blob: 2f4002f79a24b3cf242c870282d96859dc475dc9 [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 Mortonfaccbd42006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080029#include <linux/fault-inject.h>
Li Zefan55782132009-06-09 13:43:05 +080030
31#define CREATE_TRACE_POINTS
32#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Jens Axboe8324aa92008-01-29 14:51:59 +010034#include "blk.h"
35
Mike Snitzerd07335e2010-11-16 12:52:38 +010036EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020037EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Li Zefan55782132009-06-09 13:43:05 +080038EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010039
Jens Axboe165125e2007-07-24 09:28:11 +020040static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42/*
43 * For the allocated request tables
44 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010045static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47/*
48 * For queue allocation
49 */
Jens Axboe6728cb02008-01-31 13:03:55 +010050struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 * Controlling structure to kblockd
54 */
Jens Axboeff856ba2006-01-09 16:02:34 +010055static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Jens Axboe26b82562008-01-29 13:54:41 +010057static void drive_stat_acct(struct request *rq, int new_io)
58{
Jens Axboe28f13702008-05-07 10:15:46 +020059 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010060 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090061 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010062
Jens Axboec2553b52009-04-24 08:10:11 +020063 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010064 return;
65
Tejun Heo074a7ac2008-08-25 19:56:14 +090066 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +090067
Jerome Marchand09e099d2011-01-05 16:57:38 +010068 if (!new_io) {
69 part = rq->part;
Tejun Heo074a7ac2008-08-25 19:56:14 +090070 part_stat_inc(cpu, part, merges[rw]);
Jerome Marchand09e099d2011-01-05 16:57:38 +010071 } else {
72 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Jens Axboe6c23a962011-01-07 08:43:37 +010073 if (!hd_struct_try_get(part)) {
Jerome Marchand09e099d2011-01-05 16:57:38 +010074 /*
75 * The partition is already being removed,
76 * the request will be accounted on the disk only
77 *
78 * We take a reference on disk->part0 although that
79 * partition will never be deleted, so we can treat
80 * it as any other partition.
81 */
82 part = &rq->rq_disk->part0;
Jens Axboe6c23a962011-01-07 08:43:37 +010083 hd_struct_get(part);
Jerome Marchand09e099d2011-01-05 16:57:38 +010084 }
Tejun Heo074a7ac2008-08-25 19:56:14 +090085 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020086 part_inc_in_flight(part, rw);
Jerome Marchand09e099d2011-01-05 16:57:38 +010087 rq->part = part;
Jens Axboe26b82562008-01-29 13:54:41 +010088 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020089
Tejun Heo074a7ac2008-08-25 19:56:14 +090090 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010091}
92
Jens Axboe8324aa92008-01-29 14:51:59 +010093void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094{
95 int nr;
96
97 nr = q->nr_requests - (q->nr_requests / 8) + 1;
98 if (nr > q->nr_requests)
99 nr = q->nr_requests;
100 q->nr_congestion_on = nr;
101
102 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
103 if (nr < 1)
104 nr = 1;
105 q->nr_congestion_off = nr;
106}
107
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108/**
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
110 * @bdev: device
111 *
112 * Locates the passed device's request queue and returns the address of its
113 * backing_dev_info
114 *
115 * Will return NULL if the request queue cannot be located.
116 */
117struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
118{
119 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200120 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
122 if (q)
123 ret = &q->backing_dev_info;
124 return ret;
125}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126EXPORT_SYMBOL(blk_get_backing_dev_info);
127
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200128void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200130 memset(rq, 0, sizeof(*rq));
131
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700133 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200134 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100135 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900136 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200137 INIT_HLIST_NODE(&rq->hash);
138 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200139 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800140 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100141 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100142 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900143 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700144 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100145 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200147EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
NeilBrown5bb23a62007-09-27 12:46:13 +0200149static void req_bio_endio(struct request *rq, struct bio *bio,
150 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100151{
Jens Axboe165125e2007-07-24 09:28:11 +0200152 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100153
Tejun Heodd4c1332010-09-03 11:56:16 +0200154 if (&q->flush_rq != rq) {
NeilBrown5bb23a62007-09-27 12:46:13 +0200155 if (error)
156 clear_bit(BIO_UPTODATE, &bio->bi_flags);
157 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
158 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100159
NeilBrown5bb23a62007-09-27 12:46:13 +0200160 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100161 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700162 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200163 nbytes = bio->bi_size;
164 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100165
Keith Mannthey08bafc02008-11-25 10:24:35 +0100166 if (unlikely(rq->cmd_flags & REQ_QUIET))
167 set_bit(BIO_QUIET, &bio->bi_flags);
168
NeilBrown5bb23a62007-09-27 12:46:13 +0200169 bio->bi_size -= nbytes;
170 bio->bi_sector += (nbytes >> 9);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200171
172 if (bio_integrity(bio))
173 bio_integrity_advance(bio, nbytes);
174
NeilBrown5bb23a62007-09-27 12:46:13 +0200175 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200176 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200177 } else {
NeilBrown5bb23a62007-09-27 12:46:13 +0200178 /*
Tejun Heodd4c1332010-09-03 11:56:16 +0200179 * Okay, this is the sequenced flush request in
180 * progress, just record the error;
NeilBrown5bb23a62007-09-27 12:46:13 +0200181 */
Tejun Heodd4c1332010-09-03 11:56:16 +0200182 if (error && !q->flush_err)
183 q->flush_err = error;
NeilBrown5bb23a62007-09-27 12:46:13 +0200184 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100185}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187void blk_dump_rq_flags(struct request *rq, char *msg)
188{
189 int bit;
190
Jens Axboe6728cb02008-01-31 13:03:55 +0100191 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200192 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
193 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Tejun Heo83096eb2009-05-07 22:24:39 +0900195 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
196 (unsigned long long)blk_rq_pos(rq),
197 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900198 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900199 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200201 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100202 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200203 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 printk("%02x ", rq->cmd[bit]);
205 printk("\n");
206 }
207}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208EXPORT_SYMBOL(blk_dump_rq_flags);
209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210/*
211 * "plug" the device if there are no outstanding requests: this will
212 * force the transfer to start only after we have put all the requests
213 * on the list.
214 *
215 * This is called with interrupts off and no requests on the queue and
216 * with the queue lock held.
217 */
Jens Axboe165125e2007-07-24 09:28:11 +0200218void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
220 WARN_ON(!irqs_disabled());
221
222 /*
223 * don't plug a stopped queue, it must be paired with blk_start_queue()
224 * which will restart the queueing
225 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200226 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 return;
228
Jens Axboee48ec692008-07-03 13:18:54 +0200229 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100231 trace_block_plug(q);
Jens Axboe2056a782006-03-23 20:00:26 +0100232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234EXPORT_SYMBOL(blk_plug_device);
235
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200236/**
237 * blk_plug_device_unlocked - plug a device without queue lock held
238 * @q: The &struct request_queue to plug
239 *
240 * Description:
241 * Like @blk_plug_device(), but grabs the queue lock and disables
242 * interrupts.
243 **/
244void blk_plug_device_unlocked(struct request_queue *q)
245{
246 unsigned long flags;
247
248 spin_lock_irqsave(q->queue_lock, flags);
249 blk_plug_device(q);
250 spin_unlock_irqrestore(q->queue_lock, flags);
251}
252EXPORT_SYMBOL(blk_plug_device_unlocked);
253
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254/*
255 * remove the queue from the plugged list, if present. called with
256 * queue lock held and interrupts disabled.
257 */
Jens Axboe165125e2007-07-24 09:28:11 +0200258int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259{
260 WARN_ON(!irqs_disabled());
261
Jens Axboee48ec692008-07-03 13:18:54 +0200262 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 return 0;
264
265 del_timer(&q->unplug_timer);
266 return 1;
267}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268EXPORT_SYMBOL(blk_remove_plug);
269
270/*
271 * remove the plug and let it rip..
272 */
Jens Axboe165125e2007-07-24 09:28:11 +0200273void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200275 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 return;
Jens Axboea31a9732008-10-17 13:58:29 +0200277 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 return;
279
Jens Axboe22e2c502005-06-27 10:55:12 +0200280 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
283/**
284 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200285 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 *
287 * Description:
288 * Linux uses plugging to build bigger requests queues before letting
289 * the device have at them. If a queue is plugged, the I/O scheduler
290 * is still adding and merging requests on the queue. Once the queue
291 * gets unplugged, the request_fn defined for the queue is invoked and
292 * transfers started.
293 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200294void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200296 if (blk_queue_plugged(q)) {
297 spin_lock_irq(q->queue_lock);
298 __generic_unplug_device(q);
299 spin_unlock_irq(q->queue_lock);
300 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302EXPORT_SYMBOL(generic_unplug_device);
303
304static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
305 struct page *page)
306{
Jens Axboe165125e2007-07-24 09:28:11 +0200307 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500309 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310}
311
Jens Axboe86db1e22008-01-29 14:53:40 +0100312void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
Jens Axboe165125e2007-07-24 09:28:11 +0200314 struct request_queue *q =
315 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100317 trace_block_unplug_io(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 q->unplug_fn(q);
319}
320
Jens Axboe86db1e22008-01-29 14:53:40 +0100321void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322{
Jens Axboe165125e2007-07-24 09:28:11 +0200323 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100325 trace_block_unplug_timer(q);
Jens Axboe18887ad2008-07-28 13:08:45 +0200326 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327}
328
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500329void blk_unplug(struct request_queue *q)
330{
331 /*
332 * devices don't necessarily have an ->unplug_fn defined
333 */
334 if (q->unplug_fn) {
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100335 trace_block_unplug_io(q);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500336 q->unplug_fn(q);
337 }
338}
339EXPORT_SYMBOL(blk_unplug);
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341/**
342 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200343 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 *
345 * Description:
346 * blk_start_queue() will clear the stop flag on the queue, and call
347 * the request_fn for the queue if it was in a stopped state when
348 * entered. Also see blk_stop_queue(). Queue lock must be held.
