blob: 6a9cc0d22a61c1aa2300d18c9728a7546c15cf0a [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 Axboeff856ba2006-01-09 16:02:34 +010029#include <linux/interrupt.h>
30#include <linux/cpu.h>
Jens Axboe2056a782006-03-23 20:00:26 +010031#include <linux/blktrace_api.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080032#include <linux/fault-inject.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Jens Axboe8324aa92008-01-29 14:51:59 +010034#include "blk.h"
35
Jens Axboe165125e2007-07-24 09:28:11 +020036static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38/*
39 * For the allocated request tables
40 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010041static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43/*
44 * For queue allocation
45 */
Jens Axboe6728cb02008-01-31 13:03:55 +010046struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 * Controlling structure to kblockd
50 */
Jens Axboeff856ba2006-01-09 16:02:34 +010051static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Jens Axboeff856ba2006-01-09 16:02:34 +010053static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
54
Jens Axboe26b82562008-01-29 13:54:41 +010055static void drive_stat_acct(struct request *rq, int new_io)
56{
Jens Axboe28f13702008-05-07 10:15:46 +020057 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010058 int rw = rq_data_dir(rq);
59
60 if (!blk_fs_request(rq) || !rq->rq_disk)
61 return;
62
Jens Axboe28f13702008-05-07 10:15:46 +020063 part = get_part(rq->rq_disk, rq->sector);
64 if (!new_io)
65 __all_stat_inc(rq->rq_disk, part, merges[rw], rq->sector);
66 else {
Jens Axboe26b82562008-01-29 13:54:41 +010067 disk_round_stats(rq->rq_disk);
68 rq->rq_disk->in_flight++;
Jerome Marchand6f2576a2008-02-08 11:04:35 +010069 if (part) {
70 part_round_stats(part);
71 part->in_flight++;
72 }
Jens Axboe26b82562008-01-29 13:54:41 +010073 }
74}
75
Jens Axboe8324aa92008-01-29 14:51:59 +010076void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070077{
78 int nr;
79
80 nr = q->nr_requests - (q->nr_requests / 8) + 1;
81 if (nr > q->nr_requests)
82 nr = q->nr_requests;
83 q->nr_congestion_on = nr;
84
85 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
86 if (nr < 1)
87 nr = 1;
88 q->nr_congestion_off = nr;
89}
90
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/**
92 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
93 * @bdev: device
94 *
95 * Locates the passed device's request queue and returns the address of its
96 * backing_dev_info
97 *
98 * Will return NULL if the request queue cannot be located.
99 */
100struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
101{
102 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200103 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
105 if (q)
106 ret = &q->backing_dev_info;
107 return ret;
108}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109EXPORT_SYMBOL(blk_get_backing_dev_info);
110
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200111void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200113 memset(rq, 0, sizeof(*rq));
114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100116 INIT_LIST_HEAD(&rq->donelist);
Jens Axboe63a71382008-02-08 12:41:03 +0100117 rq->q = q;
118 rq->sector = rq->hard_sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200119 INIT_HLIST_NODE(&rq->hash);
120 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200121 rq->cmd = rq->__cmd;
Jens Axboe63a71382008-02-08 12:41:03 +0100122 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100123 rq->ref_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200125EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
NeilBrown5bb23a62007-09-27 12:46:13 +0200127static void req_bio_endio(struct request *rq, struct bio *bio,
128 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100129{
Jens Axboe165125e2007-07-24 09:28:11 +0200130 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100131
NeilBrown5bb23a62007-09-27 12:46:13 +0200132 if (&q->bar_rq != rq) {
133 if (error)
134 clear_bit(BIO_UPTODATE, &bio->bi_flags);
135 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
136 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100137
NeilBrown5bb23a62007-09-27 12:46:13 +0200138 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100139 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700140 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200141 nbytes = bio->bi_size;
142 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100143
NeilBrown5bb23a62007-09-27 12:46:13 +0200144 bio->bi_size -= nbytes;
145 bio->bi_sector += (nbytes >> 9);
146 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200147 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200148 } else {
149
150 /*
151 * Okay, this is the barrier request in progress, just
152 * record the error;
153 */
154 if (error && !q->orderr)
155 q->orderr = error;
156 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100157}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159void blk_dump_rq_flags(struct request *rq, char *msg)
160{
161 int bit;
162
Jens Axboe6728cb02008-01-31 13:03:55 +0100163 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200164 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
165 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Jens Axboe6728cb02008-01-31 13:03:55 +0100167 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
168 (unsigned long long)rq->sector,
169 rq->nr_sectors,
170 rq->current_nr_sectors);
171 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
172 rq->bio, rq->biotail,
173 rq->buffer, rq->data,
174 rq->data_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Jens Axboe4aff5e22006-08-10 08:44:47 +0200176 if (blk_pc_request(rq)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100177 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200178 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 printk("%02x ", rq->cmd[bit]);
180 printk("\n");
181 }
182}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183EXPORT_SYMBOL(blk_dump_rq_flags);
184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185/*
186 * "plug" the device if there are no outstanding requests: this will
187 * force the transfer to start only after we have put all the requests
188 * on the list.
