blob: 10e8a64a5a5b1b213cbc2886755b22dedce76313 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd42006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
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}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261/**
262 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200263 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 *
265 * Description:
266 * Linux uses plugging to build bigger requests queues before letting
267 * the device have at them. If a queue is plugged, the I/O scheduler
268 * is still adding and merging requests on the queue. Once the queue
269 * gets unplugged, the request_fn defined for the queue is invoked and
270 * transfers started.
271 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200272void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200274 if (blk_queue_plugged(q)) {
275 spin_lock_irq(q->queue_lock);
276 __generic_unplug_device(q);
277 spin_unlock_irq(q->queue_lock);
278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279}
280EXPORT_SYMBOL(generic_unplug_device);
281
282static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
283 struct page *page)
284{
Jens Axboe165125e2007-07-24 09:28:11 +0200285 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500287 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
Jens Axboe86db1e22008-01-29 14:53:40 +0100290void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291{
Jens Axboe165125e2007-07-24 09:28:11 +0200292 struct request_queue *q =
293 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
Jens Axboe2056a782006-03-23 20:00:26 +0100295 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
296 q->rq.count[READ] + q->rq.count[WRITE]);
297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 q->unplug_fn(q);
299}
300
Jens Axboe86db1e22008-01-29 14:53:40 +0100301void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Jens Axboe165125e2007-07-24 09:28:11 +0200303 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
Jens Axboe2056a782006-03-23 20:00:26 +0100305 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
306 q->rq.count[READ] + q->rq.count[WRITE]);
307
Jens Axboe18887ad2008-07-28 13:08:45 +0200308 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500311void blk_unplug(struct request_queue *q)
312{
313 /*
314 * devices don't necessarily have an ->unplug_fn defined
315 */
316 if (q->unplug_fn) {
317 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
318 q->rq.count[READ] + q->rq.count[WRITE]);
319
320 q->unplug_fn(q);
321 }
322}
323EXPORT_SYMBOL(blk_unplug);
324
Jens Axboec7c22e42008-09-13 20:26:01 +0200325static void blk_invoke_request_fn(struct request_queue *q)
326{
Jens Axboe80a4b582008-10-14 09:51:06 +0200327 if (unlikely(blk_queue_stopped(q)))
328 return;
329
Jens Axboec7c22e42008-09-13 20:26:01 +0200330 /*
331 * one level of recursion is ok and is much faster than kicking
332 * the unplug handling
333 */
334 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
335 q->request_fn(q);
336 queue_flag_clear(QUEUE_FLAG_REENTER, q);
337 } else {
338 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
339 kblockd_schedule_work(q, &q->unplug_work);
340 }
341}
342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343/**
344 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200345 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 *
347 * Description:
348 * blk_start_queue() will clear the stop flag on the queue, and call
349 * the request_fn for the queue if it was in a stopped state when
350 * entered. Also see blk_stop_queue(). Queue lock must be held.
351 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200352void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200354 WARN_ON(!irqs_disabled());
355
Nick Piggin75ad23b2008-04-29 14:48:33 +0200356 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Jens Axboec7c22e42008-09-13 20:26:01 +0200357 blk_invoke_request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359EXPORT_SYMBOL(blk_start_queue);
360
361/**
362 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200363 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 *
365 * Description:
366 * The Linux block layer assumes that a block driver will consume all
367 * entries on the request queue when the request_fn strategy is called.
368 * Often this will not happen, because of hardware limitations (queue
369 * depth settings). If a device driver gets a 'queue full' response,
370 * or if it simply chooses not to queue more I/O at one point, it can
371 * call this function to prevent the request_fn from being called until
372 * the driver has signalled it's ready to go again. This happens by calling
373 * blk_start_queue() to restart queue operations. Queue lock must be held.
374 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200375void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
377 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200378 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379}
380EXPORT_SYMBOL(blk_stop_queue);
381
382/**
383 * blk_sync_queue - cancel any pending callbacks on a queue
384 * @q: the queue
385 *
386 * Description:
387 * The block layer may perform asynchronous callback activity
388 * on a queue, such as calling the unplug function after a timeout.
389 * A block device may call blk_sync_queue to ensure that any
390 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200391 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 * that its ->make_request_fn will not re-add plugging prior to calling
393 * this function.
394 *
395 */
396void blk_sync_queue(struct request_queue *q)
397{
398 del_timer_sync(&q->unplug_timer);
Oleg Nesterovabbeb882007-10-23 15:08:19 +0200399 kblockd_flush_work(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400}
401EXPORT_SYMBOL(blk_sync_queue);
402
403/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200404 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200406 *
407 * Description:
408 * See @blk_run_queue. This variant must be called with the queue lock
409 * held and interrupts disabled.
410 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200412void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200415
416 /*
417 * Only recurse once to avoid overrunning the stack, let the unplug
418 * handling reinvoke the handler shortly if we already got there.
419 */
Jens Axboec7c22e42008-09-13 20:26:01 +0200420 if (!elv_queue_empty(q))
421 blk_invoke_request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200422}
423EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200424
Nick Piggin75ad23b2008-04-29 14:48:33 +0200425/**
426 * blk_run_queue - run a single device queue
427 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200428 *
429 * Description:
430 * Invoke request handling on this queue, if it has pending work to do.
431 * May be used to restart queueing when a request has completed. Also
432 * See @blk_start_queueing.
433 *
Nick Piggin75ad23b2008-04-29 14:48:33 +0200434 */
435void blk_run_queue(struct request_queue *q)
436{
437 unsigned long flags;
438
439 spin_lock_irqsave(q->queue_lock, flags);
440 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 spin_unlock_irqrestore(q->queue_lock, flags);
442}
443EXPORT_SYMBOL(blk_run_queue);
444
Jens Axboe165125e2007-07-24 09:28:11 +0200445void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500446{
447 kobject_put(&q->kobj);
448}
Al Viro483f4af2006-03-18 18:34:37 -0500449
Jens Axboe6728cb02008-01-31 13:03:55 +0100450void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500451{
Jens Axboee3335de2008-09-18 09:22:54 -0700452 /*
453 * We know we have process context here, so we can be a little
454 * cautious and ensure that pending block actions on this device
455 * are done before moving on. Going into this function, we should
456 * not have processes doing IO to this device.
457 */
458 blk_sync_queue(q);
459
Al Viro483f4af2006-03-18 18:34:37 -0500460 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200461 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500462 mutex_unlock(&q->sysfs_lock);
463
464 if (q->elevator)
465 elevator_exit(q->elevator);
466
467 blk_put_queue(q);
468}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469EXPORT_SYMBOL(blk_cleanup_queue);
470
Jens Axboe165125e2007-07-24 09:28:11 +0200471static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472{
473 struct request_list *rl = &q->rq;
474
475 rl->count[READ] = rl->count[WRITE] = 0;
476 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200477 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 init_waitqueue_head(&rl->wait[READ]);
479 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Christoph Lameter19460892005-06-23 00:08:19 -0700481 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
482 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
484 if (!rl->rq_pool)
485 return -ENOMEM;
486
487 return 0;
488}
489
Jens Axboe165125e2007-07-24 09:28:11 +0200490struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
Christoph Lameter19460892005-06-23 00:08:19 -0700492 return blk_alloc_queue_node(gfp_mask, -1);
493}
494EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
Jens Axboe165125e2007-07-24 09:28:11 +0200496struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700497{
Jens Axboe165125e2007-07-24 09:28:11 +0200498 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700499 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700500
Jens Axboe8324aa92008-01-29 14:51:59 +0100501 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700502 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 if (!q)
504 return NULL;
505
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700506 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
507 q->backing_dev_info.unplug_io_data = q;
508 err = bdi_init(&q->backing_dev_info);
509 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100510 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700511 return NULL;
512 }
513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 init_timer(&q->unplug_timer);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700515 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
516 INIT_LIST_HEAD(&q->timeout_list);
Peter Zijlstra713ada92008-10-16 13:44:57 +0200517 INIT_WORK(&q->unplug_work, blk_unplug_work);
Al Viro483f4af2006-03-18 18:34:37 -0500518
Jens Axboe8324aa92008-01-29 14:51:59 +0100519 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Al Viro483f4af2006-03-18 18:34:37 -0500521 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700522 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500523
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 return q;
525}
Christoph Lameter19460892005-06-23 00:08:19 -0700526EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528/**
529 * blk_init_queue - prepare a request queue for use with a block device
530 * @rfn: The function to be called to process requests that have been
531 * placed on the queue.
