blob: 2d053b584410255c39a2209aa90aee633083e7a0 [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 Axboe242f9dc2008-09-14 05:55:09 -0700113 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200114 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100115 rq->q = q;
116 rq->sector = rq->hard_sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200117 INIT_HLIST_NODE(&rq->hash);
118 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200119 rq->cmd = rq->__cmd;
Jens Axboe63a71382008-02-08 12:41:03 +0100120 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100121 rq->ref_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200123EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
NeilBrown5bb23a62007-09-27 12:46:13 +0200125static void req_bio_endio(struct request *rq, struct bio *bio,
126 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100127{
Jens Axboe165125e2007-07-24 09:28:11 +0200128 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100129
NeilBrown5bb23a62007-09-27 12:46:13 +0200130 if (&q->bar_rq != rq) {
131 if (error)
132 clear_bit(BIO_UPTODATE, &bio->bi_flags);
133 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
134 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100135
NeilBrown5bb23a62007-09-27 12:46:13 +0200136 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100137 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700138 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200139 nbytes = bio->bi_size;
140 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100141
NeilBrown5bb23a62007-09-27 12:46:13 +0200142 bio->bi_size -= nbytes;
143 bio->bi_sector += (nbytes >> 9);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200144
145 if (bio_integrity(bio))
146 bio_integrity_advance(bio, nbytes);
147
NeilBrown5bb23a62007-09-27 12:46:13 +0200148 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200149 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200150 } else {
151
152 /*
153 * Okay, this is the barrier request in progress, just
154 * record the error;
155 */
156 if (error && !q->orderr)
157 q->orderr = error;
158 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100159}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161void blk_dump_rq_flags(struct request *rq, char *msg)
162{
163 int bit;
164
Jens Axboe6728cb02008-01-31 13:03:55 +0100165 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200166 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
167 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Jens Axboe6728cb02008-01-31 13:03:55 +0100169 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
170 (unsigned long long)rq->sector,
171 rq->nr_sectors,
172 rq->current_nr_sectors);
173 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
174 rq->bio, rq->biotail,
175 rq->buffer, rq->data,
176 rq->data_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Jens Axboe4aff5e22006-08-10 08:44:47 +0200178 if (blk_pc_request(rq)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100179 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200180 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 printk("%02x ", rq->cmd[bit]);
182 printk("\n");
183 }
184}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185EXPORT_SYMBOL(blk_dump_rq_flags);
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187/*
188 * "plug" the device if there are no outstanding requests: this will
189 * force the transfer to start only after we have put all the requests
190 * on the list.
191 *
192 * This is called with interrupts off and no requests on the queue and
193 * with the queue lock held.
194 */
Jens Axboe165125e2007-07-24 09:28:11 +0200195void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
197 WARN_ON(!irqs_disabled());
198
199 /*
200 * don't plug a stopped queue, it must be paired with blk_start_queue()
201 * which will restart the queueing
202 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200203 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 return;
205
Jens Axboee48ec692008-07-03 13:18:54 +0200206 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +0100208 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
209 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211EXPORT_SYMBOL(blk_plug_device);
212
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200213/**
214 * blk_plug_device_unlocked - plug a device without queue lock held
215 * @q: The &struct request_queue to plug
216 *
217 * Description:
218 * Like @blk_plug_device(), but grabs the queue lock and disables
219 * interrupts.
220 **/
221void blk_plug_device_unlocked(struct request_queue *q)
222{
223 unsigned long flags;
224
225 spin_lock_irqsave(q->queue_lock, flags);
226 blk_plug_device(q);
227 spin_unlock_irqrestore(q->queue_lock, flags);
228}
229EXPORT_SYMBOL(blk_plug_device_unlocked);
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231/*
232 * remove the queue from the plugged list, if present. called with
233 * queue lock held and interrupts disabled.
234 */
Jens Axboe165125e2007-07-24 09:28:11 +0200235int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
237 WARN_ON(!irqs_disabled());
238
Jens Axboee48ec692008-07-03 13:18:54 +0200239 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 return 0;
241
242 del_timer(&q->unplug_timer);
243 return 1;
244}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245EXPORT_SYMBOL(blk_remove_plug);
246
247/*
248 * remove the plug and let it rip..
249 */
Jens Axboe165125e2007-07-24 09:28:11 +0200250void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200252 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 return;
254
255 if (!blk_remove_plug(q))
256 return;
257
Jens Axboe22e2c502005-06-27 10:55:12 +0200258 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
260EXPORT_SYMBOL(__generic_unplug_device);
261
262/**
263 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200264 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 *
266 * Description:
267 * Linux uses plugging to build bigger requests queues before letting
268 * the device have at them. If a queue is plugged, the I/O scheduler
269 * is still adding and merging requests on the queue. Once the queue
270 * gets unplugged, the request_fn defined for the queue is invoked and
271 * transfers started.
272 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200273void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200275 if (blk_queue_plugged(q)) {
276 spin_lock_irq(q->queue_lock);
277 __generic_unplug_device(q);
278 spin_unlock_irq(q->queue_lock);
279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280}
281EXPORT_SYMBOL(generic_unplug_device);
282
283static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
284 struct page *page)
285{
Jens Axboe165125e2007-07-24 09:28:11 +0200286 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500288 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290
Jens Axboe86db1e22008-01-29 14:53:40 +0100291void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292{
Jens Axboe165125e2007-07-24 09:28:11 +0200293 struct request_queue *q =
294 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
Jens Axboe2056a782006-03-23 20:00:26 +0100296 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
297 q->rq.count[READ] + q->rq.count[WRITE]);
298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 q->unplug_fn(q);
300}
301
Jens Axboe86db1e22008-01-29 14:53:40 +0100302void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
Jens Axboe165125e2007-07-24 09:28:11 +0200304 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
Jens Axboe2056a782006-03-23 20:00:26 +0100306 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
307 q->rq.count[READ] + q->rq.count[WRITE]);
308
Jens Axboe18887ad2008-07-28 13:08:45 +0200309 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310}
311
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500312void blk_unplug(struct request_queue *q)
313{
314 /*
315 * devices don't necessarily have an ->unplug_fn defined
316 */
317 if (q->unplug_fn) {
318 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
319 q->rq.count[READ] + q->rq.count[WRITE]);
320
321 q->unplug_fn(q);
322 }
323}
324EXPORT_SYMBOL(blk_unplug);
325
Jens Axboec7c22e42008-09-13 20:26:01 +0200326static void blk_invoke_request_fn(struct request_queue *q)
327{
328 /*
329 * one level of recursion is ok and is much faster than kicking
330 * the unplug handling
331 */
332 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
333 q->request_fn(q);
334 queue_flag_clear(QUEUE_FLAG_REENTER, q);
335 } else {
336 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
337 kblockd_schedule_work(q, &q->unplug_work);
338 }
339}
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341/**
342 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200343 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 *
345 * Description:
346 * blk_start_queue() will clear the stop flag on the queue, and call
347 * the request_fn for the queue if it was in a stopped state when
348 * entered. Also see blk_stop_queue(). Queue lock must be held.
349 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200350void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200352 WARN_ON(!irqs_disabled());
353
Nick Piggin75ad23b2008-04-29 14:48:33 +0200354 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Jens Axboec7c22e42008-09-13 20:26:01 +0200355 blk_invoke_request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357EXPORT_SYMBOL(blk_start_queue);
358
359/**
360 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200361 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 *
363 * Description:
364 * The Linux block layer assumes that a block driver will consume all
365 * entries on the request queue when the request_fn strategy is called.