349 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200350void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200352 WARN_ON(!irqs_disabled());
353
Nick Piggin75ad23b2008-04-29 14:48:33 +0200354 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Tejun Heoa538cd02009-04-23 11:05:17 +0900355 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357EXPORT_SYMBOL(blk_start_queue);
358
359/**
360 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200361 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 *
363 * Description:
364 * The Linux block layer assumes that a block driver will consume all
365 * entries on the request queue when the request_fn strategy is called.
366 * Often this will not happen, because of hardware limitations (queue
367 * depth settings). If a device driver gets a 'queue full' response,
368 * or if it simply chooses not to queue more I/O at one point, it can
369 * call this function to prevent the request_fn from being called until
370 * the driver has signalled it's ready to go again. This happens by calling
371 * blk_start_queue() to restart queue operations. Queue lock must be held.
372 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200373void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
375 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200376 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
378EXPORT_SYMBOL(blk_stop_queue);
379
380/**
381 * blk_sync_queue - cancel any pending callbacks on a queue
382 * @q: the queue
383 *
384 * Description:
385 * The block layer may perform asynchronous callback activity
386 * on a queue, such as calling the unplug function after a timeout.
387 * A block device may call blk_sync_queue to ensure that any
388 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200389 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 * that its ->make_request_fn will not re-add plugging prior to calling
391 * this function.
392 *
393 */
394void blk_sync_queue(struct request_queue *q)
395{
396 del_timer_sync(&q->unplug_timer);
Jens Axboe70ed28b2008-11-19 14:38:39 +0100397 del_timer_sync(&q->timeout);
Cheng Renquan64d01dc2008-12-03 12:41:39 +0100398 cancel_work_sync(&q->unplug_work);
Vivek Goyale43473b2010-09-15 17:06:35 -0400399 throtl_shutdown_timer_wq(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400}
401EXPORT_SYMBOL(blk_sync_queue);
402
403/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200404 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200406 *
407 * Description:
408 * See @blk_run_queue. This variant must be called with the queue lock
409 * held and interrupts disabled.
410 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200412void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200415
Tejun Heoa538cd02009-04-23 11:05:17 +0900416 if (unlikely(blk_queue_stopped(q)))
417 return;
418
419 if (elv_queue_empty(q))
420 return;
421
Jens Axboedac07ec2006-05-11 08:20:16 +0200422 /*
423 * Only recurse once to avoid overrunning the stack, let the unplug
424 * handling reinvoke the handler shortly if we already got there.
425 */
Tejun Heoa538cd02009-04-23 11:05:17 +0900426 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
427 q->request_fn(q);
428 queue_flag_clear(QUEUE_FLAG_REENTER, q);
429 } else {
430 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
431 kblockd_schedule_work(q, &q->unplug_work);
432 }
Nick Piggin75ad23b2008-04-29 14:48:33 +0200433}
434EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200435
Nick Piggin75ad23b2008-04-29 14:48:33 +0200436/**
437 * blk_run_queue - run a single device queue
438 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200439 *
440 * Description:
441 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900442 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200443 */
444void blk_run_queue(struct request_queue *q)
445{
446 unsigned long flags;
447
448 spin_lock_irqsave(q->queue_lock, flags);
449 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 spin_unlock_irqrestore(q->queue_lock, flags);
451}
452EXPORT_SYMBOL(blk_run_queue);
453
Jens Axboe165125e2007-07-24 09:28:11 +0200454void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500455{
456 kobject_put(&q->kobj);
457}
Al Viro483f4af2006-03-18 18:34:37 -0500458
Jens Axboe6728cb02008-01-31 13:03:55 +0100459void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500460{
Jens Axboee3335de2008-09-18 09:22:54 -0700461 /*
462 * We know we have process context here, so we can be a little
463 * cautious and ensure that pending block actions on this device
464 * are done before moving on. Going into this function, we should
465 * not have processes doing IO to this device.
466 */
467 blk_sync_queue(q);
468
Matthew Garrett31373d02010-04-06 14:25:14 +0200469 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500470 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200471 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500472 mutex_unlock(&q->sysfs_lock);
473
474 if (q->elevator)
475 elevator_exit(q->elevator);
476
477 blk_put_queue(q);
478}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479EXPORT_SYMBOL(blk_cleanup_queue);
480
Jens Axboe165125e2007-07-24 09:28:11 +0200481static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
483 struct request_list *rl = &q->rq;
484
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400485 if (unlikely(rl->rq_pool))
486 return 0;
487
Jens Axboe1faa16d2009-04-06 14:48:01 +0200488 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
489 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200490 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200491 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
492 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Christoph Lameter19460892005-06-23 00:08:19 -0700494 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
495 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
497 if (!rl->rq_pool)
498 return -ENOMEM;
499
500 return 0;
501}
502
Jens Axboe165125e2007-07-24 09:28:11 +0200503struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
Christoph Lameter19460892005-06-23 00:08:19 -0700505 return blk_alloc_queue_node(gfp_mask, -1);
506}
507EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
Jens Axboe165125e2007-07-24 09:28:11 +0200509struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700510{
Jens Axboe165125e2007-07-24 09:28:11 +0200511 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700512 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700513
Jens Axboe8324aa92008-01-29 14:51:59 +0100514 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700515 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 if (!q)
517 return NULL;
518
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700519 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
520 q->backing_dev_info.unplug_io_data = q;
Jens Axboe0989a022009-06-12 14:42:56 +0200521 q->backing_dev_info.ra_pages =
522 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
523 q->backing_dev_info.state = 0;
524 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200525 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200526
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700527 err = bdi_init(&q->backing_dev_info);
528 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100529 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700530 return NULL;
531 }
532
Vivek Goyale43473b2010-09-15 17:06:35 -0400533 if (blk_throtl_init(q)) {
534 kmem_cache_free(blk_requestq_cachep, q);
535 return NULL;
536 }
537
Matthew Garrett31373d02010-04-06 14:25:14 +0200538 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
539 laptop_mode_timer_fn, (unsigned long) q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 init_timer(&q->unplug_timer);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700541 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
542 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heodd4c1332010-09-03 11:56:16 +0200543 INIT_LIST_HEAD(&q->pending_flushes);
Peter Zijlstra713ada92008-10-16 13:44:57 +0200544 INIT_WORK(&q->unplug_work, blk_unplug_work);
Al Viro483f4af2006-03-18 18:34:37 -0500545
Jens Axboe8324aa92008-01-29 14:51:59 +0100546 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Al Viro483f4af2006-03-18 18:34:37 -0500548 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700549 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 return q;
552}
Christoph Lameter19460892005-06-23 00:08:19 -0700553EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555/**
556 * blk_init_queue - prepare a request queue for use with a block device
557 * @rfn: The function to be called to process requests that have been
558 * placed on the queue.
559 * @lock: Request queue spin lock
560 *
561 * Description:
562 * If a block device wishes to use the standard request handling procedures,
563 * which sorts requests and coalesces adjacent requests, then it must
564 * call blk_init_queue(). The function @rfn will be called when there
565 * are requests on the queue that need to be processed. If the device
566 * supports plugging, then @rfn may not be called immediately when requests
567 * are available on the queue, but may be called at some time later instead.
568 * Plugged queues are generally unplugged when a buffer belonging to one
569 * of the requests on the queue is needed, or due to memory pressure.
570 *
571 * @rfn is not required, or even expected, to remove all requests off the
572 * queue, but only as many as it can handle at a time. If it does leave
573 * requests on the queue, it is responsible for arranging that the requests
574 * get dealt with eventually.
575 *
576 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200577 * request queue; this lock will be taken also from interrupt context, so irq
578 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200580 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 * it didn't succeed.
582 *
583 * Note:
584 * blk_init_queue() must be paired with a blk_cleanup_queue() call
585 * when the block device is deactivated (such as at module unload).
586 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700587
Jens Axboe165125e2007-07-24 09:28:11 +0200588struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
Christoph Lameter19460892005-06-23 00:08:19 -0700590 return blk_init_queue_node(rfn, lock, -1);
591}
592EXPORT_SYMBOL(blk_init_queue);
593
Jens Axboe165125e2007-07-24 09:28:11 +0200594struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700595blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
596{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600597 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600599 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
600 if (!uninit_q)
601 return NULL;
602
603 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
604 if (!q)
605 blk_cleanup_queue(uninit_q);
606
607 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200608}
609EXPORT_SYMBOL(blk_init_queue_node);
610
611struct request_queue *
612blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
613 spinlock_t *lock)
614{
615 return blk_init_allocated_queue_node(q, rfn, lock, -1);
616}
617EXPORT_SYMBOL(blk_init_allocated_queue);
618
619struct request_queue *
620blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
621 spinlock_t *lock, int node_id)
622{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 if (!q)
624 return NULL;
625
Christoph Lameter19460892005-06-23 00:08:19 -0700626 q->node = node_id;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600627 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500628 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
630 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900632 q->unprep_rq_fn = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 q->unplug_fn = generic_unplug_device;
Jens Axboebc58ba92009-01-23 10:54:44 +0100634 q->queue_flags = QUEUE_FLAG_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 q->queue_lock = lock;
636
Jens Axboef3b144a2009-03-06 08:48:33 +0100637 /*
638 * This also sets hw/phys segments, boundary and size
639 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
Alan Stern44ec9542007-02-20 11:01:57 -0500642 q->sg_reserved_size = INT_MAX;
643
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 /*
645 * all done
646 */
647 if (!elevator_init(q, NULL)) {
648 blk_queue_congestion_threshold(q);
649 return q;
650 }
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 return NULL;
653}
Mike Snitzer01effb02010-05-11 08:57:42 +0200654EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Jens Axboe165125e2007-07-24 09:28:11 +0200656int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700658 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500659 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 return 0;
661 }
662
663 return 1;
664}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Jens Axboe165125e2007-07-24 09:28:11 +0200666static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200668 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200669 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 mempool_free(rq, q->rq.rq_pool);
671}
672
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200673static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200674blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
676 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
677
678 if (!rq)
679 return NULL;
680
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200681 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200682
Jerome Marchand42dad762009-04-22 14:01:49 +0200683 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Tejun Heocb98fc82005-10-28 08:29:39 +0200685 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200686 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200687 mempool_free(rq, q->rq.rq_pool);
688 return NULL;
689 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200690 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200691 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Tejun Heocb98fc82005-10-28 08:29:39 +0200693 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694}
695
696/*
697 * ioc_batching returns true if the ioc is a valid batching request and
698 * should be given priority access to a request.