532 * @lock: Request queue spin lock
533 *
534 * Description:
535 * If a block device wishes to use the standard request handling procedures,
536 * which sorts requests and coalesces adjacent requests, then it must
537 * call blk_init_queue(). The function @rfn will be called when there
538 * are requests on the queue that need to be processed. If the device
539 * supports plugging, then @rfn may not be called immediately when requests
540 * are available on the queue, but may be called at some time later instead.
541 * Plugged queues are generally unplugged when a buffer belonging to one
542 * of the requests on the queue is needed, or due to memory pressure.
543 *
544 * @rfn is not required, or even expected, to remove all requests off the
545 * queue, but only as many as it can handle at a time. If it does leave
546 * requests on the queue, it is responsible for arranging that the requests
547 * get dealt with eventually.
548 *
549 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200550 * request queue; this lock will be taken also from interrupt context, so irq
551 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200553 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 * it didn't succeed.
555 *
556 * Note:
557 * blk_init_queue() must be paired with a blk_cleanup_queue() call
558 * when the block device is deactivated (such as at module unload).
559 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700560
Jens Axboe165125e2007-07-24 09:28:11 +0200561struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
Christoph Lameter19460892005-06-23 00:08:19 -0700563 return blk_init_queue_node(rfn, lock, -1);
564}
565EXPORT_SYMBOL(blk_init_queue);
566
Jens Axboe165125e2007-07-24 09:28:11 +0200567struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700568blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
569{
Jens Axboe165125e2007-07-24 09:28:11 +0200570 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
572 if (!q)
573 return NULL;
574
Christoph Lameter19460892005-06-23 00:08:19 -0700575 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -0500576 if (blk_init_free_list(q)) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100577 kmem_cache_free(blk_requestq_cachep, q);
Al Viro8669aaf2006-03-18 13:50:00 -0500578 return NULL;
579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
152587d2005-04-12 16:22:06 -0500581 /*
582 * if caller didn't supply a lock, they get per-queue locking with
583 * our embedded lock
584 */
Neil Browne7e72bf2008-05-14 16:05:54 -0700585 if (!lock)
152587d2005-04-12 16:22:06 -0500586 lock = &q->__queue_lock;
152587d2005-04-12 16:22:06 -0500587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 q->prep_rq_fn = NULL;
590 q->unplug_fn = generic_unplug_device;
Kiyoshi Ueda4ee5eaf2008-09-18 10:46:13 -0400591 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER |
592 1 << QUEUE_FLAG_STACKABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 q->queue_lock = lock;
594
595 blk_queue_segment_boundary(q, 0xffffffff);
596
597 blk_queue_make_request(q, __make_request);
598 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
599
600 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
601 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
602
Alan Stern44ec9542007-02-20 11:01:57 -0500603 q->sg_reserved_size = INT_MAX;
604
FUJITA Tomonoriabf54392008-08-16 14:10:05 +0900605 blk_set_cmd_filter_defaults(&q->cmd_filter);
606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 /*
608 * all done
609 */
610 if (!elevator_init(q, NULL)) {
611 blk_queue_congestion_threshold(q);
612 return q;
613 }
614
Al Viro8669aaf2006-03-18 13:50:00 -0500615 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 return NULL;
617}
Christoph Lameter19460892005-06-23 00:08:19 -0700618EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Jens Axboe165125e2007-07-24 09:28:11 +0200620int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700622 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500623 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 return 0;
625 }
626
627 return 1;
628}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
Jens Axboe165125e2007-07-24 09:28:11 +0200630static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200632 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200633 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 mempool_free(rq, q->rq.rq_pool);
635}
636
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200637static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +0200638blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
640 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
641
642 if (!rq)
643 return NULL;
644
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200645 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200646
Jens Axboe49171e52006-08-10 08:59:11 +0200647 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Tejun Heocb98fc82005-10-28 08:29:39 +0200649 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200650 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200651 mempool_free(rq, q->rq.rq_pool);
652 return NULL;
653 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200654 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200655 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Tejun Heocb98fc82005-10-28 08:29:39 +0200657 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
660/*
661 * ioc_batching returns true if the ioc is a valid batching request and
662 * should be given priority access to a request.
663 */
Jens Axboe165125e2007-07-24 09:28:11 +0200664static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
666 if (!ioc)
667 return 0;
668
669 /*
670 * Make sure the process is able to allocate at least 1 request
671 * even if the batch times out, otherwise we could theoretically
672 * lose wakeups.
673 */
674 return ioc->nr_batch_requests == q->nr_batching ||
675 (ioc->nr_batch_requests > 0
676 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
677}
678
679/*
680 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
681 * will cause the process to be a "batcher" on all queues in the system. This
682 * is the behaviour we want though - once it gets a wakeup it should be given
683 * a nice run.
684 */
Jens Axboe165125e2007-07-24 09:28:11 +0200685static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
687 if (!ioc || ioc_batching(q, ioc))
688 return;
689
690 ioc->nr_batch_requests = q->nr_batching;
691 ioc->last_waited = jiffies;
692}
693
Jens Axboe165125e2007-07-24 09:28:11 +0200694static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
696 struct request_list *rl = &q->rq;
697
698 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -0700699 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
701 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 if (waitqueue_active(&rl->wait[rw]))
703 wake_up(&rl->wait[rw]);
704
705 blk_clear_queue_full(q, rw);
706 }
707}
708
709/*
710 * A request has just been released. Account for it, update the full and
711 * congestion status, wake up any waiters. Called under q->queue_lock.
712 */
Jens Axboe165125e2007-07-24 09:28:11 +0200713static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714{
715 struct request_list *rl = &q->rq;
716
717 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200718 if (priv)
719 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
721 __freed_request(q, rw);
722
723 if (unlikely(rl->starved[rw ^ 1]))
724 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
726
727#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
728/*
Nick Piggind6344532005-06-28 20:45:14 -0700729 * Get a free request, queue_lock must be held.
730 * Returns NULL on failure, with queue_lock held.
731 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 */
Jens Axboe165125e2007-07-24 09:28:11 +0200733static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100734 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
736 struct request *rq = NULL;
737 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100738 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +0100739 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100740 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Jens Axboe7749a8d2006-12-13 13:02:26 +0100742 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100743 if (may_queue == ELV_MQUEUE_NO)
744 goto rq_starved;
745
746 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
747 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200748 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100749 /*
750 * The queue will fill after this allocation, so set
751 * it as full, and mark this process as "batching".
752 * This process will be allowed to complete a batch of
753 * requests, others will be blocked.