189 *
190 * This is called with interrupts off and no requests on the queue and
191 * with the queue lock held.
192 */
Jens Axboe165125e2007-07-24 09:28:11 +0200193void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194{
195 WARN_ON(!irqs_disabled());
196
197 /*
198 * don't plug a stopped queue, it must be paired with blk_start_queue()
199 * which will restart the queueing
200 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200201 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 return;
203
Nick Piggin75ad23b2008-04-29 14:48:33 +0200204 if (!test_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
205 __set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +0100207 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210EXPORT_SYMBOL(blk_plug_device);
211
212/*
213 * remove the queue from the plugged list, if present. called with
214 * queue lock held and interrupts disabled.
215 */
Jens Axboe165125e2007-07-24 09:28:11 +0200216int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
218 WARN_ON(!irqs_disabled());
219
Nick Piggin75ad23b2008-04-29 14:48:33 +0200220 if (!test_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 return 0;
222
Nick Piggin75ad23b2008-04-29 14:48:33 +0200223 queue_flag_clear(QUEUE_FLAG_PLUGGED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 del_timer(&q->unplug_timer);
225 return 1;
226}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227EXPORT_SYMBOL(blk_remove_plug);
228
229/*
230 * remove the plug and let it rip..
231 */
Jens Axboe165125e2007-07-24 09:28:11 +0200232void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200234 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 return;
236
237 if (!blk_remove_plug(q))
238 return;
239
Jens Axboe22e2c502005-06-27 10:55:12 +0200240 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242EXPORT_SYMBOL(__generic_unplug_device);
243
244/**
245 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200246 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 *
248 * Description:
249 * Linux uses plugging to build bigger requests queues before letting
250 * the device have at them. If a queue is plugged, the I/O scheduler
251 * is still adding and merging requests on the queue. Once the queue
252 * gets unplugged, the request_fn defined for the queue is invoked and
253 * transfers started.
254 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200255void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200257 if (blk_queue_plugged(q)) {
258 spin_lock_irq(q->queue_lock);
259 __generic_unplug_device(q);
260 spin_unlock_irq(q->queue_lock);
261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262}
263EXPORT_SYMBOL(generic_unplug_device);
264
265static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
266 struct page *page)
267{
Jens Axboe165125e2007-07-24 09:28:11 +0200268 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500270 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
272
Jens Axboe86db1e22008-01-29 14:53:40 +0100273void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
Jens Axboe165125e2007-07-24 09:28:11 +0200275 struct request_queue *q =
276 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
Jens Axboe2056a782006-03-23 20:00:26 +0100278 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
279 q->rq.count[READ] + q->rq.count[WRITE]);
280
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 q->unplug_fn(q);
282}
283
Jens Axboe86db1e22008-01-29 14:53:40 +0100284void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285{
Jens Axboe165125e2007-07-24 09:28:11 +0200286 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Jens Axboe2056a782006-03-23 20:00:26 +0100288 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
289 q->rq.count[READ] + q->rq.count[WRITE]);
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 kblockd_schedule_work(&q->unplug_work);
292}
293
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500294void blk_unplug(struct request_queue *q)
295{
296 /*
297 * devices don't necessarily have an ->unplug_fn defined
298 */
299 if (q->unplug_fn) {
300 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
301 q->rq.count[READ] + q->rq.count[WRITE]);
302
303 q->unplug_fn(q);
304 }
305}
306EXPORT_SYMBOL(blk_unplug);
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308/**
309 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200310 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 *
312 * Description:
313 * blk_start_queue() will clear the stop flag on the queue, and call
314 * the request_fn for the queue if it was in a stopped state when
315 * entered. Also see blk_stop_queue(). Queue lock must be held.
316 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200317void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200319 WARN_ON(!irqs_disabled());
320
Nick Piggin75ad23b2008-04-29 14:48:33 +0200321 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
323 /*
324 * one level of recursion is ok and is much faster than kicking
325 * the unplug handling
326 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200327 if (!test_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
328 queue_flag_set(QUEUE_FLAG_REENTER, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200330 queue_flag_clear(QUEUE_FLAG_REENTER, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 } else {
332 blk_plug_device(q);
333 kblockd_schedule_work(&q->unplug_work);
334 }
335}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336EXPORT_SYMBOL(blk_start_queue);
337
338/**
339 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200340 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 *
342 * Description:
343 * The Linux block layer assumes that a block driver will consume all
344 * entries on the request queue when the request_fn strategy is called.
345 * Often this will not happen, because of hardware limitations (queue
346 * depth settings). If a device driver gets a 'queue full' response,
347 * or if it simply chooses not to queue more I/O at one point, it can
348 * call this function to prevent the request_fn from being called until
349 * the driver has signalled it's ready to go again. This happens by calling
350 * blk_start_queue() to restart queue operations. Queue lock must be held.
