blob: 9c6f818d0c33aca8f2dc1e088a12ba58dd0f2a08 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Jens Axboe2056a782006-03-23 20:00:26 +010029#include <linux/blktrace_api.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080030#include <linux/fault-inject.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Jens Axboe8324aa92008-01-29 14:51:59 +010032#include "blk.h"
33
Jens Axboe165125e2007-07-24 09:28:11 +020034static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36/*
37 * For the allocated request tables
38 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010039static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41/*
42 * For queue allocation
43 */
Jens Axboe6728cb02008-01-31 13:03:55 +010044struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 * Controlling structure to kblockd
48 */
Jens Axboeff856ba2006-01-09 16:02:34 +010049static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Jens Axboe26b82562008-01-29 13:54:41 +010051static void drive_stat_acct(struct request *rq, int new_io)
52{
Jens Axboe28f13702008-05-07 10:15:46 +020053 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010054 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090055 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010056
57 if (!blk_fs_request(rq) || !rq->rq_disk)
58 return;
59
Tejun Heo074a7ac2008-08-25 19:56:14 +090060 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +020061 part = disk_map_sector_rcu(rq->rq_disk, rq->sector);
Tejun Heoc9959052008-08-25 19:47:21 +090062
Jens Axboe28f13702008-05-07 10:15:46 +020063 if (!new_io)
Tejun Heo074a7ac2008-08-25 19:56:14 +090064 part_stat_inc(cpu, part, merges[rw]);
Jens Axboe28f13702008-05-07 10:15:46 +020065 else {
Tejun Heo074a7ac2008-08-25 19:56:14 +090066 part_round_stats(cpu, part);
67 part_inc_in_flight(part);
Jens Axboe26b82562008-01-29 13:54:41 +010068 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020069
Tejun Heo074a7ac2008-08-25 19:56:14 +090070 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010071}
72
Jens Axboe8324aa92008-01-29 14:51:59 +010073void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
75 int nr;
76
77 nr = q->nr_requests - (q->nr_requests / 8) + 1;
78 if (nr > q->nr_requests)
79 nr = q->nr_requests;
80 q->nr_congestion_on = nr;
81
82 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
83 if (nr < 1)
84 nr = 1;
85 q->nr_congestion_off = nr;
86}
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/**
89 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
90 * @bdev: device
91 *
92 * Locates the passed device's request queue and returns the address of its
93 * backing_dev_info
94 *
95 * Will return NULL if the request queue cannot be located.
96 */
97struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
98{
99 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200100 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102 if (q)
103 ret = &q->backing_dev_info;
104 return ret;
105}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106EXPORT_SYMBOL(blk_get_backing_dev_info);
107
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200108void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200110 memset(rq, 0, sizeof(*rq));
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100113 INIT_LIST_HEAD(&rq->donelist);
Jens Axboe63a71382008-02-08 12:41:03 +0100114 rq->q = q;
115 rq->sector = rq->hard_sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200116 INIT_HLIST_NODE(&rq->hash);
117 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200118 rq->cmd = rq->__cmd;
Jens Axboe63a71382008-02-08 12:41:03 +0100119 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100120 rq->ref_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200122EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
NeilBrown5bb23a62007-09-27 12:46:13 +0200124static void req_bio_endio(struct request *rq, struct bio *bio,
125 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100126{
Jens Axboe165125e2007-07-24 09:28:11 +0200127 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100128
NeilBrown5bb23a62007-09-27 12:46:13 +0200129 if (&q->bar_rq != rq) {
130 if (error)
131 clear_bit(BIO_UPTODATE, &bio->bi_flags);
132 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
133 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100134
NeilBrown5bb23a62007-09-27 12:46:13 +0200135 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100136 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700137 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200138 nbytes = bio->bi_size;
139 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100140
NeilBrown5bb23a62007-09-27 12:46:13 +0200141 bio->bi_size -= nbytes;
142 bio->bi_sector += (nbytes >> 9);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200143
144 if (bio_integrity(bio))
145 bio_integrity_advance(bio, nbytes);
146
NeilBrown5bb23a62007-09-27 12:46:13 +0200147 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200148 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200149 } else {
150
151 /*
152 * Okay, this is the barrier request in progress, just
153 * record the error;
154 */
155 if (error && !q->orderr)
156 q->orderr = error;
157 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100158}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160void blk_dump_rq_flags(struct request *rq, char *msg)
161{
162 int bit;
163
Jens Axboe6728cb02008-01-31 13:03:55 +0100164 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200165 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
166 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
Jens Axboe6728cb02008-01-31 13:03:55 +0100168 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
169 (unsigned long long)rq->sector,
170 rq->nr_sectors,
171 rq->current_nr_sectors);
172 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
173 rq->bio, rq->biotail,
174 rq->buffer, rq->data,
175 rq->data_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Jens Axboe4aff5e22006-08-10 08:44:47 +0200177 if (blk_pc_request(rq)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100178 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200179 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 printk("%02x ", rq->cmd[bit]);
181 printk("\n");
182 }
183}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184EXPORT_SYMBOL(blk_dump_rq_flags);
185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186/*
187 * "plug" the device if there are no outstanding requests: this will
188 * force the transfer to start only after we have put all the requests
189 * on the list.
190 *
191 * This is called with interrupts off and no requests on the queue and
192 * with the queue lock held.
193 */
Jens Axboe165125e2007-07-24 09:28:11 +0200194void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195{
196 WARN_ON(!irqs_disabled());
197
198 /*
199 * don't plug a stopped queue, it must be paired with blk_start_queue()
200 * which will restart the queueing
201 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200202 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 return;
204
Jens Axboee48ec692008-07-03 13:18:54 +0200205 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +0100207 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210EXPORT_SYMBOL(blk_plug_device);
211
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200212/**
213 * blk_plug_device_unlocked - plug a device without queue lock held
214 * @q: The &struct request_queue to plug
215 *
216 * Description:
217 * Like @blk_plug_device(), but grabs the queue lock and disables
218 * interrupts.
219 **/
220void blk_plug_device_unlocked(struct request_queue *q)
221{
222 unsigned long flags;
223
224 spin_lock_irqsave(q->queue_lock, flags);
225 blk_plug_device(q);
226 spin_unlock_irqrestore(q->queue_lock, flags);
227}
228EXPORT_SYMBOL(blk_plug_device_unlocked);
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230/*
231 * remove the queue from the plugged list, if present. called with
232 * queue lock held and interrupts disabled.
233 */
Jens Axboe165125e2007-07-24 09:28:11 +0200234int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
236 WARN_ON(!irqs_disabled());
237
Jens Axboee48ec692008-07-03 13:18:54 +0200238 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 return 0;
240
241 del_timer(&q->unplug_timer);
242 return 1;
243}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244EXPORT_SYMBOL(blk_remove_plug);
245
246/*
247 * remove the plug and let it rip..
248 */
Jens Axboe165125e2007-07-24 09:28:11 +0200249void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200251 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 return;
253
254 if (!blk_remove_plug(q))
255 return;
256
Jens Axboe22e2c502005-06-27 10:55:12 +0200257 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
259EXPORT_SYMBOL(__generic_unplug_device);
260
261/**
262 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200263 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 *
265 * Description:
266 * Linux uses plugging to build bigger requests queues before letting
267 * the device have at them. If a queue is plugged, the I/O scheduler
268 * is still adding and merging requests on the queue. Once the queue
269 * gets unplugged, the request_fn defined for the queue is invoked and
270 * transfers started.
