blob: 8135228e4b2907e15f8a8ff5199de3fa8e5f40cd [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
Ingo Molnar0bfc2452008-11-26 11:59:56 +010036EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
Li Zefan55782132009-06-09 13:43:05 +080037EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010038
Jens Axboe165125e2007-07-24 09:28:11 +020039static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41/*
42 * For the allocated request tables
43 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010044static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46/*
47 * For queue allocation
48 */
Jens Axboe6728cb02008-01-31 13:03:55 +010049struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
51/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 * Controlling structure to kblockd
53 */
Jens Axboeff856ba2006-01-09 16:02:34 +010054static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Jens Axboe26b82562008-01-29 13:54:41 +010056static void drive_stat_acct(struct request *rq, int new_io)
57{
Jens Axboe28f13702008-05-07 10:15:46 +020058 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010059 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090060 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010061
Jens Axboec2553b52009-04-24 08:10:11 +020062 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010063 return;
64
Tejun Heo074a7ac2008-08-25 19:56:14 +090065 cpu = part_stat_lock();
Tejun Heo83096eb2009-05-07 22:24:39 +090066 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Tejun Heoc9959052008-08-25 19:47:21 +090067
Jens Axboe28f13702008-05-07 10:15:46 +020068 if (!new_io)
Tejun Heo074a7ac2008-08-25 19:56:14 +090069 part_stat_inc(cpu, part, merges[rw]);
Jens Axboe28f13702008-05-07 10:15:46 +020070 else {
Tejun Heo074a7ac2008-08-25 19:56:14 +090071 part_round_stats(cpu, part);
Nikanth Karthikesana9327ca2009-09-11 09:18:54 +020072 part_inc_in_flight(part, rw);
Jens Axboe26b82562008-01-29 13:54:41 +010073 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020074
Tejun Heo074a7ac2008-08-25 19:56:14 +090075 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010076}
77
Jens Axboe8324aa92008-01-29 14:51:59 +010078void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079{
80 int nr;
81
82 nr = q->nr_requests - (q->nr_requests / 8) + 1;
83 if (nr > q->nr_requests)
84 nr = q->nr_requests;
85 q->nr_congestion_on = nr;
86
87 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
88 if (nr < 1)
89 nr = 1;
90 q->nr_congestion_off = nr;
91}
92
Linus Torvalds1da177e2005-04-16 15:20:36 -070093/**
94 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
95 * @bdev: device
96 *
97 * Locates the passed device's request queue and returns the address of its
98 * backing_dev_info
99 *
100 * Will return NULL if the request queue cannot be located.
101 */
102struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
103{
104 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200105 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
107 if (q)
108 ret = &q->backing_dev_info;
109 return ret;
110}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111EXPORT_SYMBOL(blk_get_backing_dev_info);
112
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200113void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200115 memset(rq, 0, sizeof(*rq));
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700118 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200119 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100120 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900121 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200122 INIT_HLIST_NODE(&rq->hash);
123 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200124 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800125 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100126 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100127 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900128 rq->start_time = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200130EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
NeilBrown5bb23a62007-09-27 12:46:13 +0200132static void req_bio_endio(struct request *rq, struct bio *bio,
133 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100134{
Jens Axboe165125e2007-07-24 09:28:11 +0200135 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100136
NeilBrown5bb23a62007-09-27 12:46:13 +0200137 if (&q->bar_rq != rq) {
138 if (error)
139 clear_bit(BIO_UPTODATE, &bio->bi_flags);
140 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
141 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100142
NeilBrown5bb23a62007-09-27 12:46:13 +0200143 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100144 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700145 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200146 nbytes = bio->bi_size;
147 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100148
Keith Mannthey08bafc02008-11-25 10:24:35 +0100149 if (unlikely(rq->cmd_flags & REQ_QUIET))
150 set_bit(BIO_QUIET, &bio->bi_flags);
151
NeilBrown5bb23a62007-09-27 12:46:13 +0200152 bio->bi_size -= nbytes;
153 bio->bi_sector += (nbytes >> 9);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200154
155 if (bio_integrity(bio))
156 bio_integrity_advance(bio, nbytes);
157
NeilBrown5bb23a62007-09-27 12:46:13 +0200158 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200159 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200160 } else {
161
162 /*
163 * Okay, this is the barrier request in progress, just
164 * record the error;
165 */
166 if (error && !q->orderr)
167 q->orderr = error;
168 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100169}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171void blk_dump_rq_flags(struct request *rq, char *msg)
172{
173 int bit;
174
Jens Axboe6728cb02008-01-31 13:03:55 +0100175 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200176 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
177 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Tejun Heo83096eb2009-05-07 22:24:39 +0900179 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
180 (unsigned long long)blk_rq_pos(rq),
181 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900182 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900183 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Jens Axboe4aff5e22006-08-10 08:44:47 +0200185 if (blk_pc_request(rq)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100186 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200187 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 printk("%02x ", rq->cmd[bit]);
189 printk("\n");
190 }
191}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192EXPORT_SYMBOL(blk_dump_rq_flags);
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194/*
195 * "plug" the device if there are no outstanding requests: this will
196 * force the transfer to start only after we have put all the requests
197 * on the list.
198 *
199 * This is called with interrupts off and no requests on the queue and
200 * with the queue lock held.
201 */
Jens Axboe165125e2007-07-24 09:28:11 +0200202void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
204 WARN_ON(!irqs_disabled());
205
206 /*
207 * don't plug a stopped queue, it must be paired with blk_start_queue()
208 * which will restart the queueing
209 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200210 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 return;
212
Jens Axboee48ec692008-07-03 13:18:54 +0200213 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100215 trace_block_plug(q);
Jens Axboe2056a782006-03-23 20:00:26 +0100216 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218EXPORT_SYMBOL(blk_plug_device);
219
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200220/**
221 * blk_plug_device_unlocked - plug a device without queue lock held
222 * @q: The &struct request_queue to plug
223 *
224 * Description:
225 * Like @blk_plug_device(), but grabs the queue lock and disables
226 * interrupts.
227 **/
228void blk_plug_device_unlocked(struct request_queue *q)
229{
230 unsigned long flags;
231
232 spin_lock_irqsave(q->queue_lock, flags);
233 blk_plug_device(q);
234 spin_unlock_irqrestore(q->queue_lock, flags);
235}
236EXPORT_SYMBOL(blk_plug_device_unlocked);
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238/*
239 * remove the queue from the plugged list, if present. called with
240 * queue lock held and interrupts disabled.
241 */
Jens Axboe165125e2007-07-24 09:28:11 +0200242int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243{
244 WARN_ON(!irqs_disabled());
245
Jens Axboee48ec692008-07-03 13:18:54 +0200246 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 return 0;
248
249 del_timer(&q->unplug_timer);
250 return 1;
251}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252EXPORT_SYMBOL(blk_remove_plug);
253
254/*
255 * remove the plug and let it rip..
256 */
Jens Axboe165125e2007-07-24 09:28:11 +0200257void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200259 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 return;
Jens Axboea31a9732008-10-17 13:58:29 +0200261 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 return;
263
Jens Axboe22e2c502005-06-27 10:55:12 +0200264 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
267/**
268 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200269 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 *
271 * Description:
272 * Linux uses plugging to build bigger requests queues before letting
273 * the device have at them. If a queue is plugged, the I/O scheduler
274 * is still adding and merging requests on the queue. Once the queue
275 * gets unplugged, the request_fn defined for the queue is invoked and
276 * transfers started.
277 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200278void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200280 if (blk_queue_plugged(q)) {
281 spin_lock_irq(q->queue_lock);
282 __generic_unplug_device(q);
283 spin_unlock_irq(q->queue_lock);
284 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285}
286EXPORT_SYMBOL(generic_unplug_device);
287
288static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
289 struct page *page)
290{
Jens Axboe165125e2007-07-24 09:28:11 +0200291 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500293 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
295
Jens Axboe86db1e22008-01-29 14:53:40 +0100296void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297{
Jens Axboe165125e2007-07-24 09:28:11 +0200298 struct request_queue *q =
299 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100301 trace_block_unplug_io(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 q->unplug_fn(q);
303}
304
Jens Axboe86db1e22008-01-29 14:53:40 +0100305void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306{
Jens Axboe165125e2007-07-24 09:28:11 +0200307 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100309 trace_block_unplug_timer(q);
Jens Axboe18887ad2008-07-28 13:08:45 +0200310 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311}
312
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500313void blk_unplug(struct request_queue *q)
314{
315 /*
316 * devices don't necessarily have an ->unplug_fn defined
317 */
318 if (q->unplug_fn) {
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100319 trace_block_unplug_io(q);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500320 q->unplug_fn(q);
321 }
322}
323EXPORT_SYMBOL(blk_unplug);
324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325/**
326 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200327 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 *
329 * Description:
330 * blk_start_queue() will clear the stop flag on the queue, and call
331 * the request_fn for the queue if it was in a stopped state when
332 * entered. Also see blk_stop_queue(). Queue lock must be held.
333 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200334void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200336 WARN_ON(!irqs_disabled());
337
Nick Piggin75ad23b2008-04-29 14:48:33 +0200338 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Tejun Heoa538cd02009-04-23 11:05:17 +0900339 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341EXPORT_SYMBOL(blk_start_queue);
342
343/**
344 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200345 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 *
347 * Description:
348 * The Linux block layer assumes that a block driver will consume all
349 * entries on the request queue when the request_fn strategy is called.
350 * Often this will not happen, because of hardware limitations (queue
351 * depth settings). If a device driver gets a 'queue full' response,
352 * or if it simply chooses not to queue more I/O at one point, it can
353 * call this function to prevent the request_fn from being called until
354 * the driver has signalled it's ready to go again. This happens by calling
355 * blk_start_queue() to restart queue operations. Queue lock must be held.
