b),
1553 bio_sectors(bio),
1554 queue_max_hw_sectors(q));
1555 goto end_io;
1556 }
1557
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001558 part = bio->bi_bdev->bd_part;
1559 if (should_fail_request(part, bio->bi_size) ||
1560 should_fail_request(&part_to_disk(part)->part0,
1561 bio->bi_size))
1562 goto end_io;
1563
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001565 * If this device has partitions, remap block n
1566 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001568 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001570 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1571 goto end_io;
1572
1573 if (bio_check_eod(bio, nr_sectors))
1574 goto end_io;
1575
1576 /*
1577 * Filter flush bio's early so that make_request based
1578 * drivers without flush support don't have to worry
1579 * about them.
1580 */
1581 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1582 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1583 if (!nr_sectors) {
1584 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001585 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001587 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001589 if ((bio->bi_rw & REQ_DISCARD) &&
1590 (!blk_queue_discard(q) ||
1591 ((bio->bi_rw & REQ_SECURE) &&
1592 !blk_queue_secdiscard(q)))) {
1593 err = -EOPNOTSUPP;
1594 goto end_io;
1595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
Tejun Heobc16a4f2011-10-19 14:33:01 +02001597 if (blk_throtl_bio(q, bio))
1598 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001599
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001600 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001601 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001602
1603end_io:
1604 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001605 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606}
1607
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001608/**
1609 * generic_make_request - hand a buffer to its device driver for I/O
1610 * @bio: The bio describing the location in memory and on the device.
1611 *
1612 * generic_make_request() is used to make I/O requests of block
1613 * devices. It is passed a &struct bio, which describes the I/O that needs
1614 * to be done.
1615 *
1616 * generic_make_request() does not return any status. The
1617 * success/failure status of the request, along with notification of
1618 * completion, is delivered asynchronously through the bio->bi_end_io
1619 * function described (one day) else where.
1620 *
1621 * The caller of generic_make_request must make sure that bi_io_vec
1622 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1623 * set to describe the device address, and the
1624 * bi_end_io and optionally bi_private are set to describe how
1625 * completion notification should be signaled.
1626 *
1627 * generic_make_request and the drivers it calls may use bi_next if this
1628 * bio happens to be merged with someone else, and may resubmit the bio to
1629 * a lower device by calling into generic_make_request recursively, which
1630 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001631 */
1632void generic_make_request(struct bio *bio)
1633{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001634 struct bio_list bio_list_on_stack;
1635
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001636 if (!generic_make_request_checks(bio))
1637 return;
1638
1639 /*
1640 * We only want one ->make_request_fn to be active at a time, else
1641 * stack usage with stacked devices could be a problem. So use
1642 * current->bio_list to keep a list of requests submited by a
1643 * make_request_fn function. current->bio_list is also used as a
1644 * flag to say if generic_make_request is currently active in this
1645 * task or not. If it is NULL, then no make_request is active. If
1646 * it is non-NULL, then a make_request is active, and new requests
1647 * should be added at the tail
1648 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001649 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001650 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001651 return;
1652 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001653
Neil Brownd89d8792007-05-01 09:53:42 +02001654 /* following loop may be a bit non-obvious, and so deserves some
1655 * explanation.
1656 * Before entering the loop, bio->bi_next is NULL (as all callers
1657 * ensure that) so we have a list with a single bio.
1658 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001659 * we assign bio_list to a pointer to the bio_list_on_stack,
1660 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001661 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001662 * through a recursive call to generic_make_request. If it
1663 * did, we find a non-NULL value in bio_list and re-enter the loop
1664 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001665 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001666 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001667 */
1668 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001669 bio_list_init(&bio_list_on_stack);
1670 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001671 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001672 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1673
1674 q->make_request_fn(q, bio);
1675
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001676 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001677 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001678 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001679}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680EXPORT_SYMBOL(generic_make_request);
1681
1682/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001683 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1685 * @bio: The &struct bio which describes the I/O
1686 *
1687 * submit_bio() is very similar in purpose to generic_make_request(), and
1688 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001689 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 *
1691 */
1692void submit_bio(int rw, struct bio *bio)
1693{
1694 int count = bio_sectors(bio);
1695
Jens Axboe22e2c502005-06-27 10:55:12 +02001696 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
Jens Axboebf2de6f2007-09-27 13:01:25 +02001698 /*
1699 * If it's a regular read/write or a barrier with data attached,
1700 * go through the normal accounting stuff before submission.
1701 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001702 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001703 if (rw & WRITE) {
1704 count_vm_events(PGPGOUT, count);
1705 } else {
1706 task_io_account_read(bio->bi_size);
1707 count_vm_events(PGPGIN, count);
1708 }
1709
1710 if (unlikely(block_dump)) {
1711 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001712 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001713 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001714 (rw & WRITE) ? "WRITE" : "READ",
1715 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001716 bdevname(bio->bi_bdev, b),
1717 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 }
1720
1721 generic_make_request(bio);
1722}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723EXPORT_SYMBOL(submit_bio);
1724
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001725/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001726 * blk_rq_check_limits - Helper function to check a request for the queue limit
1727 * @q: the queue
1728 * @rq: the request being checked
1729 *
1730 * Description:
1731 * @rq may have been made based on weaker limitations of upper-level queues
1732 * in request stacking drivers, and it may violate the limitation of @q.