366 * Often this will not happen, because of hardware limitations (queue
367 * depth settings). If a device driver gets a 'queue full' response,
368 * or if it simply chooses not to queue more I/O at one point, it can
369 * call this function to prevent the request_fn from being called until
370 * the driver has signalled it's ready to go again. This happens by calling
371 * blk_start_queue() to restart queue operations. Queue lock must be held.
372 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200373void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
375 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200376 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
378EXPORT_SYMBOL(blk_stop_queue);
379
380/**
381 * blk_sync_queue - cancel any pending callbacks on a queue
382 * @q: the queue
383 *
384 * Description:
385 * The block layer may perform asynchronous callback activity
386 * on a queue, such as calling the unplug function after a timeout.
387 * A block device may call blk_sync_queue to ensure that any
388 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200389 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 * that its ->make_request_fn will not re-add plugging prior to calling
391 * this function.
392 *
393 */
394void blk_sync_queue(struct request_queue *q)
395{
396 del_timer_sync(&q->unplug_timer);
Oleg Nesterovabbeb882007-10-23 15:08:19 +0200397 kblockd_flush_work(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
399EXPORT_SYMBOL(blk_sync_queue);
400
401/**
402 * blk_run_queue - run a single device queue
403 * @q: The queue to run
404 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200405void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200408
409 /*
410 * Only recurse once to avoid overrunning the stack, let the unplug
411 * handling reinvoke the handler shortly if we already got there.
412 */
Jens Axboec7c22e42008-09-13 20:26:01 +0200413 if (!elv_queue_empty(q))
414 blk_invoke_request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200415}
416EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200417
Nick Piggin75ad23b2008-04-29 14:48:33 +0200418/**
419 * blk_run_queue - run a single device queue
420 * @q: The queue to run
421 */
422void blk_run_queue(struct request_queue *q)
423{
424 unsigned long flags;
425
426 spin_lock_irqsave(q->queue_lock, flags);
427 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 spin_unlock_irqrestore(q->queue_lock, flags);
429}
430EXPORT_SYMBOL(blk_run_queue);
431
Jens Axboe165125e2007-07-24 09:28:11 +0200432void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500433{
434 kobject_put(&q->kobj);
435}
Al Viro483f4af2006-03-18 18:34:37 -0500436
Jens Axboe6728cb02008-01-31 13:03:55 +0100437void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500438{
Jens Axboee3335de92008-09-18 09:22:54 -0700439 /*
440 * We know we have process context here, so we can be a little
441 * cautious and ensure that pending block actions on this device
442 * are done before moving on. Going into this function, we should
443 * not have processes doing IO to this device.
444 */
445 blk_sync_queue(q);
446
Al Viro483f4af2006-03-18 18:34:37 -0500447 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200448 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500449 mutex_unlock(&q->sysfs_lock);
450
451 if (q->elevator)
452 elevator_exit(q->elevator);
453
454 blk_put_queue(q);
455}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456EXPORT_SYMBOL(blk_cleanup_queue);
457
Jens Axboe165125e2007-07-24 09:28:11 +0200458static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459{
460 struct request_list *rl = &q->rq;
461
462 rl->count[READ] = rl->count[WRITE] = 0;
463 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200464 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 init_waitqueue_head(&rl->wait[READ]);
466 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Christoph Lameter19460892005-06-23 00:08:19 -0700468 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
469 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
471 if (!rl->rq_pool)
472 return -ENOMEM;
473
474 return 0;
475}
476
Jens Axboe165125e2007-07-24 09:28:11 +0200477struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
Christoph Lameter19460892005-06-23 00:08:19 -0700479 return blk_alloc_queue_node(gfp_mask, -1);
480}
481EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
Jens Axboe165125e2007-07-24 09:28:11 +0200483struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700484{
Jens Axboe165125e2007-07-24 09:28:11 +0200485 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700486 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700487
Jens Axboe8324aa92008-01-29 14:51:59 +0100488 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700489 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 if (!q)
491 return NULL;
492
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700493 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
494 q->backing_dev_info.unplug_io_data = q;
495 err = bdi_init(&q->backing_dev_info);
496 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100497 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700498 return NULL;
499 }
500
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 init_timer(&q->unplug_timer);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700502 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
503 INIT_LIST_HEAD(&q->timeout_list);
Al Viro483f4af2006-03-18 18:34:37 -0500504
Jens Axboe8324aa92008-01-29 14:51:59 +0100505 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Al Viro483f4af2006-03-18 18:34:37 -0500507 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700508 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 return q;
511}
Christoph Lameter19460892005-06-23 00:08:19 -0700512EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
514/**
515 * blk_init_queue - prepare a request queue for use with a block device
516 * @rfn: The function to be called to process requests that have been
517 * placed on the queue.
518 * @lock: Request queue spin lock
519 *
520 * Description:
521 * If a block device wishes to use the standard request handling procedures,
522 * which sorts requests and coalesces adjacent requests, then it must
523 * call blk_init_queue(). The function @rfn will be called when there
524 * are requests on the queue that need to be processed. If the device
525 * supports plugging, then @rfn may not be called immediately when requests
526 * are available on the queue, but may be called at some time later instead.
527 * Plugged queues are generally unplugged when a buffer belonging to one
528 * of the requests on the queue is needed, or due to memory pressure.
529 *
530 * @rfn is not required, or even expected, to remove all requests off the
531 * queue, but only as many as it can handle at a time. If it does leave
532 * requests on the queue, it is responsible for arranging that the requests
533 * get dealt with eventually.
534 *
535 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200536 * request queue; this lock will be taken also from interrupt context, so irq
537 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200539 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 * it didn't succeed.
541 *
542 * Note:
543 * blk_init_queue() must be paired with a blk_cleanup_queue() call
544 * when the block device is deactivated (such as at module unload).
545 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700546
Jens Axboe165125e2007-07-24 09:28:11 +0200547struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
Christoph Lameter19460892005-06-23 00:08:19 -0700549 return blk_init_queue_node(rfn, lock, -1);
550}
551EXPORT_SYMBOL(blk_init_queue);
552
Jens Axboe165125e2007-07-24 09:28:11 +0200553struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700554blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
555{
Jens Axboe165125e2007-07-24 09:28:11 +0200556 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558 if (!q)
559 return NULL;
560
Christoph Lameter19460892005-06-23 00:08:19 -0700561 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -0500562 if (blk_init_free_list(q)) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100563 kmem_cache_free(blk_requestq_cachep, q);
Al Viro8669aaf2006-03-18 13:50:00 -0500564 return NULL;
565 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
152587d2005-04-12 16:22:06 -0500567 /*
568 * if caller didn't supply a lock, they get per-queue locking with
569 * our embedded lock
570 */
Neil Browne7e72bf2008-05-14 16:05:54 -0700571 if (!lock)
152587d2005-04-12 16:22:06 -0500572 lock = &q->__queue_lock;
152587d2005-04-12 16:22:06 -0500573
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 q->prep_rq_fn = NULL;
576 q->unplug_fn = generic_unplug_device;
Kiyoshi Ueda4ee5eaf2008-09-18 10:46:13 -0400577 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER |
578 1 << QUEUE_FLAG_STACKABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 q->queue_lock = lock;
580
581 blk_queue_segment_boundary(q, 0xffffffff);
582
583 blk_queue_make_request(q, __make_request);
584 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
585
586 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
587 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
588
Alan Stern44ec9542007-02-20 11:01:57 -0500589 q->sg_reserved_size = INT_MAX;
590
FUJITA Tomonoriabf54392008-08-16 14:10:05 +0900591 blk_set_cmd_filter_defaults(&q->cmd_filter);
592
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 /*
594 * all done
595 */
596 if (!elevator_init(q, NULL)) {
597 blk_queue_congestion_threshold(q);
598 return q;
599 }
600
Al Viro8669aaf2006-03-18 13:50:00 -0500601 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 return NULL;
603}
Christoph Lameter19460892005-06-23 00:08:19 -0700604EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
Jens Axboe165125e2007-07-24 09:28:11 +0200606int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700608 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500609 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 return 0;
611 }
612
613 return 1;
614}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
Jens Axboe165125e2007-07-24 09:28:11 +0200616static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200618 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200619 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 mempool_free(rq, q->rq.rq_pool);
621}
622
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200623static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +0200624blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625{
626 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
627
628 if (!rq)
629 return NULL;
630
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200631 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200632
Jens Axboe49171e52006-08-10 08:59:11 +0200633 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
Tejun Heocb98fc82005-10-28 08:29:39 +0200635 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200636 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200637 mempool_free(rq, q->rq.rq_pool);
638 return NULL;
639 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200640 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Tejun Heocb98fc82005-10-28 08:29:39 +0200643 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
645
646/*
647 * ioc_batching returns true if the ioc is a valid batching request and
648 * should be given priority access to a request.