356 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200357void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358{
359 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200360 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}
362EXPORT_SYMBOL(blk_stop_queue);
363
364/**
365 * blk_sync_queue - cancel any pending callbacks on a queue
366 * @q: the queue
367 *
368 * Description:
369 * The block layer may perform asynchronous callback activity
370 * on a queue, such as calling the unplug function after a timeout.
371 * A block device may call blk_sync_queue to ensure that any
372 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200373 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 * that its ->make_request_fn will not re-add plugging prior to calling
375 * this function.
376 *
377 */
378void blk_sync_queue(struct request_queue *q)
379{
380 del_timer_sync(&q->unplug_timer);
Jens Axboe70ed28b2008-11-19 14:38:39 +0100381 del_timer_sync(&q->timeout);
Cheng Renquan64d01dc2008-12-03 12:41:39 +0100382 cancel_work_sync(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384EXPORT_SYMBOL(blk_sync_queue);
385
386/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200387 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200389 *
390 * Description:
391 * See @blk_run_queue. This variant must be called with the queue lock
392 * held and interrupts disabled.
393 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200395void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200398
Tejun Heoa538cd02009-04-23 11:05:17 +0900399 if (unlikely(blk_queue_stopped(q)))
400 return;
401
402 if (elv_queue_empty(q))
403 return;
404
Jens Axboedac07ec2006-05-11 08:20:16 +0200405 /*
406 * Only recurse once to avoid overrunning the stack, let the unplug
407 * handling reinvoke the handler shortly if we already got there.
408 */
Tejun Heoa538cd02009-04-23 11:05:17 +0900409 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
410 q->request_fn(q);
411 queue_flag_clear(QUEUE_FLAG_REENTER, q);
412 } else {
413 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
414 kblockd_schedule_work(q, &q->unplug_work);
415 }
Nick Piggin75ad23b2008-04-29 14:48:33 +0200416}
417EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200418
Nick Piggin75ad23b2008-04-29 14:48:33 +0200419/**
420 * blk_run_queue - run a single device queue
421 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200422 *
423 * Description:
424 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900425 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200426 */
427void blk_run_queue(struct request_queue *q)
428{
429 unsigned long flags;
430
431 spin_lock_irqsave(q->queue_lock, flags);
432 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 spin_unlock_irqrestore(q->queue_lock, flags);
434}
435EXPORT_SYMBOL(blk_run_queue);
436
Jens Axboe165125e2007-07-24 09:28:11 +0200437void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500438{
439 kobject_put(&q->kobj);
440}
Al Viro483f4af2006-03-18 18:34:37 -0500441
Jens Axboe6728cb02008-01-31 13:03:55 +0100442void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500443{
Jens Axboee3335de2008-09-18 09:22:54 -0700444 /*
445 * We know we have process context here, so we can be a little
446 * cautious and ensure that pending block actions on this device
447 * are done before moving on. Going into this function, we should
448 * not have processes doing IO to this device.
449 */
450 blk_sync_queue(q);
451
Al Viro483f4af2006-03-18 18:34:37 -0500452 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200453 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500454 mutex_unlock(&q->sysfs_lock);
455
456 if (q->elevator)
457 elevator_exit(q->elevator);
458
459 blk_put_queue(q);
460}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461EXPORT_SYMBOL(blk_cleanup_queue);
462
Jens Axboe165125e2007-07-24 09:28:11 +0200463static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
465 struct request_list *rl = &q->rq;
466
Jens Axboe1faa16d2009-04-06 14:48:01 +0200467 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
468 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200469 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200470 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
471 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Christoph Lameter19460892005-06-23 00:08:19 -0700473 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
474 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
476 if (!rl->rq_pool)
477 return -ENOMEM;
478
479 return 0;
480}
481
Jens Axboe165125e2007-07-24 09:28:11 +0200482struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
Christoph Lameter19460892005-06-23 00:08:19 -0700484 return blk_alloc_queue_node(gfp_mask, -1);
485}
486EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
Jens Axboe165125e2007-07-24 09:28:11 +0200488struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700489{
Jens Axboe165125e2007-07-24 09:28:11 +0200490 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700491 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700492
Jens Axboe8324aa92008-01-29 14:51:59 +0100493 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700494 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 if (!q)
496 return NULL;
497
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700498 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
499 q->backing_dev_info.unplug_io_data = q;
Jens Axboe0989a022009-06-12 14:42:56 +0200500 q->backing_dev_info.ra_pages =
501 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
502 q->backing_dev_info.state = 0;
503 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200504 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200505
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700506 err = bdi_init(&q->backing_dev_info);
507 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100508 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700509 return NULL;
510 }
511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 init_timer(&q->unplug_timer);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700513 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
514 INIT_LIST_HEAD(&q->timeout_list);
Peter Zijlstra713ada92008-10-16 13:44:57 +0200515 INIT_WORK(&q->unplug_work, blk_unplug_work);
Al Viro483f4af2006-03-18 18:34:37 -0500516
Jens Axboe8324aa92008-01-29 14:51:59 +0100517 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Al Viro483f4af2006-03-18 18:34:37 -0500519 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700520 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 return q;
523}
Christoph Lameter19460892005-06-23 00:08:19 -0700524EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
526/**
527 * blk_init_queue - prepare a request queue for use with a block device
528 * @rfn: The function to be called to process requests that have been
529 * placed on the queue.
530 * @lock: Request queue spin lock
531 *
532 * Description:
533 * If a block device wishes to use the standard request handling procedures,
534 * which sorts requests and coalesces adjacent requests, then it must
535 * call blk_init_queue(). The function @rfn will be called when there
536 * are requests on the queue that need to be processed. If the device
537 * supports plugging, then @rfn may not be called immediately when requests
538 * are available on the queue, but may be called at some time later instead.
539 * Plugged queues are generally unplugged when a buffer belonging to one
540 * of the requests on the queue is needed, or due to memory pressure.
541 *
542 * @rfn is not required, or even expected, to remove all requests off the
543 * queue, but only as many as it can handle at a time. If it does leave
544 * requests on the queue, it is responsible for arranging that the requests
545 * get dealt with eventually.
546 *
547 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200548 * request queue; this lock will be taken also from interrupt context, so irq
549 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200551 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 * it didn't succeed.
553 *
554 * Note:
555 * blk_init_queue() must be paired with a blk_cleanup_queue() call
556 * when the block device is deactivated (such as at module unload).
557 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700558
Jens Axboe165125e2007-07-24 09:28:11 +0200559struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
Christoph Lameter19460892005-06-23 00:08:19 -0700561 return blk_init_queue_node(rfn, lock, -1);
562}
563EXPORT_SYMBOL(blk_init_queue);
564
Jens Axboe165125e2007-07-24 09:28:11 +0200565struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700566blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
567{
Jens Axboe165125e2007-07-24 09:28:11 +0200568 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
570 if (!q)
571 return NULL;
572
Christoph Lameter19460892005-06-23 00:08:19 -0700573 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -0500574 if (blk_init_free_list(q)) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100575 kmem_cache_free(blk_requestq_cachep, q);
Al Viro8669aaf2006-03-18 13:50:00 -0500576 return NULL;
577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
579 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 q->prep_rq_fn = NULL;
581 q->unplug_fn = generic_unplug_device;
Jens Axboebc58ba92009-01-23 10:54:44 +0100582 q->queue_flags = QUEUE_FLAG_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 q->queue_lock = lock;
584
Jens Axboef3b144a2009-03-06 08:48:33 +0100585 /*
586 * This also sets hw/phys segments, boundary and size
587 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Alan Stern44ec9542007-02-20 11:01:57 -0500590 q->sg_reserved_size = INT_MAX;
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 /*
593 * all done
594 */
595 if (!elevator_init(q, NULL)) {
596 blk_queue_congestion_threshold(q);
597 return q;
598 }
599
Al Viro8669aaf2006-03-18 13:50:00 -0500600 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 return NULL;
602}
Christoph Lameter19460892005-06-23 00:08:19 -0700603EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
Jens Axboe165125e2007-07-24 09:28:11 +0200605int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700607 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500608 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 return 0;
610 }
611
612 return 1;
613}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Jens Axboe165125e2007-07-24 09:28:11 +0200615static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200617 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200618 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 mempool_free(rq, q->rq.rq_pool);
620}
621
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200622static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200623blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624{
625 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
626
627 if (!rq)
628 return NULL;
629
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200630 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200631
Jerome Marchand42dad762009-04-22 14:01:49 +0200632 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Tejun Heocb98fc82005-10-28 08:29:39 +0200634 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200635 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200636 mempool_free(rq, q->rq.rq_pool);
637 return NULL;
638 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200639 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200640 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
Tejun Heocb98fc82005-10-28 08:29:39 +0200642 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643}
644
645/*
646 * ioc_batching returns true if the ioc is a valid batching request and
647 * should be given priority access to a request.
648 */
Jens Axboe165125e2007-07-24 09:28:11 +0200649static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
651 if (!ioc)
652 return 0;
653
654 /*
655 * Make sure the process is able to allocate at least 1 request
656 * even if the batch times out, otherwise we could theoretically
657 * lose wakeups.
658 */
659 return ioc->nr_batch_requests == q->nr_batching ||
660 (ioc->nr_batch_requests > 0
661 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
662}
663
664/*
665 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
666 * will cause the process to be a "batcher" on all queues in the system. This
667 * is the behaviour we want though - once it gets a wakeup it should be given
668 * a nice run.