699 */
Jens Axboe165125e2007-07-24 09:28:11 +0200700static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
702 if (!ioc)
703 return 0;
704
705 /*
706 * Make sure the process is able to allocate at least 1 request
707 * even if the batch times out, otherwise we could theoretically
708 * lose wakeups.
709 */
710 return ioc->nr_batch_requests == q->nr_batching ||
711 (ioc->nr_batch_requests > 0
712 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
713}
714
715/*
716 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
717 * will cause the process to be a "batcher" on all queues in the system. This
718 * is the behaviour we want though - once it gets a wakeup it should be given
719 * a nice run.
720 */
Jens Axboe165125e2007-07-24 09:28:11 +0200721static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722{
723 if (!ioc || ioc_batching(q, ioc))
724 return;
725
726 ioc->nr_batch_requests = q->nr_batching;
727 ioc->last_waited = jiffies;
728}
729
Jens Axboe1faa16d2009-04-06 14:48:01 +0200730static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
732 struct request_list *rl = &q->rq;
733
Jens Axboe1faa16d2009-04-06 14:48:01 +0200734 if (rl->count[sync] < queue_congestion_off_threshold(q))
735 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Jens Axboe1faa16d2009-04-06 14:48:01 +0200737 if (rl->count[sync] + 1 <= q->nr_requests) {
738 if (waitqueue_active(&rl->wait[sync]))
739 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Jens Axboe1faa16d2009-04-06 14:48:01 +0200741 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 }
743}
744
745/*
746 * A request has just been released. Account for it, update the full and
747 * congestion status, wake up any waiters. Called under q->queue_lock.
748 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200749static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
751 struct request_list *rl = &q->rq;
752
Jens Axboe1faa16d2009-04-06 14:48:01 +0200753 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200754 if (priv)
755 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
Jens Axboe1faa16d2009-04-06 14:48:01 +0200757 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Jens Axboe1faa16d2009-04-06 14:48:01 +0200759 if (unlikely(rl->starved[sync ^ 1]))
760 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763/*
Nick Piggind6344532005-06-28 20:45:14 -0700764 * Get a free request, queue_lock must be held.
765 * Returns NULL on failure, with queue_lock held.
766 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 */
Jens Axboe165125e2007-07-24 09:28:11 +0200768static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100769 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
771 struct request *rq = NULL;
772 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100773 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200774 const bool is_sync = rw_is_sync(rw_flags) != 0;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100775 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
Jens Axboe7749a8d2006-12-13 13:02:26 +0100777 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100778 if (may_queue == ELV_MQUEUE_NO)
779 goto rq_starved;
780
Jens Axboe1faa16d2009-04-06 14:48:01 +0200781 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
782 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200783 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100784 /*
785 * The queue will fill after this allocation, so set
786 * it as full, and mark this process as "batching".
787 * This process will be allowed to complete a batch of
788 * requests, others will be blocked.
789 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200790 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100791 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200792 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100793 } else {
794 if (may_queue != ELV_MQUEUE_MUST
795 && !ioc_batching(q, ioc)) {
796 /*
797 * The queue is full and the allocating
798 * process is not a "batcher", and not
799 * exempted by the IO scheduler
800 */
801 goto out;
802 }
803 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200805 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 }
807
Jens Axboe082cf692005-06-28 16:35:11 +0200808 /*
809 * Only allow batching queuers to allocate up to 50% over the defined
810 * limit of requests, otherwise we could have thousands of requests
811 * allocated with any setting of ->nr_requests
812 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200813 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200814 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100815
Jens Axboe1faa16d2009-04-06 14:48:01 +0200816 rl->count[is_sync]++;
817 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200818
Jens Axboe64521d12005-10-28 08:30:39 +0200819 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Jens Axboef253b862010-10-24 22:06:02 +0200820 if (priv)
Tejun Heocb98fc82005-10-28 08:29:39 +0200821 rl->elvpriv++;
822
Jens Axboef253b862010-10-24 22:06:02 +0200823 if (blk_queue_io_stat(q))
824 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 spin_unlock_irq(q->queue_lock);
826
Jens Axboe7749a8d2006-12-13 13:02:26 +0100827 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100828 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 /*
830 * Allocation failed presumably due to memory. Undo anything
831 * we might have messed up.
832 *
833 * Allocating task should really be put onto the front of the
834 * wait queue, but this is pretty rare.
835 */
836 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200837 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
839 /*
840 * in the very unlikely event that allocation failed and no
841 * requests for this direction was pending, mark us starved
842 * so that freeing of a request in the other direction will
843 * notice us. another possible fix would be to split the
844 * rq mempool into READ and WRITE
845 */
846rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200847 if (unlikely(rl->count[is_sync] == 0))
848 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 goto out;
851 }
852
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100853 /*
854 * ioc may be NULL here, and ioc_batching will be false. That's
855 * OK, if the queue is under the request limit then requests need
856 * not count toward the nr_batch_requests limit. There will always
857 * be some limit enforced by BLK_BATCH_TIME.
858 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 if (ioc_batching(q, ioc))
860 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100861
Jens Axboe1faa16d2009-04-06 14:48:01 +0200862 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 return rq;
865}
866
867/*
868 * No available requests for this queue, unplug the device and wait for some
869 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700870 *
871 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 */
Jens Axboe165125e2007-07-24 09:28:11 +0200873static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200874 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200876 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 struct request *rq;
878
Jens Axboe7749a8d2006-12-13 13:02:26 +0100879 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700880 while (!rq) {
881 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200882 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 struct request_list *rl = &q->rq;
884
Jens Axboe1faa16d2009-04-06 14:48:01 +0200885 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 TASK_UNINTERRUPTIBLE);
887
Jens Axboe1faa16d2009-04-06 14:48:01 +0200888 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200890 __generic_unplug_device(q);
891 spin_unlock_irq(q->queue_lock);
892 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200894 /*
895 * After sleeping, we become a "batching" process and
896 * will be able to allocate at least one request, and
897 * up to a big batch of them for a small period time.
898 * See ioc_batching, ioc_set_batching
899 */
900 ioc = current_io_context(GFP_NOIO, q->node);
901 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100902
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200903 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200904 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200905
906 rq = get_request(q, rw_flags, bio, GFP_NOIO);
907 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 return rq;
910}
911
Jens Axboe165125e2007-07-24 09:28:11 +0200912struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
914 struct request *rq;
915
916 BUG_ON(rw != READ && rw != WRITE);
917
Nick Piggind6344532005-06-28 20:45:14 -0700918 spin_lock_irq(q->queue_lock);
919 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200920 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700921 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200922 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700923 if (!rq)
924 spin_unlock_irq(q->queue_lock);
925 }
926 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
928 return rq;
929}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930EXPORT_SYMBOL(blk_get_request);
931
932/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300933 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700934 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300935 * @bio: The bio describing the memory mappings that will be submitted for IO.
936 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700937 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200938 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300939 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
940 * type commands. Where the struct request needs to be farther initialized by
941 * the caller. It is passed a &struct bio, which describes the memory info of
942 * the I/O transfer.
943 *
944 * The caller of blk_make_request must make sure that bi_io_vec
945 * are set to describe the memory buffers. That bio_data_dir() will return
946 * the needed direction of the request. (And all bio's in the passed bio-chain
947 * are properly set accordingly)
948 *
949 * If called under none-sleepable conditions, mapped bio buffers must not
950 * need bouncing, by calling the appropriate masked or flagged allocator,
951 * suitable for the target device. Otherwise the call to blk_queue_bounce will
952 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200953 *
954 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
955 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
956 * anything but the first bio in the chain. Otherwise you risk waiting for IO
957 * completion of a bio that hasn't been submitted yet, thus resulting in a
958 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
959 * of bio_alloc(), as that avoids the mempool deadlock.
960 * If possible a big IO should be split into smaller parts when allocation
961 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200962 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300963struct request *blk_make_request(struct request_queue *q, struct bio *bio,
964 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200965{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300966 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
967
968 if (unlikely(!rq))
969 return ERR_PTR(-ENOMEM);
970
971 for_each_bio(bio) {
972 struct bio *bounce_bio = bio;
973 int ret;
974
975 blk_queue_bounce(q, &bounce_bio);
976 ret = blk_rq_append_bio(q, rq, bounce_bio);
977 if (unlikely(ret)) {
978 blk_put_request(rq);
979 return ERR_PTR(ret);
980 }
981 }
982
983 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200984}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300985EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200986
987/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 * blk_requeue_request - put a request back on queue
989 * @q: request queue where request should be inserted
990 * @rq: request to be inserted
991 *
992 * Description:
993 * Drivers often keep queueing requests until the hardware cannot accept
994 * more, when that condition happens we need to put the request back
995 * on the queue. Must be called with queue lock held.
996 */
Jens Axboe165125e2007-07-24 09:28:11 +0200997void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700999 blk_delete_timer(rq);
1000 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001001 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001002
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 if (blk_rq_tagged(rq))
1004 blk_queue_end_tag(q, rq);
1005
James Bottomleyba396a62009-05-27 14:17:08 +02001006 BUG_ON(blk_queued_rq(rq));
1007
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 elv_requeue_request(q, rq);
1009}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010EXPORT_SYMBOL(blk_requeue_request);
1011
1012/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001013 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 * @q: request queue where request should be inserted
1015 * @rq: request to be inserted
1016 * @at_head: insert request at head or tail of queue
1017 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 *
1019 * Description:
1020 * Many block devices need to execute commands asynchronously, so they don't
1021 * block the whole kernel from preemption during request execution. This is
1022 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +02001023 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1024 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 *
1026 * We have the option of inserting the head or the tail of the queue.