754 */
755 if (!blk_queue_full(q, rw)) {
756 ioc_set_batching(q, ioc);
757 blk_set_queue_full(q, rw);
758 } else {
759 if (may_queue != ELV_MQUEUE_MUST
760 && !ioc_batching(q, ioc)) {
761 /*
762 * The queue is full and the allocating
763 * process is not a "batcher", and not
764 * exempted by the IO scheduler
765 */
766 goto out;
767 }
768 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 }
Thomas Maier79e2de42006-10-19 23:28:15 -0700770 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 }
772
Jens Axboe082cf692005-06-28 16:35:11 +0200773 /*
774 * Only allow batching queuers to allocate up to 50% over the defined
775 * limit of requests, otherwise we could have thousands of requests
776 * allocated with any setting of ->nr_requests
777 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100778 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200779 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100780
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 rl->count[rw]++;
782 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200783
Jens Axboe64521d12005-10-28 08:30:39 +0200784 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200785 if (priv)
786 rl->elvpriv++;
787
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 spin_unlock_irq(q->queue_lock);
789
Jens Axboe7749a8d2006-12-13 13:02:26 +0100790 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100791 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 /*
793 * Allocation failed presumably due to memory. Undo anything
794 * we might have messed up.
795 *
796 * Allocating task should really be put onto the front of the
797 * wait queue, but this is pretty rare.
798 */
799 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +0200800 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
802 /*
803 * in the very unlikely event that allocation failed and no
804 * requests for this direction was pending, mark us starved
805 * so that freeing of a request in the other direction will
806 * notice us. another possible fix would be to split the
807 * rq mempool into READ and WRITE
808 */
809rq_starved:
810 if (unlikely(rl->count[rw] == 0))
811 rl->starved[rw] = 1;
812
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 goto out;
814 }
815
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100816 /*
817 * ioc may be NULL here, and ioc_batching will be false. That's
818 * OK, if the queue is under the request limit then requests need
819 * not count toward the nr_batch_requests limit. There will always
820 * be some limit enforced by BLK_BATCH_TIME.
821 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 if (ioc_batching(q, ioc))
823 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100824
Jens Axboe2056a782006-03-23 20:00:26 +0100825 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 return rq;
828}
829
830/*
831 * No available requests for this queue, unplug the device and wait for some
832 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700833 *
834 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 */
Jens Axboe165125e2007-07-24 09:28:11 +0200836static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200837 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838{
Jens Axboe7749a8d2006-12-13 13:02:26 +0100839 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 struct request *rq;
841
Jens Axboe7749a8d2006-12-13 13:02:26 +0100842 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700843 while (!rq) {
844 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200845 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 struct request_list *rl = &q->rq;
847
848 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
849 TASK_UNINTERRUPTIBLE);
850
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200851 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200853 __generic_unplug_device(q);
854 spin_unlock_irq(q->queue_lock);
855 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200857 /*
858 * After sleeping, we become a "batching" process and
859 * will be able to allocate at least one request, and
860 * up to a big batch of them for a small period time.
861 * See ioc_batching, ioc_set_batching
862 */
863 ioc = current_io_context(GFP_NOIO, q->node);
864 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100865
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200866 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 finish_wait(&rl->wait[rw], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200868
869 rq = get_request(q, rw_flags, bio, GFP_NOIO);
870 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
872 return rq;
873}
874
Jens Axboe165125e2007-07-24 09:28:11 +0200875struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876{
877 struct request *rq;
878
879 BUG_ON(rw != READ && rw != WRITE);
880
Nick Piggind6344532005-06-28 20:45:14 -0700881 spin_lock_irq(q->queue_lock);
882 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200883 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700884 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200885 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700886 if (!rq)
887 spin_unlock_irq(q->queue_lock);
888 }
889 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
891 return rq;
892}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893EXPORT_SYMBOL(blk_get_request);
894
895/**
Jens Axboedc72ef42006-07-20 14:54:05 +0200896 * blk_start_queueing - initiate dispatch of requests to device
897 * @q: request queue to kick into gear
898 *
899 * This is basically a helper to remove the need to know whether a queue
900 * is plugged or not if someone just wants to initiate dispatch of requests
Jens Axboe80a4b582008-10-14 09:51:06 +0200901 * for this queue. Should be used to start queueing on a device outside
902 * of ->request_fn() context. Also see @blk_run_queue.
Jens Axboedc72ef42006-07-20 14:54:05 +0200903 *
904 * The queue lock must be held with interrupts disabled.
905 */
Jens Axboe165125e2007-07-24 09:28:11 +0200906void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +0200907{
Elias Oltmanns336c3d82008-10-01 16:02:33 +0200908 if (!blk_queue_plugged(q)) {
909 if (unlikely(blk_queue_stopped(q)))
910 return;
Jens Axboedc72ef42006-07-20 14:54:05 +0200911 q->request_fn(q);
Elias Oltmanns336c3d82008-10-01 16:02:33 +0200912 } else
Jens Axboedc72ef42006-07-20 14:54:05 +0200913 __generic_unplug_device(q);
914}
915EXPORT_SYMBOL(blk_start_queueing);
916
917/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 * blk_requeue_request - put a request back on queue
919 * @q: request queue where request should be inserted
920 * @rq: request to be inserted
921 *
922 * Description:
923 * Drivers often keep queueing requests until the hardware cannot accept
924 * more, when that condition happens we need to put the request back
925 * on the queue. Must be called with queue lock held.
926 */
Jens Axboe165125e2007-07-24 09:28:11 +0200927void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700929 blk_delete_timer(rq);
930 blk_clear_rq_complete(rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100931 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
932
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 if (blk_rq_tagged(rq))
934 blk_queue_end_tag(q, rq);
935
936 elv_requeue_request(q, rq);
937}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938EXPORT_SYMBOL(blk_requeue_request);
939
940/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200941 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 * @q: request queue where request should be inserted
943 * @rq: request to be inserted
944 * @at_head: insert request at head or tail of queue
945 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 *
947 * Description:
948 * Many block devices need to execute commands asynchronously, so they don't
949 * block the whole kernel from preemption during request execution. This is
950 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200951 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
952 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 *
954 * We have the option of inserting the head or the tail of the queue.
955 * Typically we use the tail for new ioctls and so forth. We use the head
956 * of the queue for things like a QUEUE_FULL message from a device, or a
957 * host that is unable to accept a particular command.
958 */
Jens Axboe165125e2007-07-24 09:28:11 +0200959void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500960 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961{
Tejun Heo 867d1192005-04-24 02:06:05 -0500962 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 unsigned long flags;
964
965 /*
966 * tell I/O scheduler that this isn't a regular read/write (ie it
967 * must not attempt merges on this) and that it acts as a soft
968 * barrier
969 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200970 rq->cmd_type = REQ_TYPE_SPECIAL;
971 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 rq->special = data;
974
975 spin_lock_irqsave(q->queue_lock, flags);
976
977 /*
978 * If command is tagged, release the tag
979 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500980 if (blk_rq_tagged(rq))
981 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200983 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -0500984 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +0200985 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 spin_unlock_irqrestore(q->queue_lock, flags);
987}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988EXPORT_SYMBOL(blk_insert_request);
989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990/*
991 * add-request adds a request to the linked list.
992 * queue lock is held and interrupts disabled, as we muck with the
993 * request queue list.
994 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100995static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996{
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200997 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 /*
1000 * elevator indicated where it wants this request to be
1001 * inserted at elevator_merge time
1002 */
1003 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1004}
Jens Axboe6728cb02008-01-31 13:03:55 +01001005
Tejun Heo074a7ac2008-08-25 19:56:14 +09001006static void part_round_stats_single(int cpu, struct hd_struct *part,
1007 unsigned long now)
1008{
1009 if (now == part->stamp)
1010 return;
1011
1012 if (part->in_flight) {
1013 __part_stat_add(cpu, part, time_in_queue,
1014 part->in_flight * (now - part->stamp));
1015 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1016 }
1017 part->stamp = now;
1018}
1019
1020/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001021 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1022 * @cpu: cpu number for stats access
1023 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 *
1025 * The average IO queue length and utilisation statistics are maintained
1026 * by observing the current state of the queue length and the amount of
1027 * time it has been in this state for.