649 */
Jens Axboe165125e2007-07-24 09:28:11 +0200650static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651{
652 if (!ioc)
653 return 0;
654
655 /*
656 * Make sure the process is able to allocate at least 1 request
657 * even if the batch times out, otherwise we could theoretically
658 * lose wakeups.
659 */
660 return ioc->nr_batch_requests == q->nr_batching ||
661 (ioc->nr_batch_requests > 0
662 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
663}
664
665/*
666 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
667 * will cause the process to be a "batcher" on all queues in the system. This
668 * is the behaviour we want though - once it gets a wakeup it should be given
669 * a nice run.
670 */
Jens Axboe165125e2007-07-24 09:28:11 +0200671static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672{
673 if (!ioc || ioc_batching(q, ioc))
674 return;
675
676 ioc->nr_batch_requests = q->nr_batching;
677 ioc->last_waited = jiffies;
678}
679
Jens Axboe165125e2007-07-24 09:28:11 +0200680static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
682 struct request_list *rl = &q->rq;
683
684 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -0700685 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
687 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 if (waitqueue_active(&rl->wait[rw]))
689 wake_up(&rl->wait[rw]);
690
691 blk_clear_queue_full(q, rw);
692 }
693}
694
695/*
696 * A request has just been released. Account for it, update the full and
697 * congestion status, wake up any waiters. Called under q->queue_lock.
698 */
Jens Axboe165125e2007-07-24 09:28:11 +0200699static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
701 struct request_list *rl = &q->rq;
702
703 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200704 if (priv)
705 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
707 __freed_request(q, rw);
708
709 if (unlikely(rl->starved[rw ^ 1]))
710 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711}
712
713#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
714/*
Nick Piggind6344532005-06-28 20:45:14 -0700715 * Get a free request, queue_lock must be held.
716 * Returns NULL on failure, with queue_lock held.
717 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 */
Jens Axboe165125e2007-07-24 09:28:11 +0200719static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100720 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
722 struct request *rq = NULL;
723 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100724 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +0100725 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100726 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Jens Axboe7749a8d2006-12-13 13:02:26 +0100728 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100729 if (may_queue == ELV_MQUEUE_NO)
730 goto rq_starved;
731
732 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
733 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200734 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100735 /*
736 * The queue will fill after this allocation, so set
737 * it as full, and mark this process as "batching".
738 * This process will be allowed to complete a batch of
739 * requests, others will be blocked.
740 */
741 if (!blk_queue_full(q, rw)) {
742 ioc_set_batching(q, ioc);
743 blk_set_queue_full(q, rw);
744 } else {
745 if (may_queue != ELV_MQUEUE_MUST
746 && !ioc_batching(q, ioc)) {
747 /*
748 * The queue is full and the allocating
749 * process is not a "batcher", and not
750 * exempted by the IO scheduler
751 */
752 goto out;
753 }
754 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 }
Thomas Maier79e2de42006-10-19 23:28:15 -0700756 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 }
758
Jens Axboe082cf692005-06-28 16:35:11 +0200759 /*
760 * Only allow batching queuers to allocate up to 50% over the defined
761 * limit of requests, otherwise we could have thousands of requests
762 * allocated with any setting of ->nr_requests
763 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100764 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200765 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 rl->count[rw]++;
768 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200769
Jens Axboe64521d12005-10-28 08:30:39 +0200770 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200771 if (priv)
772 rl->elvpriv++;
773
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);
Tejun Heocb98fc82005-10-28 08:29:39 +0200786 freed_request(q, rw, 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:
796 if (unlikely(rl->count[rw] == 0))
797 rl->starved[rw] = 1;
798
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 Axboe2056a782006-03-23 20:00:26 +0100811 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
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 Axboe7749a8d2006-12-13 13:02:26 +0100825 const int rw = rw_flags & 0x01;
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
834 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
835 TASK_UNINTERRUPTIBLE);
836
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200837 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 finish_wait(&rl->wait[rw], &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/**
Jens Axboedc72ef42006-07-20 14:54:05 +0200882 * blk_start_queueing - initiate dispatch of requests to device
883 * @q: request queue to kick into gear
884 *
885 * This is basically a helper to remove the need to know whether a queue
886 * is plugged or not if someone just wants to initiate dispatch of requests
887 * for this queue.
888 *
889 * The queue lock must be held with interrupts disabled.
890 */
Jens Axboe165125e2007-07-24 09:28:11 +0200891void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +0200892{
Elias Oltmanns336c3d82008-10-01 16:02:33 +0200893 if (!blk_queue_plugged(q)) {
894 if (unlikely(blk_queue_stopped(q)))
895 return;
Jens Axboedc72ef42006-07-20 14:54:05 +0200896 q->request_fn(q);
Elias Oltmanns336c3d82008-10-01 16:02:33 +0200897 } else
Jens Axboedc72ef42006-07-20 14:54:05 +0200898 __generic_unplug_device(q);
899}
900EXPORT_SYMBOL(blk_start_queueing);
901
902/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 * blk_requeue_request - put a request back on queue
904 * @q: request queue where request should be inserted
905 * @rq: request to be inserted
906 *
907 * Description:
908 * Drivers often keep queueing requests until the hardware cannot accept
909 * more, when that condition happens we need to put the request back
910 * on the queue. Must be called with queue lock held.
911 */
Jens Axboe165125e2007-07-24 09:28:11 +0200912void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700914 blk_delete_timer(rq);
915 blk_clear_rq_complete(rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100916 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
917
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 if (blk_rq_tagged(rq))
919 blk_queue_end_tag(q, rq);
920
921 elv_requeue_request(q, rq);
922}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923EXPORT_SYMBOL(blk_requeue_request);
924
925/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200926 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 * @q: request queue where request should be inserted
928 * @rq: request to be inserted
929 * @at_head: insert request at head or tail of queue
930 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 *
932 * Description:
933 * Many block devices need to execute commands asynchronously, so they don't
934 * block the whole kernel from preemption during request execution. This is
935 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200936 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
937 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 *
939 * We have the option of inserting the head or the tail of the queue.
940 * Typically we use the tail for new ioctls and so forth. We use the head
941 * of the queue for things like a QUEUE_FULL message from a device, or a
942 * host that is unable to accept a particular command.
943 */
Jens Axboe165125e2007-07-24 09:28:11 +0200944void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500945 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946{
Tejun Heo 867d1192005-04-24 02:06:05 -0500947 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 unsigned long flags;
949
950 /*
951 * tell I/O scheduler that this isn't a regular read/write (ie it
952 * must not attempt merges on this) and that it acts as a soft
953 * barrier
954 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200955 rq->cmd_type = REQ_TYPE_SPECIAL;
956 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
958 rq->special = data;
959
960 spin_lock_irqsave(q->queue_lock, flags);
961
962 /*
963 * If command is tagged, release the tag
964 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500965 if (blk_rq_tagged(rq))
966 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200968 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -0500969 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +0200970 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 spin_unlock_irqrestore(q->queue_lock, flags);
972}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973EXPORT_SYMBOL(blk_insert_request);
974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975/*
976 * add-request adds a request to the linked list.