271 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200272void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200274 if (blk_queue_plugged(q)) {
275 spin_lock_irq(q->queue_lock);
276 __generic_unplug_device(q);
277 spin_unlock_irq(q->queue_lock);
278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279}
280EXPORT_SYMBOL(generic_unplug_device);
281
282static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
283 struct page *page)
284{
Jens Axboe165125e2007-07-24 09:28:11 +0200285 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500287 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
Jens Axboe86db1e22008-01-29 14:53:40 +0100290void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291{
Jens Axboe165125e2007-07-24 09:28:11 +0200292 struct request_queue *q =
293 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
Jens Axboe2056a782006-03-23 20:00:26 +0100295 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
296 q->rq.count[READ] + q->rq.count[WRITE]);
297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 q->unplug_fn(q);
299}
300
Jens Axboe86db1e22008-01-29 14:53:40 +0100301void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Jens Axboe165125e2007-07-24 09:28:11 +0200303 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
Jens Axboe2056a782006-03-23 20:00:26 +0100305 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
306 q->rq.count[READ] + q->rq.count[WRITE]);
307
Jens Axboe18887ad2008-07-28 13:08:45 +0200308 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500311void blk_unplug(struct request_queue *q)
312{
313 /*
314 * devices don't necessarily have an ->unplug_fn defined
315 */
316 if (q->unplug_fn) {
317 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
318 q->rq.count[READ] + q->rq.count[WRITE]);
319
320 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
340 /*
341 * one level of recursion is ok and is much faster than kicking
342 * the unplug handling
343 */
Jens Axboee48ec692008-07-03 13:18:54 +0200344 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200346 queue_flag_clear(QUEUE_FLAG_REENTER, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 } else {
348 blk_plug_device(q);
Jens Axboe18887ad2008-07-28 13:08:45 +0200349 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 }
351}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352EXPORT_SYMBOL(blk_start_queue);
353
354/**
355 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200356 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 *
358 * Description:
359 * The Linux block layer assumes that a block driver will consume all
360 * entries on the request queue when the request_fn strategy is called.
361 * Often this will not happen, because of hardware limitations (queue
362 * depth settings). If a device driver gets a 'queue full' response,
363 * or if it simply chooses not to queue more I/O at one point, it can
364 * call this function to prevent the request_fn from being called until
365 * the driver has signalled it's ready to go again. This happens by calling
366 * blk_start_queue() to restart queue operations. Queue lock must be held.
367 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200368void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
370 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200371 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372}
373EXPORT_SYMBOL(blk_stop_queue);
374
375/**
376 * blk_sync_queue - cancel any pending callbacks on a queue
377 * @q: the queue
378 *
379 * Description:
380 * The block layer may perform asynchronous callback activity
381 * on a queue, such as calling the unplug function after a timeout.
382 * A block device may call blk_sync_queue to ensure that any
383 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200384 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 * that its ->make_request_fn will not re-add plugging prior to calling
386 * this function.
387 *
388 */
389void blk_sync_queue(struct request_queue *q)
390{
391 del_timer_sync(&q->unplug_timer);
Oleg Nesterovabbeb882007-10-23 15:08:19 +0200392 kblockd_flush_work(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393}
394EXPORT_SYMBOL(blk_sync_queue);
395
396/**
397 * blk_run_queue - run a single device queue
398 * @q: The queue to run
399 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200400void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200403
404 /*
405 * Only recurse once to avoid overrunning the stack, let the unplug
406 * handling reinvoke the handler shortly if we already got there.
407 */
408 if (!elv_queue_empty(q)) {
Jens Axboee48ec692008-07-03 13:18:54 +0200409 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
Jens Axboedac07ec2006-05-11 08:20:16 +0200410 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200411 queue_flag_clear(QUEUE_FLAG_REENTER, q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200412 } else {
413 blk_plug_device(q);
Jens Axboe18887ad2008-07-28 13:08:45 +0200414 kblockd_schedule_work(q, &q->unplug_work);
Jens Axboedac07ec2006-05-11 08:20:16 +0200415 }
416 }
Nick Piggin75ad23b2008-04-29 14:48:33 +0200417}
418EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200419
Nick Piggin75ad23b2008-04-29 14:48:33 +0200420/**
421 * blk_run_queue - run a single device queue
422 * @q: The queue to run
423 */
424void blk_run_queue(struct request_queue *q)
425{
426 unsigned long flags;
427
428 spin_lock_irqsave(q->queue_lock, flags);
429 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 spin_unlock_irqrestore(q->queue_lock, flags);
431}
432EXPORT_SYMBOL(blk_run_queue);
433
Jens Axboe165125e2007-07-24 09:28:11 +0200434void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500435{
436 kobject_put(&q->kobj);
437}
Al Viro483f4af2006-03-18 18:34:37 -0500438
Jens Axboe6728cb02008-01-31 13:03:55 +0100439void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500440{
441 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200442 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500443 mutex_unlock(&q->sysfs_lock);
444
445 if (q->elevator)
446 elevator_exit(q->elevator);
447
448 blk_put_queue(q);
449}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450EXPORT_SYMBOL(blk_cleanup_queue);
451
Jens Axboe165125e2007-07-24 09:28:11 +0200452static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
454 struct request_list *rl = &q->rq;
455
456 rl->count[READ] = rl->count[WRITE] = 0;
457 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200458 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 init_waitqueue_head(&rl->wait[READ]);
460 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Christoph Lameter19460892005-06-23 00:08:19 -0700462 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
463 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465 if (!rl->rq_pool)
466 return -ENOMEM;
467
468 return 0;
469}
470
Jens Axboe165125e2007-07-24 09:28:11 +0200471struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472{
Christoph Lameter19460892005-06-23 00:08:19 -0700473 return blk_alloc_queue_node(gfp_mask, -1);
474}
475EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
Jens Axboe165125e2007-07-24 09:28:11 +0200477struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700478{
Jens Axboe165125e2007-07-24 09:28:11 +0200479 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700480 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700481
Jens Axboe8324aa92008-01-29 14:51:59 +0100482 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700483 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 if (!q)
485 return NULL;
486
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700487 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
488 q->backing_dev_info.unplug_io_data = q;
489 err = bdi_init(&q->backing_dev_info);
490 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100491 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700492 return NULL;
493 }
494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500496
Jens Axboe8324aa92008-01-29 14:51:59 +0100497 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Al Viro483f4af2006-03-18 18:34:37 -0500499 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700500 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 return q;
503}
Christoph Lameter19460892005-06-23 00:08:19 -0700504EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
506/**
507 * blk_init_queue - prepare a request queue for use with a block device
508 * @rfn: The function to be called to process requests that have been
509 * placed on the queue.
510 * @lock: Request queue spin lock
511 *
512 * Description:
513 * If a block device wishes to use the standard request handling procedures,
514 * which sorts requests and coalesces adjacent requests, then it must
515 * call blk_init_queue(). The function @rfn will be called when there
516 * are requests on the queue that need to be processed. If the device
517 * supports plugging, then @rfn may not be called immediately when requests
518 * are available on the queue, but may be called at some time later instead.
519 * Plugged queues are generally unplugged when a buffer belonging to one
520 * of the requests on the queue is needed, or due to memory pressure.
521 *
522 * @rfn is not required, or even expected, to remove all requests off the
523 * queue, but only as many as it can handle at a time. If it does leave
524 * requests on the queue, it is responsible for arranging that the requests
525 * get dealt with eventually.
526 *
527 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200528 * request queue; this lock will be taken also from interrupt context, so irq
529 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200531 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 * it didn't succeed.
533 *
534 * Note:
535 * blk_init_queue() must be paired with a blk_cleanup_queue() call
536 * when the block device is deactivated (such as at module unload).