669 */
Jens Axboe165125e2007-07-24 09:28:11 +0200670static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671{
672 if (!ioc || ioc_batching(q, ioc))
673 return;
674
675 ioc->nr_batch_requests = q->nr_batching;
676 ioc->last_waited = jiffies;
677}
678
Jens Axboe1faa16d2009-04-06 14:48:01 +0200679static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
681 struct request_list *rl = &q->rq;
682
Jens Axboe1faa16d2009-04-06 14:48:01 +0200683 if (rl->count[sync] < queue_congestion_off_threshold(q))
684 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Jens Axboe1faa16d2009-04-06 14:48:01 +0200686 if (rl->count[sync] + 1 <= q->nr_requests) {
687 if (waitqueue_active(&rl->wait[sync]))
688 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
Jens Axboe1faa16d2009-04-06 14:48:01 +0200690 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 }
692}
693
694/*
695 * A request has just been released. Account for it, update the full and
696 * congestion status, wake up any waiters. Called under q->queue_lock.
697 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200698static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
700 struct request_list *rl = &q->rq;
701
Jens Axboe1faa16d2009-04-06 14:48:01 +0200702 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200703 if (priv)
704 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
Jens Axboe1faa16d2009-04-06 14:48:01 +0200706 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Jens Axboe1faa16d2009-04-06 14:48:01 +0200708 if (unlikely(rl->starved[sync ^ 1]))
709 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712/*
Nick Piggind6344532005-06-28 20:45:14 -0700713 * Get a free request, queue_lock must be held.
714 * Returns NULL on failure, with queue_lock held.
715 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 */
Jens Axboe165125e2007-07-24 09:28:11 +0200717static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100718 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719{
720 struct request *rq = NULL;
721 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100722 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200723 const bool is_sync = rw_is_sync(rw_flags) != 0;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100724 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Jens Axboe7749a8d2006-12-13 13:02:26 +0100726 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100727 if (may_queue == ELV_MQUEUE_NO)
728 goto rq_starved;
729
Jens Axboe1faa16d2009-04-06 14:48:01 +0200730 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
731 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200732 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100733 /*
734 * The queue will fill after this allocation, so set
735 * it as full, and mark this process as "batching".
736 * This process will be allowed to complete a batch of
737 * requests, others will be blocked.
738 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200739 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100740 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200741 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100742 } else {
743 if (may_queue != ELV_MQUEUE_MUST
744 && !ioc_batching(q, ioc)) {
745 /*
746 * The queue is full and the allocating
747 * process is not a "batcher", and not
748 * exempted by the IO scheduler
749 */
750 goto out;
751 }
752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200754 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 }
756
Jens Axboe082cf692005-06-28 16:35:11 +0200757 /*
758 * Only allow batching queuers to allocate up to 50% over the defined
759 * limit of requests, otherwise we could have thousands of requests
760 * allocated with any setting of ->nr_requests
761 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200762 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200763 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100764
Jens Axboe1faa16d2009-04-06 14:48:01 +0200765 rl->count[is_sync]++;
766 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200767
Jens Axboe64521d12005-10-28 08:30:39 +0200768 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200769 if (priv)
770 rl->elvpriv++;
771
Jerome Marchand42dad762009-04-22 14:01:49 +0200772 if (blk_queue_io_stat(q))
773 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 spin_unlock_irq(q->queue_lock);
775
Jens Axboe7749a8d2006-12-13 13:02:26 +0100776 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100777 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 /*
779 * Allocation failed presumably due to memory. Undo anything
780 * we might have messed up.
781 *
782 * Allocating task should really be put onto the front of the
783 * wait queue, but this is pretty rare.
784 */
785 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200786 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
788 /*
789 * in the very unlikely event that allocation failed and no
790 * requests for this direction was pending, mark us starved
791 * so that freeing of a request in the other direction will
792 * notice us. another possible fix would be to split the
793 * rq mempool into READ and WRITE
794 */
795rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200796 if (unlikely(rl->count[is_sync] == 0))
797 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 goto out;
800 }
801
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100802 /*
803 * ioc may be NULL here, and ioc_batching will be false. That's
804 * OK, if the queue is under the request limit then requests need
805 * not count toward the nr_batch_requests limit. There will always
806 * be some limit enforced by BLK_BATCH_TIME.
807 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if (ioc_batching(q, ioc))
809 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100810
Jens Axboe1faa16d2009-04-06 14:48:01 +0200811 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 return rq;
814}
815
816/*
817 * No available requests for this queue, unplug the device and wait for some
818 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700819 *
820 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 */
Jens Axboe165125e2007-07-24 09:28:11 +0200822static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200823 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200825 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 struct request *rq;
827
Jens Axboe7749a8d2006-12-13 13:02:26 +0100828 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700829 while (!rq) {
830 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200831 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 struct request_list *rl = &q->rq;
833
Jens Axboe1faa16d2009-04-06 14:48:01 +0200834 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 TASK_UNINTERRUPTIBLE);
836
Jens Axboe1faa16d2009-04-06 14:48:01 +0200837 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200839 __generic_unplug_device(q);
840 spin_unlock_irq(q->queue_lock);
841 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200843 /*
844 * After sleeping, we become a "batching" process and
845 * will be able to allocate at least one request, and
846 * up to a big batch of them for a small period time.
847 * See ioc_batching, ioc_set_batching
848 */
849 ioc = current_io_context(GFP_NOIO, q->node);
850 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100851
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200852 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200853 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200854
855 rq = get_request(q, rw_flags, bio, GFP_NOIO);
856 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 return rq;
859}
860
Jens Axboe165125e2007-07-24 09:28:11 +0200861struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862{
863 struct request *rq;
864
865 BUG_ON(rw != READ && rw != WRITE);
866
Nick Piggind6344532005-06-28 20:45:14 -0700867 spin_lock_irq(q->queue_lock);
868 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200869 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700870 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200871 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700872 if (!rq)
873 spin_unlock_irq(q->queue_lock);
874 }
875 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
877 return rq;
878}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879EXPORT_SYMBOL(blk_get_request);
880
881/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300882 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700883 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300884 * @bio: The bio describing the memory mappings that will be submitted for IO.
885 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700886 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200887 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300888 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
889 * type commands. Where the struct request needs to be farther initialized by
890 * the caller. It is passed a &struct bio, which describes the memory info of
891 * the I/O transfer.
892 *
893 * The caller of blk_make_request must make sure that bi_io_vec
894 * are set to describe the memory buffers. That bio_data_dir() will return
895 * the needed direction of the request. (And all bio's in the passed bio-chain
896 * are properly set accordingly)
897 *
898 * If called under none-sleepable conditions, mapped bio buffers must not
899 * need bouncing, by calling the appropriate masked or flagged allocator,
900 * suitable for the target device. Otherwise the call to blk_queue_bounce will
901 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200902 *
903 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
904 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
905 * anything but the first bio in the chain. Otherwise you risk waiting for IO
906 * completion of a bio that hasn't been submitted yet, thus resulting in a
907 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
908 * of bio_alloc(), as that avoids the mempool deadlock.
909 * If possible a big IO should be split into smaller parts when allocation
910 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200911 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300912struct request *blk_make_request(struct request_queue *q, struct bio *bio,
913 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200914{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300915 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
916
917 if (unlikely(!rq))
918 return ERR_PTR(-ENOMEM);
919
920 for_each_bio(bio) {
921 struct bio *bounce_bio = bio;
922 int ret;
923
924 blk_queue_bounce(q, &bounce_bio);
925 ret = blk_rq_append_bio(q, rq, bounce_bio);
926 if (unlikely(ret)) {
927 blk_put_request(rq);
928 return ERR_PTR(ret);
929 }
930 }
931
932 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200933}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300934EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200935
936/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 * blk_requeue_request - put a request back on queue
938 * @q: request queue where request should be inserted
939 * @rq: request to be inserted
940 *
941 * Description:
942 * Drivers often keep queueing requests until the hardware cannot accept
943 * more, when that condition happens we need to put the request back
944 * on the queue. Must be called with queue lock held.
945 */
Jens Axboe165125e2007-07-24 09:28:11 +0200946void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700948 blk_delete_timer(rq);
949 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100950 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 if (blk_rq_tagged(rq))
953 blk_queue_end_tag(q, rq);
954
James Bottomleyba396a62009-05-27 14:17:08 +0200955 BUG_ON(blk_queued_rq(rq));
956
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 elv_requeue_request(q, rq);
958}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959EXPORT_SYMBOL(blk_requeue_request);
960
961/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200962 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 * @q: request queue where request should be inserted
964 * @rq: request to be inserted
965 * @at_head: insert request at head or tail of queue
966 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 *
968 * Description:
969 * Many block devices need to execute commands asynchronously, so they don't
970 * block the whole kernel from preemption during request execution. This is
971 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200972 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
973 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 *
975 * We have the option of inserting the head or the tail of the queue.
976 * Typically we use the tail for new ioctls and so forth. We use the head
977 * of the queue for things like a QUEUE_FULL message from a device, or a
978 * host that is unable to accept a particular command.
979 */
Jens Axboe165125e2007-07-24 09:28:11 +0200980void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500981 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982{
Tejun Heo 867d1192005-04-24 02:06:05 -0500983 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 unsigned long flags;
985
986 /*
987 * tell I/O scheduler that this isn't a regular read/write (ie it
988 * must not attempt merges on this) and that it acts as a soft
989 * barrier
990 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200991 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
993 rq->special = data;
994
995 spin_lock_irqsave(q->queue_lock, flags);
996
997 /*
998 * If command is tagged, release the tag
999 */
Tejun Heo 867d1192005-04-24 02:06:05 -05001000 if (blk_rq_tagged(rq))
1001 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Jerome Marchandb238b3d2007-10-23 15:05:46 +02001003 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -05001004 __elv_add_request(q, rq, where, 0);
Tejun Heoa7f55792009-04-23 11:05:17 +09001005 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 spin_unlock_irqrestore(q->queue_lock, flags);
1007}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008EXPORT_SYMBOL(blk_insert_request);
1009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010/*
1011 * add-request adds a request to the linked list.