977 * queue lock is held and interrupts disabled, as we muck with the
978 * request queue list.
979 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100980static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981{
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200982 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 /*
985 * elevator indicated where it wants this request to be
986 * inserted at elevator_merge time
987 */
988 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
989}
Jens Axboe6728cb02008-01-31 13:03:55 +0100990
Tejun Heo074a7ac2008-08-25 19:56:14 +0900991static void part_round_stats_single(int cpu, struct hd_struct *part,
992 unsigned long now)
993{
994 if (now == part->stamp)
995 return;
996
997 if (part->in_flight) {
998 __part_stat_add(cpu, part, time_in_queue,
999 part->in_flight * (now - part->stamp));
1000 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1001 }
1002 part->stamp = now;
1003}
1004
1005/**
1006 * part_round_stats() - Round off the performance stats on a struct
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 * disk_stats.
1008 *
1009 * The average IO queue length and utilisation statistics are maintained
1010 * by observing the current state of the queue length and the amount of
1011 * time it has been in this state for.
1012 *
1013 * Normally, that accounting is done on IO completion, but that can result
1014 * in more than a second's worth of IO being accounted for within any one
1015 * second, leading to >100% utilisation. To deal with that, we call this
1016 * function to do a round-off before returning the results when reading
1017 * /proc/diskstats. This accounts immediately for all queue usage up to
1018 * the current jiffies and restarts the counters again.
1019 */
Tejun Heoc9959052008-08-25 19:47:21 +09001020void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001021{
1022 unsigned long now = jiffies;
1023
Tejun Heo074a7ac2008-08-25 19:56:14 +09001024 if (part->partno)
1025 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1026 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001027}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001028EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030/*
1031 * queue lock must be held
1032 */
Jens Axboe165125e2007-07-24 09:28:11 +02001033void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 if (unlikely(!q))
1036 return;
1037 if (unlikely(--req->ref_count))
1038 return;
1039
Tejun Heo8922e162005-10-20 16:23:44 +02001040 elv_completed_request(q, req);
1041
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 /*
1043 * Request may not have originated from ll_rw_blk. if not,
1044 * it didn't come out of our reserved rq pools
1045 */
Jens Axboe49171e52006-08-10 08:59:11 +02001046 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001048 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001051 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
1053 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02001054 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 }
1056}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001057EXPORT_SYMBOL_GPL(__blk_put_request);
1058
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059void blk_put_request(struct request *req)
1060{
Tejun Heo8922e162005-10-20 16:23:44 +02001061 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001062 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001064 spin_lock_irqsave(q->queue_lock, flags);
1065 __blk_put_request(q, req);
1066 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068EXPORT_SYMBOL(blk_put_request);
1069
Jens Axboe86db1e22008-01-29 14:53:40 +01001070void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001071{
Jens Axboec7c22e42008-09-13 20:26:01 +02001072 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001073 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001074
1075 /*
1076 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1077 */
1078 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001079 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01001080
1081 /*
1082 * REQ_BARRIER implies no merging, but lets make it explicit
1083 */
David Woodhousefb2dce82008-08-05 18:01:53 +01001084 if (unlikely(bio_discard(bio))) {
David Woodhousee17fc0a2008-08-09 16:42:20 +01001085 req->cmd_flags |= REQ_DISCARD;
1086 if (bio_barrier(bio))
1087 req->cmd_flags |= REQ_SOFTBARRIER;
David Woodhousefb2dce82008-08-05 18:01:53 +01001088 req->q->prepare_discard_fn(req->q, req);
David Woodhousee17fc0a2008-08-09 16:42:20 +01001089 } else if (unlikely(bio_barrier(bio)))
1090 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01001091
Jens Axboeb31dc662006-06-13 08:26:10 +02001092 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001093 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02001094 if (bio_rw_meta(bio))
1095 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02001096
Tejun Heo52d9e672006-01-06 09:49:58 +01001097 req->errors = 0;
1098 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001099 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001100 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02001101 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001102}
1103
Jens Axboe165125e2007-07-24 09:28:11 +02001104static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
Nick Piggin450991b2005-06-28 20:45:13 -07001106 struct request *req;
David Woodhousefb2dce82008-08-05 18:01:53 +01001107 int el_ret, nr_sectors, barrier, discard, err;
Jens Axboe51da90f2006-07-18 04:14:45 +02001108 const unsigned short prio = bio_prio(bio);
1109 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01001110 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
1114 /*
1115 * low level driver can indicate that it wants pages above a
1116 * certain limit bounced to low memory (ie for highmem, or even
1117 * ISA dma in theory)
1118 */
1119 blk_queue_bounce(q, &bio);
1120
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 barrier = bio_barrier(bio);
David Woodhousee17fc0a2008-08-09 16:42:20 +01001122 if (unlikely(barrier) && bio_has_data(bio) &&
1123 (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 err = -EOPNOTSUPP;
1125 goto end_io;
1126 }
1127
David Woodhousefb2dce82008-08-05 18:01:53 +01001128 discard = bio_discard(bio);
1129 if (unlikely(discard) && !q->prepare_discard_fn) {
1130 err = -EOPNOTSUPP;
1131 goto end_io;
1132 }
1133
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 spin_lock_irq(q->queue_lock);
1135
Nick Piggin450991b2005-06-28 20:45:13 -07001136 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 goto get_rq;
1138
1139 el_ret = elv_merge(q, &req, bio);
1140 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001141 case ELEVATOR_BACK_MERGE:
1142 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Jens Axboe6728cb02008-01-31 13:03:55 +01001144 if (!ll_back_merge_fn(q, req, bio))
1145 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Jens Axboe6728cb02008-01-31 13:03:55 +01001147 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001148
Jens Axboe6728cb02008-01-31 13:03:55 +01001149 req->biotail->bi_next = bio;
1150 req->biotail = bio;
1151 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1152 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001153 if (!blk_rq_cpu_valid(req))
1154 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001155 drive_stat_acct(req, 0);
1156 if (!attempt_back_merge(q, req))
1157 elv_merged_request(q, req, el_ret);
1158 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
Jens Axboe6728cb02008-01-31 13:03:55 +01001160 case ELEVATOR_FRONT_MERGE:
1161 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
Jens Axboe6728cb02008-01-31 13:03:55 +01001163 if (!ll_front_merge_fn(q, req, bio))
1164 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
Jens Axboe6728cb02008-01-31 13:03:55 +01001166 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001167
Jens Axboe6728cb02008-01-31 13:03:55 +01001168 bio->bi_next = req->bio;
1169 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
Jens Axboe6728cb02008-01-31 13:03:55 +01001171 /*
1172 * may not be valid. if the low level driver said
1173 * it didn't need a bounce buffer then it better
1174 * not touch req->buffer either...
1175 */
1176 req->buffer = bio_data(bio);
1177 req->current_nr_sectors = bio_cur_sectors(bio);
1178 req->hard_cur_sectors = req->current_nr_sectors;
1179 req->sector = req->hard_sector = bio->bi_sector;
1180 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1181 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001182 if (!blk_rq_cpu_valid(req))
1183 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001184 drive_stat_acct(req, 0);
1185 if (!attempt_front_merge(q, req))
1186 elv_merged_request(q, req, el_ret);
1187 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
Jens Axboe6728cb02008-01-31 13:03:55 +01001189 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1190 default:
1191 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 }
1193
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001195 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001196 * This sync check and mask will be re-done in init_request_from_bio(),
1197 * but we need to set it earlier to expose the sync flag to the
1198 * rq allocator and io schedulers.