537 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700538
Jens Axboe165125e2007-07-24 09:28:11 +0200539struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540{
Christoph Lameter19460892005-06-23 00:08:19 -0700541 return blk_init_queue_node(rfn, lock, -1);
542}
543EXPORT_SYMBOL(blk_init_queue);
544
Jens Axboe165125e2007-07-24 09:28:11 +0200545struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700546blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
547{
Jens Axboe165125e2007-07-24 09:28:11 +0200548 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550 if (!q)
551 return NULL;
552
Christoph Lameter19460892005-06-23 00:08:19 -0700553 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -0500554 if (blk_init_free_list(q)) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100555 kmem_cache_free(blk_requestq_cachep, q);
Al Viro8669aaf2006-03-18 13:50:00 -0500556 return NULL;
557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
152587d2005-04-12 16:22:06 -0500559 /*
560 * if caller didn't supply a lock, they get per-queue locking with
561 * our embedded lock
562 */
Neil Browne7e72bf2008-05-14 16:05:54 -0700563 if (!lock)
152587d2005-04-12 16:22:06 -0500564 lock = &q->__queue_lock;
152587d2005-04-12 16:22:06 -0500565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 q->prep_rq_fn = NULL;
568 q->unplug_fn = generic_unplug_device;
569 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
570 q->queue_lock = lock;
571
572 blk_queue_segment_boundary(q, 0xffffffff);
573
574 blk_queue_make_request(q, __make_request);
575 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
576
577 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
578 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
579
Alan Stern44ec9542007-02-20 11:01:57 -0500580 q->sg_reserved_size = INT_MAX;
581
FUJITA Tomonoriabf54392008-08-16 14:10:05 +0900582 blk_set_cmd_filter_defaults(&q->cmd_filter);
583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 /*
585 * all done
586 */
587 if (!elevator_init(q, NULL)) {
588 blk_queue_congestion_threshold(q);
589 return q;
590 }
591
Al Viro8669aaf2006-03-18 13:50:00 -0500592 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 return NULL;
594}
Christoph Lameter19460892005-06-23 00:08:19 -0700595EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
Jens Axboe165125e2007-07-24 09:28:11 +0200597int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700599 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500600 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 return 0;
602 }
603
604 return 1;
605}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
Jens Axboe165125e2007-07-24 09:28:11 +0200607static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200609 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200610 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 mempool_free(rq, q->rq.rq_pool);
612}
613
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200614static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +0200615blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
617 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
618
619 if (!rq)
620 return NULL;
621
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200622 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200623
Jens Axboe49171e52006-08-10 08:59:11 +0200624 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
Tejun Heocb98fc82005-10-28 08:29:39 +0200626 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200627 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200628 mempool_free(rq, q->rq.rq_pool);
629 return NULL;
630 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200631 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Tejun Heocb98fc82005-10-28 08:29:39 +0200634 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}
636
637/*
638 * ioc_batching returns true if the ioc is a valid batching request and
639 * should be given priority access to a request.
640 */
Jens Axboe165125e2007-07-24 09:28:11 +0200641static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
643 if (!ioc)
644 return 0;
645
646 /*
647 * Make sure the process is able to allocate at least 1 request
648 * even if the batch times out, otherwise we could theoretically
649 * lose wakeups.
650 */
651 return ioc->nr_batch_requests == q->nr_batching ||
652 (ioc->nr_batch_requests > 0
653 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
654}
655
656/*
657 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
658 * will cause the process to be a "batcher" on all queues in the system. This
659 * is the behaviour we want though - once it gets a wakeup it should be given
660 * a nice run.
661 */
Jens Axboe165125e2007-07-24 09:28:11 +0200662static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663{
664 if (!ioc || ioc_batching(q, ioc))
665 return;
666
667 ioc->nr_batch_requests = q->nr_batching;
668 ioc->last_waited = jiffies;
669}
670
Jens Axboe165125e2007-07-24 09:28:11 +0200671static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672{
673 struct request_list *rl = &q->rq;
674
675 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -0700676 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
678 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 if (waitqueue_active(&rl->wait[rw]))
680 wake_up(&rl->wait[rw]);
681
682 blk_clear_queue_full(q, rw);
683 }
684}
685
686/*
687 * A request has just been released. Account for it, update the full and
688 * congestion status, wake up any waiters. Called under q->queue_lock.
689 */
Jens Axboe165125e2007-07-24 09:28:11 +0200690static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
692 struct request_list *rl = &q->rq;
693
694 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200695 if (priv)
696 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698 __freed_request(q, rw);
699
700 if (unlikely(rl->starved[rw ^ 1]))
701 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702}
703
704#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
705/*
Nick Piggind6344532005-06-28 20:45:14 -0700706 * Get a free request, queue_lock must be held.
707 * Returns NULL on failure, with queue_lock held.
708 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 */
Jens Axboe165125e2007-07-24 09:28:11 +0200710static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100711 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712{
713 struct request *rq = NULL;
714 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100715 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +0100716 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100717 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Jens Axboe7749a8d2006-12-13 13:02:26 +0100719 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100720 if (may_queue == ELV_MQUEUE_NO)
721 goto rq_starved;
722
723 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
724 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200725 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100726 /*
727 * The queue will fill after this allocation, so set
728 * it as full, and mark this process as "batching".
729 * This process will be allowed to complete a batch of
730 * requests, others will be blocked.
731 */
732 if (!blk_queue_full(q, rw)) {
733 ioc_set_batching(q, ioc);
734 blk_set_queue_full(q, rw);
735 } else {
736 if (may_queue != ELV_MQUEUE_MUST
737 && !ioc_batching(q, ioc)) {
738 /*
739 * The queue is full and the allocating
740 * process is not a "batcher", and not
741 * exempted by the IO scheduler
742 */
743 goto out;
744 }
745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 }
Thomas Maier79e2de42006-10-19 23:28:15 -0700747 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 }
749
Jens Axboe082cf692005-06-28 16:35:11 +0200750 /*
751 * Only allow batching queuers to allocate up to 50% over the defined
752 * limit of requests, otherwise we could have thousands of requests
753 * allocated with any setting of ->nr_requests
754 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100755 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200756 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100757
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 rl->count[rw]++;
759 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200760
Jens Axboe64521d12005-10-28 08:30:39 +0200761 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200762 if (priv)
763 rl->elvpriv++;
764
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 spin_unlock_irq(q->queue_lock);
766
Jens Axboe7749a8d2006-12-13 13:02:26 +0100767 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100768 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 /*
770 * Allocation failed presumably due to memory. Undo anything
771 * we might have messed up.
772 *
773 * Allocating task should really be put onto the front of the
774 * wait queue, but this is pretty rare.
775 */
776 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +0200777 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
779 /*
780 * in the very unlikely event that allocation failed and no
781 * requests for this direction was pending, mark us starved
782 * so that freeing of a request in the other direction will
783 * notice us. another possible fix would be to split the
784 * rq mempool into READ and WRITE
785 */
786rq_starved:
787 if (unlikely(rl->count[rw] == 0))
788 rl->starved[rw] = 1;
789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 goto out;
791 }
792
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100793 /*
794 * ioc may be NULL here, and ioc_batching will be false. That's
795 * OK, if the queue is under the request limit then requests need
796 * not count toward the nr_batch_requests limit. There will always
797 * be some limit enforced by BLK_BATCH_TIME.
798 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 if (ioc_batching(q, ioc))
800 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100801
Jens Axboe2056a782006-03-23 20:00:26 +0100802 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 return rq;
805}
806
807/*
808 * No available requests for this queue, unplug the device and wait for some
809 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700810 *
811 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 */
Jens Axboe165125e2007-07-24 09:28:11 +0200813static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200814 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
Jens Axboe7749a8d2006-12-13 13:02:26 +0100816 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 struct request *rq;
818
Jens Axboe7749a8d2006-12-13 13:02:26 +0100819 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700820 while (!rq) {
821 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200822 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 struct request_list *rl = &q->rq;
824
825 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
826 TASK_UNINTERRUPTIBLE);
827
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200828 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200830 __generic_unplug_device(q);
831 spin_unlock_irq(q->queue_lock);
832 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200834 /*
835 * After sleeping, we become a "batching" process and
836 * will be able to allocate at least one request, and
837 * up to a big batch of them for a small period time.