1733 * Since the block layer and the underlying device driver trust @rq
1734 * after it is inserted to @q, it should be checked against @q before
1735 * the insertion using this generic function.
1736 *
1737 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001738 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001739 * Request stacking drivers like request-based dm may change the queue
1740 * limits while requests are in the queue (e.g. dm's table swapping).
1741 * Such request stacking drivers should check those requests agaist
1742 * the new queue limits again when they dispatch those requests,
1743 * although such checkings are also done against the old queue limits
1744 * when submitting requests.
1745 */
1746int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1747{
ike Snitzer33839772010-08-08 12:11:33 -04001748 if (rq->cmd_flags & REQ_DISCARD)
1749 return 0;
1750
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001751 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1752 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001753 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1754 return -EIO;
1755 }
1756
1757 /*
1758 * queue's settings related to segment counting like q->bounce_pfn
1759 * may differ from that of other stacking queues.
1760 * Recalculate it to check the request correctly on this queue's
1761 * limitation.
1762 */
1763 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001764 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001765 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1766 return -EIO;
1767 }
1768
1769 return 0;
1770}
1771EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1772
1773/**
1774 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1775 * @q: the queue to submit the request
1776 * @rq: the request being queued
1777 */
1778int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1779{
1780 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001781 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001782
1783 if (blk_rq_check_limits(q, rq))
1784 return -EIO;
1785
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001786 if (rq->rq_disk &&
1787 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001788 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001789
1790 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo8ba61432011-12-14 00:33:37 +01001791 if (unlikely(blk_queue_dead(q))) {
1792 spin_unlock_irqrestore(q->queue_lock, flags);
1793 return -ENODEV;
1794 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001795
1796 /*
1797 * Submitting request must be dequeued before calling this function
1798 * because it will be linked to another request_queue
1799 */
1800 BUG_ON(blk_queued_rq(rq));
1801
Jeff Moyer4853aba2011-08-15 21:37:25 +02001802 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1803 where = ELEVATOR_INSERT_FLUSH;
1804
1805 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001806 if (where == ELEVATOR_INSERT_FLUSH)
1807 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001808 spin_unlock_irqrestore(q->queue_lock, flags);
1809
1810 return 0;
1811}
1812EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1813
Tejun Heo80a761f2009-07-03 17:48:17 +09001814/**
1815 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1816 * @rq: request to examine
1817 *
1818 * Description:
1819 * A request could be merge of IOs which require different failure
1820 * handling. This function determines the number of bytes which
1821 * can be failed from the beginning of the request without
1822 * crossing into area which need to be retried further.
1823 *
1824 * Return:
1825 * The number of bytes to fail.
1826 *
1827 * Context:
1828 * queue_lock must be held.
1829 */
1830unsigned int blk_rq_err_bytes(const struct request *rq)
1831{
1832 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1833 unsigned int bytes = 0;
1834 struct bio *bio;
1835
1836 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1837 return blk_rq_bytes(rq);
1838
1839 /*
1840 * Currently the only 'mixing' which can happen is between
1841 * different fastfail types. We can safely fail portions
1842 * which have all the failfast bits that the first one has -
1843 * the ones which are at least as eager to fail as the first
1844 * one.
1845 */
1846 for (bio = rq->bio; bio; bio = bio->bi_next) {
1847 if ((bio->bi_rw & ff) != ff)
1848 break;
1849 bytes += bio->bi_size;
1850 }
1851
1852 /* this could lead to infinite loop */
1853 BUG_ON(blk_rq_bytes(rq) && !bytes);
1854 return bytes;
1855}
1856EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1857
Jens Axboebc58ba92009-01-23 10:54:44 +01001858static void blk_account_io_completion(struct request *req, unsigned int bytes)
1859{
Jens Axboec2553b52009-04-24 08:10:11 +02001860 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001861 const int rw = rq_data_dir(req);
1862 struct hd_struct *part;
1863 int cpu;
1864
1865 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001866 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001867 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1868 part_stat_unlock();
1869 }
1870}
1871
1872static void blk_account_io_done(struct request *req)
1873{
Jens Axboebc58ba92009-01-23 10:54:44 +01001874 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001875 * Account IO completion. flush_rq isn't accounted as a
1876 * normal IO on queueing nor completion. Accounting the
1877 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001878 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001879 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001880 unsigned long duration = jiffies - req->start_time;
1881 const int rw = rq_data_dir(req);
1882 struct hd_struct *part;
1883 int cpu;
1884
1885 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001886 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001887
1888 part_stat_inc(cpu, part, ios[rw]);
1889 part_stat_add(cpu, part, ticks[rw], duration);
1890 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001891 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001892
Jens Axboe6c23a962011-01-07 08:43:37 +01001893 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001894 part_stat_unlock();
1895 }
1896}
1897
Tejun Heo53a08802008-12-03 12:41:26 +01001898/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001899 * blk_peek_request - peek at the top of a request queue
1900 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001901 *
1902 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001903 * Return the request at the top of @q. The returned request
1904 * should be started using blk_start_request() before LLD starts
1905 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001906 *
1907 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001908 * Pointer to the request at the top of @q if available. Null
1909 * otherwise.