351 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200352void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
354 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200355 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356}
357EXPORT_SYMBOL(blk_stop_queue);
358
359/**
360 * blk_sync_queue - cancel any pending callbacks on a queue
361 * @q: the queue
362 *
363 * Description:
364 * The block layer may perform asynchronous callback activity
365 * on a queue, such as calling the unplug function after a timeout.
366 * A block device may call blk_sync_queue to ensure that any
367 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200368 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 * that its ->make_request_fn will not re-add plugging prior to calling
370 * this function.
371 *
372 */
373void blk_sync_queue(struct request_queue *q)
374{
375 del_timer_sync(&q->unplug_timer);
Oleg Nesterovabbeb882007-10-23 15:08:19 +0200376 kblockd_flush_work(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
378EXPORT_SYMBOL(blk_sync_queue);
379
380/**
381 * blk_run_queue - run a single device queue
382 * @q: The queue to run
383 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200384void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200387
388 /*
389 * Only recurse once to avoid overrunning the stack, let the unplug
390 * handling reinvoke the handler shortly if we already got there.
391 */
392 if (!elv_queue_empty(q)) {
Nick Piggin75ad23b2008-04-29 14:48:33 +0200393 if (!test_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
394 queue_flag_set(QUEUE_FLAG_REENTER, q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200395 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200396 queue_flag_clear(QUEUE_FLAG_REENTER, q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200397 } else {
398 blk_plug_device(q);
399 kblockd_schedule_work(&q->unplug_work);
400 }
401 }
Nick Piggin75ad23b2008-04-29 14:48:33 +0200402}
403EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200404
Nick Piggin75ad23b2008-04-29 14:48:33 +0200405/**
406 * blk_run_queue - run a single device queue
407 * @q: The queue to run
408 */
409void blk_run_queue(struct request_queue *q)
410{
411 unsigned long flags;
412
413 spin_lock_irqsave(q->queue_lock, flags);
414 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 spin_unlock_irqrestore(q->queue_lock, flags);
416}
417EXPORT_SYMBOL(blk_run_queue);
418
Jens Axboe165125e2007-07-24 09:28:11 +0200419void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500420{
421 kobject_put(&q->kobj);
422}
Al Viro483f4af2006-03-18 18:34:37 -0500423
Jens Axboe6728cb02008-01-31 13:03:55 +0100424void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500425{
426 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200427 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500428 mutex_unlock(&q->sysfs_lock);
429
430 if (q->elevator)
431 elevator_exit(q->elevator);
432
433 blk_put_queue(q);
434}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435EXPORT_SYMBOL(blk_cleanup_queue);
436
Jens Axboe165125e2007-07-24 09:28:11 +0200437static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438{
439 struct request_list *rl = &q->rq;
440
441 rl->count[READ] = rl->count[WRITE] = 0;
442 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200443 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 init_waitqueue_head(&rl->wait[READ]);
445 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Christoph Lameter19460892005-06-23 00:08:19 -0700447 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
448 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
450 if (!rl->rq_pool)
451 return -ENOMEM;
452
453 return 0;
454}
455
Jens Axboe165125e2007-07-24 09:28:11 +0200456struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
Christoph Lameter19460892005-06-23 00:08:19 -0700458 return blk_alloc_queue_node(gfp_mask, -1);
459}
460EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Jens Axboe165125e2007-07-24 09:28:11 +0200462struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700463{
Jens Axboe165125e2007-07-24 09:28:11 +0200464 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700465 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700466
Jens Axboe8324aa92008-01-29 14:51:59 +0100467 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700468 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 if (!q)
470 return NULL;
471
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700472 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
473 q->backing_dev_info.unplug_io_data = q;
474 err = bdi_init(&q->backing_dev_info);
475 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100476 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700477 return NULL;
478 }
479
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500481
Jens Axboe8324aa92008-01-29 14:51:59 +0100482 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
Al Viro483f4af2006-03-18 18:34:37 -0500484 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700485 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500486
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 return q;
488}
Christoph Lameter19460892005-06-23 00:08:19 -0700489EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
491/**
492 * blk_init_queue - prepare a request queue for use with a block device
493 * @rfn: The function to be called to process requests that have been
494 * placed on the queue.
495 * @lock: Request queue spin lock
496 *
497 * Description:
498 * If a block device wishes to use the standard request handling procedures,
499 * which sorts requests and coalesces adjacent requests, then it must
500 * call blk_init_queue(). The function @rfn will be called when there
501 * are requests on the queue that need to be processed. If the device
502 * supports plugging, then @rfn may not be called immediately when requests
503 * are available on the queue, but may be called at some time later instead.
504 * Plugged queues are generally unplugged when a buffer belonging to one
505 * of the requests on the queue is needed, or due to memory pressure.