1028 *
1029 * Normally, that accounting is done on IO completion, but that can result
1030 * in more than a second's worth of IO being accounted for within any one
1031 * second, leading to >100% utilisation. To deal with that, we call this
1032 * function to do a round-off before returning the results when reading
1033 * /proc/diskstats. This accounts immediately for all queue usage up to
1034 * the current jiffies and restarts the counters again.
1035 */
Tejun Heoc9959052008-08-25 19:47:21 +09001036void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001037{
1038 unsigned long now = jiffies;
1039
Tejun Heo074a7ac2008-08-25 19:56:14 +09001040 if (part->partno)
1041 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1042 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001043}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001044EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001045
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046/*
1047 * queue lock must be held
1048 */
Jens Axboe165125e2007-07-24 09:28:11 +02001049void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 if (unlikely(!q))
1052 return;
1053 if (unlikely(--req->ref_count))
1054 return;
1055
Tejun Heo8922e162005-10-20 16:23:44 +02001056 elv_completed_request(q, req);
1057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 /*
1059 * Request may not have originated from ll_rw_blk. if not,
1060 * it didn't come out of our reserved rq pools
1061 */
Jens Axboe49171e52006-08-10 08:59:11 +02001062 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001064 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001067 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
1069 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02001070 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 }
1072}
Mike Christie6e39b692005-11-11 05:30:24 -06001073EXPORT_SYMBOL_GPL(__blk_put_request);
1074
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075void blk_put_request(struct request *req)
1076{
Tejun Heo8922e162005-10-20 16:23:44 +02001077 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001078 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001080 spin_lock_irqsave(q->queue_lock, flags);
1081 __blk_put_request(q, req);
1082 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084EXPORT_SYMBOL(blk_put_request);
1085
Jens Axboe86db1e22008-01-29 14:53:40 +01001086void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001087{
Jens Axboec7c22e42008-09-13 20:26:01 +02001088 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001089 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001090
1091 /*
1092 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1093 */
Mike Christie6000a362008-08-19 18:45:30 -05001094 if (bio_rw_ahead(bio))
1095 req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
1096 REQ_FAILFAST_DRIVER);
1097 if (bio_failfast_dev(bio))
1098 req->cmd_flags |= REQ_FAILFAST_DEV;
1099 if (bio_failfast_transport(bio))
1100 req->cmd_flags |= REQ_FAILFAST_TRANSPORT;
1101 if (bio_failfast_driver(bio))
1102 req->cmd_flags |= REQ_FAILFAST_DRIVER;
Tejun Heo52d9e672006-01-06 09:49:58 +01001103
1104 /*
1105 * REQ_BARRIER implies no merging, but lets make it explicit
1106 */
David Woodhousefb2dce82008-08-05 18:01:53 +01001107 if (unlikely(bio_discard(bio))) {
David Woodhousee17fc0a2008-08-09 16:42:20 +01001108 req->cmd_flags |= REQ_DISCARD;
1109 if (bio_barrier(bio))
1110 req->cmd_flags |= REQ_SOFTBARRIER;
David Woodhousefb2dce82008-08-05 18:01:53 +01001111 req->q->prepare_discard_fn(req->q, req);
David Woodhousee17fc0a2008-08-09 16:42:20 +01001112 } else if (unlikely(bio_barrier(bio)))
1113 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01001114
Jens Axboeb31dc662006-06-13 08:26:10 +02001115 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001116 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02001117 if (bio_rw_meta(bio))
1118 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02001119
Tejun Heo52d9e672006-01-06 09:49:58 +01001120 req->errors = 0;
1121 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001122 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001123 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02001124 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001125}
1126
Jens Axboe165125e2007-07-24 09:28:11 +02001127static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128{
Nick Piggin450991b2005-06-28 20:45:13 -07001129 struct request *req;
David Woodhousefb2dce82008-08-05 18:01:53 +01001130 int el_ret, nr_sectors, barrier, discard, err;
Jens Axboe51da90f2006-07-18 04:14:45 +02001131 const unsigned short prio = bio_prio(bio);
1132 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01001133 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
1137 /*
1138 * low level driver can indicate that it wants pages above a
1139 * certain limit bounced to low memory (ie for highmem, or even
1140 * ISA dma in theory)
1141 */
1142 blk_queue_bounce(q, &bio);
1143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 barrier = bio_barrier(bio);
David Woodhousee17fc0a2008-08-09 16:42:20 +01001145 if (unlikely(barrier) && bio_has_data(bio) &&
1146 (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 err = -EOPNOTSUPP;
1148 goto end_io;
1149 }
1150
David Woodhousefb2dce82008-08-05 18:01:53 +01001151 discard = bio_discard(bio);
1152 if (unlikely(discard) && !q->prepare_discard_fn) {
1153 err = -EOPNOTSUPP;
1154 goto end_io;
1155 }
1156
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 spin_lock_irq(q->queue_lock);
1158
Nick Piggin450991b2005-06-28 20:45:13 -07001159 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 goto get_rq;
1161
1162 el_ret = elv_merge(q, &req, bio);
1163 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001164 case ELEVATOR_BACK_MERGE:
1165 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
Jens Axboe6728cb02008-01-31 13:03:55 +01001167 if (!ll_back_merge_fn(q, req, bio))
1168 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
Jens Axboe6728cb02008-01-31 13:03:55 +01001170 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001171
Jens Axboe6728cb02008-01-31 13:03:55 +01001172 req->biotail->bi_next = bio;
1173 req->biotail = bio;
1174 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1175 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001176 if (!blk_rq_cpu_valid(req))
1177 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001178 drive_stat_acct(req, 0);
1179 if (!attempt_back_merge(q, req))
1180 elv_merged_request(q, req, el_ret);
1181 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
Jens Axboe6728cb02008-01-31 13:03:55 +01001183 case ELEVATOR_FRONT_MERGE:
1184 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Jens Axboe6728cb02008-01-31 13:03:55 +01001186 if (!ll_front_merge_fn(q, req, bio))
1187 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
Jens Axboe6728cb02008-01-31 13:03:55 +01001189 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001190
Jens Axboe6728cb02008-01-31 13:03:55 +01001191 bio->bi_next = req->bio;
1192 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Jens Axboe6728cb02008-01-31 13:03:55 +01001194 /*
1195 * may not be valid. if the low level driver said
1196 * it didn't need a bounce buffer then it better
1197 * not touch req->buffer either...
1198 */
1199 req->buffer = bio_data(bio);
1200 req->current_nr_sectors = bio_cur_sectors(bio);
1201 req->hard_cur_sectors = req->current_nr_sectors;
1202 req->sector = req->hard_sector = bio->bi_sector;
1203 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1204 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001205 if (!blk_rq_cpu_valid(req))
1206 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001207 drive_stat_acct(req, 0);
1208 if (!attempt_front_merge(q, req))
1209 elv_merged_request(q, req, el_ret);
1210 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
Jens Axboe6728cb02008-01-31 13:03:55 +01001212 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1213 default:
1214 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001218 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001219 * This sync check and mask will be re-done in init_request_from_bio(),
1220 * but we need to set it earlier to expose the sync flag to the
1221 * rq allocator and io schedulers.