1910 *
1911 * Context:
1912 * queue_lock must be held.
1913 */
1914struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001915{
1916 struct request *rq;
1917 int ret;
1918
1919 while ((rq = __elv_next_request(q)) != NULL) {
1920 if (!(rq->cmd_flags & REQ_STARTED)) {
1921 /*
1922 * This is the first time the device driver
1923 * sees this request (possibly after
1924 * requeueing). Notify IO scheduler.
1925 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001926 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001927 elv_activate_rq(q, rq);
1928
1929 /*
1930 * just mark as started even if we don't start
1931 * it, a request that has been delayed should
1932 * not be passed by new incoming requests
1933 */
1934 rq->cmd_flags |= REQ_STARTED;
1935 trace_block_rq_issue(q, rq);
1936 }
1937
1938 if (!q->boundary_rq || q->boundary_rq == rq) {
1939 q->end_sector = rq_end_sector(rq);
1940 q->boundary_rq = NULL;
1941 }
1942
1943 if (rq->cmd_flags & REQ_DONTPREP)
1944 break;
1945
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001946 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001947 /*
1948 * make sure space for the drain appears we
1949 * know we can do this because max_hw_segments
1950 * has been adjusted to be one fewer than the
1951 * device can handle
1952 */
1953 rq->nr_phys_segments++;
1954 }
1955
1956 if (!q->prep_rq_fn)
1957 break;
1958
1959 ret = q->prep_rq_fn(q, rq);
1960 if (ret == BLKPREP_OK) {
1961 break;
1962 } else if (ret == BLKPREP_DEFER) {
1963 /*
1964 * the request may have been (partially) prepped.
1965 * we need to keep this request in the front to
1966 * avoid resource deadlock. REQ_STARTED will
1967 * prevent other fs requests from passing this one.
1968 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001969 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001970 !(rq->cmd_flags & REQ_DONTPREP)) {
1971 /*
1972 * remove the space for the drain we added
1973 * so that we don't add it again
1974 */
1975 --rq->nr_phys_segments;
1976 }
1977
1978 rq = NULL;
1979 break;
1980 } else if (ret == BLKPREP_KILL) {
1981 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001982 /*
1983 * Mark this request as started so we don't trigger
1984 * any debug logic in the end I/O path.
1985 */
1986 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001987 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001988 } else {
1989 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1990 break;
1991 }
1992 }
1993
1994 return rq;
1995}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001996EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001997
Tejun Heo9934c8c2009-05-08 11:54:16 +09001998void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001999{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002000 struct request_queue *q = rq->q;
2001
Tejun Heo158dbda2009-04-23 11:05:18 +09002002 BUG_ON(list_empty(&rq->queuelist));
2003 BUG_ON(ELV_ON_HASH(rq));
2004
2005 list_del_init(&rq->queuelist);
2006
2007 /*
2008 * the time frame between a request being removed from the lists
2009 * and to it is freed is accounted as io that is in progress at
2010 * the driver side.
2011 */
Divyesh Shah91952912010-04-01 15:01:41 -07002012 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002013 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002014 set_io_start_time_ns(rq);
2015 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002016}
2017
Tejun Heo5efccd12009-04-23 11:05:18 +09002018/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002019 * blk_start_request - start request processing on the driver
2020 * @req: request to dequeue
2021 *
2022 * Description:
2023 * Dequeue @req and start timeout timer on it. This hands off the
2024 * request to the driver.
2025 *
2026 * Block internal functions which don't want to start timer should
2027 * call blk_dequeue_request().
2028 *
2029 * Context:
2030 * queue_lock must be held.
2031 */
2032void blk_start_request(struct request *req)
2033{
2034 blk_dequeue_request(req);
2035
2036 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002037 * We are now handing the request to the hardware, initialize
2038 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002039 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002040 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002041 if (unlikely(blk_bidi_rq(req)))
2042 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2043
Tejun Heo9934c8c2009-05-08 11:54:16 +09002044 blk_add_timer(req);
2045}
2046EXPORT_SYMBOL(blk_start_request);
2047
2048/**
2049 * blk_fetch_request - fetch a request from a request queue
2050 * @q: request queue to fetch a request from
2051 *
2052 * Description:
2053 * Return the request at the top of @q. The request is started on
2054 * return and LLD can start processing it immediately.