506 *
507 * @rfn is not required, or even expected, to remove all requests off the
508 * queue, but only as many as it can handle at a time. If it does leave
509 * requests on the queue, it is responsible for arranging that the requests
510 * get dealt with eventually.
511 *
512 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200513 * request queue; this lock will be taken also from interrupt context, so irq
514 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 *
516 * Function returns a pointer to the initialized request queue, or NULL if
517 * it didn't succeed.
518 *
519 * Note:
520 * blk_init_queue() must be paired with a blk_cleanup_queue() call
521 * when the block device is deactivated (such as at module unload).
522 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700523
Jens Axboe165125e2007-07-24 09:28:11 +0200524struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
Christoph Lameter19460892005-06-23 00:08:19 -0700526 return blk_init_queue_node(rfn, lock, -1);
527}
528EXPORT_SYMBOL(blk_init_queue);
529
Jens Axboe165125e2007-07-24 09:28:11 +0200530struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700531blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
532{
Jens Axboe165125e2007-07-24 09:28:11 +0200533 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535 if (!q)
536 return NULL;
537
Christoph Lameter19460892005-06-23 00:08:19 -0700538 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -0500539 if (blk_init_free_list(q)) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100540 kmem_cache_free(blk_requestq_cachep, q);
Al Viro8669aaf2006-03-18 13:50:00 -0500541 return NULL;
542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
152587d2005-04-12 16:22:06 -0500544 /*
545 * if caller didn't supply a lock, they get per-queue locking with
546 * our embedded lock
547 */
Neil Browne7e72bf2008-05-14 16:05:54 -0700548 if (!lock)
152587d2005-04-12 16:22:06 -0500549 lock = &q->__queue_lock;
152587d2005-04-12 16:22:06 -0500550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 q->prep_rq_fn = NULL;
553 q->unplug_fn = generic_unplug_device;
554 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
555 q->queue_lock = lock;
556
557 blk_queue_segment_boundary(q, 0xffffffff);
558
559 blk_queue_make_request(q, __make_request);
560 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
561
562 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
563 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
564
Alan Stern44ec9542007-02-20 11:01:57 -0500565 q->sg_reserved_size = INT_MAX;
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 /*
568 * all done
569 */
570 if (!elevator_init(q, NULL)) {
571 blk_queue_congestion_threshold(q);
572 return q;
573 }
574
Al Viro8669aaf2006-03-18 13:50:00 -0500575 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 return NULL;
577}
Christoph Lameter19460892005-06-23 00:08:19 -0700578EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
Jens Axboe165125e2007-07-24 09:28:11 +0200580int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700582 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500583 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 return 0;
585 }
586
587 return 1;
588}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Jens Axboe165125e2007-07-24 09:28:11 +0200590static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200592 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200593 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 mempool_free(rq, q->rq.rq_pool);
595}
596
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200597static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +0200598blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599{
600 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
601
602 if (!rq)
603 return NULL;
604
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200605 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +0200608 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 * see bio.h and blkdev.h
610 */
Jens Axboe49171e52006-08-10 08:59:11 +0200611 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Tejun Heocb98fc82005-10-28 08:29:39 +0200613 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200614 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200615 mempool_free(rq, q->rq.rq_pool);
616 return NULL;
617 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200618 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Tejun Heocb98fc82005-10-28 08:29:39 +0200621 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622}
623
624/*
625 * ioc_batching returns true if the ioc is a valid batching request and
626 * should be given priority access to a request.
627 */
Jens Axboe165125e2007-07-24 09:28:11 +0200628static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
630 if (!ioc)
631 return 0;
632
633 /*
634 * Make sure the process is able to allocate at least 1 request
635 * even if the batch times out, otherwise we could theoretically
636 * lose wakeups.
637 */
638 return ioc->nr_batch_requests == q->nr_batching ||
639 (ioc->nr_batch_requests > 0
640 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
641}
642
643/*
644 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
645 * will cause the process to be a "batcher" on all queues in the system. This
646 * is the behaviour we want though - once it gets a wakeup it should be given
647 * a nice run.
648 */
Jens Axboe165125e2007-07-24 09:28:11 +0200649static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
651 if (!ioc || ioc_batching(q, ioc))
652 return;
653
654 ioc->nr_batch_requests = q->nr_batching;
655 ioc->last_waited = jiffies;
656}
657
Jens Axboe165125e2007-07-24 09:28:11 +0200658static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
660 struct request_list *rl = &q->rq;
661
662 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -0700663 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
665 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 if (waitqueue_active(&rl->wait[rw]))
667 wake_up(&rl->wait[rw]);
668
669 blk_clear_queue_full(q, rw);
670 }
671}
672
673/*
674 * A request has just been released. Account for it, update the full and
675 * congestion status, wake up any waiters. Called under q->queue_lock.