1027 * Typically we use the tail for new ioctls and so forth. We use the head
1028 * of the queue for things like a QUEUE_FULL message from a device, or a
1029 * host that is unable to accept a particular command.
1030 */
Jens Axboe165125e2007-07-24 09:28:11 +02001031void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05001032 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
Tejun Heo 867d1192005-04-24 02:06:05 -05001034 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 unsigned long flags;
1036
1037 /*
1038 * tell I/O scheduler that this isn't a regular read/write (ie it
1039 * must not attempt merges on this) and that it acts as a soft
1040 * barrier
1041 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02001042 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 rq->special = data;
1045
1046 spin_lock_irqsave(q->queue_lock, flags);
1047
1048 /*
1049 * If command is tagged, release the tag
1050 */
Tejun Heo 867d1192005-04-24 02:06:05 -05001051 if (blk_rq_tagged(rq))
1052 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Jerome Marchandb238b3d2007-10-23 15:05:46 +02001054 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -05001055 __elv_add_request(q, rq, where, 0);
Tejun Heoa7f55792009-04-23 11:05:17 +09001056 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 spin_unlock_irqrestore(q->queue_lock, flags);
1058}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059EXPORT_SYMBOL(blk_insert_request);
1060
Tejun Heo074a7ac2008-08-25 19:56:14 +09001061static void part_round_stats_single(int cpu, struct hd_struct *part,
1062 unsigned long now)
1063{
1064 if (now == part->stamp)
1065 return;
1066
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001067 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001068 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001069 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001070 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1071 }
1072 part->stamp = now;
1073}
1074
1075/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001076 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1077 * @cpu: cpu number for stats access
1078 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 *
1080 * The average IO queue length and utilisation statistics are maintained
1081 * by observing the current state of the queue length and the amount of
1082 * time it has been in this state for.
1083 *
1084 * Normally, that accounting is done on IO completion, but that can result
1085 * in more than a second's worth of IO being accounted for within any one
1086 * second, leading to >100% utilisation. To deal with that, we call this
1087 * function to do a round-off before returning the results when reading
1088 * /proc/diskstats. This accounts immediately for all queue usage up to
1089 * the current jiffies and restarts the counters again.
1090 */
Tejun Heoc9959052008-08-25 19:47:21 +09001091void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001092{
1093 unsigned long now = jiffies;
1094
Tejun Heo074a7ac2008-08-25 19:56:14 +09001095 if (part->partno)
1096 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1097 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001098}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001099EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101/*
1102 * queue lock must be held
1103 */
Jens Axboe165125e2007-07-24 09:28:11 +02001104void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 if (unlikely(!q))
1107 return;
1108 if (unlikely(--req->ref_count))
1109 return;
1110
Tejun Heo8922e162005-10-20 16:23:44 +02001111 elv_completed_request(q, req);
1112
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001113 /* this is a bio leak */
1114 WARN_ON(req->bio != NULL);
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 /*
1117 * Request may not have originated from ll_rw_blk. if not,
1118 * it didn't come out of our reserved rq pools
1119 */
Jens Axboe49171e52006-08-10 08:59:11 +02001120 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001121 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001122 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001125 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001128 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 }
1130}
Mike Christie6e39b692005-11-11 05:30:24 -06001131EXPORT_SYMBOL_GPL(__blk_put_request);
1132
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133void blk_put_request(struct request *req)
1134{
Tejun Heo8922e162005-10-20 16:23:44 +02001135 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001136 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001138 spin_lock_irqsave(q->queue_lock, flags);
1139 __blk_put_request(q, req);
1140 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142EXPORT_SYMBOL(blk_put_request);
1143
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001144/**
1145 * blk_add_request_payload - add a payload to a request
1146 * @rq: request to update
1147 * @page: page backing the payload
1148 * @len: length of the payload.
1149 *
1150 * This allows to later add a payload to an already submitted request by
1151 * a block driver. The driver needs to take care of freeing the payload
1152 * itself.
1153 *
1154 * Note that this is a quite horrible hack and nothing but handling of
1155 * discard requests should ever use it.
1156 */
1157void blk_add_request_payload(struct request *rq, struct page *page,
1158 unsigned int len)
1159{
1160 struct bio *bio = rq->bio;
1161
1162 bio->bi_io_vec->bv_page = page;
1163 bio->bi_io_vec->bv_offset = 0;
1164 bio->bi_io_vec->bv_len = len;
1165
1166 bio->bi_size = len;
1167 bio->bi_vcnt = 1;
1168 bio->bi_phys_segments = 1;
1169
1170 rq->__data_len = rq->resid_len = len;
1171 rq->nr_phys_segments = 1;
1172 rq->buffer = bio_data(bio);
1173}
1174EXPORT_SYMBOL_GPL(blk_add_request_payload);
1175
Jens Axboe86db1e22008-01-29 14:53:40 +01001176void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001177{
Jens Axboec7c22e42008-09-13 20:26:01 +02001178 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001179 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001180
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001181 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1182 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001183 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001184
Tejun Heo52d9e672006-01-06 09:49:58 +01001185 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001186 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001187 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001188 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001189}
1190
Jens Axboe644b2d92009-04-06 14:48:06 +02001191/*
1192 * Only disabling plugging for non-rotational devices if it does tagging
1193 * as well, otherwise we do need the proper merging
1194 */
1195static inline bool queue_should_plug(struct request_queue *q)
1196{
Jens Axboe79da0642010-02-23 08:40:43 +01001197 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
Jens Axboe644b2d92009-04-06 14:48:06 +02001198}
1199
Jens Axboe165125e2007-07-24 09:28:11 +02001200static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201{
Nick Piggin450991b2005-06-28 20:45:13 -07001202 struct request *req;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001203 int el_ret;
1204 unsigned int bytes = bio->bi_size;
Jens Axboe51da90f2006-07-18 04:14:45 +02001205 const unsigned short prio = bio_prio(bio);
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001206 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1207 const bool unplug = !!(bio->bi_rw & REQ_UNPLUG);
1208 const unsigned long ff = bio->bi_rw & REQ_FAILFAST_MASK;
Tejun Heo28e7d182010-09-03 11:56:16 +02001209 int where = ELEVATOR_INSERT_SORT;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001210 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 /*
1213 * low level driver can indicate that it wants pages above a
1214 * certain limit bounced to low memory (ie for highmem, or even
1215 * ISA dma in theory)
1216 */
1217 blk_queue_bounce(q, &bio);
1218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 spin_lock_irq(q->queue_lock);
1220
Tejun Heo4fed9472010-09-03 11:56:17 +02001221 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Tejun Heo28e7d182010-09-03 11:56:16 +02001222 where = ELEVATOR_INSERT_FRONT;
1223 goto get_rq;
1224 }
1225
1226 if (elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 goto get_rq;
1228
1229 el_ret = elv_merge(q, &req, bio);
1230 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001231 case ELEVATOR_BACK_MERGE:
1232 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Jens Axboe6728cb02008-01-31 13:03:55 +01001234 if (!ll_back_merge_fn(q, req, bio))
1235 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001237 trace_block_bio_backmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001238
Tejun Heo80a761f2009-07-03 17:48:17 +09001239 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1240 blk_rq_set_mixed_merge(req);
1241
Jens Axboe6728cb02008-01-31 13:03:55 +01001242 req->biotail->bi_next = bio;
1243 req->biotail = bio;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001244 req->__data_len += bytes;
Jens Axboe6728cb02008-01-31 13:03:55 +01001245 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001246 if (!blk_rq_cpu_valid(req))
1247 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001248 drive_stat_acct(req, 0);
Divyesh Shah812d4022010-04-08 21:14:23 -07001249 elv_bio_merged(q, req, bio);
Jens Axboe6728cb02008-01-31 13:03:55 +01001250 if (!attempt_back_merge(q, req))
1251 elv_merged_request(q, req, el_ret);
1252 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Jens Axboe6728cb02008-01-31 13:03:55 +01001254 case ELEVATOR_FRONT_MERGE:
1255 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
Jens Axboe6728cb02008-01-31 13:03:55 +01001257 if (!ll_front_merge_fn(q, req, bio))
1258 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001260 trace_block_bio_frontmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001261
Tejun Heo80a761f2009-07-03 17:48:17 +09001262 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1263 blk_rq_set_mixed_merge(req);
1264 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1265 req->cmd_flags |= ff;
1266 }
1267
Jens Axboe6728cb02008-01-31 13:03:55 +01001268 bio->bi_next = req->bio;
1269 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Jens Axboe6728cb02008-01-31 13:03:55 +01001271 /*
1272 * may not be valid. if the low level driver said
1273 * it didn't need a bounce buffer then it better
1274 * not touch req->buffer either...
1275 */
1276 req->buffer = bio_data(bio);
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001277 req->__sector = bio->bi_sector;
1278 req->__data_len += bytes;
Jens Axboe6728cb02008-01-31 13:03:55 +01001279 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001280 if (!blk_rq_cpu_valid(req))
1281 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001282 drive_stat_acct(req, 0);
Divyesh Shah812d4022010-04-08 21:14:23 -07001283 elv_bio_merged(q, req, bio);
Jens Axboe6728cb02008-01-31 13:03:55 +01001284 if (!attempt_front_merge(q, req))
1285 elv_merged_request(q, req, el_ret);
1286 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
Jens Axboe6728cb02008-01-31 13:03:55 +01001288 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1289 default:
1290 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 }
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001294 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001295 * This sync check and mask will be re-done in init_request_from_bio(),
1296 * but we need to set it earlier to expose the sync flag to the
1297 * rq allocator and io schedulers.