2055 *
2056 * Return:
2057 * Pointer to the request at the top of @q if available. Null
2058 * otherwise.
2059 *
2060 * Context:
2061 * queue_lock must be held.
2062 */
2063struct request *blk_fetch_request(struct request_queue *q)
2064{
2065 struct request *rq;
2066
2067 rq = blk_peek_request(q);
2068 if (rq)
2069 blk_start_request(rq);
2070 return rq;
2071}
2072EXPORT_SYMBOL(blk_fetch_request);
2073
2074/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002075 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002076 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002077 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002078 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002079 *
2080 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002081 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2082 * the request structure even if @req doesn't have leftover.
2083 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002084 *
2085 * This special helper function is only for request stacking drivers
2086 * (e.g. request-based dm) so that they can handle partial completion.
2087 * Actual device drivers should use blk_end_request instead.
2088 *
2089 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2090 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002091 *
2092 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002093 * %false - this request doesn't have any more data
2094 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002095 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002096bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002098 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 struct bio *bio;
2100
Tejun Heo2e60e022009-04-23 11:05:18 +09002101 if (!req->bio)
2102 return false;
2103
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002104 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002105
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002107 * For fs requests, rq is just carrier of independent bio's
2108 * and each partial completion should be handled separately.
2109 * Reset per-request error on each partial completion.
2110 *
2111 * TODO: tj: This is too subtle. It would be better to let
2112 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002114 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 req->errors = 0;
2116
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002117 if (error && req->cmd_type == REQ_TYPE_FS &&
2118 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002119 char *error_type;
2120
2121 switch (error) {
2122 case -ENOLINK:
2123 error_type = "recoverable transport";
2124 break;
2125 case -EREMOTEIO:
2126 error_type = "critical target";
2127 break;
2128 case -EBADE:
2129 error_type = "critical nexus";
2130 break;
2131 case -EIO:
2132 default:
2133 error_type = "I/O";
2134 break;
2135 }
2136 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2137 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2138 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 }
2140
Jens Axboebc58ba92009-01-23 10:54:44 +01002141 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002142
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 total_bytes = bio_nbytes = 0;
2144 while ((bio = req->bio) != NULL) {
2145 int nbytes;
2146
2147 if (nr_bytes >= bio->bi_size) {
2148 req->bio = bio->bi_next;
2149 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002150 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 next_idx = 0;
2152 bio_nbytes = 0;
2153 } else {
2154 int idx = bio->bi_idx + next_idx;
2155
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002156 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002158 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002159 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 break;
2161 }
2162
2163 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2164 BIO_BUG_ON(nbytes > bio->bi_size);
2165
2166 /*
2167 * not a complete bvec done
2168 */
2169 if (unlikely(nbytes > nr_bytes)) {
2170 bio_nbytes += nr_bytes;
2171 total_bytes += nr_bytes;
2172 break;
2173 }
2174
2175 /*
2176 * advance to the next vector
2177 */
2178 next_idx++;
2179 bio_nbytes += nbytes;
2180 }
2181
2182 total_bytes += nbytes;
2183 nr_bytes -= nbytes;
2184
Jens Axboe6728cb02008-01-31 13:03:55 +01002185 bio = req->bio;
2186 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 /*
2188 * end more in this run, or just return 'not-done'
2189 */
2190 if (unlikely(nr_bytes <= 0))
2191 break;
2192 }
2193 }
2194
2195 /*
2196 * completely done
2197 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002198 if (!req->bio) {
2199 /*
2200 * Reset counters so that the request stacking driver
2201 * can find how many bytes remain in the request
2202 * later.
2203 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002204 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002205 return false;
2206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
2208 /*
2209 * if the request wasn't completed, update state
2210 */
2211 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002212 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 bio->bi_idx += next_idx;
2214 bio_iovec(bio)->bv_offset += nr_bytes;
2215 bio_iovec(bio)->bv_len -= nr_bytes;
2216 }
2217
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002218 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002219 req->buffer = bio_data(req->bio);
2220
2221 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002222 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002223 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002224
Tejun Heo80a761f2009-07-03 17:48:17 +09002225 /* mixed attributes always follow the first bio */
2226 if (req->cmd_flags & REQ_MIXED_MERGE) {
2227 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2228 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2229 }
2230
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002231 /*
2232 * If total number of sectors is less than the first segment
2233 * size, something has gone terribly wrong.