676 */
Jens Axboe165125e2007-07-24 09:28:11 +0200677static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
679 struct request_list *rl = &q->rq;
680
681 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200682 if (priv)
683 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
685 __freed_request(q, rw);
686
687 if (unlikely(rl->starved[rw ^ 1]))
688 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689}
690
691#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
692/*
Nick Piggind6344532005-06-28 20:45:14 -0700693 * Get a free request, queue_lock must be held.
694 * Returns NULL on failure, with queue_lock held.
695 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 */
Jens Axboe165125e2007-07-24 09:28:11 +0200697static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100698 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
700 struct request *rq = NULL;
701 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100702 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +0100703 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100704 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
Jens Axboe7749a8d2006-12-13 13:02:26 +0100706 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100707 if (may_queue == ELV_MQUEUE_NO)
708 goto rq_starved;
709
710 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
711 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200712 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100713 /*
714 * The queue will fill after this allocation, so set
715 * it as full, and mark this process as "batching".
716 * This process will be allowed to complete a batch of
717 * requests, others will be blocked.
718 */
719 if (!blk_queue_full(q, rw)) {
720 ioc_set_batching(q, ioc);
721 blk_set_queue_full(q, rw);
722 } else {
723 if (may_queue != ELV_MQUEUE_MUST
724 && !ioc_batching(q, ioc)) {
725 /*
726 * The queue is full and the allocating
727 * process is not a "batcher", and not
728 * exempted by the IO scheduler
729 */
730 goto out;
731 }
732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 }
Thomas Maier79e2de42006-10-19 23:28:15 -0700734 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 }
736
Jens Axboe082cf692005-06-28 16:35:11 +0200737 /*
738 * Only allow batching queuers to allocate up to 50% over the defined
739 * limit of requests, otherwise we could have thousands of requests
740 * allocated with any setting of ->nr_requests
741 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100742 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200743 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100744
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 rl->count[rw]++;
746 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200747
Jens Axboe64521d12005-10-28 08:30:39 +0200748 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200749 if (priv)
750 rl->elvpriv++;
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 spin_unlock_irq(q->queue_lock);
753
Jens Axboe7749a8d2006-12-13 13:02:26 +0100754 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100755 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 /*
757 * Allocation failed presumably due to memory. Undo anything
758 * we might have messed up.
759 *
760 * Allocating task should really be put onto the front of the
761 * wait queue, but this is pretty rare.
762 */
763 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +0200764 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766 /*
767 * in the very unlikely event that allocation failed and no
768 * requests for this direction was pending, mark us starved
769 * so that freeing of a request in the other direction will
770 * notice us. another possible fix would be to split the
771 * rq mempool into READ and WRITE
772 */
773rq_starved:
774 if (unlikely(rl->count[rw] == 0))
775 rl->starved[rw] = 1;
776
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 goto out;
778 }
779
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100780 /*
781 * ioc may be NULL here, and ioc_batching will be false. That's
782 * OK, if the queue is under the request limit then requests need
783 * not count toward the nr_batch_requests limit. There will always
784 * be some limit enforced by BLK_BATCH_TIME.
785 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 if (ioc_batching(q, ioc))
787 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100788
Jens Axboe2056a782006-03-23 20:00:26 +0100789 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 return rq;
792}
793
794/*
795 * No available requests for this queue, unplug the device and wait for some
796 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700797 *
798 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 */
Jens Axboe165125e2007-07-24 09:28:11 +0200800static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200801 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Jens Axboe7749a8d2006-12-13 13:02:26 +0100803 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 struct request *rq;
805
Jens Axboe7749a8d2006-12-13 13:02:26 +0100806 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700807 while (!rq) {
808 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 struct request_list *rl = &q->rq;
810
811 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
812 TASK_UNINTERRUPTIBLE);
813
Jens Axboe7749a8d2006-12-13 13:02:26 +0100814 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
816 if (!rq) {
817 struct io_context *ioc;
818
Jens Axboe2056a782006-03-23 20:00:26 +0100819 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
820
Nick Piggind6344532005-06-28 20:45:14 -0700821 __generic_unplug_device(q);
822 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 io_schedule();
824
825 /*
826 * After sleeping, we become a "batching" process and
827 * will be able to allocate at least one request, and
828 * up to a big batch of them for a small period time.
829 * See ioc_batching, ioc_set_batching
830 */
Jens Axboeb5deef92006-07-19 23:39:40 +0200831 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -0700833
834 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 }
836 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -0700837 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
839 return rq;
840}
841
Jens Axboe165125e2007-07-24 09:28:11 +0200842struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
844 struct request *rq;
845
846 BUG_ON(rw != READ && rw != WRITE);
847
Nick Piggind6344532005-06-28 20:45:14 -0700848 spin_lock_irq(q->queue_lock);
849 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200850 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700851 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200852 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700853 if (!rq)
854 spin_unlock_irq(q->queue_lock);
855 }
856 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 return rq;
859}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860EXPORT_SYMBOL(blk_get_request);
861
862/**
Jens Axboedc72ef42006-07-20 14:54:05 +0200863 * blk_start_queueing - initiate dispatch of requests to device
864 * @q: request queue to kick into gear
865 *
866 * This is basically a helper to remove the need to know whether a queue
867 * is plugged or not if someone just wants to initiate dispatch of requests
868 * for this queue.