1298 */
1299 rw_flags = bio_data_dir(bio);
1300 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001301 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001302
1303 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001304 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001305 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001306 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001307 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001308
Nick Piggin450991b2005-06-28 20:45:13 -07001309 /*
1310 * After dropping the lock and possibly sleeping here, our request
1311 * may now be mergeable after it had proven unmergeable (above).
1312 * We don't worry about that case for efficiency. It won't happen
1313 * often, and the elevators are able to handle it.
1314 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001315 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Nick Piggin450991b2005-06-28 20:45:13 -07001317 spin_lock_irq(q->queue_lock);
Jens Axboec7c22e42008-09-13 20:26:01 +02001318 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1319 bio_flagged(bio, BIO_CPU_AFFINE))
1320 req->cpu = blk_cpu_to_group(smp_processor_id());
Jens Axboe644b2d92009-04-06 14:48:06 +02001321 if (queue_should_plug(q) && elv_queue_empty(q))
Nick Piggin450991b2005-06-28 20:45:13 -07001322 blk_plug_device(q);
Tejun Heodd831002010-09-03 11:56:16 +02001323
1324 /* insert the request into the elevator */
1325 drive_stat_acct(req, 1);
Tejun Heo28e7d182010-09-03 11:56:16 +02001326 __elv_add_request(q, req, where, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327out:
Jens Axboe644b2d92009-04-06 14:48:06 +02001328 if (unplug || !queue_should_plug(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 __generic_unplug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 spin_unlock_irq(q->queue_lock);
1331 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332}
1333
1334/*
1335 * If bio->bi_dev is a partition, remap the location
1336 */
1337static inline void blk_partition_remap(struct bio *bio)
1338{
1339 struct block_device *bdev = bio->bi_bdev;
1340
Jens Axboebf2de6f2007-09-27 13:01:25 +02001341 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 bio->bi_sector += p->start_sect;
1345 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001346
Mike Snitzerd07335e2010-11-16 12:52:38 +01001347 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1348 bdev->bd_dev,
1349 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 }
1351}
1352
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353static void handle_bad_sector(struct bio *bio)
1354{
1355 char b[BDEVNAME_SIZE];
1356
1357 printk(KERN_INFO "attempt to access beyond end of device\n");
1358 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1359 bdevname(bio->bi_bdev, b),
1360 bio->bi_rw,
1361 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001362 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
1364 set_bit(BIO_EOF, &bio->bi_flags);
1365}
1366
Akinobu Mitac17bb492006-12-08 02:39:46 -08001367#ifdef CONFIG_FAIL_MAKE_REQUEST
1368
1369static DECLARE_FAULT_ATTR(fail_make_request);
1370
1371static int __init setup_fail_make_request(char *str)
1372{
1373 return setup_fault_attr(&fail_make_request, str);
1374}
1375__setup("fail_make_request=", setup_fail_make_request);
1376
1377static int should_fail_request(struct bio *bio)
1378{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001379 struct hd_struct *part = bio->bi_bdev->bd_part;
1380
1381 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001382 return should_fail(&fail_make_request, bio->bi_size);
1383
1384 return 0;
1385}
1386
1387static int __init fail_make_request_debugfs(void)
1388{
1389 return init_fault_attr_dentries(&fail_make_request,
1390 "fail_make_request");
1391}
1392
1393late_initcall(fail_make_request_debugfs);
1394
1395#else /* CONFIG_FAIL_MAKE_REQUEST */
1396
1397static inline int should_fail_request(struct bio *bio)
1398{
1399 return 0;
1400}
1401
1402#endif /* CONFIG_FAIL_MAKE_REQUEST */
1403
Jens Axboec07e2b42007-07-18 13:27:58 +02001404/*
1405 * Check whether this bio extends beyond the end of the device.
1406 */
1407static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1408{
1409 sector_t maxsector;
1410
1411 if (!nr_sectors)
1412 return 0;
1413
1414 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001415 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001416 if (maxsector) {
1417 sector_t sector = bio->bi_sector;
1418
1419 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1420 /*
1421 * This may well happen - the kernel calls bread()
1422 * without checking the size of the device, e.g., when
1423 * mounting a device.
1424 */
1425 handle_bad_sector(bio);
1426 return 1;
1427 }
1428 }
1429
1430 return 0;
1431}
1432
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001434 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 * @bio: The bio describing the location in memory and on the device.
1436 *
1437 * generic_make_request() is used to make I/O requests of block
1438 * devices. It is passed a &struct bio, which describes the I/O that needs
1439 * to be done.
1440 *
1441 * generic_make_request() does not return any status. The
1442 * success/failure status of the request, along with notification of
1443 * completion, is delivered asynchronously through the bio->bi_end_io
1444 * function described (one day) else where.
1445 *
1446 * The caller of generic_make_request must make sure that bi_io_vec
1447 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1448 * set to describe the device address, and the
1449 * bi_end_io and optionally bi_private are set to describe how
1450 * completion notification should be signaled.
1451 *
1452 * generic_make_request and the drivers it calls may use bi_next if this
1453 * bio happens to be merged with someone else, and may change bi_dev and
1454 * bi_sector for remaps as it sees fit. So the values of these fields
1455 * should NOT be depended on after the call to generic_make_request.
1456 */
Neil Brownd89d8792007-05-01 09:53:42 +02001457static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458{
Jens Axboe165125e2007-07-24 09:28:11 +02001459 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001460 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001462 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001463 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
1465 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
Jens Axboec07e2b42007-07-18 13:27:58 +02001467 if (bio_check_eod(bio, nr_sectors))
1468 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
1470 /*
1471 * Resolve the mapping until finished. (drivers are
1472 * still free to implement/resolve their own stacking
1473 * by explicitly returning 0)
1474 *
1475 * NOTE: we don't repeat the blk_size check for each new device.
1476 * Stacking drivers are expected to know what they are doing.
1477 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001478 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001479 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 do {
1481 char b[BDEVNAME_SIZE];
1482
1483 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001484 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 printk(KERN_ERR
1486 "generic_make_request: Trying to access "
1487 "nonexistent block-device %s (%Lu)\n",
1488 bdevname(bio->bi_bdev, b),
1489 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001490 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 }
1492
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001493 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
Christoph Hellwig67efc922009-09-30 13:54:20 +02001494 nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001495 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001496 bdevname(bio->bi_bdev, b),
1497 bio_sectors(bio),
1498 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 goto end_io;
1500 }
1501
Nick Pigginfde6ad22005-06-23 00:08:53 -07001502 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 goto end_io;
1504
Akinobu Mitac17bb492006-12-08 02:39:46 -08001505 if (should_fail_request(bio))
1506 goto end_io;
1507
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 /*
1509 * If this device has partitions, remap block n
1510 * of partition p to block n+start(p) of the disk.
1511 */
1512 blk_partition_remap(bio);
1513
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001514 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1515 goto end_io;
1516
NeilBrown5ddfe962006-10-30 22:07:21 -08001517 if (old_sector != -1)
Mike Snitzerd07335e2010-11-16 12:52:38 +01001518 trace_block_bio_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001519
NeilBrown5ddfe962006-10-30 22:07:21 -08001520 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001521 old_dev = bio->bi_bdev->bd_dev;
1522
Jens Axboec07e2b42007-07-18 13:27:58 +02001523 if (bio_check_eod(bio, nr_sectors))
1524 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001525
Tejun Heo1e879012010-09-03 11:56:17 +02001526 /*
1527 * Filter flush bio's early so that make_request based
1528 * drivers without flush support don't have to worry
1529 * about them.
1530 */
1531 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1532 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1533 if (!nr_sectors) {
1534 err = 0;
1535 goto end_io;
1536 }
1537 }
1538
Adrian Hunter8d57a982010-08-11 14:17:49 -07001539 if ((bio->bi_rw & REQ_DISCARD) &&
1540 (!blk_queue_discard(q) ||
1541 ((bio->bi_rw & REQ_SECURE) &&
1542 !blk_queue_secdiscard(q)))) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001543 err = -EOPNOTSUPP;
1544 goto end_io;
1545 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001546
Vivek Goyale43473b2010-09-15 17:06:35 -04001547 blk_throtl_bio(q, &bio);
1548
1549 /*
1550 * If bio = NULL, bio has been throttled and will be submitted
1551 * later.
1552 */
1553 if (!bio)
1554 break;
1555
Minchan Kim01edede2009-09-08 21:56:38 +02001556 trace_block_bio_queue(q, bio);
1557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 ret = q->make_request_fn(q, bio);
1559 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001560
1561 return;
1562
1563end_io:
1564 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565}
1566
Neil Brownd89d8792007-05-01 09:53:42 +02001567/*
1568 * We only want one ->make_request_fn to be active at a time,
1569 * else stack usage with stacked devices could be a problem.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001570 * So use current->bio_list to keep a list of requests
Neil Brownd89d8792007-05-01 09:53:42 +02001571 * submited by a make_request_fn function.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001572 * current->bio_list is also used as a flag to say if
Neil Brownd89d8792007-05-01 09:53:42 +02001573 * generic_make_request is currently active in this task or not.
1574 * If it is NULL, then no make_request is active. If it is non-NULL,
1575 * then a make_request is active, and new requests should be added
1576 * at the tail
1577 */
1578void generic_make_request(struct bio *bio)
1579{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001580 struct bio_list bio_list_on_stack;
1581
1582 if (current->bio_list) {
Neil Brownd89d8792007-05-01 09:53:42 +02001583 /* make_request is active */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001584 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001585 return;
1586 }
1587 /* following loop may be a bit non-obvious, and so deserves some
1588 * explanation.
1589 * Before entering the loop, bio->bi_next is NULL (as all callers
1590 * ensure that) so we have a list with a single bio.