2234 */
2235 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002236 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002237 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002238 }
2239
2240 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002242
Tejun Heo2e60e022009-04-23 11:05:18 +09002243 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244}
Tejun Heo2e60e022009-04-23 11:05:18 +09002245EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
Tejun Heo2e60e022009-04-23 11:05:18 +09002247static bool blk_update_bidi_request(struct request *rq, int error,
2248 unsigned int nr_bytes,
2249 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002250{
Tejun Heo2e60e022009-04-23 11:05:18 +09002251 if (blk_update_request(rq, error, nr_bytes))
2252 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002253
Tejun Heo2e60e022009-04-23 11:05:18 +09002254 /* Bidi request must be completed as a whole */
2255 if (unlikely(blk_bidi_rq(rq)) &&
2256 blk_update_request(rq->next_rq, error, bidi_bytes))
2257 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002258
Jens Axboee2e1a142010-06-09 10:42:09 +02002259 if (blk_queue_add_random(rq->q))
2260 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002261
2262 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263}
2264
James Bottomley28018c22010-07-01 19:49:17 +09002265/**
2266 * blk_unprep_request - unprepare a request
2267 * @req: the request
2268 *
2269 * This function makes a request ready for complete resubmission (or
2270 * completion). It happens only after all error handling is complete,
2271 * so represents the appropriate moment to deallocate any resources
2272 * that were allocated to the request in the prep_rq_fn. The queue
2273 * lock is held when calling this.
2274 */
2275void blk_unprep_request(struct request *req)
2276{
2277 struct request_queue *q = req->q;
2278
2279 req->cmd_flags &= ~REQ_DONTPREP;
2280 if (q->unprep_rq_fn)
2281 q->unprep_rq_fn(q, req);
2282}
2283EXPORT_SYMBOL_GPL(blk_unprep_request);
2284
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285/*
2286 * queue lock must be held
2287 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002288static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002290 if (blk_rq_tagged(req))
2291 blk_queue_end_tag(req->q, req);
2292
James Bottomleyba396a62009-05-27 14:17:08 +02002293 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002295 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002296 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
Mike Andersone78042e2008-10-30 02:16:20 -07002298 blk_delete_timer(req);
2299
James Bottomley28018c22010-07-01 19:49:17 +09002300 if (req->cmd_flags & REQ_DONTPREP)
2301 blk_unprep_request(req);
2302
2303
Jens Axboebc58ba92009-01-23 10:54:44 +01002304 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002305
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002307 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002308 else {
2309 if (blk_bidi_rq(req))
2310 __blk_put_request(req->next_rq->q, req->next_rq);
2311
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 }
2314}
2315
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002316/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002317 * blk_end_bidi_request - Complete a bidi request
2318 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002319 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002320 * @nr_bytes: number of bytes to complete @rq
2321 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002322 *
2323 * Description:
2324 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002325 * Drivers that supports bidi can safely call this member for any
2326 * type of request, bidi or uni. In the later case @bidi_bytes is
2327 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002328 *
2329 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002330 * %false - we are done with this request
2331 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002332 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002333static bool blk_end_bidi_request(struct request *rq, int error,
2334 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002335{
2336 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002337 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002338
Tejun Heo2e60e022009-04-23 11:05:18 +09002339 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2340 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002341
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002342 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002343 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002344 spin_unlock_irqrestore(q->queue_lock, flags);
2345
Tejun Heo2e60e022009-04-23 11:05:18 +09002346 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002347}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002348
2349/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002350 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2351 * @rq: the request to complete
2352 * @error: %0 for success, < %0 for error
2353 * @nr_bytes: number of bytes to complete @rq
2354 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002355 *
2356 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002357 * Identical to blk_end_bidi_request() except that queue lock is
2358 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002359 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002360 * Return:
2361 * %false - we are done with this request
2362 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002363 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002364bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002365 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002366{
Tejun Heo2e60e022009-04-23 11:05:18 +09002367 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2368 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002369
Tejun Heo2e60e022009-04-23 11:05:18 +09002370 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002371
Tejun Heo2e60e022009-04-23 11:05:18 +09002372 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002373}
2374
2375/**
2376 * blk_end_request - Helper function for drivers to complete the request.
2377 * @rq: the request being processed
2378 * @error: %0 for success, < %0 for error
2379 * @nr_bytes: number of bytes to complete
2380 *
2381 * Description:
2382 * Ends I/O on a number of bytes attached to @rq.
2383 * If @rq has leftover, sets it up for the next range of segments.
2384 *
2385 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002386 * %false - we are done with this request
2387 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002388 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002389bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002390{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002391 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002392}
Jens Axboe56ad1742009-07-28 22:11:24 +02002393EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002394
2395/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002396 * blk_end_request_all - Helper function for drives to finish the request.
2397 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002398 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002399 *
2400 * Description:
2401 * Completely finish @rq.
2402 */
2403void blk_end_request_all(struct request *rq, int error)
2404{
2405 bool pending;
2406 unsigned int bidi_bytes = 0;
2407
2408 if (unlikely(blk_bidi_rq(rq)))
2409 bidi_bytes = blk_rq_bytes(rq->next_rq);
2410
2411 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2412 BUG_ON(pending);
2413}
Jens Axboe56ad1742009-07-28 22:11:24 +02002414EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002415
2416/**
2417 * blk_end_request_cur - Helper function to finish the current request chunk.