838 * See ioc_batching, ioc_set_batching
839 */
840 ioc = current_io_context(GFP_NOIO, q->node);
841 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100842
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200843 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 finish_wait(&rl->wait[rw], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200845
846 rq = get_request(q, rw_flags, bio, GFP_NOIO);
847 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849 return rq;
850}
851
Jens Axboe165125e2007-07-24 09:28:11 +0200852struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853{
854 struct request *rq;
855
856 BUG_ON(rw != READ && rw != WRITE);
857
Nick Piggind6344532005-06-28 20:45:14 -0700858 spin_lock_irq(q->queue_lock);
859 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200860 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700861 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200862 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700863 if (!rq)
864 spin_unlock_irq(q->queue_lock);
865 }
866 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
868 return rq;
869}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870EXPORT_SYMBOL(blk_get_request);
871
872/**
Jens Axboedc72ef42006-07-20 14:54:05 +0200873 * blk_start_queueing - initiate dispatch of requests to device
874 * @q: request queue to kick into gear
875 *
876 * This is basically a helper to remove the need to know whether a queue
877 * is plugged or not if someone just wants to initiate dispatch of requests
878 * for this queue.
879 *
880 * The queue lock must be held with interrupts disabled.
881 */
Jens Axboe165125e2007-07-24 09:28:11 +0200882void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +0200883{
884 if (!blk_queue_plugged(q))
885 q->request_fn(q);
886 else
887 __generic_unplug_device(q);
888}
889EXPORT_SYMBOL(blk_start_queueing);
890
891/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 * blk_requeue_request - put a request back on queue
893 * @q: request queue where request should be inserted
894 * @rq: request to be inserted
895 *
896 * Description:
897 * Drivers often keep queueing requests until the hardware cannot accept
898 * more, when that condition happens we need to put the request back
899 * on the queue. Must be called with queue lock held.
900 */
Jens Axboe165125e2007-07-24 09:28:11 +0200901void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902{
Jens Axboe2056a782006-03-23 20:00:26 +0100903 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
904
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 if (blk_rq_tagged(rq))
906 blk_queue_end_tag(q, rq);
907
908 elv_requeue_request(q, rq);
909}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910EXPORT_SYMBOL(blk_requeue_request);
911
912/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200913 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 * @q: request queue where request should be inserted
915 * @rq: request to be inserted
916 * @at_head: insert request at head or tail of queue
917 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 *
919 * Description:
920 * Many block devices need to execute commands asynchronously, so they don't
921 * block the whole kernel from preemption during request execution. This is
922 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200923 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
924 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 *
926 * We have the option of inserting the head or the tail of the queue.
927 * Typically we use the tail for new ioctls and so forth. We use the head
928 * of the queue for things like a QUEUE_FULL message from a device, or a
929 * host that is unable to accept a particular command.
930 */
Jens Axboe165125e2007-07-24 09:28:11 +0200931void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500932 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933{
Tejun Heo 867d1192005-04-24 02:06:05 -0500934 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 unsigned long flags;
936
937 /*
938 * tell I/O scheduler that this isn't a regular read/write (ie it
939 * must not attempt merges on this) and that it acts as a soft
940 * barrier
941 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200942 rq->cmd_type = REQ_TYPE_SPECIAL;
943 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
945 rq->special = data;
946
947 spin_lock_irqsave(q->queue_lock, flags);
948
949 /*
950 * If command is tagged, release the tag
951 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500952 if (blk_rq_tagged(rq))
953 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200955 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -0500956 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +0200957 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 spin_unlock_irqrestore(q->queue_lock, flags);
959}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960EXPORT_SYMBOL(blk_insert_request);
961
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962/*
963 * add-request adds a request to the linked list.
964 * queue lock is held and interrupts disabled, as we muck with the
965 * request queue list.
966 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100967static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968{
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200969 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 /*
972 * elevator indicated where it wants this request to be
973 * inserted at elevator_merge time
974 */
975 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
976}
Jens Axboe6728cb02008-01-31 13:03:55 +0100977
Tejun Heo074a7ac2008-08-25 19:56:14 +0900978static void part_round_stats_single(int cpu, struct hd_struct *part,
979 unsigned long now)
980{
981 if (now == part->stamp)
982 return;
983
984 if (part->in_flight) {
985 __part_stat_add(cpu, part, time_in_queue,
986 part->in_flight * (now - part->stamp));
987 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
988 }
989 part->stamp = now;
990}
991
992/**
993 * part_round_stats() - Round off the performance stats on a struct
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 * disk_stats.
995 *
996 * The average IO queue length and utilisation statistics are maintained
997 * by observing the current state of the queue length and the amount of
998 * time it has been in this state for.
999 *
1000 * Normally, that accounting is done on IO completion, but that can result
1001 * in more than a second's worth of IO being accounted for within any one
1002 * second, leading to >100% utilisation. To deal with that, we call this
1003 * function to do a round-off before returning the results when reading
1004 * /proc/diskstats. This accounts immediately for all queue usage up to
1005 * the current jiffies and restarts the counters again.
1006 */
Tejun Heoc9959052008-08-25 19:47:21 +09001007void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576af2008-02-08 11:04:35 +01001008{
1009 unsigned long now = jiffies;
1010
Tejun Heo074a7ac2008-08-25 19:56:14 +09001011 if (part->partno)
1012 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1013 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576af2008-02-08 11:04:35 +01001014}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001015EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576af2008-02-08 11:04:35 +01001016
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017/*
1018 * queue lock must be held
1019 */
Jens Axboe165125e2007-07-24 09:28:11 +02001020void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 if (unlikely(!q))
1023 return;
1024 if (unlikely(--req->ref_count))
1025 return;
1026
Tejun Heo8922e162005-10-20 16:23:44 +02001027 elv_completed_request(q, req);
1028
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 /*
1030 * Request may not have originated from ll_rw_blk. if not,
1031 * it didn't come out of our reserved rq pools
1032 */
Jens Axboe49171e52006-08-10 08:59:11 +02001033 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001035 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001038 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02001041 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 }
1043}
Mike Christie6e39b692005-11-11 05:30:24 -06001044EXPORT_SYMBOL_GPL(__blk_put_request);
1045
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046void blk_put_request(struct request *req)
1047{
Tejun Heo8922e162005-10-20 16:23:44 +02001048 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001049 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001051 spin_lock_irqsave(q->queue_lock, flags);
1052 __blk_put_request(q, req);
1053 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055EXPORT_SYMBOL(blk_put_request);
1056
Jens Axboe86db1e22008-01-29 14:53:40 +01001057void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001058{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001059 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001060
1061 /*
1062 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1063 */
1064 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001065 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01001066
1067 /*
1068 * REQ_BARRIER implies no merging, but lets make it explicit
1069 */
David Woodhousefb2dce82008-08-05 18:01:53 +01001070 if (unlikely(bio_discard(bio))) {