869 *
870 * The queue lock must be held with interrupts disabled.
871 */
Jens Axboe165125e2007-07-24 09:28:11 +0200872void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +0200873{
874 if (!blk_queue_plugged(q))
875 q->request_fn(q);
876 else
877 __generic_unplug_device(q);
878}
879EXPORT_SYMBOL(blk_start_queueing);
880
881/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 * blk_requeue_request - put a request back on queue
883 * @q: request queue where request should be inserted
884 * @rq: request to be inserted
885 *
886 * Description:
887 * Drivers often keep queueing requests until the hardware cannot accept
888 * more, when that condition happens we need to put the request back
889 * on the queue. Must be called with queue lock held.
890 */
Jens Axboe165125e2007-07-24 09:28:11 +0200891void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892{
Jens Axboe2056a782006-03-23 20:00:26 +0100893 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 if (blk_rq_tagged(rq))
896 blk_queue_end_tag(q, rq);
897
898 elv_requeue_request(q, rq);
899}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900EXPORT_SYMBOL(blk_requeue_request);
901
902/**
903 * blk_insert_request - insert a special request in to a request queue
904 * @q: request queue where request should be inserted
905 * @rq: request to be inserted
906 * @at_head: insert request at head or tail of queue
907 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 *
909 * Description:
910 * Many block devices need to execute commands asynchronously, so they don't
911 * block the whole kernel from preemption during request execution. This is
912 * accomplished normally by inserting aritficial requests tagged as
913 * REQ_SPECIAL in to the corresponding request queue, and letting them be
914 * scheduled for actual execution by the request queue.
915 *
916 * We have the option of inserting the head or the tail of the queue.
917 * Typically we use the tail for new ioctls and so forth. We use the head
918 * of the queue for things like a QUEUE_FULL message from a device, or a
919 * host that is unable to accept a particular command.
920 */
Jens Axboe165125e2007-07-24 09:28:11 +0200921void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500922 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923{
Tejun Heo 867d1192005-04-24 02:06:05 -0500924 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 unsigned long flags;
926
927 /*
928 * tell I/O scheduler that this isn't a regular read/write (ie it
929 * must not attempt merges on this) and that it acts as a soft
930 * barrier
931 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200932 rq->cmd_type = REQ_TYPE_SPECIAL;
933 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
935 rq->special = data;
936
937 spin_lock_irqsave(q->queue_lock, flags);
938
939 /*
940 * If command is tagged, release the tag
941 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500942 if (blk_rq_tagged(rq))
943 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200945 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -0500946 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +0200947 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 spin_unlock_irqrestore(q->queue_lock, flags);
949}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950EXPORT_SYMBOL(blk_insert_request);
951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952/*
953 * add-request adds a request to the linked list.
954 * queue lock is held and interrupts disabled, as we muck with the
955 * request queue list.
956 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100957static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958{
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200959 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 /*
962 * elevator indicated where it wants this request to be
963 * inserted at elevator_merge time
964 */
965 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
966}
Jens Axboe6728cb02008-01-31 13:03:55 +0100967
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968/*
969 * disk_round_stats() - Round off the performance stats on a struct
970 * disk_stats.
971 *
972 * The average IO queue length and utilisation statistics are maintained
973 * by observing the current state of the queue length and the amount of
974 * time it has been in this state for.
975 *
976 * Normally, that accounting is done on IO completion, but that can result
977 * in more than a second's worth of IO being accounted for within any one
978 * second, leading to >100% utilisation. To deal with that, we call this
979 * function to do a round-off before returning the results when reading
980 * /proc/diskstats. This accounts immediately for all queue usage up to
981 * the current jiffies and restarts the counters again.