2418 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002419 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002420 *
2421 * Description:
2422 * Complete the current consecutively mapped chunk from @rq.
2423 *
2424 * Return:
2425 * %false - we are done with this request
2426 * %true - still buffers pending for this request
2427 */
2428bool blk_end_request_cur(struct request *rq, int error)
2429{
2430 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2431}
Jens Axboe56ad1742009-07-28 22:11:24 +02002432EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002433
2434/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002435 * blk_end_request_err - Finish a request till the next failure boundary.
2436 * @rq: the request to finish till the next failure boundary for
2437 * @error: must be negative errno
2438 *
2439 * Description:
2440 * Complete @rq till the next failure boundary.
2441 *
2442 * Return:
2443 * %false - we are done with this request
2444 * %true - still buffers pending for this request
2445 */
2446bool blk_end_request_err(struct request *rq, int error)
2447{
2448 WARN_ON(error >= 0);
2449 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2450}
2451EXPORT_SYMBOL_GPL(blk_end_request_err);
2452
2453/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002454 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002455 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002456 * @error: %0 for success, < %0 for error
2457 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002458 *
2459 * Description:
2460 * Must be called with queue lock held unlike blk_end_request().
2461 *
2462 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002463 * %false - we are done with this request
2464 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002465 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002466bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002467{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002468 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002469}
Jens Axboe56ad1742009-07-28 22:11:24 +02002470EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002471
2472/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002473 * __blk_end_request_all - Helper function for drives to finish the request.
2474 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002475 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002476 *
2477 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002478 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002479 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002480void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002481{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002482 bool pending;
2483 unsigned int bidi_bytes = 0;
2484
2485 if (unlikely(blk_bidi_rq(rq)))
2486 bidi_bytes = blk_rq_bytes(rq->next_rq);
2487
2488 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2489 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002490}
Jens Axboe56ad1742009-07-28 22:11:24 +02002491EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002492
2493/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002494 * __blk_end_request_cur - Helper function to finish the current request chunk.
2495 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002496 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002497 *
2498 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002499 * Complete the current consecutively mapped chunk from @rq. Must
2500 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002501 *
2502 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002503 * %false - we are done with this request
2504 * %true - still buffers pending for this request
2505 */
2506bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002507{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002508 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002509}
Jens Axboe56ad1742009-07-28 22:11:24 +02002510EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002511
Tejun Heo80a761f2009-07-03 17:48:17 +09002512/**
2513 * __blk_end_request_err - Finish a request till the next failure boundary.
2514 * @rq: the request to finish till the next failure boundary for
2515 * @error: must be negative errno
2516 *
2517 * Description:
2518 * Complete @rq till the next failure boundary. Must be called
2519 * with queue lock held.
2520 *
2521 * Return:
2522 * %false - we are done with this request
2523 * %true - still buffers pending for this request
2524 */
2525bool __blk_end_request_err(struct request *rq, int error)
2526{
2527 WARN_ON(error >= 0);
2528 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2529}
2530EXPORT_SYMBOL_GPL(__blk_end_request_err);
2531
Jens Axboe86db1e22008-01-29 14:53:40 +01002532void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2533 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002535 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002536 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537
David Woodhousefb2dce82008-08-05 18:01:53 +01002538 if (bio_has_data(bio)) {
2539 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002540 rq->buffer = bio_data(bio);
2541 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002542 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
NeilBrown66846572007-08-16 13:31:28 +02002545 if (bio->bi_bdev)
2546 rq->rq_disk = bio->bi_bdev->bd_disk;
2547}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002549#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2550/**
2551 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2552 * @rq: the request to be flushed
2553 *
2554 * Description:
2555 * Flush all pages in @rq.
2556 */
2557void rq_flush_dcache_pages(struct request *rq)
2558{
2559 struct req_iterator iter;
2560 struct bio_vec *bvec;
2561
2562 rq_for_each_segment(bvec, rq, iter)
2563 flush_dcache_page(bvec->bv_page);
2564}
2565EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2566#endif
2567
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002568/**
2569 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2570 * @q : the queue of the device being checked
2571 *
2572 * Description:
2573 * Check if underlying low-level drivers of a device are busy.
2574 * If the drivers want to export their busy state, they must set own
2575 * exporting function using blk_queue_lld_busy() first.
2576 *
2577 * Basically, this function is used only by request stacking drivers
2578 * to stop dispatching requests to underlying devices when underlying
2579 * devices are busy. This behavior helps more I/O merging on the queue
2580 * of the request stacking driver and prevents I/O throughput regression
2581 * on burst I/O load.
2582 *
2583 * Return:
2584 * 0 - Not busy (The request stacking driver should dispatch request)
2585 * 1 - Busy (The request stacking driver should stop dispatching request)
2586 */
2587int blk_lld_busy(struct request_queue *q)
2588{
2589 if (q->lld_busy_fn)
2590 return q->lld_busy_fn(q);
2591
2592 return 0;
2593}
2594EXPORT_SYMBOL_GPL(blk_lld_busy);
2595
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002596/**
2597 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2598 * @rq: the clone request to be cleaned up
2599 *
2600 * Description:
2601 * Free all bios in @rq for a cloned request.