1199 */
1200 rw_flags = bio_data_dir(bio);
1201 if (sync)
1202 rw_flags |= REQ_RW_SYNC;
1203
1204 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001205 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001206 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001207 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001208 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001209
Nick Piggin450991b2005-06-28 20:45:13 -07001210 /*
1211 * After dropping the lock and possibly sleeping here, our request
1212 * may now be mergeable after it had proven unmergeable (above).
1213 * We don't worry about that case for efficiency. It won't happen
1214 * often, and the elevators are able to handle it.
1215 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001216 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
Nick Piggin450991b2005-06-28 20:45:13 -07001218 spin_lock_irq(q->queue_lock);
Jens Axboec7c22e42008-09-13 20:26:01 +02001219 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1220 bio_flagged(bio, BIO_CPU_AFFINE))
1221 req->cpu = blk_cpu_to_group(smp_processor_id());
Nick Piggin450991b2005-06-28 20:45:13 -07001222 if (elv_queue_empty(q))
1223 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 add_request(q, req);
1225out:
Jens Axboe4a534f92005-04-16 15:25:40 -07001226 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 __generic_unplug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 spin_unlock_irq(q->queue_lock);
1229 return 0;
1230
1231end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02001232 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 return 0;
1234}
1235
1236/*
1237 * If bio->bi_dev is a partition, remap the location
1238 */
1239static inline void blk_partition_remap(struct bio *bio)
1240{
1241 struct block_device *bdev = bio->bi_bdev;
1242
Jens Axboebf2de6f2007-09-27 13:01:25 +02001243 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001245
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 bio->bi_sector += p->start_sect;
1247 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001248
1249 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
1250 bdev->bd_dev, bio->bi_sector,
1251 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 }
1253}
1254
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255static void handle_bad_sector(struct bio *bio)
1256{
1257 char b[BDEVNAME_SIZE];
1258
1259 printk(KERN_INFO "attempt to access beyond end of device\n");
1260 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1261 bdevname(bio->bi_bdev, b),
1262 bio->bi_rw,
1263 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1264 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1265
1266 set_bit(BIO_EOF, &bio->bi_flags);
1267}
1268
Akinobu Mitac17bb492006-12-08 02:39:46 -08001269#ifdef CONFIG_FAIL_MAKE_REQUEST
1270
1271static DECLARE_FAULT_ATTR(fail_make_request);
1272
1273static int __init setup_fail_make_request(char *str)
1274{
1275 return setup_fault_attr(&fail_make_request, str);
1276}
1277__setup("fail_make_request=", setup_fail_make_request);
1278
1279static int should_fail_request(struct bio *bio)
1280{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001281 struct hd_struct *part = bio->bi_bdev->bd_part;
1282
1283 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001284 return should_fail(&fail_make_request, bio->bi_size);
1285
1286 return 0;
1287}
1288
1289static int __init fail_make_request_debugfs(void)
1290{
1291 return init_fault_attr_dentries(&fail_make_request,
1292 "fail_make_request");
1293}
1294
1295late_initcall(fail_make_request_debugfs);
1296
1297#else /* CONFIG_FAIL_MAKE_REQUEST */
1298
1299static inline int should_fail_request(struct bio *bio)
1300{
1301 return 0;
1302}
1303
1304#endif /* CONFIG_FAIL_MAKE_REQUEST */
1305
Jens Axboec07e2b42007-07-18 13:27:58 +02001306/*
1307 * Check whether this bio extends beyond the end of the device.
1308 */
1309static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1310{
1311 sector_t maxsector;
1312
1313 if (!nr_sectors)
1314 return 0;
1315
1316 /* Test device or partition size, when known. */
1317 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1318 if (maxsector) {
1319 sector_t sector = bio->bi_sector;
1320
1321 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1322 /*
1323 * This may well happen - the kernel calls bread()
1324 * without checking the size of the device, e.g., when
1325 * mounting a device.
1326 */
1327 handle_bad_sector(bio);
1328 return 1;
1329 }
1330 }
1331
1332 return 0;
1333}
1334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001336 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 * @bio: The bio describing the location in memory and on the device.
1338 *
1339 * generic_make_request() is used to make I/O requests of block
1340 * devices. It is passed a &struct bio, which describes the I/O that needs
1341 * to be done.
1342 *
1343 * generic_make_request() does not return any status. The
1344 * success/failure status of the request, along with notification of
1345 * completion, is delivered asynchronously through the bio->bi_end_io
1346 * function described (one day) else where.
1347 *
1348 * The caller of generic_make_request must make sure that bi_io_vec
1349 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1350 * set to describe the device address, and the
1351 * bi_end_io and optionally bi_private are set to describe how
1352 * completion notification should be signaled.
1353 *
1354 * generic_make_request and the drivers it calls may use bi_next if this
1355 * bio happens to be merged with someone else, and may change bi_dev and
1356 * bi_sector for remaps as it sees fit. So the values of these fields
1357 * should NOT be depended on after the call to generic_make_request.
1358 */
Neil Brownd89d8792007-05-01 09:53:42 +02001359static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360{
Jens Axboe165125e2007-07-24 09:28:11 +02001361 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001362 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001364 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001365 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
1367 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Jens Axboec07e2b42007-07-18 13:27:58 +02001369 if (bio_check_eod(bio, nr_sectors))
1370 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
1372 /*
1373 * Resolve the mapping until finished. (drivers are
1374 * still free to implement/resolve their own stacking
1375 * by explicitly returning 0)
1376 *
1377 * NOTE: we don't repeat the blk_size check for each new device.
1378 * Stacking drivers are expected to know what they are doing.
1379 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001380 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001381 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 do {
1383 char b[BDEVNAME_SIZE];
1384
1385 q = bdev_get_queue(bio->bi_bdev);
1386 if (!q) {
1387 printk(KERN_ERR
1388 "generic_make_request: Trying to access "
1389 "nonexistent block-device %s (%Lu)\n",
1390 bdevname(bio->bi_bdev, b),
1391 (long long) bio->bi_sector);
1392end_io:
Jens Axboe51fd77b2007-11-02 08:49:08 +01001393 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 break;
1395 }
1396
Jens Axboe4fa253f2007-07-18 13:13:10 +02001397 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001398 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 bdevname(bio->bi_bdev, b),
1400 bio_sectors(bio),
1401 q->max_hw_sectors);
1402 goto end_io;
1403 }
1404
Nick Pigginfde6ad22005-06-23 00:08:53 -07001405 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 goto end_io;
1407
Akinobu Mitac17bb492006-12-08 02:39:46 -08001408 if (should_fail_request(bio))
1409 goto end_io;
1410
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 /*
1412 * If this device has partitions, remap block n
1413 * of partition p to block n+start(p) of the disk.
1414 */
1415 blk_partition_remap(bio);
1416
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001417 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1418 goto end_io;
1419
NeilBrown5ddfe962006-10-30 22:07:21 -08001420 if (old_sector != -1)
Jens Axboe4fa253f2007-07-18 13:13:10 +02001421 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08001422 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001423
1424 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
1425
NeilBrown5ddfe962006-10-30 22:07:21 -08001426 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001427 old_dev = bio->bi_bdev->bd_dev;
1428
Jens Axboec07e2b42007-07-18 13:27:58 +02001429 if (bio_check_eod(bio, nr_sectors))
1430 goto end_io;
David Woodhousefb2dce82008-08-05 18:01:53 +01001431 if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) ||
1432 (bio_discard(bio) && !q->prepare_discard_fn)) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001433 err = -EOPNOTSUPP;
1434 goto end_io;
1435 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001436
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 ret = q->make_request_fn(q, bio);
1438 } while (ret);
1439}
1440
Neil Brownd89d8792007-05-01 09:53:42 +02001441/*
1442 * We only want one ->make_request_fn to be active at a time,
1443 * else stack usage with stacked devices could be a problem.