982 */
983void disk_round_stats(struct gendisk *disk)
984{
985 unsigned long now = jiffies;
986
Chen, Kenneth Wb2982642005-10-13 21:49:29 +0200987 if (now == disk->stamp)
988 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Chen, Kenneth W20e5c812005-10-13 21:48:42 +0200990 if (disk->in_flight) {
991 __disk_stat_add(disk, time_in_queue,
992 disk->in_flight * (now - disk->stamp));
993 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
994 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996}
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800997EXPORT_SYMBOL_GPL(disk_round_stats);
998
Jerome Marchand6f2576a2008-02-08 11:04:35 +0100999void part_round_stats(struct hd_struct *part)
1000{
1001 unsigned long now = jiffies;
1002
1003 if (now == part->stamp)
1004 return;
1005
1006 if (part->in_flight) {
1007 __part_stat_add(part, time_in_queue,
1008 part->in_flight * (now - part->stamp));
1009 __part_stat_add(part, io_ticks, (now - part->stamp));
1010 }
1011 part->stamp = now;
1012}
1013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014/*
1015 * queue lock must be held
1016 */
Jens Axboe165125e2007-07-24 09:28:11 +02001017void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 if (unlikely(!q))
1020 return;
1021 if (unlikely(--req->ref_count))
1022 return;
1023
Tejun Heo8922e162005-10-20 16:23:44 +02001024 elv_completed_request(q, req);
1025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 /*
1027 * Request may not have originated from ll_rw_blk. if not,
1028 * it didn't come out of our reserved rq pools
1029 */
Jens Axboe49171e52006-08-10 08:59:11 +02001030 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001032 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001035 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
1037 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02001038 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 }
1040}
Mike Christie6e39b692005-11-11 05:30:24 -06001041EXPORT_SYMBOL_GPL(__blk_put_request);
1042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043void blk_put_request(struct request *req)
1044{
Tejun Heo8922e162005-10-20 16:23:44 +02001045 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001046 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Tejun Heo8922e162005-10-20 16:23:44 +02001048 /*
1049 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
1050 * following if (q) test.
1051 */
1052 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 spin_lock_irqsave(q->queue_lock, flags);
1054 __blk_put_request(q, req);
1055 spin_unlock_irqrestore(q->queue_lock, flags);
1056 }
1057}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058EXPORT_SYMBOL(blk_put_request);
1059
Jens Axboe86db1e22008-01-29 14:53:40 +01001060void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001061{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001062 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001063
1064 /*
1065 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1066 */
1067 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001068 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01001069
1070 /*
1071 * REQ_BARRIER implies no merging, but lets make it explicit
1072 */
1073 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001074 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01001075
Jens Axboeb31dc662006-06-13 08:26:10 +02001076 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001077 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02001078 if (bio_rw_meta(bio))
1079 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02001080
Tejun Heo52d9e672006-01-06 09:49:58 +01001081 req->errors = 0;
1082 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001083 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001084 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02001085 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001086}
1087
Jens Axboe165125e2007-07-24 09:28:11 +02001088static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089{
Nick Piggin450991b2005-06-28 20:45:13 -07001090 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02001091 int el_ret, nr_sectors, barrier, err;
1092 const unsigned short prio = bio_prio(bio);
1093 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01001094 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
1098 /*
1099 * low level driver can indicate that it wants pages above a
1100 * certain limit bounced to low memory (ie for highmem, or even
1101 * ISA dma in theory)
1102 */
1103 blk_queue_bounce(q, &bio);
1104
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01001106 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 err = -EOPNOTSUPP;
1108 goto end_io;
1109 }
1110
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 spin_lock_irq(q->queue_lock);
1112
Nick Piggin450991b2005-06-28 20:45:13 -07001113 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 goto get_rq;
1115
1116 el_ret = elv_merge(q, &req, bio);
1117 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001118 case ELEVATOR_BACK_MERGE:
1119 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Jens Axboe6728cb02008-01-31 13:03:55 +01001121 if (!ll_back_merge_fn(q, req, bio))
1122 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Jens Axboe6728cb02008-01-31 13:03:55 +01001124 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001125
Jens Axboe6728cb02008-01-31 13:03:55 +01001126 req->biotail->bi_next = bio;
1127 req->biotail = bio;
1128 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1129 req->ioprio = ioprio_best(req->ioprio, prio);
1130 drive_stat_acct(req, 0);
1131 if (!attempt_back_merge(q, req))
1132 elv_merged_request(q, req, el_ret);
1133 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
Jens Axboe6728cb02008-01-31 13:03:55 +01001135 case ELEVATOR_FRONT_MERGE:
1136 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Jens Axboe6728cb02008-01-31 13:03:55 +01001138 if (!ll_front_merge_fn(q, req, bio))
1139 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Jens Axboe6728cb02008-01-31 13:03:55 +01001141 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001142
Jens Axboe6728cb02008-01-31 13:03:55 +01001143 bio->bi_next = req->bio;
1144 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Jens Axboe6728cb02008-01-31 13:03:55 +01001146 /*
1147 * may not be valid. if the low level driver said
1148 * it didn't need a bounce buffer then it better
1149 * not touch req->buffer either...
1150 */
1151 req->buffer = bio_data(bio);
1152 req->current_nr_sectors = bio_cur_sectors(bio);
1153 req->hard_cur_sectors = req->current_nr_sectors;
1154 req->sector = req->hard_sector = bio->bi_sector;
1155 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1156 req->ioprio = ioprio_best(req->ioprio, prio);
1157 drive_stat_acct(req, 0);
1158 if (!attempt_front_merge(q, req))
1159 elv_merged_request(q, req, el_ret);
1160 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
Jens Axboe6728cb02008-01-31 13:03:55 +01001162 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1163 default:
1164 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 }
1166
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001168 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001169 * This sync check and mask will be re-done in init_request_from_bio(),
1170 * but we need to set it earlier to expose the sync flag to the
1171 * rq allocator and io schedulers.