2602 */
2603void blk_rq_unprep_clone(struct request *rq)
2604{
2605 struct bio *bio;
2606
2607 while ((bio = rq->bio) != NULL) {
2608 rq->bio = bio->bi_next;
2609
2610 bio_put(bio);
2611 }
2612}
2613EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2614
2615/*
2616 * Copy attributes of the original request to the clone request.
2617 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2618 */
2619static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2620{
2621 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002622 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002623 dst->cmd_type = src->cmd_type;
2624 dst->__sector = blk_rq_pos(src);
2625 dst->__data_len = blk_rq_bytes(src);
2626 dst->nr_phys_segments = src->nr_phys_segments;
2627 dst->ioprio = src->ioprio;
2628 dst->extra_len = src->extra_len;
2629}
2630
2631/**
2632 * blk_rq_prep_clone - Helper function to setup clone request
2633 * @rq: the request to be setup
2634 * @rq_src: original request to be cloned
2635 * @bs: bio_set that bios for clone are allocated from
2636 * @gfp_mask: memory allocation mask for bio
2637 * @bio_ctr: setup function to be called for each clone bio.
2638 * Returns %0 for success, non %0 for failure.
2639 * @data: private data to be passed to @bio_ctr
2640 *
2641 * Description:
2642 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2643 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2644 * are not copied, and copying such parts is the caller's responsibility.
2645 * Also, pages which the original bios are pointing to are not copied
2646 * and the cloned bios just point same pages.
2647 * So cloned bios must be completed before original bios, which means
2648 * the caller must complete @rq before @rq_src.
2649 */
2650int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2651 struct bio_set *bs, gfp_t gfp_mask,
2652 int (*bio_ctr)(struct bio *, struct bio *, void *),
2653 void *data)
2654{
2655 struct bio *bio, *bio_src;
2656
2657 if (!bs)
2658 bs = fs_bio_set;
2659
2660 blk_rq_init(NULL, rq);
2661
2662 __rq_for_each_bio(bio_src, rq_src) {
2663 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2664 if (!bio)
2665 goto free_and_out;
2666
2667 __bio_clone(bio, bio_src);
2668
2669 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002670 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002671 goto free_and_out;
2672
2673 if (bio_ctr && bio_ctr(bio, bio_src, data))
2674 goto free_and_out;
2675
2676 if (rq->bio) {
2677 rq->biotail->bi_next = bio;
2678 rq->biotail = bio;
2679 } else
2680 rq->bio = rq->biotail = bio;
2681 }
2682
2683 __blk_rq_prep_clone(rq, rq_src);
2684
2685 return 0;
2686
2687free_and_out:
2688 if (bio)
2689 bio_free(bio, bs);
2690 blk_rq_unprep_clone(rq);
2691
2692 return -ENOMEM;
2693}
2694EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2695
Jens Axboe18887ad2008-07-28 13:08:45 +02002696int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697{
2698 return queue_work(kblockd_workqueue, work);
2699}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700EXPORT_SYMBOL(kblockd_schedule_work);
2701
Vivek Goyale43473b2010-09-15 17:06:35 -04002702int kblockd_schedule_delayed_work(struct request_queue *q,
2703 struct delayed_work *dwork, unsigned long delay)
2704{
2705 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2706}
2707EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2708
Jens Axboe73c10102011-03-08 13:19:51 +01002709#define PLUG_MAGIC 0x91827364
2710
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002711/**
2712 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2713 * @plug: The &struct blk_plug that needs to be initialized
2714 *
2715 * Description:
2716 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2717 * pending I/O should the task end up blocking between blk_start_plug() and
2718 * blk_finish_plug(). This is important from a performance perspective, but
2719 * also ensures that we don't deadlock. For instance, if the task is blocking
2720 * for a memory allocation, memory reclaim could end up wanting to free a
2721 * page belonging to that request that is currently residing in our private
2722 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2723 * this kind of deadlock.
2724 */
Jens Axboe73c10102011-03-08 13:19:51 +01002725void blk_start_plug(struct blk_plug *plug)
2726{
2727 struct task_struct *tsk = current;
2728
2729 plug->magic = PLUG_MAGIC;
2730 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002731 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002732 plug->should_sort = 0;
2733
2734 /*
2735 * If this is a nested plug, don't actually assign it. It will be
2736 * flushed on its own.
2737 */
2738 if (!tsk->plug) {
2739 /*
2740 * Store ordering should not be needed here, since a potential
2741 * preempt will imply a full memory barrier
2742 */
2743 tsk->plug = plug;
2744 }
2745}
2746EXPORT_SYMBOL(blk_start_plug);
2747
2748static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2749{
2750 struct request *rqa = container_of(a, struct request, queuelist);
2751 struct request *rqb = container_of(b, struct request, queuelist);
2752
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002753 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002754}
2755
Jens Axboe49cac012011-04-16 13:51:05 +02002756/*
2757 * If 'from_schedule' is true, then postpone the dispatch of requests
2758 * until a safe kblockd context. We due this to avoid accidental big
2759 * additional stack usage in driver dispatch, in places where the originally
2760 * plugger did not intend it.