1012 * queue lock is held and interrupts disabled, as we muck with the
1013 * request queue list.
1014 */
Jens Axboe6728cb02008-01-31 13:03:55 +01001015static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016{
Jerome Marchandb238b3d2007-10-23 15:05:46 +02001017 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 /*
1020 * elevator indicated where it wants this request to be
1021 * inserted at elevator_merge time
1022 */
1023 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1024}
Jens Axboe6728cb02008-01-31 13:03:55 +01001025
Tejun Heo074a7ac2008-08-25 19:56:14 +09001026static void part_round_stats_single(int cpu, struct hd_struct *part,
1027 unsigned long now)
1028{
1029 if (now == part->stamp)
1030 return;
1031
1032 if (part->in_flight) {
1033 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesana9327ca2009-09-11 09:18:54 +02001034 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001035 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1036 }
1037 part->stamp = now;
1038}
1039
1040/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001041 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1042 * @cpu: cpu number for stats access
1043 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 *
1045 * The average IO queue length and utilisation statistics are maintained
1046 * by observing the current state of the queue length and the amount of
1047 * time it has been in this state for.
1048 *
1049 * Normally, that accounting is done on IO completion, but that can result
1050 * in more than a second's worth of IO being accounted for within any one
1051 * second, leading to >100% utilisation. To deal with that, we call this
1052 * function to do a round-off before returning the results when reading
1053 * /proc/diskstats. This accounts immediately for all queue usage up to
1054 * the current jiffies and restarts the counters again.
1055 */
Tejun Heoc9959052008-08-25 19:47:21 +09001056void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001057{
1058 unsigned long now = jiffies;
1059
Tejun Heo074a7ac2008-08-25 19:56:14 +09001060 if (part->partno)
1061 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1062 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001063}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001064EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001065
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066/*
1067 * queue lock must be held
1068 */
Jens Axboe165125e2007-07-24 09:28:11 +02001069void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 if (unlikely(!q))
1072 return;
1073 if (unlikely(--req->ref_count))
1074 return;
1075
Tejun Heo8922e162005-10-20 16:23:44 +02001076 elv_completed_request(q, req);
1077
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001078 /* this is a bio leak */
1079 WARN_ON(req->bio != NULL);
1080
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 /*
1082 * Request may not have originated from ll_rw_blk. if not,
1083 * it didn't come out of our reserved rq pools
1084 */
Jens Axboe49171e52006-08-10 08:59:11 +02001085 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001086 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001087 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001090 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
1092 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001093 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 }
1095}
Mike Christie6e39b692005-11-11 05:30:24 -06001096EXPORT_SYMBOL_GPL(__blk_put_request);
1097
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098void blk_put_request(struct request *req)
1099{
Tejun Heo8922e162005-10-20 16:23:44 +02001100 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001101 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001103 spin_lock_irqsave(q->queue_lock, flags);
1104 __blk_put_request(q, req);
1105 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107EXPORT_SYMBOL(blk_put_request);
1108
Jens Axboe86db1e22008-01-29 14:53:40 +01001109void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001110{
Jens Axboec7c22e42008-09-13 20:26:01 +02001111 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001112 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001113
1114 /*
Tejun Heoa82afdf2009-07-03 17:48:16 +09001115 * Inherit FAILFAST from bio (for read-ahead, and explicit
1116 * FAILFAST). FAILFAST flags are identical for req and bio.
Tejun Heo52d9e672006-01-06 09:49:58 +01001117 */
Jens Axboe1f98a132009-09-11 14:32:04 +02001118 if (bio_rw_flagged(bio, BIO_RW_AHEAD))
Tejun Heoa82afdf2009-07-03 17:48:16 +09001119 req->cmd_flags |= REQ_FAILFAST_MASK;
1120 else
1121 req->cmd_flags |= bio->bi_rw & REQ_FAILFAST_MASK;
Tejun Heo52d9e672006-01-06 09:49:58 +01001122
Jens Axboe1f98a132009-09-11 14:32:04 +02001123 if (unlikely(bio_rw_flagged(bio, BIO_RW_DISCARD))) {
David Woodhousee17fc0a2008-08-09 16:42:20 +01001124 req->cmd_flags |= REQ_DISCARD;
Jens Axboe1f98a132009-09-11 14:32:04 +02001125 if (bio_rw_flagged(bio, BIO_RW_BARRIER))
David Woodhousee17fc0a2008-08-09 16:42:20 +01001126 req->cmd_flags |= REQ_SOFTBARRIER;
David Woodhousefb2dce82008-08-05 18:01:53 +01001127 req->q->prepare_discard_fn(req->q, req);
Jens Axboe1f98a132009-09-11 14:32:04 +02001128 } else if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)))
Tejun Heoe4025f62009-04-23 11:05:17 +09001129 req->cmd_flags |= REQ_HARDBARRIER;
Tejun Heo52d9e672006-01-06 09:49:58 +01001130
Jens Axboe1f98a132009-09-11 14:32:04 +02001131 if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001132 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe1f98a132009-09-11 14:32:04 +02001133 if (bio_rw_flagged(bio, BIO_RW_META))
Jens Axboe5404bc72006-08-10 09:01:02 +02001134 req->cmd_flags |= REQ_RW_META;
Jens Axboe1f98a132009-09-11 14:32:04 +02001135 if (bio_rw_flagged(bio, BIO_RW_NOIDLE))
Jens Axboeaeb6faf2009-04-06 14:48:07 +02001136 req->cmd_flags |= REQ_NOIDLE;
Jens Axboeb31dc662006-06-13 08:26:10 +02001137
Tejun Heo52d9e672006-01-06 09:49:58 +01001138 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001139 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001140 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001141 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001142}
1143
Jens Axboe644b2d92009-04-06 14:48:06 +02001144/*
1145 * Only disabling plugging for non-rotational devices if it does tagging
1146 * as well, otherwise we do need the proper merging
1147 */
1148static inline bool queue_should_plug(struct request_queue *q)
1149{
Jens Axboefb1e7532009-07-30 08:18:24 +02001150 return !(blk_queue_nonrot(q) && blk_queue_queuing(q));
Jens Axboe644b2d92009-04-06 14:48:06 +02001151}
1152
Jens Axboe165125e2007-07-24 09:28:11 +02001153static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154{
Nick Piggin450991b2005-06-28 20:45:13 -07001155 struct request *req;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001156 int el_ret;
1157 unsigned int bytes = bio->bi_size;
Jens Axboe51da90f2006-07-18 04:14:45 +02001158 const unsigned short prio = bio_prio(bio);
Jens Axboe1f98a132009-09-11 14:32:04 +02001159 const bool sync = bio_rw_flagged(bio, BIO_RW_SYNCIO);
1160 const bool unplug = bio_rw_flagged(bio, BIO_RW_UNPLUG);
Tejun Heo80a761f2009-07-03 17:48:17 +09001161 const unsigned int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001162 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
Jens Axboe1f98a132009-09-11 14:32:04 +02001164 if (bio_rw_flagged(bio, BIO_RW_BARRIER) && bio_has_data(bio) &&
NeilBrowndb64f682009-06-30 09:35:44 +02001165 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1166 bio_endio(bio, -EOPNOTSUPP);
1167 return 0;
1168 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 /*
1170 * low level driver can indicate that it wants pages above a
1171 * certain limit bounced to low memory (ie for highmem, or even
1172 * ISA dma in theory)
1173 */
1174 blk_queue_bounce(q, &bio);
1175
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 spin_lock_irq(q->queue_lock);
1177
Jens Axboe1f98a132009-09-11 14:32:04 +02001178 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 goto get_rq;
1180
1181 el_ret = elv_merge(q, &req, bio);
1182 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001183 case ELEVATOR_BACK_MERGE:
1184 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Jens Axboe6728cb02008-01-31 13:03:55 +01001186 if (!ll_back_merge_fn(q, req, bio))
1187 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001189 trace_block_bio_backmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001190
Tejun Heo80a761f2009-07-03 17:48:17 +09001191 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1192 blk_rq_set_mixed_merge(req);
1193
Jens Axboe6728cb02008-01-31 13:03:55 +01001194 req->biotail->bi_next = bio;
1195 req->biotail = bio;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001196 req->__data_len += bytes;
Jens Axboe6728cb02008-01-31 13:03:55 +01001197 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001198 if (!blk_rq_cpu_valid(req))
1199 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001200 drive_stat_acct(req, 0);
1201 if (!attempt_back_merge(q, req))
1202 elv_merged_request(q, req, el_ret);
1203 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
Jens Axboe6728cb02008-01-31 13:03:55 +01001205 case ELEVATOR_FRONT_MERGE:
1206 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
Jens Axboe6728cb02008-01-31 13:03:55 +01001208 if (!ll_front_merge_fn(q, req, bio))
1209 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001211 trace_block_bio_frontmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001212
Tejun Heo80a761f2009-07-03 17:48:17 +09001213 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1214 blk_rq_set_mixed_merge(req);
1215 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1216 req->cmd_flags |= ff;
1217 }
1218
Jens Axboe6728cb02008-01-31 13:03:55 +01001219 bio->bi_next = req->bio;
1220 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Jens Axboe6728cb02008-01-31 13:03:55 +01001222 /*
1223 * may not be valid. if the low level driver said
1224 * it didn't need a bounce buffer then it better
1225 * not touch req->buffer either...