David Woodhousee17fc0a2008-08-09 16:42:20 +01001071 req->cmd_flags |= REQ_DISCARD;
1072 if (bio_barrier(bio))
1073 req->cmd_flags |= REQ_SOFTBARRIER;
David Woodhousefb2dce82008-08-05 18:01:53 +01001074 req->q->prepare_discard_fn(req->q, req);
David Woodhousee17fc0a2008-08-09 16:42:20 +01001075 } else if (unlikely(bio_barrier(bio)))
1076 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01001077
Jens Axboeb31dc662006-06-13 08:26:10 +02001078 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001079 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02001080 if (bio_rw_meta(bio))
1081 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02001082
Tejun Heo52d9e672006-01-06 09:49:58 +01001083 req->errors = 0;
1084 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001085 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001086 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02001087 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001088}
1089
Jens Axboe165125e2007-07-24 09:28:11 +02001090static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091{
Nick Piggin450991b2005-06-28 20:45:13 -07001092 struct request *req;
David Woodhousefb2dce82008-08-05 18:01:53 +01001093 int el_ret, nr_sectors, barrier, discard, err;
Jens Axboe51da90f2006-07-18 04:14:45 +02001094 const unsigned short prio = bio_prio(bio);
1095 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01001096 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
1100 /*
1101 * low level driver can indicate that it wants pages above a
1102 * certain limit bounced to low memory (ie for highmem, or even
1103 * ISA dma in theory)
1104 */
1105 blk_queue_bounce(q, &bio);
1106
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 barrier = bio_barrier(bio);
David Woodhousee17fc0a2008-08-09 16:42:20 +01001108 if (unlikely(barrier) && bio_has_data(bio) &&
1109 (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 err = -EOPNOTSUPP;
1111 goto end_io;
1112 }
1113
David Woodhousefb2dce82008-08-05 18:01:53 +01001114 discard = bio_discard(bio);
1115 if (unlikely(discard) && !q->prepare_discard_fn) {
1116 err = -EOPNOTSUPP;
1117 goto end_io;
1118 }
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 spin_lock_irq(q->queue_lock);
1121
Nick Piggin450991b2005-06-28 20:45:13 -07001122 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 goto get_rq;
1124
1125 el_ret = elv_merge(q, &req, bio);
1126 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001127 case ELEVATOR_BACK_MERGE:
1128 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129
Jens Axboe6728cb02008-01-31 13:03:55 +01001130 if (!ll_back_merge_fn(q, req, bio))
1131 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Jens Axboe6728cb02008-01-31 13:03:55 +01001133 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001134
Jens Axboe6728cb02008-01-31 13:03:55 +01001135 req->biotail->bi_next = bio;
1136 req->biotail = bio;
1137 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1138 req->ioprio = ioprio_best(req->ioprio, prio);
1139 drive_stat_acct(req, 0);
1140 if (!attempt_back_merge(q, req))
1141 elv_merged_request(q, req, el_ret);
1142 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Jens Axboe6728cb02008-01-31 13:03:55 +01001144 case ELEVATOR_FRONT_MERGE:
1145 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Jens Axboe6728cb02008-01-31 13:03:55 +01001147 if (!ll_front_merge_fn(q, req, bio))
1148 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Jens Axboe6728cb02008-01-31 13:03:55 +01001150 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001151
Jens Axboe6728cb02008-01-31 13:03:55 +01001152 bio->bi_next = req->bio;
1153 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154
Jens Axboe6728cb02008-01-31 13:03:55 +01001155 /*
1156 * may not be valid. if the low level driver said
1157 * it didn't need a bounce buffer then it better
1158 * not touch req->buffer either...
1159 */
1160 req->buffer = bio_data(bio);
1161 req->current_nr_sectors = bio_cur_sectors(bio);
1162 req->hard_cur_sectors = req->current_nr_sectors;
1163 req->sector = req->hard_sector = bio->bi_sector;
1164 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1165 req->ioprio = ioprio_best(req->ioprio, prio);
1166 drive_stat_acct(req, 0);
1167 if (!attempt_front_merge(q, req))
1168 elv_merged_request(q, req, el_ret);
1169 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
Jens Axboe6728cb02008-01-31 13:03:55 +01001171 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1172 default:
1173 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 }
1175
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001177 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001178 * This sync check and mask will be re-done in init_request_from_bio(),
1179 * but we need to set it earlier to expose the sync flag to the
1180 * rq allocator and io schedulers.
1181 */
1182 rw_flags = bio_data_dir(bio);
1183 if (sync)
1184 rw_flags |= REQ_RW_SYNC;
1185
1186 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001187 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001188 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001189 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001190 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001191
Nick Piggin450991b2005-06-28 20:45:13 -07001192 /*
1193 * After dropping the lock and possibly sleeping here, our request
1194 * may now be mergeable after it had proven unmergeable (above).
1195 * We don't worry about that case for efficiency. It won't happen
1196 * often, and the elevators are able to handle it.
1197 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001198 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
Nick Piggin450991b2005-06-28 20:45:13 -07001200 spin_lock_irq(q->queue_lock);
1201 if (elv_queue_empty(q))
1202 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 add_request(q, req);
1204out:
Jens Axboe4a534f92005-04-16 15:25:40 -07001205 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 __generic_unplug_device(q);
1207
1208 spin_unlock_irq(q->queue_lock);
1209 return 0;
1210
1211end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02001212 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 return 0;
1214}
1215
1216/*
1217 * If bio->bi_dev is a partition, remap the location
1218 */
1219static inline void blk_partition_remap(struct bio *bio)
1220{
1221 struct block_device *bdev = bio->bi_bdev;
1222
Jens Axboebf2de6f2007-09-27 13:01:25 +02001223 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 bio->bi_sector += p->start_sect;
1227 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001228
1229 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
1230 bdev->bd_dev, bio->bi_sector,
1231 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 }
1233}
1234
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235static void handle_bad_sector(struct bio *bio)
1236{
1237 char b[BDEVNAME_SIZE];
1238
1239 printk(KERN_INFO "attempt to access beyond end of device\n");
1240 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1241 bdevname(bio->bi_bdev, b),
1242 bio->bi_rw,
1243 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1244 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1245
1246 set_bit(BIO_EOF, &bio->bi_flags);
1247}
1248
Akinobu Mitac17bb492006-12-08 02:39:46 -08001249#ifdef CONFIG_FAIL_MAKE_REQUEST
1250
1251static DECLARE_FAULT_ATTR(fail_make_request);
1252
1253static int __init setup_fail_make_request(char *str)
1254{
1255 return setup_fault_attr(&fail_make_request, str);
1256}
1257__setup("fail_make_request=", setup_fail_make_request);
1258
1259static int should_fail_request(struct bio *bio)
1260{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001261 struct hd_struct *part = bio->bi_bdev->bd_part;
1262
1263 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001264 return should_fail(&fail_make_request, bio->bi_size);
1265
1266 return 0;
1267}
1268
1269static int __init fail_make_request_debugfs(void)
1270{
1271 return init_fault_attr_dentries(&fail_make_request,
1272 "fail_make_request");
1273}
1274
1275late_initcall(fail_make_request_debugfs);
1276
1277#else /* CONFIG_FAIL_MAKE_REQUEST */
1278
1279static inline int should_fail_request(struct bio *bio)
1280{
1281 return 0;
1282}
1283
1284#endif /* CONFIG_FAIL_MAKE_REQUEST */
1285
Jens Axboec07e2b42007-07-18 13:27:58 +02001286/*
1287 * Check whether this bio extends beyond the end of the device.
1288 */
1289static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1290{
1291 sector_t maxsector;
1292
1293 if (!nr_sectors)
1294 return 0;
1295
1296 /* Test device or partition size, when known. */
1297 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1298 if (maxsector) {
1299 sector_t sector = bio->bi_sector;
1300
1301 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1302 /*
1303 * This may well happen - the kernel calls bread()
1304 * without checking the size of the device, e.g., when
1305 * mounting a device.
1306 */
1307 handle_bad_sector(bio);
1308 return 1;
1309 }
1310 }
1311
1312 return 0;
1313}
1314
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001316 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 * @bio: The bio describing the location in memory and on the device.
1318 *
1319 * generic_make_request() is used to make I/O requests of block
1320 * devices. It is passed a &struct bio, which describes the I/O that needs
1321 * to be done.