1591 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001592 * we assign bio_list to a pointer to the bio_list_on_stack,
1593 * thus initialising the bio_list of new bios to be
Neil Brownd89d8792007-05-01 09:53:42 +02001594 * added. __generic_make_request may indeed add some more bios
1595 * through a recursive call to generic_make_request. If it
1596 * did, we find a non-NULL value in bio_list and re-enter the loop
1597 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001598 * of the top of the list (no pretending) and so remove it from
1599 * bio_list, and call into __generic_make_request again.
Neil Brownd89d8792007-05-01 09:53:42 +02001600 *
1601 * The loop was structured like this to make only one call to
1602 * __generic_make_request (which is important as it is large and
1603 * inlined) and to keep the structure simple.
1604 */
1605 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001606 bio_list_init(&bio_list_on_stack);
1607 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001608 do {
Neil Brownd89d8792007-05-01 09:53:42 +02001609 __generic_make_request(bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001610 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001611 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001612 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001613}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614EXPORT_SYMBOL(generic_make_request);
1615
1616/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001617 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1619 * @bio: The &struct bio which describes the I/O
1620 *
1621 * submit_bio() is very similar in purpose to generic_make_request(), and
1622 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001623 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 *
1625 */
1626void submit_bio(int rw, struct bio *bio)
1627{
1628 int count = bio_sectors(bio);
1629
Jens Axboe22e2c502005-06-27 10:55:12 +02001630 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
Jens Axboebf2de6f2007-09-27 13:01:25 +02001632 /*
1633 * If it's a regular read/write or a barrier with data attached,
1634 * go through the normal accounting stuff before submission.
1635 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001636 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001637 if (rw & WRITE) {
1638 count_vm_events(PGPGOUT, count);
1639 } else {
1640 task_io_account_read(bio->bi_size);
1641 count_vm_events(PGPGIN, count);
1642 }
1643
1644 if (unlikely(block_dump)) {
1645 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001646 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001647 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001648 (rw & WRITE) ? "WRITE" : "READ",
1649 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001650 bdevname(bio->bi_bdev, b),
1651 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001652 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 }
1654
1655 generic_make_request(bio);
1656}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657EXPORT_SYMBOL(submit_bio);
1658
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001659/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001660 * blk_rq_check_limits - Helper function to check a request for the queue limit
1661 * @q: the queue
1662 * @rq: the request being checked
1663 *
1664 * Description:
1665 * @rq may have been made based on weaker limitations of upper-level queues
1666 * in request stacking drivers, and it may violate the limitation of @q.
1667 * Since the block layer and the underlying device driver trust @rq
1668 * after it is inserted to @q, it should be checked against @q before
1669 * the insertion using this generic function.
1670 *
1671 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001672 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001673 * Request stacking drivers like request-based dm may change the queue
1674 * limits while requests are in the queue (e.g. dm's table swapping).
1675 * Such request stacking drivers should check those requests agaist
1676 * the new queue limits again when they dispatch those requests,
1677 * although such checkings are also done against the old queue limits
1678 * when submitting requests.
1679 */
1680int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1681{
ike Snitzer33839772010-08-08 12:11:33 -04001682 if (rq->cmd_flags & REQ_DISCARD)
1683 return 0;
1684
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001685 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1686 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001687 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1688 return -EIO;
1689 }
1690
1691 /*
1692 * queue's settings related to segment counting like q->bounce_pfn
1693 * may differ from that of other stacking queues.
1694 * Recalculate it to check the request correctly on this queue's
1695 * limitation.
1696 */
1697 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001698 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001699 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1700 return -EIO;
1701 }
1702
1703 return 0;
1704}
1705EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1706
1707/**
1708 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1709 * @q: the queue to submit the request
1710 * @rq: the request being queued
1711 */
1712int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1713{
1714 unsigned long flags;
1715
1716 if (blk_rq_check_limits(q, rq))
1717 return -EIO;
1718
1719#ifdef CONFIG_FAIL_MAKE_REQUEST
1720 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1721 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1722 return -EIO;
1723#endif
1724
1725 spin_lock_irqsave(q->queue_lock, flags);
1726
1727 /*
1728 * Submitting request must be dequeued before calling this function
1729 * because it will be linked to another request_queue
1730 */
1731 BUG_ON(blk_queued_rq(rq));
1732
1733 drive_stat_acct(rq, 1);
1734 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1735
1736 spin_unlock_irqrestore(q->queue_lock, flags);
1737
1738 return 0;
1739}
1740EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1741
Tejun Heo80a761f2009-07-03 17:48:17 +09001742/**
1743 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1744 * @rq: request to examine
1745 *
1746 * Description:
1747 * A request could be merge of IOs which require different failure
1748 * handling. This function determines the number of bytes which
1749 * can be failed from the beginning of the request without
1750 * crossing into area which need to be retried further.
1751 *
1752 * Return:
1753 * The number of bytes to fail.
1754 *
1755 * Context:
1756 * queue_lock must be held.
1757 */
1758unsigned int blk_rq_err_bytes(const struct request *rq)
1759{
1760 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1761 unsigned int bytes = 0;
1762 struct bio *bio;
1763
1764 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1765 return blk_rq_bytes(rq);
1766
1767 /*
1768 * Currently the only 'mixing' which can happen is between
1769 * different fastfail types. We can safely fail portions
1770 * which have all the failfast bits that the first one has -
1771 * the ones which are at least as eager to fail as the first
1772 * one.
1773 */
1774 for (bio = rq->bio; bio; bio = bio->bi_next) {
1775 if ((bio->bi_rw & ff) != ff)
1776 break;
1777 bytes += bio->bi_size;
1778 }
1779
1780 /* this could lead to infinite loop */
1781 BUG_ON(blk_rq_bytes(rq) && !bytes);
1782 return bytes;
1783}
1784EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1785
Jens Axboebc58ba92009-01-23 10:54:44 +01001786static void blk_account_io_completion(struct request *req, unsigned int bytes)
1787{
Jens Axboec2553b52009-04-24 08:10:11 +02001788 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001789 const int rw = rq_data_dir(req);
1790 struct hd_struct *part;
1791 int cpu;
1792
1793 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001794 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001795 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1796 part_stat_unlock();
1797 }
1798}
1799
1800static void blk_account_io_done(struct request *req)
1801{
Jens Axboebc58ba92009-01-23 10:54:44 +01001802 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001803 * Account IO completion. flush_rq isn't accounted as a
1804 * normal IO on queueing nor completion. Accounting the
1805 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001806 */
Tejun Heodd4c1332010-09-03 11:56:16 +02001807 if (blk_do_io_stat(req) && req != &req->q->flush_rq) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001808 unsigned long duration = jiffies - req->start_time;
1809 const int rw = rq_data_dir(req);
1810 struct hd_struct *part;
1811 int cpu;
1812
1813 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001814 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001815
1816 part_stat_inc(cpu, part, ios[rw]);
1817 part_stat_add(cpu, part, ticks[rw], duration);
1818 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001819 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001820
Jens Axboe6c23a962011-01-07 08:43:37 +01001821 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001822 part_stat_unlock();
1823 }
1824}
1825
Tejun Heo53a08802008-12-03 12:41:26 +01001826/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001827 * blk_peek_request - peek at the top of a request queue
1828 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001829 *
1830 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001831 * Return the request at the top of @q. The returned request
1832 * should be started using blk_start_request() before LLD starts
1833 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001834 *
1835 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001836 * Pointer to the request at the top of @q if available. Null
1837 * otherwise.
1838 *
1839 * Context:
1840 * queue_lock must be held.
1841 */
1842struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001843{
1844 struct request *rq;
1845 int ret;
1846
1847 while ((rq = __elv_next_request(q)) != NULL) {
1848 if (!(rq->cmd_flags & REQ_STARTED)) {
1849 /*
1850 * This is the first time the device driver
1851 * sees this request (possibly after
1852 * requeueing). Notify IO scheduler.
1853 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001854 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001855 elv_activate_rq(q, rq);
1856
1857 /*
1858 * just mark as started even if we don't start
1859 * it, a request that has been delayed should
1860 * not be passed by new incoming requests
1861 */
1862 rq->cmd_flags |= REQ_STARTED;
1863 trace_block_rq_issue(q, rq);
1864 }
1865
1866 if (!q->boundary_rq || q->boundary_rq == rq) {
1867 q->end_sector = rq_end_sector(rq);
1868 q->boundary_rq = NULL;
1869 }
1870
1871 if (rq->cmd_flags & REQ_DONTPREP)
1872 break;
1873
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001874 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001875 /*
1876 * make sure space for the drain appears we
1877 * know we can do this because max_hw_segments
1878 * has been adjusted to be one fewer than the
1879 * device can handle
1880 */
1881 rq->nr_phys_segments++;
1882 }
1883
1884 if (!q->prep_rq_fn)
1885 break;
1886
1887 ret = q->prep_rq_fn(q, rq);
1888 if (ret == BLKPREP_OK) {
1889 break;
1890 } else if (ret == BLKPREP_DEFER) {
1891 /*
1892 * the request may have been (partially) prepped.
1893 * we need to keep this request in the front to
1894 * avoid resource deadlock. REQ_STARTED will
1895 * prevent other fs requests from passing this one.
1896 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001897 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001898 !(rq->cmd_flags & REQ_DONTPREP)) {
1899 /*
1900 * remove the space for the drain we added
1901 * so that we don't add it again
1902 */
1903 --rq->nr_phys_segments;
1904 }
1905
1906 rq = NULL;
1907 break;
1908 } else if (ret == BLKPREP_KILL) {
1909 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001910 /*
1911 * Mark this request as started so we don't trigger
1912 * any debug logic in the end I/O path.
1913 */
1914 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001915 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001916 } else {
1917 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1918 break;
1919 }
1920 }
1921
1922 return rq;
1923}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001924EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001925
Tejun Heo9934c8c2009-05-08 11:54:16 +09001926void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001927{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001928 struct request_queue *q = rq->q;
1929
Tejun Heo158dbda2009-04-23 11:05:18 +09001930 BUG_ON(list_empty(&rq->queuelist));
1931 BUG_ON(ELV_ON_HASH(rq));
1932
1933 list_del_init(&rq->queuelist);
1934
1935 /*
1936 * the time frame between a request being removed from the lists
1937 * and to it is freed is accounted as io that is in progress at
1938 * the driver side.