1226 */
1227 req->buffer = bio_data(bio);
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001228 req->__sector = bio->bi_sector;
1229 req->__data_len += bytes;
Jens Axboe6728cb02008-01-31 13:03:55 +01001230 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001231 if (!blk_rq_cpu_valid(req))
1232 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001233 drive_stat_acct(req, 0);
1234 if (!attempt_front_merge(q, req))
1235 elv_merged_request(q, req, el_ret);
1236 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Jens Axboe6728cb02008-01-31 13:03:55 +01001238 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1239 default:
1240 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 }
1242
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001244 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001245 * This sync check and mask will be re-done in init_request_from_bio(),
1246 * but we need to set it earlier to expose the sync flag to the
1247 * rq allocator and io schedulers.
1248 */
1249 rw_flags = bio_data_dir(bio);
1250 if (sync)
1251 rw_flags |= REQ_RW_SYNC;
1252
1253 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001254 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001255 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001256 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001257 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001258
Nick Piggin450991b2005-06-28 20:45:13 -07001259 /*
1260 * After dropping the lock and possibly sleeping here, our request
1261 * may now be mergeable after it had proven unmergeable (above).
1262 * We don't worry about that case for efficiency. It won't happen
1263 * often, and the elevators are able to handle it.
1264 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001265 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Nick Piggin450991b2005-06-28 20:45:13 -07001267 spin_lock_irq(q->queue_lock);
Jens Axboec7c22e42008-09-13 20:26:01 +02001268 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1269 bio_flagged(bio, BIO_CPU_AFFINE))
1270 req->cpu = blk_cpu_to_group(smp_processor_id());
Jens Axboe644b2d92009-04-06 14:48:06 +02001271 if (queue_should_plug(q) && elv_queue_empty(q))
Nick Piggin450991b2005-06-28 20:45:13 -07001272 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 add_request(q, req);
1274out:
Jens Axboe644b2d92009-04-06 14:48:06 +02001275 if (unplug || !queue_should_plug(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 __generic_unplug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 spin_unlock_irq(q->queue_lock);
1278 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279}
1280
1281/*
1282 * If bio->bi_dev is a partition, remap the location
1283 */
1284static inline void blk_partition_remap(struct bio *bio)
1285{
1286 struct block_device *bdev = bio->bi_bdev;
1287
Jens Axboebf2de6f2007-09-27 13:01:25 +02001288 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001290
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 bio->bi_sector += p->start_sect;
1292 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001293
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001294 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
Alan D. Brunelle22a7c312009-05-04 16:35:08 -04001295 bdev->bd_dev,
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001296 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 }
1298}
1299
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300static void handle_bad_sector(struct bio *bio)
1301{
1302 char b[BDEVNAME_SIZE];
1303
1304 printk(KERN_INFO "attempt to access beyond end of device\n");
1305 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1306 bdevname(bio->bi_bdev, b),
1307 bio->bi_rw,
1308 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1309 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1310
1311 set_bit(BIO_EOF, &bio->bi_flags);
1312}
1313
Akinobu Mitac17bb492006-12-08 02:39:46 -08001314#ifdef CONFIG_FAIL_MAKE_REQUEST
1315
1316static DECLARE_FAULT_ATTR(fail_make_request);
1317
1318static int __init setup_fail_make_request(char *str)
1319{
1320 return setup_fault_attr(&fail_make_request, str);
1321}
1322__setup("fail_make_request=", setup_fail_make_request);
1323
1324static int should_fail_request(struct bio *bio)
1325{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001326 struct hd_struct *part = bio->bi_bdev->bd_part;
1327
1328 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001329 return should_fail(&fail_make_request, bio->bi_size);
1330
1331 return 0;
1332}
1333
1334static int __init fail_make_request_debugfs(void)
1335{
1336 return init_fault_attr_dentries(&fail_make_request,
1337 "fail_make_request");
1338}
1339
1340late_initcall(fail_make_request_debugfs);
1341
1342#else /* CONFIG_FAIL_MAKE_REQUEST */
1343
1344static inline int should_fail_request(struct bio *bio)
1345{
1346 return 0;
1347}
1348
1349#endif /* CONFIG_FAIL_MAKE_REQUEST */
1350
Jens Axboec07e2b42007-07-18 13:27:58 +02001351/*
1352 * Check whether this bio extends beyond the end of the device.
1353 */
1354static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1355{
1356 sector_t maxsector;
1357
1358 if (!nr_sectors)
1359 return 0;
1360
1361 /* Test device or partition size, when known. */
1362 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1363 if (maxsector) {
1364 sector_t sector = bio->bi_sector;
1365
1366 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1367 /*
1368 * This may well happen - the kernel calls bread()
1369 * without checking the size of the device, e.g., when
1370 * mounting a device.
1371 */
1372 handle_bad_sector(bio);
1373 return 1;
1374 }
1375 }
1376
1377 return 0;
1378}
1379
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001381 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 * @bio: The bio describing the location in memory and on the device.
1383 *
1384 * generic_make_request() is used to make I/O requests of block
1385 * devices. It is passed a &struct bio, which describes the I/O that needs
1386 * to be done.
1387 *
1388 * generic_make_request() does not return any status. The
1389 * success/failure status of the request, along with notification of
1390 * completion, is delivered asynchronously through the bio->bi_end_io
1391 * function described (one day) else where.
1392 *
1393 * The caller of generic_make_request must make sure that bi_io_vec
1394 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1395 * set to describe the device address, and the
1396 * bi_end_io and optionally bi_private are set to describe how
1397 * completion notification should be signaled.
1398 *
1399 * generic_make_request and the drivers it calls may use bi_next if this
1400 * bio happens to be merged with someone else, and may change bi_dev and
1401 * bi_sector for remaps as it sees fit. So the values of these fields
1402 * should NOT be depended on after the call to generic_make_request.
1403 */
Neil Brownd89d8792007-05-01 09:53:42 +02001404static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405{
Jens Axboe165125e2007-07-24 09:28:11 +02001406 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001407 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001409 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001410 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
1412 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
Jens Axboec07e2b42007-07-18 13:27:58 +02001414 if (bio_check_eod(bio, nr_sectors))
1415 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
1417 /*
1418 * Resolve the mapping until finished. (drivers are
1419 * still free to implement/resolve their own stacking
1420 * by explicitly returning 0)
1421 *
1422 * NOTE: we don't repeat the blk_size check for each new device.
1423 * Stacking drivers are expected to know what they are doing.
1424 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001425 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001426 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 do {
1428 char b[BDEVNAME_SIZE];
1429
1430 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001431 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 printk(KERN_ERR
1433 "generic_make_request: Trying to access "
1434 "nonexistent block-device %s (%Lu)\n",
1435 bdevname(bio->bi_bdev, b),
1436 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001437 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 }
1439
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001440 if (unlikely(nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001441 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001442 bdevname(bio->bi_bdev, b),
1443 bio_sectors(bio),
1444 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 goto end_io;
1446 }
1447
Nick Pigginfde6ad22005-06-23 00:08:53 -07001448 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 goto end_io;
1450
Akinobu Mitac17bb492006-12-08 02:39:46 -08001451 if (should_fail_request(bio))
1452 goto end_io;
1453
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 /*
1455 * If this device has partitions, remap block n
1456 * of partition p to block n+start(p) of the disk.
1457 */
1458 blk_partition_remap(bio);
1459
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001460 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1461 goto end_io;
1462
NeilBrown5ddfe962006-10-30 22:07:21 -08001463 if (old_sector != -1)
Alan D. Brunelle22a7c312009-05-04 16:35:08 -04001464 trace_block_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001465
NeilBrown5ddfe962006-10-30 22:07:21 -08001466 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001467 old_dev = bio->bi_bdev->bd_dev;
1468
Jens Axboec07e2b42007-07-18 13:27:58 +02001469 if (bio_check_eod(bio, nr_sectors))
1470 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001471
Jens Axboe1f98a132009-09-11 14:32:04 +02001472 if (bio_rw_flagged(bio, BIO_RW_DISCARD) &&
1473 !q->prepare_discard_fn) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001474 err = -EOPNOTSUPP;
1475 goto end_io;
1476 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001477
Minchan Kim01edede2009-09-08 21:56:38 +02001478 trace_block_bio_queue(q, bio);
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 ret = q->make_request_fn(q, bio);
1481 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001482
1483 return;
1484
1485end_io:
1486 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487}
1488
Neil Brownd89d8792007-05-01 09:53:42 +02001489/*
1490 * We only want one ->make_request_fn to be active at a time,
1491 * else stack usage with stacked devices could be a problem.
1492 * So use current->bio_{list,tail} to keep a list of requests
1493 * submited by a make_request_fn function.
1494 * current->bio_tail is also used as a flag to say if
1495 * generic_make_request is currently active in this task or not.
1496 * If it is NULL, then no make_request is active. If it is non-NULL,
1497 * then a make_request is active, and new requests should be added
1498 * at the tail
1499 */
1500void generic_make_request(struct bio *bio)
1501{
1502 if (current->bio_tail) {
1503 /* make_request is active */
1504 *(current->bio_tail) = bio;
1505 bio->bi_next = NULL;
1506 current->bio_tail = &bio->bi_next;
1507 return;
1508 }
1509 /* following loop may be a bit non-obvious, and so deserves some
1510 * explanation.
1511 * Before entering the loop, bio->bi_next is NULL (as all callers
1512 * ensure that) so we have a list with a single bio.
1513 * We pretend that we have just taken it off a longer list, so
1514 * we assign bio_list to the next (which is NULL) and bio_tail
1515 * to &bio_list, thus initialising the bio_list of new bios to be
1516 * added. __generic_make_request may indeed add some more bios
1517 * through a recursive call to generic_make_request. If it
1518 * did, we find a non-NULL value in bio_list and re-enter the loop
1519 * from the top. In this case we really did just take the bio
1520 * of the top of the list (no pretending) and so fixup bio_list and
1521 * bio_tail or bi_next, and call into __generic_make_request again.