1322 *
1323 * generic_make_request() does not return any status. The
1324 * success/failure status of the request, along with notification of
1325 * completion, is delivered asynchronously through the bio->bi_end_io
1326 * function described (one day) else where.
1327 *
1328 * The caller of generic_make_request must make sure that bi_io_vec
1329 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1330 * set to describe the device address, and the
1331 * bi_end_io and optionally bi_private are set to describe how
1332 * completion notification should be signaled.
1333 *
1334 * generic_make_request and the drivers it calls may use bi_next if this
1335 * bio happens to be merged with someone else, and may change bi_dev and
1336 * bi_sector for remaps as it sees fit. So the values of these fields
1337 * should NOT be depended on after the call to generic_make_request.
1338 */
Neil Brownd89d8792007-05-01 09:53:42 +02001339static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340{
Jens Axboe165125e2007-07-24 09:28:11 +02001341 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001342 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001344 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001345 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Jens Axboec07e2b42007-07-18 13:27:58 +02001349 if (bio_check_eod(bio, nr_sectors))
1350 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
1352 /*
1353 * Resolve the mapping until finished. (drivers are
1354 * still free to implement/resolve their own stacking
1355 * by explicitly returning 0)
1356 *
1357 * NOTE: we don't repeat the blk_size check for each new device.
1358 * Stacking drivers are expected to know what they are doing.
1359 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001360 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001361 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 do {
1363 char b[BDEVNAME_SIZE];
1364
1365 q = bdev_get_queue(bio->bi_bdev);
1366 if (!q) {
1367 printk(KERN_ERR
1368 "generic_make_request: Trying to access "
1369 "nonexistent block-device %s (%Lu)\n",
1370 bdevname(bio->bi_bdev, b),
1371 (long long) bio->bi_sector);
1372end_io:
Jens Axboe51fd77b2007-11-02 08:49:08 +01001373 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 break;
1375 }
1376
Jens Axboe4fa253f2007-07-18 13:13:10 +02001377 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001378 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 bdevname(bio->bi_bdev, b),
1380 bio_sectors(bio),
1381 q->max_hw_sectors);
1382 goto end_io;
1383 }
1384
Nick Pigginfde6ad22005-06-23 00:08:53 -07001385 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 goto end_io;
1387
Akinobu Mitac17bb492006-12-08 02:39:46 -08001388 if (should_fail_request(bio))
1389 goto end_io;
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 /*
1392 * If this device has partitions, remap block n
1393 * of partition p to block n+start(p) of the disk.
1394 */
1395 blk_partition_remap(bio);
1396
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001397 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1398 goto end_io;
1399
NeilBrown5ddfe962006-10-30 22:07:21 -08001400 if (old_sector != -1)
Jens Axboe4fa253f2007-07-18 13:13:10 +02001401 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08001402 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001403
1404 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
1405
NeilBrown5ddfe962006-10-30 22:07:21 -08001406 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001407 old_dev = bio->bi_bdev->bd_dev;
1408
Jens Axboec07e2b42007-07-18 13:27:58 +02001409 if (bio_check_eod(bio, nr_sectors))
1410 goto end_io;
David Woodhousefb2dce82008-08-05 18:01:53 +01001411 if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) ||
1412 (bio_discard(bio) && !q->prepare_discard_fn)) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001413 err = -EOPNOTSUPP;
1414 goto end_io;
1415 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001416
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 ret = q->make_request_fn(q, bio);
1418 } while (ret);
1419}
1420
Neil Brownd89d8792007-05-01 09:53:42 +02001421/*
1422 * We only want one ->make_request_fn to be active at a time,
1423 * else stack usage with stacked devices could be a problem.
1424 * So use current->bio_{list,tail} to keep a list of requests
1425 * submited by a make_request_fn function.
1426 * current->bio_tail is also used as a flag to say if
1427 * generic_make_request is currently active in this task or not.
1428 * If it is NULL, then no make_request is active. If it is non-NULL,
1429 * then a make_request is active, and new requests should be added
1430 * at the tail
1431 */
1432void generic_make_request(struct bio *bio)
1433{
1434 if (current->bio_tail) {
1435 /* make_request is active */
1436 *(current->bio_tail) = bio;
1437 bio->bi_next = NULL;
1438 current->bio_tail = &bio->bi_next;
1439 return;
1440 }
1441 /* following loop may be a bit non-obvious, and so deserves some
1442 * explanation.
1443 * Before entering the loop, bio->bi_next is NULL (as all callers
1444 * ensure that) so we have a list with a single bio.
1445 * We pretend that we have just taken it off a longer list, so
1446 * we assign bio_list to the next (which is NULL) and bio_tail
1447 * to &bio_list, thus initialising the bio_list of new bios to be
1448 * added. __generic_make_request may indeed add some more bios
1449 * through a recursive call to generic_make_request. If it
1450 * did, we find a non-NULL value in bio_list and re-enter the loop
1451 * from the top. In this case we really did just take the bio
1452 * of the top of the list (no pretending) and so fixup bio_list and
1453 * bio_tail or bi_next, and call into __generic_make_request again.
1454 *
1455 * The loop was structured like this to make only one call to
1456 * __generic_make_request (which is important as it is large and
1457 * inlined) and to keep the structure simple.
1458 */
1459 BUG_ON(bio->bi_next);
1460 do {
1461 current->bio_list = bio->bi_next;
1462 if (bio->bi_next == NULL)
1463 current->bio_tail = &current->bio_list;
1464 else
1465 bio->bi_next = NULL;
1466 __generic_make_request(bio);
1467 bio = current->bio_list;
1468 } while (bio);
1469 current->bio_tail = NULL; /* deactivate */
1470}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471EXPORT_SYMBOL(generic_make_request);
1472
1473/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001474 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1476 * @bio: The &struct bio which describes the I/O
1477 *
1478 * submit_bio() is very similar in purpose to generic_make_request(), and
1479 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001480 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 *
1482 */
1483void submit_bio(int rw, struct bio *bio)
1484{
1485 int count = bio_sectors(bio);
1486
Jens Axboe22e2c502005-06-27 10:55:12 +02001487 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
Jens Axboebf2de6f2007-09-27 13:01:25 +02001489 /*
1490 * If it's a regular read/write or a barrier with data attached,
1491 * go through the normal accounting stuff before submission.
1492 */
Jens Axboea9c701e2008-08-08 11:04:44 +02001493 if (bio_has_data(bio)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001494 if (rw & WRITE) {
1495 count_vm_events(PGPGOUT, count);
1496 } else {
1497 task_io_account_read(bio->bi_size);
1498 count_vm_events(PGPGIN, count);
1499 }
1500
1501 if (unlikely(block_dump)) {
1502 char b[BDEVNAME_SIZE];
1503 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001504 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001505 (rw & WRITE) ? "WRITE" : "READ",
1506 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001507 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 }
1510
1511 generic_make_request(bio);
1512}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513EXPORT_SYMBOL(submit_bio);
1514
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001515/**
1516 * __end_that_request_first - end I/O on a request
1517 * @req: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001518 * @error: %0 for success, < %0 for error
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001519 * @nr_bytes: number of bytes to complete
1520 *
1521 * Description:
1522 * Ends I/O on a number of bytes attached to @req, and sets it up
1523 * for the next range of segments (if any) in the cluster.