1939 */
Divyesh Shah91952912010-04-01 15:01:41 -07001940 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001941 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001942 set_io_start_time_ns(rq);
1943 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001944}
1945
Tejun Heo5efccd12009-04-23 11:05:18 +09001946/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001947 * blk_start_request - start request processing on the driver
1948 * @req: request to dequeue
1949 *
1950 * Description:
1951 * Dequeue @req and start timeout timer on it. This hands off the
1952 * request to the driver.
1953 *
1954 * Block internal functions which don't want to start timer should
1955 * call blk_dequeue_request().
1956 *
1957 * Context:
1958 * queue_lock must be held.
1959 */
1960void blk_start_request(struct request *req)
1961{
1962 blk_dequeue_request(req);
1963
1964 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001965 * We are now handing the request to the hardware, initialize
1966 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001967 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001968 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001969 if (unlikely(blk_bidi_rq(req)))
1970 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1971
Tejun Heo9934c8c2009-05-08 11:54:16 +09001972 blk_add_timer(req);
1973}
1974EXPORT_SYMBOL(blk_start_request);
1975
1976/**
1977 * blk_fetch_request - fetch a request from a request queue
1978 * @q: request queue to fetch a request from
1979 *
1980 * Description:
1981 * Return the request at the top of @q. The request is started on
1982 * return and LLD can start processing it immediately.
1983 *
1984 * Return:
1985 * Pointer to the request at the top of @q if available. Null
1986 * otherwise.
1987 *
1988 * Context:
1989 * queue_lock must be held.
1990 */
1991struct request *blk_fetch_request(struct request_queue *q)
1992{
1993 struct request *rq;
1994
1995 rq = blk_peek_request(q);
1996 if (rq)
1997 blk_start_request(rq);
1998 return rq;
1999}
2000EXPORT_SYMBOL(blk_fetch_request);
2001
2002/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002003 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002004 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002005 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002006 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002007 *
2008 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002009 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2010 * the request structure even if @req doesn't have leftover.
2011 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002012 *
2013 * This special helper function is only for request stacking drivers
2014 * (e.g. request-based dm) so that they can handle partial completion.
2015 * Actual device drivers should use blk_end_request instead.
2016 *
2017 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2018 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002019 *
2020 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002021 * %false - this request doesn't have any more data
2022 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002023 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002024bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002026 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 struct bio *bio;
2028
Tejun Heo2e60e022009-04-23 11:05:18 +09002029 if (!req->bio)
2030 return false;
2031
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002032 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002033
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002035 * For fs requests, rq is just carrier of independent bio's
2036 * and each partial completion should be handled separately.
2037 * Reset per-request error on each partial completion.
2038 *
2039 * TODO: tj: This is too subtle. It would be better to let
2040 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002042 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 req->errors = 0;
2044
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002045 if (error && req->cmd_type == REQ_TYPE_FS &&
2046 !(req->cmd_flags & REQ_QUIET)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01002047 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 req->rq_disk ? req->rq_disk->disk_name : "?",
Tejun Heo83096eb2009-05-07 22:24:39 +09002049 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 }
2051
Jens Axboebc58ba92009-01-23 10:54:44 +01002052 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002053
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 total_bytes = bio_nbytes = 0;
2055 while ((bio = req->bio) != NULL) {
2056 int nbytes;
2057
2058 if (nr_bytes >= bio->bi_size) {
2059 req->bio = bio->bi_next;
2060 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002061 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 next_idx = 0;
2063 bio_nbytes = 0;
2064 } else {
2065 int idx = bio->bi_idx + next_idx;
2066
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002067 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002069 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002070 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 break;
2072 }
2073
2074 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2075 BIO_BUG_ON(nbytes > bio->bi_size);
2076
2077 /*
2078 * not a complete bvec done
2079 */
2080 if (unlikely(nbytes > nr_bytes)) {
2081 bio_nbytes += nr_bytes;
2082 total_bytes += nr_bytes;
2083 break;
2084 }
2085
2086 /*
2087 * advance to the next vector
2088 */
2089 next_idx++;
2090 bio_nbytes += nbytes;
2091 }
2092
2093 total_bytes += nbytes;
2094 nr_bytes -= nbytes;
2095
Jens Axboe6728cb02008-01-31 13:03:55 +01002096 bio = req->bio;
2097 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 /*
2099 * end more in this run, or just return 'not-done'
2100 */
2101 if (unlikely(nr_bytes <= 0))
2102 break;
2103 }
2104 }
2105
2106 /*
2107 * completely done
2108 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002109 if (!req->bio) {
2110 /*
2111 * Reset counters so that the request stacking driver
2112 * can find how many bytes remain in the request
2113 * later.
2114 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002115 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002116 return false;
2117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
2119 /*
2120 * if the request wasn't completed, update state
2121 */
2122 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002123 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 bio->bi_idx += next_idx;
2125 bio_iovec(bio)->bv_offset += nr_bytes;
2126 bio_iovec(bio)->bv_len -= nr_bytes;
2127 }
2128
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002129 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002130 req->buffer = bio_data(req->bio);
2131
2132 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002133 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002134 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002135
Tejun Heo80a761f2009-07-03 17:48:17 +09002136 /* mixed attributes always follow the first bio */
2137 if (req->cmd_flags & REQ_MIXED_MERGE) {
2138 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2139 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2140 }
2141
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002142 /*
2143 * If total number of sectors is less than the first segment
2144 * size, something has gone terribly wrong.
2145 */
2146 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2147 printk(KERN_ERR "blk: request botched\n");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002148 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002149 }
2150
2151 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002153
Tejun Heo2e60e022009-04-23 11:05:18 +09002154 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155}
Tejun Heo2e60e022009-04-23 11:05:18 +09002156EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
Tejun Heo2e60e022009-04-23 11:05:18 +09002158static bool blk_update_bidi_request(struct request *rq, int error,
2159 unsigned int nr_bytes,
2160 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002161{
Tejun Heo2e60e022009-04-23 11:05:18 +09002162 if (blk_update_request(rq, error, nr_bytes))
2163 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002164
Tejun Heo2e60e022009-04-23 11:05:18 +09002165 /* Bidi request must be completed as a whole */
2166 if (unlikely(blk_bidi_rq(rq)) &&
2167 blk_update_request(rq->next_rq, error, bidi_bytes))
2168 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002169
Jens Axboee2e1a142010-06-09 10:42:09 +02002170 if (blk_queue_add_random(rq->q))
2171 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002172
2173 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174}
2175
James Bottomley28018c22010-07-01 19:49:17 +09002176/**
2177 * blk_unprep_request - unprepare a request
2178 * @req: the request
2179 *
2180 * This function makes a request ready for complete resubmission (or
2181 * completion). It happens only after all error handling is complete,
2182 * so represents the appropriate moment to deallocate any resources
2183 * that were allocated to the request in the prep_rq_fn. The queue
2184 * lock is held when calling this.
2185 */
2186void blk_unprep_request(struct request *req)
2187{
2188 struct request_queue *q = req->q;
2189
2190 req->cmd_flags &= ~REQ_DONTPREP;
2191 if (q->unprep_rq_fn)
2192 q->unprep_rq_fn(q, req);
2193}
2194EXPORT_SYMBOL_GPL(blk_unprep_request);
2195
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196/*
2197 * queue lock must be held
2198 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002199static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002201 if (blk_rq_tagged(req))
2202 blk_queue_end_tag(req->q, req);
2203
James Bottomleyba396a62009-05-27 14:17:08 +02002204 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002206 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002207 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208
Mike Andersone78042e2008-10-30 02:16:20 -07002209 blk_delete_timer(req);
2210
James Bottomley28018c22010-07-01 19:49:17 +09002211 if (req->cmd_flags & REQ_DONTPREP)
2212 blk_unprep_request(req);
2213
2214
Jens Axboebc58ba92009-01-23 10:54:44 +01002215 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002216
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002218 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002219 else {
2220 if (blk_bidi_rq(req))
2221 __blk_put_request(req->next_rq->q, req->next_rq);
2222
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 }
2225}
2226
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002227/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002228 * blk_end_bidi_request - Complete a bidi request
2229 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002230 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002231 * @nr_bytes: number of bytes to complete @rq
2232 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002233 *
2234 * Description:
2235 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002236 * Drivers that supports bidi can safely call this member for any
2237 * type of request, bidi or uni. In the later case @bidi_bytes is
2238 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002239 *
2240 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002241 * %false - we are done with this request
2242 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002243 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002244static bool blk_end_bidi_request(struct request *rq, int error,
2245 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002246{
2247 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002248 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002249
Tejun Heo2e60e022009-04-23 11:05:18 +09002250 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2251 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002252
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002253 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002254 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002255 spin_unlock_irqrestore(q->queue_lock, flags);
2256
Tejun Heo2e60e022009-04-23 11:05:18 +09002257 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002258}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002259
2260/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002261 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2262 * @rq: the request to complete
2263 * @error: %0 for success, < %0 for error
2264 * @nr_bytes: number of bytes to complete @rq
2265 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002266 *
2267 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002268 * Identical to blk_end_bidi_request() except that queue lock is
2269 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002270 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002271 * Return:
2272 * %false - we are done with this request
2273 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002274 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002275static bool __blk_end_bidi_request(struct request *rq, int error,
2276 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002277{
Tejun Heo2e60e022009-04-23 11:05:18 +09002278 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2279 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002280
Tejun Heo2e60e022009-04-23 11:05:18 +09002281 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002282
Tejun Heo2e60e022009-04-23 11:05:18 +09002283 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002284}
2285
2286/**
2287 * blk_end_request - Helper function for drivers to complete the request.