1222 */
1223 rw_flags = bio_data_dir(bio);
1224 if (sync)
1225 rw_flags |= REQ_RW_SYNC;
1226
1227 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001228 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001229 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001230 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001231 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001232
Nick Piggin450991b2005-06-28 20:45:13 -07001233 /*
1234 * After dropping the lock and possibly sleeping here, our request
1235 * may now be mergeable after it had proven unmergeable (above).
1236 * We don't worry about that case for efficiency. It won't happen
1237 * often, and the elevators are able to handle it.
1238 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001239 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
Nick Piggin450991b2005-06-28 20:45:13 -07001241 spin_lock_irq(q->queue_lock);
Jens Axboec7c22e42008-09-13 20:26:01 +02001242 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1243 bio_flagged(bio, BIO_CPU_AFFINE))
1244 req->cpu = blk_cpu_to_group(smp_processor_id());
Nick Piggin450991b2005-06-28 20:45:13 -07001245 if (elv_queue_empty(q))
1246 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 add_request(q, req);
1248out:
Jens Axboe4a534f92005-04-16 15:25:40 -07001249 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 __generic_unplug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 spin_unlock_irq(q->queue_lock);
1252 return 0;
1253
1254end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02001255 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 return 0;
1257}
1258
1259/*
1260 * If bio->bi_dev is a partition, remap the location
1261 */
1262static inline void blk_partition_remap(struct bio *bio)
1263{
1264 struct block_device *bdev = bio->bi_bdev;
1265
Jens Axboebf2de6f2007-09-27 13:01:25 +02001266 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001268
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 bio->bi_sector += p->start_sect;
1270 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001271
1272 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
1273 bdev->bd_dev, bio->bi_sector,
1274 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 }
1276}
1277
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278static void handle_bad_sector(struct bio *bio)
1279{
1280 char b[BDEVNAME_SIZE];
1281
1282 printk(KERN_INFO "attempt to access beyond end of device\n");
1283 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1284 bdevname(bio->bi_bdev, b),
1285 bio->bi_rw,
1286 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1287 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1288
1289 set_bit(BIO_EOF, &bio->bi_flags);
1290}
1291
Akinobu Mitac17bb492006-12-08 02:39:46 -08001292#ifdef CONFIG_FAIL_MAKE_REQUEST
1293
1294static DECLARE_FAULT_ATTR(fail_make_request);
1295
1296static int __init setup_fail_make_request(char *str)
1297{
1298 return setup_fault_attr(&fail_make_request, str);
1299}
1300__setup("fail_make_request=", setup_fail_make_request);
1301
1302static int should_fail_request(struct bio *bio)
1303{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001304 struct hd_struct *part = bio->bi_bdev->bd_part;
1305
1306 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001307 return should_fail(&fail_make_request, bio->bi_size);
1308
1309 return 0;
1310}
1311
1312static int __init fail_make_request_debugfs(void)
1313{
1314 return init_fault_attr_dentries(&fail_make_request,
1315 "fail_make_request");
1316}
1317
1318late_initcall(fail_make_request_debugfs);
1319
1320#else /* CONFIG_FAIL_MAKE_REQUEST */
1321
1322static inline int should_fail_request(struct bio *bio)
1323{
1324 return 0;
1325}
1326
1327#endif /* CONFIG_FAIL_MAKE_REQUEST */
1328
Jens Axboec07e2b42007-07-18 13:27:58 +02001329/*
1330 * Check whether this bio extends beyond the end of the device.
1331 */
1332static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1333{
1334 sector_t maxsector;
1335
1336 if (!nr_sectors)
1337 return 0;
1338
1339 /* Test device or partition size, when known. */
1340 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1341 if (maxsector) {
1342 sector_t sector = bio->bi_sector;
1343
1344 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1345 /*
1346 * This may well happen - the kernel calls bread()
1347 * without checking the size of the device, e.g., when
1348 * mounting a device.
1349 */
1350 handle_bad_sector(bio);
1351 return 1;
1352 }
1353 }
1354
1355 return 0;
1356}
1357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001359 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 * @bio: The bio describing the location in memory and on the device.
1361 *
1362 * generic_make_request() is used to make I/O requests of block
1363 * devices. It is passed a &struct bio, which describes the I/O that needs
1364 * to be done.
1365 *
1366 * generic_make_request() does not return any status. The
1367 * success/failure status of the request, along with notification of
1368 * completion, is delivered asynchronously through the bio->bi_end_io
1369 * function described (one day) else where.
1370 *
1371 * The caller of generic_make_request must make sure that bi_io_vec
1372 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1373 * set to describe the device address, and the
1374 * bi_end_io and optionally bi_private are set to describe how
1375 * completion notification should be signaled.
1376 *
1377 * generic_make_request and the drivers it calls may use bi_next if this
1378 * bio happens to be merged with someone else, and may change bi_dev and
1379 * bi_sector for remaps as it sees fit. So the values of these fields
1380 * should NOT be depended on after the call to generic_make_request.
1381 */
Neil Brownd89d8792007-05-01 09:53:42 +02001382static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383{
Jens Axboe165125e2007-07-24 09:28:11 +02001384 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001385 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001387 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001388 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
1390 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
Jens Axboec07e2b42007-07-18 13:27:58 +02001392 if (bio_check_eod(bio, nr_sectors))
1393 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
1395 /*
1396 * Resolve the mapping until finished. (drivers are
1397 * still free to implement/resolve their own stacking
1398 * by explicitly returning 0)
1399 *
1400 * NOTE: we don't repeat the blk_size check for each new device.
1401 * Stacking drivers are expected to know what they are doing.
1402 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001403 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001404 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 do {
1406 char b[BDEVNAME_SIZE];
1407
1408 q = bdev_get_queue(bio->bi_bdev);
1409 if (!q) {
1410 printk(KERN_ERR
1411 "generic_make_request: Trying to access "
1412 "nonexistent block-device %s (%Lu)\n",
1413 bdevname(bio->bi_bdev, b),
1414 (long long) bio->bi_sector);
1415end_io:
Jens Axboe51fd77b2007-11-02 08:49:08 +01001416 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 break;
1418 }
1419
Jens Axboe4fa253f2007-07-18 13:13:10 +02001420 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001421 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 bdevname(bio->bi_bdev, b),
1423 bio_sectors(bio),
1424 q->max_hw_sectors);
1425 goto end_io;
1426 }
1427
Nick Pigginfde6ad22005-06-23 00:08:53 -07001428 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 goto end_io;
1430
Akinobu Mitac17bb492006-12-08 02:39:46 -08001431 if (should_fail_request(bio))
1432 goto end_io;
1433
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 /*
1435 * If this device has partitions, remap block n
1436 * of partition p to block n+start(p) of the disk.
1437 */
1438 blk_partition_remap(bio);
1439
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001440 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1441 goto end_io;
1442
NeilBrown5ddfe962006-10-30 22:07:21 -08001443 if (old_sector != -1)
Jens Axboe4fa253f2007-07-18 13:13:10 +02001444 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08001445 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001446
1447 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
1448
NeilBrown5ddfe962006-10-30 22:07:21 -08001449 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001450 old_dev = bio->bi_bdev->bd_dev;
1451
Jens Axboec07e2b42007-07-18 13:27:58 +02001452 if (bio_check_eod(bio, nr_sectors))
1453 goto end_io;
David Woodhousefb2dce82008-08-05 18:01:53 +01001454 if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) ||
1455 (bio_discard(bio) && !q->prepare_discard_fn)) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001456 err = -EOPNOTSUPP;
1457 goto end_io;
1458 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001459
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 ret = q->make_request_fn(q, bio);
1461 } while (ret);
1462}
1463
Neil Brownd89d8792007-05-01 09:53:42 +02001464/*
1465 * We only want one ->make_request_fn to be active at a time,
1466 * else stack usage with stacked devices could be a problem.