1444 * So use current->bio_{list,tail} to keep a list of requests
1445 * submited by a make_request_fn function.
1446 * current->bio_tail is also used as a flag to say if
1447 * generic_make_request is currently active in this task or not.
1448 * If it is NULL, then no make_request is active. If it is non-NULL,
1449 * then a make_request is active, and new requests should be added
1450 * at the tail
1451 */
1452void generic_make_request(struct bio *bio)
1453{
1454 if (current->bio_tail) {
1455 /* make_request is active */
1456 *(current->bio_tail) = bio;
1457 bio->bi_next = NULL;
1458 current->bio_tail = &bio->bi_next;
1459 return;
1460 }
1461 /* following loop may be a bit non-obvious, and so deserves some
1462 * explanation.
1463 * Before entering the loop, bio->bi_next is NULL (as all callers
1464 * ensure that) so we have a list with a single bio.
1465 * We pretend that we have just taken it off a longer list, so
1466 * we assign bio_list to the next (which is NULL) and bio_tail
1467 * to &bio_list, thus initialising the bio_list of new bios to be
1468 * added. __generic_make_request may indeed add some more bios
1469 * through a recursive call to generic_make_request. If it
1470 * did, we find a non-NULL value in bio_list and re-enter the loop
1471 * from the top. In this case we really did just take the bio
1472 * of the top of the list (no pretending) and so fixup bio_list and
1473 * bio_tail or bi_next, and call into __generic_make_request again.
1474 *
1475 * The loop was structured like this to make only one call to
1476 * __generic_make_request (which is important as it is large and
1477 * inlined) and to keep the structure simple.
1478 */
1479 BUG_ON(bio->bi_next);
1480 do {
1481 current->bio_list = bio->bi_next;
1482 if (bio->bi_next == NULL)
1483 current->bio_tail = &current->bio_list;
1484 else
1485 bio->bi_next = NULL;
1486 __generic_make_request(bio);
1487 bio = current->bio_list;
1488 } while (bio);
1489 current->bio_tail = NULL; /* deactivate */
1490}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491EXPORT_SYMBOL(generic_make_request);
1492
1493/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001494 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1496 * @bio: The &struct bio which describes the I/O
1497 *
1498 * submit_bio() is very similar in purpose to generic_make_request(), and
1499 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001500 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 *
1502 */
1503void submit_bio(int rw, struct bio *bio)
1504{
1505 int count = bio_sectors(bio);
1506
Jens Axboe22e2c502005-06-27 10:55:12 +02001507 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
Jens Axboebf2de6f2007-09-27 13:01:25 +02001509 /*
1510 * If it's a regular read/write or a barrier with data attached,
1511 * go through the normal accounting stuff before submission.
1512 */
Jens Axboea9c701e2008-08-08 11:04:44 +02001513 if (bio_has_data(bio)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001514 if (rw & WRITE) {
1515 count_vm_events(PGPGOUT, count);
1516 } else {
1517 task_io_account_read(bio->bi_size);
1518 count_vm_events(PGPGIN, count);
1519 }
1520
1521 if (unlikely(block_dump)) {
1522 char b[BDEVNAME_SIZE];
1523 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001524 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001525 (rw & WRITE) ? "WRITE" : "READ",
1526 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001527 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 }
1530
1531 generic_make_request(bio);
1532}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533EXPORT_SYMBOL(submit_bio);
1534
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001535/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001536 * blk_rq_check_limits - Helper function to check a request for the queue limit
1537 * @q: the queue
1538 * @rq: the request being checked
1539 *
1540 * Description:
1541 * @rq may have been made based on weaker limitations of upper-level queues
1542 * in request stacking drivers, and it may violate the limitation of @q.
1543 * Since the block layer and the underlying device driver trust @rq
1544 * after it is inserted to @q, it should be checked against @q before
1545 * the insertion using this generic function.
1546 *
1547 * This function should also be useful for request stacking drivers
1548 * in some cases below, so export this fuction.
1549 * Request stacking drivers like request-based dm may change the queue
1550 * limits while requests are in the queue (e.g. dm's table swapping).
1551 * Such request stacking drivers should check those requests agaist
1552 * the new queue limits again when they dispatch those requests,
1553 * although such checkings are also done against the old queue limits
1554 * when submitting requests.
1555 */
1556int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1557{
1558 if (rq->nr_sectors > q->max_sectors ||
1559 rq->data_len > q->max_hw_sectors << 9) {
1560 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1561 return -EIO;
1562 }
1563
1564 /*
1565 * queue's settings related to segment counting like q->bounce_pfn
1566 * may differ from that of other stacking queues.
1567 * Recalculate it to check the request correctly on this queue's
1568 * limitation.
1569 */
1570 blk_recalc_rq_segments(rq);
1571 if (rq->nr_phys_segments > q->max_phys_segments ||
1572 rq->nr_phys_segments > q->max_hw_segments) {
1573 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1574 return -EIO;
1575 }
1576
1577 return 0;
1578}
1579EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1580
1581/**
1582 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1583 * @q: the queue to submit the request
1584 * @rq: the request being queued
1585 */
1586int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1587{
1588 unsigned long flags;
1589
1590 if (blk_rq_check_limits(q, rq))
1591 return -EIO;
1592
1593#ifdef CONFIG_FAIL_MAKE_REQUEST
1594 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1595 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1596 return -EIO;
1597#endif
1598
1599 spin_lock_irqsave(q->queue_lock, flags);
1600
1601 /*
1602 * Submitting request must be dequeued before calling this function
1603 * because it will be linked to another request_queue
1604 */
1605 BUG_ON(blk_queued_rq(rq));
1606
1607 drive_stat_acct(rq, 1);
1608 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1609
1610 spin_unlock_irqrestore(q->queue_lock, flags);
1611
1612 return 0;
1613}
1614EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1615
1616/**
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001617 * __end_that_request_first - end I/O on a request
1618 * @req: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001619 * @error: %0 for success, < %0 for error
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001620 * @nr_bytes: number of bytes to complete
1621 *
1622 * Description:
1623 * Ends I/O on a number of bytes attached to @req, and sets it up
1624 * for the next range of segments (if any) in the cluster.
1625 *
1626 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001627 * %0 - we are done with this request, call end_that_request_last()
1628 * %1 - still buffers pending for this request
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001629 **/
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001630static int __end_that_request_first(struct request *req, int error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 int nr_bytes)
1632{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001633 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 struct bio *bio;
1635
Jens Axboe2056a782006-03-23 20:00:26 +01001636 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
1637
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 /*
Randy Dunlap710027a2008-08-19 20:13:11 +02001639 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 * sense key with us all the way through
1641 */
1642 if (!blk_pc_request(req))
1643 req->errors = 0;
1644
Jens Axboe6728cb02008-01-31 13:03:55 +01001645 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1646 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 req->rq_disk ? req->rq_disk->disk_name : "?",
1648 (unsigned long long)req->sector);
1649 }
1650
Jens Axboed72d9042005-11-01 08:35:42 +01001651 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01001652 const int rw = rq_data_dir(req);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001653 struct hd_struct *part;
Tejun Heoc9959052008-08-25 19:47:21 +09001654 int cpu;
Jens Axboea3623572005-11-01 09:26:16 +01001655
Tejun Heo074a7ac2008-08-25 19:56:14 +09001656 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001657 part = disk_map_sector_rcu(req->rq_disk, req->sector);
Tejun Heo074a7ac2008-08-25 19:56:14 +09001658 part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9);
1659 part_stat_unlock();
Jens Axboed72d9042005-11-01 08:35:42 +01001660 }
1661
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 total_bytes = bio_nbytes = 0;
1663 while ((bio = req->bio) != NULL) {
1664 int nbytes;
1665
Jens Axboebf2de6f2007-09-27 13:01:25 +02001666 /*
1667 * For an empty barrier request, the low level driver must
1668 * store a potential error location in ->sector. We pass
1669 * that back up in ->bi_sector.