1522 *
1523 * The loop was structured like this to make only one call to
1524 * __generic_make_request (which is important as it is large and
1525 * inlined) and to keep the structure simple.
1526 */
1527 BUG_ON(bio->bi_next);
1528 do {
1529 current->bio_list = bio->bi_next;
1530 if (bio->bi_next == NULL)
1531 current->bio_tail = &current->bio_list;
1532 else
1533 bio->bi_next = NULL;
1534 __generic_make_request(bio);
1535 bio = current->bio_list;
1536 } while (bio);
1537 current->bio_tail = NULL; /* deactivate */
1538}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539EXPORT_SYMBOL(generic_make_request);
1540
1541/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001542 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1544 * @bio: The &struct bio which describes the I/O
1545 *
1546 * submit_bio() is very similar in purpose to generic_make_request(), and
1547 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001548 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 *
1550 */
1551void submit_bio(int rw, struct bio *bio)
1552{
1553 int count = bio_sectors(bio);
1554
Jens Axboe22e2c502005-06-27 10:55:12 +02001555 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
Jens Axboebf2de6f2007-09-27 13:01:25 +02001557 /*
1558 * If it's a regular read/write or a barrier with data attached,
1559 * go through the normal accounting stuff before submission.
1560 */
Jens Axboea9c701e2008-08-08 11:04:44 +02001561 if (bio_has_data(bio)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001562 if (rw & WRITE) {
1563 count_vm_events(PGPGOUT, count);
1564 } else {
1565 task_io_account_read(bio->bi_size);
1566 count_vm_events(PGPGIN, count);
1567 }
1568
1569 if (unlikely(block_dump)) {
1570 char b[BDEVNAME_SIZE];
1571 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001572 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001573 (rw & WRITE) ? "WRITE" : "READ",
1574 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001575 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 }
1578
1579 generic_make_request(bio);
1580}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581EXPORT_SYMBOL(submit_bio);
1582
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001583/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001584 * blk_rq_check_limits - Helper function to check a request for the queue limit
1585 * @q: the queue
1586 * @rq: the request being checked
1587 *
1588 * Description:
1589 * @rq may have been made based on weaker limitations of upper-level queues
1590 * in request stacking drivers, and it may violate the limitation of @q.
1591 * Since the block layer and the underlying device driver trust @rq
1592 * after it is inserted to @q, it should be checked against @q before
1593 * the insertion using this generic function.
1594 *
1595 * This function should also be useful for request stacking drivers
1596 * in some cases below, so export this fuction.
1597 * Request stacking drivers like request-based dm may change the queue
1598 * limits while requests are in the queue (e.g. dm's table swapping).
1599 * Such request stacking drivers should check those requests agaist
1600 * the new queue limits again when they dispatch those requests,
1601 * although such checkings are also done against the old queue limits
1602 * when submitting requests.
1603 */
1604int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1605{
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001606 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1607 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001608 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1609 return -EIO;
1610 }
1611
1612 /*
1613 * queue's settings related to segment counting like q->bounce_pfn
1614 * may differ from that of other stacking queues.
1615 * Recalculate it to check the request correctly on this queue's
1616 * limitation.
1617 */
1618 blk_recalc_rq_segments(rq);
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001619 if (rq->nr_phys_segments > queue_max_phys_segments(q) ||
1620 rq->nr_phys_segments > queue_max_hw_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001621 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1622 return -EIO;
1623 }
1624
1625 return 0;
1626}
1627EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1628
1629/**
1630 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1631 * @q: the queue to submit the request
1632 * @rq: the request being queued
1633 */
1634int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1635{
1636 unsigned long flags;
1637
1638 if (blk_rq_check_limits(q, rq))
1639 return -EIO;
1640
1641#ifdef CONFIG_FAIL_MAKE_REQUEST
1642 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1643 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1644 return -EIO;
1645#endif
1646
1647 spin_lock_irqsave(q->queue_lock, flags);
1648
1649 /*
1650 * Submitting request must be dequeued before calling this function
1651 * because it will be linked to another request_queue
1652 */
1653 BUG_ON(blk_queued_rq(rq));
1654
1655 drive_stat_acct(rq, 1);
1656 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1657
1658 spin_unlock_irqrestore(q->queue_lock, flags);
1659
1660 return 0;
1661}
1662EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1663
Tejun Heo80a761f2009-07-03 17:48:17 +09001664/**
1665 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1666 * @rq: request to examine
1667 *
1668 * Description:
1669 * A request could be merge of IOs which require different failure
1670 * handling. This function determines the number of bytes which
1671 * can be failed from the beginning of the request without
1672 * crossing into area which need to be retried further.
1673 *
1674 * Return:
1675 * The number of bytes to fail.
1676 *
1677 * Context:
1678 * queue_lock must be held.
1679 */
1680unsigned int blk_rq_err_bytes(const struct request *rq)
1681{
1682 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1683 unsigned int bytes = 0;
1684 struct bio *bio;
1685
1686 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1687 return blk_rq_bytes(rq);
1688
1689 /*
1690 * Currently the only 'mixing' which can happen is between
1691 * different fastfail types. We can safely fail portions
1692 * which have all the failfast bits that the first one has -
1693 * the ones which are at least as eager to fail as the first
1694 * one.
1695 */
1696 for (bio = rq->bio; bio; bio = bio->bi_next) {
1697 if ((bio->bi_rw & ff) != ff)
1698 break;
1699 bytes += bio->bi_size;
1700 }
1701
1702 /* this could lead to infinite loop */
1703 BUG_ON(blk_rq_bytes(rq) && !bytes);
1704 return bytes;
1705}
1706EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1707
Jens Axboebc58ba92009-01-23 10:54:44 +01001708static void blk_account_io_completion(struct request *req, unsigned int bytes)
1709{
Jens Axboec2553b52009-04-24 08:10:11 +02001710 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001711 const int rw = rq_data_dir(req);
1712 struct hd_struct *part;
1713 int cpu;
1714
1715 cpu = part_stat_lock();
Tejun Heo83096eb2009-05-07 22:24:39 +09001716 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001717 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1718 part_stat_unlock();
1719 }
1720}
1721
1722static void blk_account_io_done(struct request *req)
1723{
Jens Axboebc58ba92009-01-23 10:54:44 +01001724 /*
1725 * Account IO completion. bar_rq isn't accounted as a normal
1726 * IO on queueing nor completion. Accounting the containing
1727 * request is enough.
1728 */
Jens Axboec2553b52009-04-24 08:10:11 +02001729 if (blk_do_io_stat(req) && req != &req->q->bar_rq) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001730 unsigned long duration = jiffies - req->start_time;
1731 const int rw = rq_data_dir(req);
1732 struct hd_struct *part;
1733 int cpu;
1734
1735 cpu = part_stat_lock();
Tejun Heo83096eb2009-05-07 22:24:39 +09001736 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001737
1738 part_stat_inc(cpu, part, ios[rw]);
1739 part_stat_add(cpu, part, ticks[rw], duration);
1740 part_round_stats(cpu, part);
Nikanth Karthikesana9327ca2009-09-11 09:18:54 +02001741 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001742
1743 part_stat_unlock();
1744 }
1745}
1746
Tejun Heo53a08802008-12-03 12:41:26 +01001747/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001748 * blk_peek_request - peek at the top of a request queue
1749 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001750 *
1751 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001752 * Return the request at the top of @q. The returned request
1753 * should be started using blk_start_request() before LLD starts
1754 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001755 *
1756 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001757 * Pointer to the request at the top of @q if available. Null
1758 * otherwise.
1759 *
1760 * Context:
1761 * queue_lock must be held.
1762 */
1763struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001764{
1765 struct request *rq;
1766 int ret;
1767
1768 while ((rq = __elv_next_request(q)) != NULL) {
1769 if (!(rq->cmd_flags & REQ_STARTED)) {
1770 /*
1771 * This is the first time the device driver
1772 * sees this request (possibly after
1773 * requeueing). Notify IO scheduler.
1774 */
1775 if (blk_sorted_rq(rq))
1776 elv_activate_rq(q, rq);
1777
1778 /*
1779 * just mark as started even if we don't start
1780 * it, a request that has been delayed should
1781 * not be passed by new incoming requests
1782 */
1783 rq->cmd_flags |= REQ_STARTED;
1784 trace_block_rq_issue(q, rq);
1785 }
1786
1787 if (!q->boundary_rq || q->boundary_rq == rq) {
1788 q->end_sector = rq_end_sector(rq);
1789 q->boundary_rq = NULL;
1790 }
1791
1792 if (rq->cmd_flags & REQ_DONTPREP)
1793 break;
1794
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001795 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001796 /*
1797 * make sure space for the drain appears we
1798 * know we can do this because max_hw_segments
1799 * has been adjusted to be one fewer than the
1800 * device can handle
1801 */
1802 rq->nr_phys_segments++;
1803 }
1804
1805 if (!q->prep_rq_fn)
1806 break;
1807
1808 ret = q->prep_rq_fn(q, rq);
1809 if (ret == BLKPREP_OK) {
1810 break;
1811 } else if (ret == BLKPREP_DEFER) {
1812 /*
1813 * the request may have been (partially) prepped.
1814 * we need to keep this request in the front to
1815 * avoid resource deadlock. REQ_STARTED will
1816 * prevent other fs requests from passing this one.
1817 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001818 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001819 !(rq->cmd_flags & REQ_DONTPREP)) {
1820 /*
1821 * remove the space for the drain we added
1822 * so that we don't add it again
1823 */
1824 --rq->nr_phys_segments;
1825 }
1826
1827 rq = NULL;
1828 break;
1829 } else if (ret == BLKPREP_KILL) {
1830 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001831 /*
1832 * Mark this request as started so we don't trigger
1833 * any debug logic in the end I/O path.