1467 * So use current->bio_{list,tail} to keep a list of requests
1468 * submited by a make_request_fn function.
1469 * current->bio_tail is also used as a flag to say if
1470 * generic_make_request is currently active in this task or not.
1471 * If it is NULL, then no make_request is active. If it is non-NULL,
1472 * then a make_request is active, and new requests should be added
1473 * at the tail
1474 */
1475void generic_make_request(struct bio *bio)
1476{
1477 if (current->bio_tail) {
1478 /* make_request is active */
1479 *(current->bio_tail) = bio;
1480 bio->bi_next = NULL;
1481 current->bio_tail = &bio->bi_next;
1482 return;
1483 }
1484 /* following loop may be a bit non-obvious, and so deserves some
1485 * explanation.
1486 * Before entering the loop, bio->bi_next is NULL (as all callers
1487 * ensure that) so we have a list with a single bio.
1488 * We pretend that we have just taken it off a longer list, so
1489 * we assign bio_list to the next (which is NULL) and bio_tail
1490 * to &bio_list, thus initialising the bio_list of new bios to be
1491 * added. __generic_make_request may indeed add some more bios
1492 * through a recursive call to generic_make_request. If it
1493 * did, we find a non-NULL value in bio_list and re-enter the loop
1494 * from the top. In this case we really did just take the bio
1495 * of the top of the list (no pretending) and so fixup bio_list and
1496 * bio_tail or bi_next, and call into __generic_make_request again.
1497 *
1498 * The loop was structured like this to make only one call to
1499 * __generic_make_request (which is important as it is large and
1500 * inlined) and to keep the structure simple.
1501 */
1502 BUG_ON(bio->bi_next);
1503 do {
1504 current->bio_list = bio->bi_next;
1505 if (bio->bi_next == NULL)
1506 current->bio_tail = &current->bio_list;
1507 else
1508 bio->bi_next = NULL;
1509 __generic_make_request(bio);
1510 bio = current->bio_list;
1511 } while (bio);
1512 current->bio_tail = NULL; /* deactivate */
1513}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514EXPORT_SYMBOL(generic_make_request);
1515
1516/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001517 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1519 * @bio: The &struct bio which describes the I/O
1520 *
1521 * submit_bio() is very similar in purpose to generic_make_request(), and
1522 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001523 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 *
1525 */
1526void submit_bio(int rw, struct bio *bio)
1527{
1528 int count = bio_sectors(bio);
1529
Jens Axboe22e2c502005-06-27 10:55:12 +02001530 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
Jens Axboebf2de6f2007-09-27 13:01:25 +02001532 /*
1533 * If it's a regular read/write or a barrier with data attached,
1534 * go through the normal accounting stuff before submission.
1535 */
Jens Axboea9c701e2008-08-08 11:04:44 +02001536 if (bio_has_data(bio)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001537 if (rw & WRITE) {
1538 count_vm_events(PGPGOUT, count);
1539 } else {
1540 task_io_account_read(bio->bi_size);
1541 count_vm_events(PGPGIN, count);
1542 }
1543
1544 if (unlikely(block_dump)) {
1545 char b[BDEVNAME_SIZE];
1546 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001547 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001548 (rw & WRITE) ? "WRITE" : "READ",
1549 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001550 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 }
1553
1554 generic_make_request(bio);
1555}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556EXPORT_SYMBOL(submit_bio);
1557
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001558/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001559 * blk_rq_check_limits - Helper function to check a request for the queue limit
1560 * @q: the queue
1561 * @rq: the request being checked
1562 *
1563 * Description:
1564 * @rq may have been made based on weaker limitations of upper-level queues
1565 * in request stacking drivers, and it may violate the limitation of @q.
1566 * Since the block layer and the underlying device driver trust @rq
1567 * after it is inserted to @q, it should be checked against @q before
1568 * the insertion using this generic function.
1569 *
1570 * This function should also be useful for request stacking drivers
1571 * in some cases below, so export this fuction.
1572 * Request stacking drivers like request-based dm may change the queue
1573 * limits while requests are in the queue (e.g. dm's table swapping).
1574 * Such request stacking drivers should check those requests agaist
1575 * the new queue limits again when they dispatch those requests,
1576 * although such checkings are also done against the old queue limits
1577 * when submitting requests.
1578 */
1579int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1580{
1581 if (rq->nr_sectors > q->max_sectors ||
1582 rq->data_len > q->max_hw_sectors << 9) {
1583 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1584 return -EIO;
1585 }
1586
1587 /*
1588 * queue's settings related to segment counting like q->bounce_pfn
1589 * may differ from that of other stacking queues.
1590 * Recalculate it to check the request correctly on this queue's
1591 * limitation.
1592 */
1593 blk_recalc_rq_segments(rq);
1594 if (rq->nr_phys_segments > q->max_phys_segments ||
1595 rq->nr_phys_segments > q->max_hw_segments) {
1596 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1597 return -EIO;
1598 }
1599
1600 return 0;
1601}
1602EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1603
1604/**
1605 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1606 * @q: the queue to submit the request
1607 * @rq: the request being queued
1608 */
1609int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1610{
1611 unsigned long flags;
1612
1613 if (blk_rq_check_limits(q, rq))
1614 return -EIO;
1615
1616#ifdef CONFIG_FAIL_MAKE_REQUEST
1617 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1618 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1619 return -EIO;
1620#endif
1621
1622 spin_lock_irqsave(q->queue_lock, flags);
1623
1624 /*
1625 * Submitting request must be dequeued before calling this function
1626 * because it will be linked to another request_queue
1627 */
1628 BUG_ON(blk_queued_rq(rq));
1629
1630 drive_stat_acct(rq, 1);
1631 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1632
1633 spin_unlock_irqrestore(q->queue_lock, flags);
1634
1635 return 0;
1636}
1637EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1638
1639/**
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001640 * __end_that_request_first - end I/O on a request
1641 * @req: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001642 * @error: %0 for success, < %0 for error
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001643 * @nr_bytes: number of bytes to complete
1644 *
1645 * Description:
1646 * Ends I/O on a number of bytes attached to @req, and sets it up
1647 * for the next range of segments (if any) in the cluster.
1648 *
1649 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001650 * %0 - we are done with this request, call end_that_request_last()
1651 * %1 - still buffers pending for this request
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001652 **/
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001653static int __end_that_request_first(struct request *req, int error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 int nr_bytes)
1655{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001656 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 struct bio *bio;
1658
Jens Axboe2056a782006-03-23 20:00:26 +01001659 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 /*
Randy Dunlap710027a2008-08-19 20:13:11 +02001662 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 * sense key with us all the way through
1664 */
1665 if (!blk_pc_request(req))
1666 req->errors = 0;
1667
Jens Axboe6728cb02008-01-31 13:03:55 +01001668 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1669 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 req->rq_disk ? req->rq_disk->disk_name : "?",
1671 (unsigned long long)req->sector);
1672 }
1673
Jens Axboed72d9042005-11-01 08:35:42 +01001674 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01001675 const int rw = rq_data_dir(req);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001676 struct hd_struct *part;
Tejun Heoc9959052008-08-25 19:47:21 +09001677 int cpu;
Jens Axboea3623572005-11-01 09:26:16 +01001678
Tejun Heo074a7ac2008-08-25 19:56:14 +09001679 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001680 part = disk_map_sector_rcu(req->rq_disk, req->sector);
Tejun Heo074a7ac2008-08-25 19:56:14 +09001681 part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9);
1682 part_stat_unlock();
Jens Axboed72d9042005-11-01 08:35:42 +01001683 }
1684
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 total_bytes = bio_nbytes = 0;
1686 while ((bio = req->bio) != NULL) {
1687 int nbytes;
1688
Jens Axboebf2de6f2007-09-27 13:01:25 +02001689 /*
1690 * For an empty barrier request, the low level driver must
1691 * store a potential error location in ->sector. We pass
1692 * that back up in ->bi_sector.