1670 */
1671 if (blk_empty_barrier(req))
1672 bio->bi_sector = req->sector;
1673
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 if (nr_bytes >= bio->bi_size) {
1675 req->bio = bio->bi_next;
1676 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02001677 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 next_idx = 0;
1679 bio_nbytes = 0;
1680 } else {
1681 int idx = bio->bi_idx + next_idx;
1682
1683 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1684 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01001685 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -07001686 __func__, bio->bi_idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 break;
1688 }
1689
1690 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1691 BIO_BUG_ON(nbytes > bio->bi_size);
1692
1693 /*
1694 * not a complete bvec done
1695 */
1696 if (unlikely(nbytes > nr_bytes)) {
1697 bio_nbytes += nr_bytes;
1698 total_bytes += nr_bytes;
1699 break;
1700 }
1701
1702 /*
1703 * advance to the next vector
1704 */
1705 next_idx++;
1706 bio_nbytes += nbytes;
1707 }
1708
1709 total_bytes += nbytes;
1710 nr_bytes -= nbytes;
1711
Jens Axboe6728cb02008-01-31 13:03:55 +01001712 bio = req->bio;
1713 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 /*
1715 * end more in this run, or just return 'not-done'
1716 */
1717 if (unlikely(nr_bytes <= 0))
1718 break;
1719 }
1720 }
1721
1722 /*
1723 * completely done
1724 */
1725 if (!req->bio)
1726 return 0;
1727
1728 /*
1729 * if the request wasn't completed, update state
1730 */
1731 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02001732 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 bio->bi_idx += next_idx;
1734 bio_iovec(bio)->bv_offset += nr_bytes;
1735 bio_iovec(bio)->bv_len -= nr_bytes;
1736 }
1737
1738 blk_recalc_rq_sectors(req, total_bytes >> 9);
1739 blk_recalc_rq_segments(req);
1740 return 1;
1741}
1742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743/*
1744 * queue lock must be held
1745 */
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001746static void end_that_request_last(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747{
1748 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01001749
Jens Axboe242f9dc2008-09-14 05:55:09 -07001750 blk_delete_timer(req);
1751
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001752 if (blk_rq_tagged(req))
1753 blk_queue_end_tag(req->q, req);
1754
1755 if (blk_queued_rq(req))
1756 blkdev_dequeue_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
1758 if (unlikely(laptop_mode) && blk_fs_request(req))
1759 laptop_io_completion();
1760
Jens Axboefd0ff8a2006-05-23 11:23:49 +02001761 /*
1762 * Account IO completion. bar_rq isn't accounted as a normal
1763 * IO on queueing nor completion. Accounting the containing
1764 * request is enough.
1765 */
1766 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01001768 const int rw = rq_data_dir(req);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001769 struct hd_struct *part;
Tejun Heoc9959052008-08-25 19:47:21 +09001770 int cpu;
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001771
Tejun Heo074a7ac2008-08-25 19:56:14 +09001772 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001773 part = disk_map_sector_rcu(disk, req->sector);
Jens Axboea3623572005-11-01 09:26:16 +01001774
Tejun Heo074a7ac2008-08-25 19:56:14 +09001775 part_stat_inc(cpu, part, ios[rw]);
1776 part_stat_add(cpu, part, ticks[rw], duration);
1777 part_round_stats(cpu, part);
1778 part_dec_in_flight(part);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001779
Tejun Heo074a7ac2008-08-25 19:56:14 +09001780 part_stat_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 }
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001782
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01001784 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001785 else {
1786 if (blk_bidi_rq(req))
1787 __blk_put_request(req->next_rq->q, req->next_rq);
1788
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 }
1791}
1792
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001793/**
1794 * blk_rq_bytes - Returns bytes left to complete in the entire request
Randy Dunlap5d87a052008-02-20 09:01:22 +01001795 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001796 **/
1797unsigned int blk_rq_bytes(struct request *rq)
Jens Axboea0cd1282007-09-21 10:41:07 +02001798{
1799 if (blk_fs_request(rq))
1800 return rq->hard_nr_sectors << 9;
1801
1802 return rq->data_len;
1803}
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001804EXPORT_SYMBOL_GPL(blk_rq_bytes);
1805
1806/**
1807 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
Randy Dunlap5d87a052008-02-20 09:01:22 +01001808 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001809 **/
1810unsigned int blk_rq_cur_bytes(struct request *rq)
1811{
1812 if (blk_fs_request(rq))
1813 return rq->current_nr_sectors << 9;
1814
1815 if (rq->bio)
1816 return rq->bio->bi_size;
1817
1818 return rq->data_len;
1819}
1820EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
Jens Axboea0cd1282007-09-21 10:41:07 +02001821
1822/**
Jens Axboea0cd1282007-09-21 10:41:07 +02001823 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07001824 * @req: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001825 * @uptodate: error value or %0/%1 uptodate flag
Jens Axboea0cd1282007-09-21 10:41:07 +02001826 *
1827 * Description:
1828 * Ends I/O on the current segment of a request. If that is the only
1829 * remaining segment, the request is also completed and freed.
1830 *
Randy Dunlap710027a2008-08-19 20:13:11 +02001831 * This is a remnant of how older block drivers handled I/O completions.
1832 * Modern drivers typically end I/O on the full request in one go, unless
Jens Axboea0cd1282007-09-21 10:41:07 +02001833 * they have a residual value to account for. For that case this function
1834 * isn't really useful, unless the residual just happens to be the
1835 * full current segment. In other words, don't use this function in new
Kiyoshi Uedad00e29f2008-10-01 10:14:46 -04001836 * code. Use blk_end_request() or __blk_end_request() to end a request.
Jens Axboea0cd1282007-09-21 10:41:07 +02001837 **/
1838void end_request(struct request *req, int uptodate)
1839{
Kiyoshi Uedad00e29f2008-10-01 10:14:46 -04001840 int error = 0;
1841
1842 if (uptodate <= 0)
1843 error = uptodate ? uptodate : -EIO;
1844
1845 __blk_end_request(req, error, req->hard_cur_sectors << 9);
Jens Axboea0cd1282007-09-21 10:41:07 +02001846}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847EXPORT_SYMBOL(end_request);
1848
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04001849static int end_that_request_data(struct request *rq, int error,
1850 unsigned int nr_bytes, unsigned int bidi_bytes)
1851{
1852 if (rq->bio) {
1853 if (__end_that_request_first(rq, error, nr_bytes))
1854 return 1;
1855
1856 /* Bidi request must be completed as a whole */
1857 if (blk_bidi_rq(rq) &&
1858 __end_that_request_first(rq->next_rq, error, bidi_bytes))
1859 return 1;
1860 }
1861
1862 return 0;
1863}
1864
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001865/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001866 * blk_end_io - Generic end_io function to complete a request.
1867 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001868 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001869 * @nr_bytes: number of bytes to complete @rq
1870 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001871 * @drv_callback: function called between completion of bios in the request
1872 * and completion of the request.
Randy Dunlap710027a2008-08-19 20:13:11 +02001873 * If the callback returns non %0, this helper returns without
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001874 * completion of the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001875 *
1876 * Description:
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001877 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001878 * If @rq has leftover, sets it up for the next range of segments.