1834 */
1835 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001836 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001837 } else {
1838 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1839 break;
1840 }
1841 }
1842
1843 return rq;
1844}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001845EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001846
Tejun Heo9934c8c2009-05-08 11:54:16 +09001847void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001848{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001849 struct request_queue *q = rq->q;
1850
Tejun Heo158dbda2009-04-23 11:05:18 +09001851 BUG_ON(list_empty(&rq->queuelist));
1852 BUG_ON(ELV_ON_HASH(rq));
1853
1854 list_del_init(&rq->queuelist);
1855
1856 /*
1857 * the time frame between a request being removed from the lists
1858 * and to it is freed is accounted as io that is in progress at
1859 * the driver side.
1860 */
Jens Axboefb1e7532009-07-30 08:18:24 +02001861 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001862 q->in_flight[rq_is_sync(rq)]++;
Jens Axboefb1e7532009-07-30 08:18:24 +02001863 /*
1864 * Mark this device as supporting hardware queuing, if
1865 * we have more IOs in flight than 4.
1866 */
1867 if (!blk_queue_queuing(q) && queue_in_flight(q) > 4)
1868 set_bit(QUEUE_FLAG_CQ, &q->queue_flags);
1869 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001870}
1871
Tejun Heo5efccd12009-04-23 11:05:18 +09001872/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001873 * blk_start_request - start request processing on the driver
1874 * @req: request to dequeue
1875 *
1876 * Description:
1877 * Dequeue @req and start timeout timer on it. This hands off the
1878 * request to the driver.
1879 *
1880 * Block internal functions which don't want to start timer should
1881 * call blk_dequeue_request().
1882 *
1883 * Context:
1884 * queue_lock must be held.
1885 */
1886void blk_start_request(struct request *req)
1887{
1888 blk_dequeue_request(req);
1889
1890 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001891 * We are now handing the request to the hardware, initialize
1892 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001893 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001894 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001895 if (unlikely(blk_bidi_rq(req)))
1896 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1897
Tejun Heo9934c8c2009-05-08 11:54:16 +09001898 blk_add_timer(req);
1899}
1900EXPORT_SYMBOL(blk_start_request);
1901
1902/**
1903 * blk_fetch_request - fetch a request from a request queue
1904 * @q: request queue to fetch a request from
1905 *
1906 * Description:
1907 * Return the request at the top of @q. The request is started on
1908 * return and LLD can start processing it immediately.
1909 *
1910 * Return:
1911 * Pointer to the request at the top of @q if available. Null
1912 * otherwise.
1913 *
1914 * Context:
1915 * queue_lock must be held.
1916 */
1917struct request *blk_fetch_request(struct request_queue *q)
1918{
1919 struct request *rq;
1920
1921 rq = blk_peek_request(q);
1922 if (rq)
1923 blk_start_request(rq);
1924 return rq;
1925}
1926EXPORT_SYMBOL(blk_fetch_request);
1927
1928/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001929 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001930 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001931 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001932 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001933 *
1934 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001935 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1936 * the request structure even if @req doesn't have leftover.
1937 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09001938 *
1939 * This special helper function is only for request stacking drivers
1940 * (e.g. request-based dm) so that they can handle partial completion.
1941 * Actual device drivers should use blk_end_request instead.
1942 *
1943 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1944 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001945 *
1946 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09001947 * %false - this request doesn't have any more data
1948 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001949 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09001950bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001952 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 struct bio *bio;
1954
Tejun Heo2e60e022009-04-23 11:05:18 +09001955 if (!req->bio)
1956 return false;
1957
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001958 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01001959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 /*
Tejun Heo6f414692009-04-19 07:00:41 +09001961 * For fs requests, rq is just carrier of independent bio's
1962 * and each partial completion should be handled separately.
1963 * Reset per-request error on each partial completion.
1964 *
1965 * TODO: tj: This is too subtle. It would be better to let
1966 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 */
Tejun Heo6f414692009-04-19 07:00:41 +09001968 if (blk_fs_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 req->errors = 0;
1970
Jens Axboe6728cb02008-01-31 13:03:55 +01001971 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1972 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 req->rq_disk ? req->rq_disk->disk_name : "?",
Tejun Heo83096eb2009-05-07 22:24:39 +09001974 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 }
1976
Jens Axboebc58ba92009-01-23 10:54:44 +01001977 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01001978
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 total_bytes = bio_nbytes = 0;
1980 while ((bio = req->bio) != NULL) {
1981 int nbytes;
1982
1983 if (nr_bytes >= bio->bi_size) {
1984 req->bio = bio->bi_next;
1985 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02001986 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 next_idx = 0;
1988 bio_nbytes = 0;
1989 } else {
1990 int idx = bio->bi_idx + next_idx;
1991
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02001992 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01001994 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02001995 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 break;
1997 }
1998
1999 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2000 BIO_BUG_ON(nbytes > bio->bi_size);
2001
2002 /*
2003 * not a complete bvec done
2004 */
2005 if (unlikely(nbytes > nr_bytes)) {
2006 bio_nbytes += nr_bytes;
2007 total_bytes += nr_bytes;
2008 break;
2009 }
2010
2011 /*
2012 * advance to the next vector
2013 */
2014 next_idx++;
2015 bio_nbytes += nbytes;
2016 }
2017
2018 total_bytes += nbytes;
2019 nr_bytes -= nbytes;
2020
Jens Axboe6728cb02008-01-31 13:03:55 +01002021 bio = req->bio;
2022 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 /*
2024 * end more in this run, or just return 'not-done'
2025 */
2026 if (unlikely(nr_bytes <= 0))
2027 break;
2028 }
2029 }
2030
2031 /*
2032 * completely done
2033 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002034 if (!req->bio) {
2035 /*
2036 * Reset counters so that the request stacking driver
2037 * can find how many bytes remain in the request
2038 * later.
2039 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002040 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002041 return false;
2042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
2044 /*
2045 * if the request wasn't completed, update state
2046 */
2047 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002048 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 bio->bi_idx += next_idx;
2050 bio_iovec(bio)->bv_offset += nr_bytes;
2051 bio_iovec(bio)->bv_len -= nr_bytes;
2052 }
2053
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002054 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002055 req->buffer = bio_data(req->bio);
2056
2057 /* update sector only for requests with clear definition of sector */
2058 if (blk_fs_request(req) || blk_discard_rq(req))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002059 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002060
Tejun Heo80a761f2009-07-03 17:48:17 +09002061 /* mixed attributes always follow the first bio */
2062 if (req->cmd_flags & REQ_MIXED_MERGE) {
2063 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2064 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2065 }
2066
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002067 /*
2068 * If total number of sectors is less than the first segment
2069 * size, something has gone terribly wrong.
2070 */
2071 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2072 printk(KERN_ERR "blk: request botched\n");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002073 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002074 }
2075
2076 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002078
Tejun Heo2e60e022009-04-23 11:05:18 +09002079 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080}
Tejun Heo2e60e022009-04-23 11:05:18 +09002081EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082
Tejun Heo2e60e022009-04-23 11:05:18 +09002083static bool blk_update_bidi_request(struct request *rq, int error,
2084 unsigned int nr_bytes,
2085 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002086{
Tejun Heo2e60e022009-04-23 11:05:18 +09002087 if (blk_update_request(rq, error, nr_bytes))
2088 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002089
Tejun Heo2e60e022009-04-23 11:05:18 +09002090 /* Bidi request must be completed as a whole */
2091 if (unlikely(blk_bidi_rq(rq)) &&
2092 blk_update_request(rq->next_rq, error, bidi_bytes))
2093 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002094
Tejun Heo2e60e022009-04-23 11:05:18 +09002095 add_disk_randomness(rq->rq_disk);
2096
2097 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098}
2099
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100/*
2101 * queue lock must be held
2102 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002103static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002105 if (blk_rq_tagged(req))
2106 blk_queue_end_tag(req->q, req);
2107
James Bottomleyba396a62009-05-27 14:17:08 +02002108 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
2110 if (unlikely(laptop_mode) && blk_fs_request(req))
2111 laptop_io_completion();
2112
Mike Andersone78042e2008-10-30 02:16:20 -07002113 blk_delete_timer(req);
2114
Jens Axboebc58ba92009-01-23 10:54:44 +01002115 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002116
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002118 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002119 else {
2120 if (blk_bidi_rq(req))
2121 __blk_put_request(req->next_rq->q, req->next_rq);
2122
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 }
2125}
2126
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002127/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002128 * blk_end_bidi_request - Complete a bidi request
2129 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002130 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002131 * @nr_bytes: number of bytes to complete @rq
2132 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002133 *
2134 * Description:
2135 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002136 * Drivers that supports bidi can safely call this member for any
2137 * type of request, bidi or uni. In the later case @bidi_bytes is
2138 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002139 *
2140 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002141 * %false - we are done with this request
2142 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002143 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002144static bool blk_end_bidi_request(struct request *rq, int error,
2145 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002146{
2147 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002148 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002149
Tejun Heo2e60e022009-04-23 11:05:18 +09002150 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2151 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002152
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002153 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002154 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002155 spin_unlock_irqrestore(q->queue_lock, flags);
2156
Tejun Heo2e60e022009-04-23 11:05:18 +09002157 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002158}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002159
2160/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002161 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2162 * @rq: the request to complete
2163 * @error: %0 for success, < %0 for error
2164 * @nr_bytes: number of bytes to complete @rq
2165 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002166 *
2167 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002168 * Identical to blk_end_bidi_request() except that queue lock is
2169 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002170 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002171 * Return:
2172 * %false - we are done with this request
2173 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002174 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002175static bool __blk_end_bidi_request(struct request *rq, int error,
2176 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002177{
Tejun Heo2e60e022009-04-23 11:05:18 +09002178 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2179 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002180
Tejun Heo2e60e022009-04-23 11:05:18 +09002181 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002182
Tejun Heo2e60e022009-04-23 11:05:18 +09002183 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002184}
2185
2186/**
2187 * blk_end_request - Helper function for drivers to complete the request.