2288 * @rq: the request being processed
2289 * @error: %0 for success, < %0 for error
2290 * @nr_bytes: number of bytes to complete
2291 *
2292 * Description:
2293 * Ends I/O on a number of bytes attached to @rq.
2294 * If @rq has leftover, sets it up for the next range of segments.
2295 *
2296 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002297 * %false - we are done with this request
2298 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002299 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002300bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002301{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002302 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002303}
Jens Axboe56ad1742009-07-28 22:11:24 +02002304EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002305
2306/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002307 * blk_end_request_all - Helper function for drives to finish the request.
2308 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002309 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002310 *
2311 * Description:
2312 * Completely finish @rq.
2313 */
2314void blk_end_request_all(struct request *rq, int error)
2315{
2316 bool pending;
2317 unsigned int bidi_bytes = 0;
2318
2319 if (unlikely(blk_bidi_rq(rq)))
2320 bidi_bytes = blk_rq_bytes(rq->next_rq);
2321
2322 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2323 BUG_ON(pending);
2324}
Jens Axboe56ad1742009-07-28 22:11:24 +02002325EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002326
2327/**
2328 * blk_end_request_cur - Helper function to finish the current request chunk.
2329 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002330 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002331 *
2332 * Description:
2333 * Complete the current consecutively mapped chunk from @rq.
2334 *
2335 * Return:
2336 * %false - we are done with this request
2337 * %true - still buffers pending for this request
2338 */
2339bool blk_end_request_cur(struct request *rq, int error)
2340{
2341 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2342}
Jens Axboe56ad1742009-07-28 22:11:24 +02002343EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002344
2345/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002346 * blk_end_request_err - Finish a request till the next failure boundary.
2347 * @rq: the request to finish till the next failure boundary for
2348 * @error: must be negative errno
2349 *
2350 * Description:
2351 * Complete @rq till the next failure boundary.
2352 *
2353 * Return:
2354 * %false - we are done with this request
2355 * %true - still buffers pending for this request
2356 */
2357bool blk_end_request_err(struct request *rq, int error)
2358{
2359 WARN_ON(error >= 0);
2360 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2361}
2362EXPORT_SYMBOL_GPL(blk_end_request_err);
2363
2364/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002365 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002366 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002367 * @error: %0 for success, < %0 for error
2368 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002369 *
2370 * Description:
2371 * Must be called with queue lock held unlike blk_end_request().
2372 *
2373 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002374 * %false - we are done with this request
2375 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002376 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002377bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002378{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002379 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002380}
Jens Axboe56ad1742009-07-28 22:11:24 +02002381EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002382
2383/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002384 * __blk_end_request_all - Helper function for drives to finish the request.
2385 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002386 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002387 *
2388 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002389 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002390 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002391void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002392{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002393 bool pending;
2394 unsigned int bidi_bytes = 0;
2395
2396 if (unlikely(blk_bidi_rq(rq)))
2397 bidi_bytes = blk_rq_bytes(rq->next_rq);
2398
2399 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2400 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002401}
Jens Axboe56ad1742009-07-28 22:11:24 +02002402EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002403
2404/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002405 * __blk_end_request_cur - Helper function to finish the current request chunk.
2406 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002407 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002408 *
2409 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002410 * Complete the current consecutively mapped chunk from @rq. Must
2411 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002412 *
2413 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002414 * %false - we are done with this request
2415 * %true - still buffers pending for this request
2416 */
2417bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002418{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002419 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002420}
Jens Axboe56ad1742009-07-28 22:11:24 +02002421EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002422
Tejun Heo80a761f2009-07-03 17:48:17 +09002423/**
2424 * __blk_end_request_err - Finish a request till the next failure boundary.
2425 * @rq: the request to finish till the next failure boundary for
2426 * @error: must be negative errno
2427 *
2428 * Description:
2429 * Complete @rq till the next failure boundary. Must be called
2430 * with queue lock held.
2431 *
2432 * Return:
2433 * %false - we are done with this request
2434 * %true - still buffers pending for this request
2435 */
2436bool __blk_end_request_err(struct request *rq, int error)
2437{
2438 WARN_ON(error >= 0);
2439 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2440}
2441EXPORT_SYMBOL_GPL(__blk_end_request_err);
2442
Jens Axboe86db1e22008-01-29 14:53:40 +01002443void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2444 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002446 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002447 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448
David Woodhousefb2dce82008-08-05 18:01:53 +01002449 if (bio_has_data(bio)) {
2450 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002451 rq->buffer = bio_data(bio);
2452 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002453 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455
NeilBrown66846572007-08-16 13:31:28 +02002456 if (bio->bi_bdev)
2457 rq->rq_disk = bio->bi_bdev->bd_disk;
2458}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002460#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2461/**
2462 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2463 * @rq: the request to be flushed
2464 *
2465 * Description:
2466 * Flush all pages in @rq.
2467 */
2468void rq_flush_dcache_pages(struct request *rq)
2469{
2470 struct req_iterator iter;
2471 struct bio_vec *bvec;
2472
2473 rq_for_each_segment(bvec, rq, iter)
2474 flush_dcache_page(bvec->bv_page);
2475}
2476EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2477#endif
2478
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002479/**
2480 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2481 * @q : the queue of the device being checked
2482 *
2483 * Description:
2484 * Check if underlying low-level drivers of a device are busy.
2485 * If the drivers want to export their busy state, they must set own
2486 * exporting function using blk_queue_lld_busy() first.
2487 *
2488 * Basically, this function is used only by request stacking drivers
2489 * to stop dispatching requests to underlying devices when underlying
2490 * devices are busy. This behavior helps more I/O merging on the queue
2491 * of the request stacking driver and prevents I/O throughput regression
2492 * on burst I/O load.
2493 *
2494 * Return:
2495 * 0 - Not busy (The request stacking driver should dispatch request)
2496 * 1 - Busy (The request stacking driver should stop dispatching request)
2497 */
2498int blk_lld_busy(struct request_queue *q)
2499{
2500 if (q->lld_busy_fn)
2501 return q->lld_busy_fn(q);
2502
2503 return 0;
2504}
2505EXPORT_SYMBOL_GPL(blk_lld_busy);
2506
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002507/**
2508 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2509 * @rq: the clone request to be cleaned up
2510 *
2511 * Description:
2512 * Free all bios in @rq for a cloned request.
2513 */
2514void blk_rq_unprep_clone(struct request *rq)
2515{
2516 struct bio *bio;
2517
2518 while ((bio = rq->bio) != NULL) {
2519 rq->bio = bio->bi_next;
2520
2521 bio_put(bio);
2522 }
2523}
2524EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2525
2526/*
2527 * Copy attributes of the original request to the clone request.
2528 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2529 */
2530static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2531{
2532 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002533 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002534 dst->cmd_type = src->cmd_type;
2535 dst->__sector = blk_rq_pos(src);
2536 dst->__data_len = blk_rq_bytes(src);
2537 dst->nr_phys_segments = src->nr_phys_segments;
2538 dst->ioprio = src->ioprio;
2539 dst->extra_len = src->extra_len;
2540}
2541
2542/**
2543 * blk_rq_prep_clone - Helper function to setup clone request
2544 * @rq: the request to be setup
2545 * @rq_src: original request to be cloned
2546 * @bs: bio_set that bios for clone are allocated from
2547 * @gfp_mask: memory allocation mask for bio
2548 * @bio_ctr: setup function to be called for each clone bio.
2549 * Returns %0 for success, non %0 for failure.
2550 * @data: private data to be passed to @bio_ctr
2551 *
2552 * Description:
2553 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2554 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2555 * are not copied, and copying such parts is the caller's responsibility.
2556 * Also, pages which the original bios are pointing to are not copied
2557 * and the cloned bios just point same pages.
2558 * So cloned bios must be completed before original bios, which means
2559 * the caller must complete @rq before @rq_src.
2560 */
2561int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2562 struct bio_set *bs, gfp_t gfp_mask,
2563 int (*bio_ctr)(struct bio *, struct bio *, void *),
2564 void *data)
2565{
2566 struct bio *bio, *bio_src;
2567
2568 if (!bs)
2569 bs = fs_bio_set;
2570
2571 blk_rq_init(NULL, rq);
2572
2573 __rq_for_each_bio(bio_src, rq_src) {
2574 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2575 if (!bio)
2576 goto free_and_out;
2577
2578 __bio_clone(bio, bio_src);
2579
2580 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002581 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002582 goto free_and_out;
2583
2584 if (bio_ctr && bio_ctr(bio, bio_src, data))
2585 goto free_and_out;
2586
2587 if (rq->bio) {
2588 rq->biotail->bi_next = bio;
2589 rq->biotail = bio;
2590 } else
2591 rq->bio = rq->biotail = bio;
2592 }
2593
2594 __blk_rq_prep_clone(rq, rq_src);
2595
2596 return 0;
2597
2598free_and_out:
2599 if (bio)
2600 bio_free(bio, bs);
2601 blk_rq_unprep_clone(rq);
2602
2603 return -ENOMEM;
2604}
2605EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2606
Jens Axboe18887ad2008-07-28 13:08:45 +02002607int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608{
2609 return queue_work(kblockd_workqueue, work);
2610}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611EXPORT_SYMBOL(kblockd_schedule_work);
2612
Vivek Goyale43473b2010-09-15 17:06:35 -04002613int kblockd_schedule_delayed_work(struct request_queue *q,
2614 struct delayed_work *dwork, unsigned long delay)
2615{
2616 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2617}
2618EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2619
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620int __init blk_dev_init(void)
2621{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002622 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2623 sizeof(((struct request *)0)->cmd_flags));
2624
Tejun Heo89b90be2011-01-03 15:01:47 +01002625 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2626 kblockd_workqueue = alloc_workqueue("kblockd",
2627 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 if (!kblockd_workqueue)
2629 panic("Failed to create kblockd\n");
2630
2631 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002632 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633
Jens Axboe8324aa92008-01-29 14:51:59 +01002634 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002635 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 return 0;
2638}