1524 *
1525 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001526 * %0 - we are done with this request, call end_that_request_last()
1527 * %1 - still buffers pending for this request
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001528 **/
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001529static int __end_that_request_first(struct request *req, int error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 int nr_bytes)
1531{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001532 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 struct bio *bio;
1534
Jens Axboe2056a782006-03-23 20:00:26 +01001535 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
1536
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 /*
Randy Dunlap710027a2008-08-19 20:13:11 +02001538 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 * sense key with us all the way through
1540 */
1541 if (!blk_pc_request(req))
1542 req->errors = 0;
1543
Jens Axboe6728cb02008-01-31 13:03:55 +01001544 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1545 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 req->rq_disk ? req->rq_disk->disk_name : "?",
1547 (unsigned long long)req->sector);
1548 }
1549
Jens Axboed72d9042005-11-01 08:35:42 +01001550 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01001551 const int rw = rq_data_dir(req);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001552 struct hd_struct *part;
Tejun Heoc9959052008-08-25 19:47:21 +09001553 int cpu;
Jens Axboea3623572005-11-01 09:26:16 +01001554
Tejun Heo074a7ac2008-08-25 19:56:14 +09001555 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001556 part = disk_map_sector_rcu(req->rq_disk, req->sector);
Tejun Heo074a7ac2008-08-25 19:56:14 +09001557 part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9);
1558 part_stat_unlock();
Jens Axboed72d9042005-11-01 08:35:42 +01001559 }
1560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 total_bytes = bio_nbytes = 0;
1562 while ((bio = req->bio) != NULL) {
1563 int nbytes;
1564
Jens Axboebf2de6f2007-09-27 13:01:25 +02001565 /*
1566 * For an empty barrier request, the low level driver must
1567 * store a potential error location in ->sector. We pass
1568 * that back up in ->bi_sector.
1569 */
1570 if (blk_empty_barrier(req))
1571 bio->bi_sector = req->sector;
1572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 if (nr_bytes >= bio->bi_size) {
1574 req->bio = bio->bi_next;
1575 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02001576 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 next_idx = 0;
1578 bio_nbytes = 0;
1579 } else {
1580 int idx = bio->bi_idx + next_idx;
1581
1582 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1583 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01001584 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -07001585 __func__, bio->bi_idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 break;
1587 }
1588
1589 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1590 BIO_BUG_ON(nbytes > bio->bi_size);
1591
1592 /*
1593 * not a complete bvec done
1594 */
1595 if (unlikely(nbytes > nr_bytes)) {
1596 bio_nbytes += nr_bytes;
1597 total_bytes += nr_bytes;
1598 break;
1599 }
1600
1601 /*
1602 * advance to the next vector
1603 */
1604 next_idx++;
1605 bio_nbytes += nbytes;
1606 }
1607
1608 total_bytes += nbytes;
1609 nr_bytes -= nbytes;
1610
Jens Axboe6728cb02008-01-31 13:03:55 +01001611 bio = req->bio;
1612 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 /*
1614 * end more in this run, or just return 'not-done'
1615 */
1616 if (unlikely(nr_bytes <= 0))
1617 break;
1618 }
1619 }
1620
1621 /*
1622 * completely done
1623 */
1624 if (!req->bio)
1625 return 0;
1626
1627 /*
1628 * if the request wasn't completed, update state
1629 */
1630 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02001631 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 bio->bi_idx += next_idx;
1633 bio_iovec(bio)->bv_offset += nr_bytes;
1634 bio_iovec(bio)->bv_len -= nr_bytes;
1635 }
1636
1637 blk_recalc_rq_sectors(req, total_bytes >> 9);
1638 blk_recalc_rq_segments(req);
1639 return 1;
1640}
1641
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642/*
1643 * queue lock must be held
1644 */
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001645static void end_that_request_last(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646{
1647 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01001648
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001649 if (blk_rq_tagged(req))
1650 blk_queue_end_tag(req->q, req);
1651
1652 if (blk_queued_rq(req))
1653 blkdev_dequeue_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654
1655 if (unlikely(laptop_mode) && blk_fs_request(req))
1656 laptop_io_completion();
1657
Jens Axboefd0ff8a2006-05-23 11:23:49 +02001658 /*
1659 * Account IO completion. bar_rq isn't accounted as a normal
1660 * IO on queueing nor completion. Accounting the containing
1661 * request is enough.
1662 */
1663 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01001665 const int rw = rq_data_dir(req);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001666 struct hd_struct *part;
Tejun Heoc9959052008-08-25 19:47:21 +09001667 int cpu;
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001668
Tejun Heo074a7ac2008-08-25 19:56:14 +09001669 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001670 part = disk_map_sector_rcu(disk, req->sector);
Jens Axboea3623572005-11-01 09:26:16 +01001671
Tejun Heo074a7ac2008-08-25 19:56:14 +09001672 part_stat_inc(cpu, part, ios[rw]);
1673 part_stat_add(cpu, part, ticks[rw], duration);
1674 part_round_stats(cpu, part);
1675 part_dec_in_flight(part);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001676
Tejun Heo074a7ac2008-08-25 19:56:14 +09001677 part_stat_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 }
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001679
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01001681 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001682 else {
1683 if (blk_bidi_rq(req))
1684 __blk_put_request(req->next_rq->q, req->next_rq);
1685
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 }
1688}
1689
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690static inline void __end_request(struct request *rq, int uptodate,
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001691 unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001693 int error = 0;
1694
1695 if (uptodate <= 0)
1696 error = uptodate ? uptodate : -EIO;
1697
1698 __blk_end_request(rq, error, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699}
1700
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001701/**
1702 * blk_rq_bytes - Returns bytes left to complete in the entire request
Randy Dunlap5d87a052008-02-20 09:01:22 +01001703 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001704 **/
1705unsigned int blk_rq_bytes(struct request *rq)
Jens Axboea0cd1282007-09-21 10:41:07 +02001706{
1707 if (blk_fs_request(rq))
1708 return rq->hard_nr_sectors << 9;
1709
1710 return rq->data_len;
1711}
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001712EXPORT_SYMBOL_GPL(blk_rq_bytes);
1713
1714/**
1715 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
Randy Dunlap5d87a052008-02-20 09:01:22 +01001716 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001717 **/
1718unsigned int blk_rq_cur_bytes(struct request *rq)
1719{
1720 if (blk_fs_request(rq))
1721 return rq->current_nr_sectors << 9;
1722
1723 if (rq->bio)
1724 return rq->bio->bi_size;
1725
1726 return rq->data_len;
1727}
1728EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
Jens Axboea0cd1282007-09-21 10:41:07 +02001729
1730/**
1731 * end_queued_request - end all I/O on a queued request
1732 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001733 * @uptodate: error value or %0/%1 uptodate flag
Jens Axboea0cd1282007-09-21 10:41:07 +02001734 *
1735 * Description:
1736 * Ends all I/O on a request, and removes it from the block layer queues.
Randy Dunlap710027a2008-08-19 20:13:11 +02001737 * Not suitable for normal I/O completion, unless the driver still has
Jens Axboea0cd1282007-09-21 10:41:07 +02001738 * the request attached to the block layer.
1739 *
1740 **/
1741void end_queued_request(struct request *rq, int uptodate)
1742{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001743 __end_request(rq, uptodate, blk_rq_bytes(rq));
Jens Axboea0cd1282007-09-21 10:41:07 +02001744}
1745EXPORT_SYMBOL(end_queued_request);
1746
1747/**
1748 * end_dequeued_request - end all I/O on a dequeued request
1749 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001750 * @uptodate: error value or %0/%1 uptodate flag
Jens Axboea0cd1282007-09-21 10:41:07 +02001751 *
1752 * Description:
1753 * Ends all I/O on a request. The request must already have been
1754 * dequeued using blkdev_dequeue_request(), as is normally the case
1755 * for most drivers.