2188 * @rq: the request being processed
2189 * @error: %0 for success, < %0 for error
2190 * @nr_bytes: number of bytes to complete
2191 *
2192 * Description:
2193 * Ends I/O on a number of bytes attached to @rq.
2194 * If @rq has leftover, sets it up for the next range of segments.
2195 *
2196 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002197 * %false - we are done with this request
2198 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002199 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002200bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002201{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002202 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002203}
Jens Axboe56ad1742009-07-28 22:11:24 +02002204EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002205
2206/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002207 * blk_end_request_all - Helper function for drives to finish the request.
2208 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002209 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002210 *
2211 * Description:
2212 * Completely finish @rq.
2213 */
2214void blk_end_request_all(struct request *rq, int error)
2215{
2216 bool pending;
2217 unsigned int bidi_bytes = 0;
2218
2219 if (unlikely(blk_bidi_rq(rq)))
2220 bidi_bytes = blk_rq_bytes(rq->next_rq);
2221
2222 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2223 BUG_ON(pending);
2224}
Jens Axboe56ad1742009-07-28 22:11:24 +02002225EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002226
2227/**
2228 * blk_end_request_cur - Helper function to finish the current request chunk.
2229 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002230 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002231 *
2232 * Description:
2233 * Complete the current consecutively mapped chunk from @rq.
2234 *
2235 * Return:
2236 * %false - we are done with this request
2237 * %true - still buffers pending for this request
2238 */
2239bool blk_end_request_cur(struct request *rq, int error)
2240{
2241 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2242}
Jens Axboe56ad1742009-07-28 22:11:24 +02002243EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002244
2245/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002246 * blk_end_request_err - Finish a request till the next failure boundary.
2247 * @rq: the request to finish till the next failure boundary for
2248 * @error: must be negative errno
2249 *
2250 * Description:
2251 * Complete @rq till the next failure boundary.
2252 *
2253 * Return:
2254 * %false - we are done with this request
2255 * %true - still buffers pending for this request
2256 */
2257bool blk_end_request_err(struct request *rq, int error)
2258{
2259 WARN_ON(error >= 0);
2260 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2261}
2262EXPORT_SYMBOL_GPL(blk_end_request_err);
2263
2264/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002265 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002266 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002267 * @error: %0 for success, < %0 for error
2268 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002269 *
2270 * Description:
2271 * Must be called with queue lock held unlike blk_end_request().
2272 *
2273 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002274 * %false - we are done with this request
2275 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002276 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002277bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002278{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002279 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002280}
Jens Axboe56ad1742009-07-28 22:11:24 +02002281EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002282
2283/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002284 * __blk_end_request_all - Helper function for drives to finish the request.
2285 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002286 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002287 *
2288 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002289 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002290 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002291void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002292{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002293 bool pending;
2294 unsigned int bidi_bytes = 0;
2295
2296 if (unlikely(blk_bidi_rq(rq)))
2297 bidi_bytes = blk_rq_bytes(rq->next_rq);
2298
2299 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2300 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002301}
Jens Axboe56ad1742009-07-28 22:11:24 +02002302EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002303
2304/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002305 * __blk_end_request_cur - Helper function to finish the current request chunk.
2306 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002307 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002308 *
2309 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002310 * Complete the current consecutively mapped chunk from @rq. Must
2311 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002312 *
2313 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002314 * %false - we are done with this request
2315 * %true - still buffers pending for this request
2316 */
2317bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002318{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002319 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002320}
Jens Axboe56ad1742009-07-28 22:11:24 +02002321EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002322
Tejun Heo80a761f2009-07-03 17:48:17 +09002323/**
2324 * __blk_end_request_err - Finish a request till the next failure boundary.
2325 * @rq: the request to finish till the next failure boundary for
2326 * @error: must be negative errno
2327 *
2328 * Description:
2329 * Complete @rq till the next failure boundary. Must be called
2330 * with queue lock held.
2331 *
2332 * Return:
2333 * %false - we are done with this request
2334 * %true - still buffers pending for this request
2335 */
2336bool __blk_end_request_err(struct request *rq, int error)
2337{
2338 WARN_ON(error >= 0);
2339 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2340}
2341EXPORT_SYMBOL_GPL(__blk_end_request_err);
2342
Jens Axboe86db1e22008-01-29 14:53:40 +01002343void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2344 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002346 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2347 rq->cmd_flags |= bio->bi_rw & REQ_RW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
David Woodhousefb2dce82008-08-05 18:01:53 +01002349 if (bio_has_data(bio)) {
2350 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002351 rq->buffer = bio_data(bio);
2352 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002353 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355
NeilBrown66846572007-08-16 13:31:28 +02002356 if (bio->bi_bdev)
2357 rq->rq_disk = bio->bi_bdev->bd_disk;
2358}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002360/**
2361 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2362 * @q : the queue of the device being checked
2363 *
2364 * Description:
2365 * Check if underlying low-level drivers of a device are busy.
2366 * If the drivers want to export their busy state, they must set own
2367 * exporting function using blk_queue_lld_busy() first.
2368 *
2369 * Basically, this function is used only by request stacking drivers
2370 * to stop dispatching requests to underlying devices when underlying
2371 * devices are busy. This behavior helps more I/O merging on the queue
2372 * of the request stacking driver and prevents I/O throughput regression
2373 * on burst I/O load.
2374 *
2375 * Return:
2376 * 0 - Not busy (The request stacking driver should dispatch request)
2377 * 1 - Busy (The request stacking driver should stop dispatching request)
2378 */
2379int blk_lld_busy(struct request_queue *q)
2380{
2381 if (q->lld_busy_fn)
2382 return q->lld_busy_fn(q);
2383
2384 return 0;
2385}
2386EXPORT_SYMBOL_GPL(blk_lld_busy);
2387
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002388/**
2389 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2390 * @rq: the clone request to be cleaned up
2391 *
2392 * Description:
2393 * Free all bios in @rq for a cloned request.
2394 */
2395void blk_rq_unprep_clone(struct request *rq)
2396{
2397 struct bio *bio;
2398
2399 while ((bio = rq->bio) != NULL) {
2400 rq->bio = bio->bi_next;
2401
2402 bio_put(bio);
2403 }
2404}
2405EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2406
2407/*
2408 * Copy attributes of the original request to the clone request.
2409 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2410 */
2411static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2412{
2413 dst->cpu = src->cpu;
2414 dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE);
2415 dst->cmd_type = src->cmd_type;
2416 dst->__sector = blk_rq_pos(src);
2417 dst->__data_len = blk_rq_bytes(src);
2418 dst->nr_phys_segments = src->nr_phys_segments;
2419 dst->ioprio = src->ioprio;
2420 dst->extra_len = src->extra_len;
2421}
2422
2423/**
2424 * blk_rq_prep_clone - Helper function to setup clone request
2425 * @rq: the request to be setup
2426 * @rq_src: original request to be cloned
2427 * @bs: bio_set that bios for clone are allocated from
2428 * @gfp_mask: memory allocation mask for bio
2429 * @bio_ctr: setup function to be called for each clone bio.
2430 * Returns %0 for success, non %0 for failure.
2431 * @data: private data to be passed to @bio_ctr
2432 *
2433 * Description:
2434 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2435 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2436 * are not copied, and copying such parts is the caller's responsibility.
2437 * Also, pages which the original bios are pointing to are not copied
2438 * and the cloned bios just point same pages.
2439 * So cloned bios must be completed before original bios, which means
2440 * the caller must complete @rq before @rq_src.
2441 */
2442int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2443 struct bio_set *bs, gfp_t gfp_mask,
2444 int (*bio_ctr)(struct bio *, struct bio *, void *),
2445 void *data)
2446{
2447 struct bio *bio, *bio_src;
2448
2449 if (!bs)
2450 bs = fs_bio_set;
2451
2452 blk_rq_init(NULL, rq);
2453
2454 __rq_for_each_bio(bio_src, rq_src) {
2455 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2456 if (!bio)
2457 goto free_and_out;
2458
2459 __bio_clone(bio, bio_src);
2460
2461 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002462 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002463 goto free_and_out;
2464
2465 if (bio_ctr && bio_ctr(bio, bio_src, data))
2466 goto free_and_out;
2467
2468 if (rq->bio) {
2469 rq->biotail->bi_next = bio;
2470 rq->biotail = bio;
2471 } else
2472 rq->bio = rq->biotail = bio;
2473 }
2474
2475 __blk_rq_prep_clone(rq, rq_src);
2476
2477 return 0;
2478
2479free_and_out:
2480 if (bio)
2481 bio_free(bio, bs);
2482 blk_rq_unprep_clone(rq);
2483
2484 return -ENOMEM;
2485}
2486EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2487
Jens Axboe18887ad2008-07-28 13:08:45 +02002488int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489{
2490 return queue_work(kblockd_workqueue, work);
2491}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492EXPORT_SYMBOL(kblockd_schedule_work);
2493
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494int __init blk_dev_init(void)
2495{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002496 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2497 sizeof(((struct request *)0)->cmd_flags));
2498
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 kblockd_workqueue = create_workqueue("kblockd");
2500 if (!kblockd_workqueue)
2501 panic("Failed to create kblockd\n");
2502
2503 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002504 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505
Jens Axboe8324aa92008-01-29 14:51:59 +01002506 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002507 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 return 0;
2510}
2511