1693 */
1694 if (blk_empty_barrier(req))
1695 bio->bi_sector = req->sector;
1696
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 if (nr_bytes >= bio->bi_size) {
1698 req->bio = bio->bi_next;
1699 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02001700 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 next_idx = 0;
1702 bio_nbytes = 0;
1703 } else {
1704 int idx = bio->bi_idx + next_idx;
1705
1706 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1707 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01001708 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -07001709 __func__, bio->bi_idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 break;
1711 }
1712
1713 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1714 BIO_BUG_ON(nbytes > bio->bi_size);
1715
1716 /*
1717 * not a complete bvec done
1718 */
1719 if (unlikely(nbytes > nr_bytes)) {
1720 bio_nbytes += nr_bytes;
1721 total_bytes += nr_bytes;
1722 break;
1723 }
1724
1725 /*
1726 * advance to the next vector
1727 */
1728 next_idx++;
1729 bio_nbytes += nbytes;
1730 }
1731
1732 total_bytes += nbytes;
1733 nr_bytes -= nbytes;
1734
Jens Axboe6728cb02008-01-31 13:03:55 +01001735 bio = req->bio;
1736 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 /*
1738 * end more in this run, or just return 'not-done'
1739 */
1740 if (unlikely(nr_bytes <= 0))
1741 break;
1742 }
1743 }
1744
1745 /*
1746 * completely done
1747 */
1748 if (!req->bio)
1749 return 0;
1750
1751 /*
1752 * if the request wasn't completed, update state
1753 */
1754 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02001755 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 bio->bi_idx += next_idx;
1757 bio_iovec(bio)->bv_offset += nr_bytes;
1758 bio_iovec(bio)->bv_len -= nr_bytes;
1759 }
1760
1761 blk_recalc_rq_sectors(req, total_bytes >> 9);
1762 blk_recalc_rq_segments(req);
1763 return 1;
1764}
1765
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766/*
1767 * queue lock must be held
1768 */
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001769static void end_that_request_last(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770{
1771 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01001772
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001773 if (blk_rq_tagged(req))
1774 blk_queue_end_tag(req->q, req);
1775
1776 if (blk_queued_rq(req))
1777 blkdev_dequeue_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
1779 if (unlikely(laptop_mode) && blk_fs_request(req))
1780 laptop_io_completion();
1781
Mike Andersone78042e2008-10-30 02:16:20 -07001782 blk_delete_timer(req);
1783
Jens Axboefd0ff8a2006-05-23 11:23:49 +02001784 /*
1785 * Account IO completion. bar_rq isn't accounted as a normal
1786 * IO on queueing nor completion. Accounting the containing
1787 * request is enough.
1788 */
1789 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01001791 const int rw = rq_data_dir(req);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001792 struct hd_struct *part;
Tejun Heoc9959052008-08-25 19:47:21 +09001793 int cpu;
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001794
Tejun Heo074a7ac2008-08-25 19:56:14 +09001795 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001796 part = disk_map_sector_rcu(disk, req->sector);
Jens Axboea3623572005-11-01 09:26:16 +01001797
Tejun Heo074a7ac2008-08-25 19:56:14 +09001798 part_stat_inc(cpu, part, ios[rw]);
1799 part_stat_add(cpu, part, ticks[rw], duration);
1800 part_round_stats(cpu, part);
1801 part_dec_in_flight(part);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001802
Tejun Heo074a7ac2008-08-25 19:56:14 +09001803 part_stat_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 }
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001805
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01001807 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001808 else {
1809 if (blk_bidi_rq(req))
1810 __blk_put_request(req->next_rq->q, req->next_rq);
1811
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 }
1814}
1815
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001816/**
1817 * blk_rq_bytes - Returns bytes left to complete in the entire request
Randy Dunlap5d87a052008-02-20 09:01:22 +01001818 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001819 **/
1820unsigned int blk_rq_bytes(struct request *rq)
Jens Axboea0cd1282007-09-21 10:41:07 +02001821{
1822 if (blk_fs_request(rq))
1823 return rq->hard_nr_sectors << 9;
1824
1825 return rq->data_len;
1826}
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001827EXPORT_SYMBOL_GPL(blk_rq_bytes);
1828
1829/**
1830 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
Randy Dunlap5d87a052008-02-20 09:01:22 +01001831 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001832 **/
1833unsigned int blk_rq_cur_bytes(struct request *rq)
1834{
1835 if (blk_fs_request(rq))
1836 return rq->current_nr_sectors << 9;
1837
1838 if (rq->bio)
1839 return rq->bio->bi_size;
1840
1841 return rq->data_len;
1842}
1843EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
Jens Axboea0cd1282007-09-21 10:41:07 +02001844
1845/**
Jens Axboea0cd1282007-09-21 10:41:07 +02001846 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07001847 * @req: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001848 * @uptodate: error value or %0/%1 uptodate flag
Jens Axboea0cd1282007-09-21 10:41:07 +02001849 *
1850 * Description:
1851 * Ends I/O on the current segment of a request. If that is the only
1852 * remaining segment, the request is also completed and freed.
1853 *
Randy Dunlap710027a2008-08-19 20:13:11 +02001854 * This is a remnant of how older block drivers handled I/O completions.
1855 * Modern drivers typically end I/O on the full request in one go, unless
Jens Axboea0cd1282007-09-21 10:41:07 +02001856 * they have a residual value to account for. For that case this function
1857 * isn't really useful, unless the residual just happens to be the
1858 * full current segment. In other words, don't use this function in new
Kiyoshi Uedad00e29f2008-10-01 10:14:46 -04001859 * code. Use blk_end_request() or __blk_end_request() to end a request.
Jens Axboea0cd1282007-09-21 10:41:07 +02001860 **/
1861void end_request(struct request *req, int uptodate)
1862{
Kiyoshi Uedad00e29f2008-10-01 10:14:46 -04001863 int error = 0;
1864
1865 if (uptodate <= 0)
1866 error = uptodate ? uptodate : -EIO;
1867
1868 __blk_end_request(req, error, req->hard_cur_sectors << 9);
Jens Axboea0cd1282007-09-21 10:41:07 +02001869}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870EXPORT_SYMBOL(end_request);
1871
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04001872static int end_that_request_data(struct request *rq, int error,
1873 unsigned int nr_bytes, unsigned int bidi_bytes)
1874{
1875 if (rq->bio) {
1876 if (__end_that_request_first(rq, error, nr_bytes))
1877 return 1;
1878
1879 /* Bidi request must be completed as a whole */
1880 if (blk_bidi_rq(rq) &&
1881 __end_that_request_first(rq->next_rq, error, bidi_bytes))
1882 return 1;
1883 }
1884
1885 return 0;
1886}
1887
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001888/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001889 * blk_end_io - Generic end_io function to complete a request.
1890 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001891 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001892 * @nr_bytes: number of bytes to complete @rq
1893 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001894 * @drv_callback: function called between completion of bios in the request
1895 * and completion of the request.
Randy Dunlap710027a2008-08-19 20:13:11 +02001896 * If the callback returns non %0, this helper returns without
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001897 * completion of the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001898 *
1899 * Description:
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001900 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001901 * If @rq has leftover, sets it up for the next range of segments.