1756 *
1757 **/
1758void end_dequeued_request(struct request *rq, int uptodate)
1759{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001760 __end_request(rq, uptodate, blk_rq_bytes(rq));
Jens Axboea0cd1282007-09-21 10:41:07 +02001761}
1762EXPORT_SYMBOL(end_dequeued_request);
1763
1764
1765/**
1766 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07001767 * @req: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001768 * @uptodate: error value or %0/%1 uptodate flag
Jens Axboea0cd1282007-09-21 10:41:07 +02001769 *
1770 * Description:
1771 * Ends I/O on the current segment of a request. If that is the only
1772 * remaining segment, the request is also completed and freed.
1773 *
Randy Dunlap710027a2008-08-19 20:13:11 +02001774 * This is a remnant of how older block drivers handled I/O completions.
1775 * Modern drivers typically end I/O on the full request in one go, unless
Jens Axboea0cd1282007-09-21 10:41:07 +02001776 * they have a residual value to account for. For that case this function
1777 * isn't really useful, unless the residual just happens to be the
1778 * full current segment. In other words, don't use this function in new
1779 * code. Either use end_request_completely(), or the
1780 * end_that_request_chunk() (along with end_that_request_last()) for
1781 * partial completions.
1782 *
1783 **/
1784void end_request(struct request *req, int uptodate)
1785{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001786 __end_request(req, uptodate, req->hard_cur_sectors << 9);
Jens Axboea0cd1282007-09-21 10:41:07 +02001787}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788EXPORT_SYMBOL(end_request);
1789
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001790/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001791 * blk_end_io - Generic end_io function to complete a request.
1792 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001793 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001794 * @nr_bytes: number of bytes to complete @rq
1795 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001796 * @drv_callback: function called between completion of bios in the request
1797 * and completion of the request.
Randy Dunlap710027a2008-08-19 20:13:11 +02001798 * If the callback returns non %0, this helper returns without
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001799 * completion of the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001800 *
1801 * Description:
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001802 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001803 * If @rq has leftover, sets it up for the next range of segments.
1804 *
1805 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001806 * %0 - we are done with this request
1807 * %1 - this request is not freed yet, it still has pending buffers.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001808 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001809static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1810 unsigned int bidi_bytes,
1811 int (drv_callback)(struct request *))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001812{
1813 struct request_queue *q = rq->q;
1814 unsigned long flags = 0UL;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001815
David Woodhousefb2dce82008-08-05 18:01:53 +01001816 if (bio_has_data(rq->bio) || blk_discard_rq(rq)) {
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001817 if (__end_that_request_first(rq, error, nr_bytes))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001818 return 1;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001819
1820 /* Bidi request must be completed as a whole */
1821 if (blk_bidi_rq(rq) &&
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001822 __end_that_request_first(rq->next_rq, error, bidi_bytes))
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001823 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001824 }
1825
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001826 /* Special feature for tricky drivers */
1827 if (drv_callback && drv_callback(rq))
1828 return 1;
1829
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001830 add_disk_randomness(rq->rq_disk);
1831
1832 spin_lock_irqsave(q->queue_lock, flags);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001833 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001834 spin_unlock_irqrestore(q->queue_lock, flags);
1835
1836 return 0;
1837}
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001838
1839/**
1840 * blk_end_request - Helper function for drivers to complete the request.
1841 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001842 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001843 * @nr_bytes: number of bytes to complete
1844 *
1845 * Description:
1846 * Ends I/O on a number of bytes attached to @rq.
1847 * If @rq has leftover, sets it up for the next range of segments.
1848 *
1849 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001850 * %0 - we are done with this request
1851 * %1 - still buffers pending for this request
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001852 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001853int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001854{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001855 return blk_end_io(rq, error, nr_bytes, 0, NULL);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001856}
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001857EXPORT_SYMBOL_GPL(blk_end_request);
1858
1859/**
1860 * __blk_end_request - Helper function for drivers to complete the request.
1861 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001862 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001863 * @nr_bytes: number of bytes to complete
1864 *
1865 * Description:
1866 * Must be called with queue lock held unlike blk_end_request().
1867 *
1868 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001869 * %0 - we are done with this request
1870 * %1 - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001871 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001872int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001873{
David Woodhousefb2dce82008-08-05 18:01:53 +01001874 if ((bio_has_data(rq->bio) || blk_discard_rq(rq)) &&
Jens Axboe051cc392008-08-08 11:06:45 +02001875 __end_that_request_first(rq, error, nr_bytes))
1876 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001877
1878 add_disk_randomness(rq->rq_disk);
1879
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001880 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001881
1882 return 0;
1883}
1884EXPORT_SYMBOL_GPL(__blk_end_request);
1885
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001886/**
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001887 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1888 * @rq: the bidi request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001889 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001890 * @nr_bytes: number of bytes to complete @rq
1891 * @bidi_bytes: number of bytes to complete @rq->next_rq
1892 *
1893 * Description:
1894 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1895 *
1896 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001897 * %0 - we are done with this request
1898 * %1 - still buffers pending for this request
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001899 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001900int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
1901 unsigned int bidi_bytes)
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001902{
1903 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
1904}
1905EXPORT_SYMBOL_GPL(blk_end_bidi_request);
1906
1907/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001908 * blk_end_request_callback - Special helper function for tricky drivers
1909 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001910 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001911 * @nr_bytes: number of bytes to complete
1912 * @drv_callback: function called between completion of bios in the request
1913 * and completion of the request.
Randy Dunlap710027a2008-08-19 20:13:11 +02001914 * If the callback returns non %0, this helper returns without
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001915 * completion of the request.
1916 *
1917 * Description:
1918 * Ends I/O on a number of bytes attached to @rq.
1919 * If @rq has leftover, sets it up for the next range of segments.
1920 *
1921 * This special helper function is used only for existing tricky drivers.
1922 * (e.g. cdrom_newpc_intr() of ide-cd)
1923 * This interface will be removed when such drivers are rewritten.
1924 * Don't use this interface in other places anymore.
1925 *
1926 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001927 * %0 - we are done with this request
1928 * %1 - this request is not freed yet.
1929 * this request still has pending buffers or
1930 * the driver doesn't want to finish this request yet.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001931 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001932int blk_end_request_callback(struct request *rq, int error,
1933 unsigned int nr_bytes,
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001934 int (drv_callback)(struct request *))
1935{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001936 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001937}
1938EXPORT_SYMBOL_GPL(blk_end_request_callback);
1939
Jens Axboe86db1e22008-01-29 14:53:40 +01001940void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
1941 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942{
David Woodhoused628eae2008-08-09 16:22:17 +01001943 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
1944 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
Jens Axboe4aff5e22006-08-10 08:44:47 +02001945 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946
David Woodhousefb2dce82008-08-05 18:01:53 +01001947 if (bio_has_data(bio)) {
1948 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01001949 rq->buffer = bio_data(bio);
1950 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 rq->current_nr_sectors = bio_cur_sectors(bio);
1952 rq->hard_cur_sectors = rq->current_nr_sectors;
1953 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01001954 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
1956 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
NeilBrown66846572007-08-16 13:31:28 +02001958 if (bio->bi_bdev)
1959 rq->rq_disk = bio->bi_bdev->bd_disk;
1960}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961
Jens Axboe18887ad2008-07-28 13:08:45 +02001962int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963{
1964 return queue_work(kblockd_workqueue, work);
1965}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966EXPORT_SYMBOL(kblockd_schedule_work);
1967
Andrew Morton19a75d82007-05-09 02:33:56 -07001968void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07001970 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971}
Andrew Morton19a75d82007-05-09 02:33:56 -07001972EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
1974int __init blk_dev_init(void)
1975{
1976 kblockd_workqueue = create_workqueue("kblockd");
1977 if (!kblockd_workqueue)
1978 panic("Failed to create kblockd\n");
1979
1980 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09001981 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
Jens Axboe8324aa92008-01-29 14:51:59 +01001983 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02001984 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 return 0;
1987}
1988