1879 *
1880 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001881 * %0 - we are done with this request
1882 * %1 - this request is not freed yet, it still has pending buffers.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001883 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001884static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1885 unsigned int bidi_bytes,
1886 int (drv_callback)(struct request *))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001887{
1888 struct request_queue *q = rq->q;
1889 unsigned long flags = 0UL;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001890
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04001891 if (end_that_request_data(rq, error, nr_bytes, bidi_bytes))
1892 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001893
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001894 /* Special feature for tricky drivers */
1895 if (drv_callback && drv_callback(rq))
1896 return 1;
1897
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001898 add_disk_randomness(rq->rq_disk);
1899
1900 spin_lock_irqsave(q->queue_lock, flags);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001901 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001902 spin_unlock_irqrestore(q->queue_lock, flags);
1903
1904 return 0;
1905}
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001906
1907/**
1908 * blk_end_request - Helper function for drivers to complete the request.
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 *
1913 * Description:
1914 * Ends I/O on a number of bytes attached to @rq.
1915 * If @rq has leftover, sets it up for the next range of segments.
1916 *
1917 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001918 * %0 - we are done with this request
1919 * %1 - still buffers pending for this request
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001920 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001921int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001922{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001923 return blk_end_io(rq, error, nr_bytes, 0, NULL);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001924}
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001925EXPORT_SYMBOL_GPL(blk_end_request);
1926
1927/**
1928 * __blk_end_request - Helper function for drivers to complete the request.
1929 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001930 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001931 * @nr_bytes: number of bytes to complete
1932 *
1933 * Description:
1934 * Must be called with queue lock held unlike blk_end_request().
1935 *
1936 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001937 * %0 - we are done with this request
1938 * %1 - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001939 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001940int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001941{
Jens Axboe60540162008-08-26 13:34:34 +02001942 if (rq->bio && __end_that_request_first(rq, error, nr_bytes))
Jens Axboe051cc392008-08-08 11:06:45 +02001943 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001944
1945 add_disk_randomness(rq->rq_disk);
1946
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001947 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001948
1949 return 0;
1950}
1951EXPORT_SYMBOL_GPL(__blk_end_request);
1952
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001953/**
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001954 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1955 * @rq: the bidi request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001956 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001957 * @nr_bytes: number of bytes to complete @rq
1958 * @bidi_bytes: number of bytes to complete @rq->next_rq
1959 *
1960 * Description:
1961 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1962 *
1963 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001964 * %0 - we are done with this request
1965 * %1 - still buffers pending for this request
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001966 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001967int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
1968 unsigned int bidi_bytes)
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001969{
1970 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
1971}
1972EXPORT_SYMBOL_GPL(blk_end_bidi_request);
1973
1974/**
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04001975 * blk_update_request - Special helper function for request stacking drivers
1976 * @rq: the request being processed
1977 * @error: %0 for success, < %0 for error
1978 * @nr_bytes: number of bytes to complete @rq
1979 *
1980 * Description:
1981 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
1982 * the request structure even if @rq doesn't have leftover.
1983 * If @rq has leftover, sets it up for the next range of segments.
1984 *
1985 * This special helper function is only for request stacking drivers
1986 * (e.g. request-based dm) so that they can handle partial completion.
1987 * Actual device drivers should use blk_end_request instead.
1988 */
1989void blk_update_request(struct request *rq, int error, unsigned int nr_bytes)
1990{
1991 if (!end_that_request_data(rq, error, nr_bytes, 0)) {
1992 /*
1993 * These members are not updated in end_that_request_data()
1994 * when all bios are completed.
1995 * Update them so that the request stacking driver can find
1996 * how many bytes remain in the request later.
1997 */
1998 rq->nr_sectors = rq->hard_nr_sectors = 0;
1999 rq->current_nr_sectors = rq->hard_cur_sectors = 0;
2000 }
2001}
2002EXPORT_SYMBOL_GPL(blk_update_request);
2003
2004/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002005 * blk_end_request_callback - Special helper function for tricky drivers
2006 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02002007 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002008 * @nr_bytes: number of bytes to complete
2009 * @drv_callback: function called between completion of bios in the request
2010 * and completion of the request.
Randy Dunlap710027a2008-08-19 20:13:11 +02002011 * If the callback returns non %0, this helper returns without
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002012 * completion of the request.
2013 *
2014 * Description:
2015 * Ends I/O on a number of bytes attached to @rq.
2016 * If @rq has leftover, sets it up for the next range of segments.
2017 *
2018 * This special helper function is used only for existing tricky drivers.
2019 * (e.g. cdrom_newpc_intr() of ide-cd)
2020 * This interface will be removed when such drivers are rewritten.
2021 * Don't use this interface in other places anymore.
2022 *
2023 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02002024 * %0 - we are done with this request
2025 * %1 - this request is not freed yet.
2026 * this request still has pending buffers or
2027 * the driver doesn't want to finish this request yet.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002028 **/
Jens Axboe22b13212008-01-31 12:36:19 +01002029int blk_end_request_callback(struct request *rq, int error,
2030 unsigned int nr_bytes,
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002031 int (drv_callback)(struct request *))
2032{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002033 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002034}
2035EXPORT_SYMBOL_GPL(blk_end_request_callback);
2036
Jens Axboe86db1e22008-01-29 14:53:40 +01002037void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2038 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039{
David Woodhoused628eae2008-08-09 16:22:17 +01002040 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2041 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002042 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
David Woodhousefb2dce82008-08-05 18:01:53 +01002044 if (bio_has_data(bio)) {
2045 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002046 rq->buffer = bio_data(bio);
2047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 rq->current_nr_sectors = bio_cur_sectors(bio);
2049 rq->hard_cur_sectors = rq->current_nr_sectors;
2050 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002051 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
2053 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
NeilBrown66846572007-08-16 13:31:28 +02002055 if (bio->bi_bdev)
2056 rq->rq_disk = bio->bi_bdev->bd_disk;
2057}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002059/**
2060 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2061 * @q : the queue of the device being checked
2062 *
2063 * Description:
2064 * Check if underlying low-level drivers of a device are busy.
2065 * If the drivers want to export their busy state, they must set own
2066 * exporting function using blk_queue_lld_busy() first.
2067 *
2068 * Basically, this function is used only by request stacking drivers
2069 * to stop dispatching requests to underlying devices when underlying
2070 * devices are busy. This behavior helps more I/O merging on the queue
2071 * of the request stacking driver and prevents I/O throughput regression
2072 * on burst I/O load.
2073 *
2074 * Return:
2075 * 0 - Not busy (The request stacking driver should dispatch request)
2076 * 1 - Busy (The request stacking driver should stop dispatching request)
2077 */
2078int blk_lld_busy(struct request_queue *q)
2079{
2080 if (q->lld_busy_fn)
2081 return q->lld_busy_fn(q);
2082
2083 return 0;
2084}
2085EXPORT_SYMBOL_GPL(blk_lld_busy);
2086
Jens Axboe18887ad2008-07-28 13:08:45 +02002087int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088{
2089 return queue_work(kblockd_workqueue, work);
2090}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091EXPORT_SYMBOL(kblockd_schedule_work);
2092
Andrew Morton19a75d82007-05-09 02:33:56 -07002093void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07002095 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096}
Andrew Morton19a75d82007-05-09 02:33:56 -07002097EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
2099int __init blk_dev_init(void)
2100{
2101 kblockd_workqueue = create_workqueue("kblockd");
2102 if (!kblockd_workqueue)
2103 panic("Failed to create kblockd\n");
2104
2105 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002106 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Jens Axboe8324aa92008-01-29 14:51:59 +01002108 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002109 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 return 0;
2112}
2113