1902 *
1903 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001904 * %0 - we are done with this request
1905 * %1 - this request is not freed yet, it still has pending buffers.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001906 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001907static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1908 unsigned int bidi_bytes,
1909 int (drv_callback)(struct request *))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001910{
1911 struct request_queue *q = rq->q;
1912 unsigned long flags = 0UL;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001913
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04001914 if (end_that_request_data(rq, error, nr_bytes, bidi_bytes))
1915 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001916
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001917 /* Special feature for tricky drivers */
1918 if (drv_callback && drv_callback(rq))
1919 return 1;
1920
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001921 add_disk_randomness(rq->rq_disk);
1922
1923 spin_lock_irqsave(q->queue_lock, flags);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001924 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001925 spin_unlock_irqrestore(q->queue_lock, flags);
1926
1927 return 0;
1928}
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001929
1930/**
1931 * blk_end_request - Helper function for drivers to complete the request.
1932 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001933 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001934 * @nr_bytes: number of bytes to complete
1935 *
1936 * Description:
1937 * Ends I/O on a number of bytes attached to @rq.
1938 * If @rq has leftover, sets it up for the next range of segments.
1939 *
1940 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001941 * %0 - we are done with this request
1942 * %1 - still buffers pending for this request
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001943 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001944int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001945{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001946 return blk_end_io(rq, error, nr_bytes, 0, NULL);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001947}
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001948EXPORT_SYMBOL_GPL(blk_end_request);
1949
1950/**
1951 * __blk_end_request - Helper function for drivers to complete the request.
1952 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001953 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001954 * @nr_bytes: number of bytes to complete
1955 *
1956 * Description:
1957 * Must be called with queue lock held unlike blk_end_request().
1958 *
1959 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001960 * %0 - we are done with this request
1961 * %1 - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001962 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001963int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001964{
Jens Axboe60540162008-08-26 13:34:34 +02001965 if (rq->bio && __end_that_request_first(rq, error, nr_bytes))
Jens Axboe051cc392008-08-08 11:06:45 +02001966 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001967
1968 add_disk_randomness(rq->rq_disk);
1969
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001970 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001971
1972 return 0;
1973}
1974EXPORT_SYMBOL_GPL(__blk_end_request);
1975
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001976/**
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001977 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1978 * @rq: the bidi request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001979 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001980 * @nr_bytes: number of bytes to complete @rq
1981 * @bidi_bytes: number of bytes to complete @rq->next_rq
1982 *
1983 * Description:
1984 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1985 *
1986 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001987 * %0 - we are done with this request
1988 * %1 - still buffers pending for this request
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001989 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001990int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
1991 unsigned int bidi_bytes)
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001992{
1993 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
1994}
1995EXPORT_SYMBOL_GPL(blk_end_bidi_request);
1996
1997/**
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04001998 * blk_update_request - Special helper function for request stacking drivers
1999 * @rq: the request being processed
2000 * @error: %0 for success, < %0 for error
2001 * @nr_bytes: number of bytes to complete @rq
2002 *
2003 * Description:
2004 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
2005 * the request structure even if @rq doesn't have leftover.
2006 * If @rq has leftover, sets it up for the next range of segments.
2007 *
2008 * This special helper function is only for request stacking drivers
2009 * (e.g. request-based dm) so that they can handle partial completion.
2010 * Actual device drivers should use blk_end_request instead.
2011 */
2012void blk_update_request(struct request *rq, int error, unsigned int nr_bytes)
2013{
2014 if (!end_that_request_data(rq, error, nr_bytes, 0)) {
2015 /*
2016 * These members are not updated in end_that_request_data()
2017 * when all bios are completed.
2018 * Update them so that the request stacking driver can find
2019 * how many bytes remain in the request later.
2020 */
2021 rq->nr_sectors = rq->hard_nr_sectors = 0;
2022 rq->current_nr_sectors = rq->hard_cur_sectors = 0;
2023 }
2024}
2025EXPORT_SYMBOL_GPL(blk_update_request);
2026
2027/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002028 * blk_end_request_callback - Special helper function for tricky drivers
2029 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02002030 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002031 * @nr_bytes: number of bytes to complete
2032 * @drv_callback: function called between completion of bios in the request
2033 * and completion of the request.
Randy Dunlap710027a2008-08-19 20:13:11 +02002034 * If the callback returns non %0, this helper returns without
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002035 * completion of the request.
2036 *
2037 * Description:
2038 * Ends I/O on a number of bytes attached to @rq.
2039 * If @rq has leftover, sets it up for the next range of segments.
2040 *
2041 * This special helper function is used only for existing tricky drivers.
2042 * (e.g. cdrom_newpc_intr() of ide-cd)
2043 * This interface will be removed when such drivers are rewritten.
2044 * Don't use this interface in other places anymore.
2045 *
2046 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02002047 * %0 - we are done with this request
2048 * %1 - this request is not freed yet.
2049 * this request still has pending buffers or
2050 * the driver doesn't want to finish this request yet.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002051 **/
Jens Axboe22b13212008-01-31 12:36:19 +01002052int blk_end_request_callback(struct request *rq, int error,
2053 unsigned int nr_bytes,
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002054 int (drv_callback)(struct request *))
2055{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002056 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002057}
2058EXPORT_SYMBOL_GPL(blk_end_request_callback);
2059
Jens Axboe86db1e22008-01-29 14:53:40 +01002060void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2061 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062{
David Woodhoused628eae2008-08-09 16:22:17 +01002063 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2064 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002065 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066
David Woodhousefb2dce82008-08-05 18:01:53 +01002067 if (bio_has_data(bio)) {
2068 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002069 rq->buffer = bio_data(bio);
2070 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 rq->current_nr_sectors = bio_cur_sectors(bio);
2072 rq->hard_cur_sectors = rq->current_nr_sectors;
2073 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002074 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
2076 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
NeilBrown66846572007-08-16 13:31:28 +02002078 if (bio->bi_bdev)
2079 rq->rq_disk = bio->bi_bdev->bd_disk;
2080}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002082/**
2083 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2084 * @q : the queue of the device being checked
2085 *
2086 * Description:
2087 * Check if underlying low-level drivers of a device are busy.
2088 * If the drivers want to export their busy state, they must set own
2089 * exporting function using blk_queue_lld_busy() first.
2090 *
2091 * Basically, this function is used only by request stacking drivers
2092 * to stop dispatching requests to underlying devices when underlying
2093 * devices are busy. This behavior helps more I/O merging on the queue
2094 * of the request stacking driver and prevents I/O throughput regression
2095 * on burst I/O load.
2096 *
2097 * Return:
2098 * 0 - Not busy (The request stacking driver should dispatch request)
2099 * 1 - Busy (The request stacking driver should stop dispatching request)
2100 */
2101int blk_lld_busy(struct request_queue *q)
2102{
2103 if (q->lld_busy_fn)
2104 return q->lld_busy_fn(q);
2105
2106 return 0;
2107}
2108EXPORT_SYMBOL_GPL(blk_lld_busy);
2109
Jens Axboe18887ad2008-07-28 13:08:45 +02002110int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111{
2112 return queue_work(kblockd_workqueue, work);
2113}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114EXPORT_SYMBOL(kblockd_schedule_work);
2115
Andrew Morton19a75d82007-05-09 02:33:56 -07002116void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07002118 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119}
Andrew Morton19a75d82007-05-09 02:33:56 -07002120EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121
2122int __init blk_dev_init(void)
2123{
2124 kblockd_workqueue = create_workqueue("kblockd");
2125 if (!kblockd_workqueue)
2126 panic("Failed to create kblockd\n");
2127
2128 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002129 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130
Jens Axboe8324aa92008-01-29 14:51:59 +01002131 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002132 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 return 0;
2135}
2136