2761 */
Jens Axboef6603782011-04-15 15:49:07 +02002762static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002763 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002764 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002765{
Jens Axboe49cac012011-04-16 13:51:05 +02002766 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002767
2768 /*
Tejun Heo8ba61432011-12-14 00:33:37 +01002769 * Don't mess with dead queue.
2770 */
2771 if (unlikely(blk_queue_dead(q))) {
2772 spin_unlock(q->queue_lock);
2773 return;
2774 }
2775
2776 /*
Jens Axboe99e22592011-04-18 09:59:55 +02002777 * If we are punting this to kblockd, then we can safely drop
2778 * the queue_lock before waking kblockd (which needs to take
2779 * this lock).
2780 */
2781 if (from_schedule) {
2782 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002783 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002784 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002785 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002786 spin_unlock(q->queue_lock);
2787 }
2788
Jens Axboe94b5eb22011-04-12 10:12:19 +02002789}
2790
NeilBrown048c9372011-04-18 09:52:22 +02002791static void flush_plug_callbacks(struct blk_plug *plug)
2792{
2793 LIST_HEAD(callbacks);
2794
2795 if (list_empty(&plug->cb_list))
2796 return;
2797
2798 list_splice_init(&plug->cb_list, &callbacks);
2799
2800 while (!list_empty(&callbacks)) {
2801 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2802 struct blk_plug_cb,
2803 list);
2804 list_del(&cb->list);
2805 cb->callback(cb);
2806 }
2807}
2808
Jens Axboe49cac012011-04-16 13:51:05 +02002809void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002810{
2811 struct request_queue *q;
2812 unsigned long flags;
2813 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002814 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002815 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002816
2817 BUG_ON(plug->magic != PLUG_MAGIC);
2818
NeilBrown048c9372011-04-18 09:52:22 +02002819 flush_plug_callbacks(plug);
Jens Axboe73c10102011-03-08 13:19:51 +01002820 if (list_empty(&plug->list))
2821 return;
2822
NeilBrown109b8122011-04-11 14:13:10 +02002823 list_splice_init(&plug->list, &list);
2824
2825 if (plug->should_sort) {
2826 list_sort(NULL, &list, plug_rq_cmp);
2827 plug->should_sort = 0;
2828 }
Jens Axboe73c10102011-03-08 13:19:51 +01002829
2830 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002831 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002832
2833 /*
2834 * Save and disable interrupts here, to avoid doing it for every
2835 * queue lock we have to take.
2836 */
Jens Axboe73c10102011-03-08 13:19:51 +01002837 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002838 while (!list_empty(&list)) {
2839 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002840 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002841 BUG_ON(!rq->q);
2842 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002843 /*
2844 * This drops the queue lock
2845 */
2846 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002847 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002848 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002849 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002850 spin_lock(q->queue_lock);
2851 }
Tejun Heo8ba61432011-12-14 00:33:37 +01002852
2853 /*
2854 * Short-circuit if @q is dead
2855 */
2856 if (unlikely(blk_queue_dead(q))) {
2857 __blk_end_request_all(rq, -ENODEV);
2858 continue;
2859 }
2860
Jens Axboe73c10102011-03-08 13:19:51 +01002861 /*
2862 * rq is already accounted, so use raw insert
2863 */
Jens Axboe401a18e2011-03-25 16:57:52 +01002864 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2865 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2866 else
2867 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002868
2869 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01002870 }
2871
Jens Axboe99e22592011-04-18 09:59:55 +02002872 /*
2873 * This drops the queue lock
2874 */
2875 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002876 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002877
Jens Axboe73c10102011-03-08 13:19:51 +01002878 local_irq_restore(flags);
2879}
Jens Axboe73c10102011-03-08 13:19:51 +01002880
2881void blk_finish_plug(struct blk_plug *plug)
2882{
Jens Axboef6603782011-04-15 15:49:07 +02002883 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02002884
2885 if (plug == current->plug)
2886 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01002887}
2888EXPORT_SYMBOL(blk_finish_plug);
2889
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890int __init blk_dev_init(void)
2891{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002892 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2893 sizeof(((struct request *)0)->cmd_flags));
2894
Tejun Heo89b90be2011-01-03 15:01:47 +01002895 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2896 kblockd_workqueue = alloc_workqueue("kblockd",
2897 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 if (!kblockd_workqueue)
2899 panic("Failed to create kblockd\n");
2900
2901 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002902 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903
Jens Axboe8324aa92008-01-29 14:51:59 +01002904 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002905 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 return 0;
2908}