1172 */
1173 rw_flags = bio_data_dir(bio);
1174 if (sync)
1175 rw_flags |= REQ_RW_SYNC;
1176
1177 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001178 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001179 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001180 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001181 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001182
Nick Piggin450991b2005-06-28 20:45:13 -07001183 /*
1184 * After dropping the lock and possibly sleeping here, our request
1185 * may now be mergeable after it had proven unmergeable (above).
1186 * We don't worry about that case for efficiency. It won't happen
1187 * often, and the elevators are able to handle it.
1188 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001189 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
Nick Piggin450991b2005-06-28 20:45:13 -07001191 spin_lock_irq(q->queue_lock);
1192 if (elv_queue_empty(q))
1193 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 add_request(q, req);
1195out:
Jens Axboe4a534f92005-04-16 15:25:40 -07001196 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 __generic_unplug_device(q);
1198
1199 spin_unlock_irq(q->queue_lock);
1200 return 0;
1201
1202end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02001203 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 return 0;
1205}
1206
1207/*
1208 * If bio->bi_dev is a partition, remap the location
1209 */
1210static inline void blk_partition_remap(struct bio *bio)
1211{
1212 struct block_device *bdev = bio->bi_bdev;
1213
Jens Axboebf2de6f2007-09-27 13:01:25 +02001214 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 bio->bi_sector += p->start_sect;
1218 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001219
1220 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
1221 bdev->bd_dev, bio->bi_sector,
1222 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 }
1224}
1225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226static void handle_bad_sector(struct bio *bio)
1227{
1228 char b[BDEVNAME_SIZE];
1229
1230 printk(KERN_INFO "attempt to access beyond end of device\n");
1231 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1232 bdevname(bio->bi_bdev, b),
1233 bio->bi_rw,
1234 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1235 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1236
1237 set_bit(BIO_EOF, &bio->bi_flags);
1238}
1239
Akinobu Mitac17bb492006-12-08 02:39:46 -08001240#ifdef CONFIG_FAIL_MAKE_REQUEST
1241
1242static DECLARE_FAULT_ATTR(fail_make_request);
1243
1244static int __init setup_fail_make_request(char *str)
1245{
1246 return setup_fault_attr(&fail_make_request, str);
1247}
1248__setup("fail_make_request=", setup_fail_make_request);
1249
1250static int should_fail_request(struct bio *bio)
1251{
1252 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
1253 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
1254 return should_fail(&fail_make_request, bio->bi_size);
1255
1256 return 0;
1257}
1258
1259static int __init fail_make_request_debugfs(void)
1260{
1261 return init_fault_attr_dentries(&fail_make_request,
1262 "fail_make_request");
1263}
1264
1265late_initcall(fail_make_request_debugfs);
1266
1267#else /* CONFIG_FAIL_MAKE_REQUEST */
1268
1269static inline int should_fail_request(struct bio *bio)
1270{
1271 return 0;
1272}
1273
1274#endif /* CONFIG_FAIL_MAKE_REQUEST */
1275
Jens Axboec07e2b42007-07-18 13:27:58 +02001276/*
1277 * Check whether this bio extends beyond the end of the device.
1278 */
1279static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1280{
1281 sector_t maxsector;
1282
1283 if (!nr_sectors)
1284 return 0;
1285
1286 /* Test device or partition size, when known. */
1287 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1288 if (maxsector) {
1289 sector_t sector = bio->bi_sector;
1290
1291 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1292 /*
1293 * This may well happen - the kernel calls bread()
1294 * without checking the size of the device, e.g., when
1295 * mounting a device.
1296 */
1297 handle_bad_sector(bio);
1298 return 1;
1299 }
1300 }
1301
1302 return 0;
1303}
1304
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305/**
1306 * generic_make_request: hand a buffer to its device driver for I/O
1307 * @bio: The bio describing the location in memory and on the device.
1308 *
1309 * generic_make_request() is used to make I/O requests of block
1310 * devices. It is passed a &struct bio, which describes the I/O that needs
1311 * to be done.
1312 *
1313 * generic_make_request() does not return any status. The
1314 * success/failure status of the request, along with notification of
1315 * completion, is delivered asynchronously through the bio->bi_end_io
1316 * function described (one day) else where.
1317 *
1318 * The caller of generic_make_request must make sure that bi_io_vec
1319 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1320 * set to describe the device address, and the
1321 * bi_end_io and optionally bi_private are set to describe how
1322 * completion notification should be signaled.
1323 *
1324 * generic_make_request and the drivers it calls may use bi_next if this
1325 * bio happens to be merged with someone else, and may change bi_dev and
1326 * bi_sector for remaps as it sees fit. So the values of these fields
1327 * should NOT be depended on after the call to generic_make_request.
1328 */
Neil Brownd89d8792007-05-01 09:53:42 +02001329static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330{
Jens Axboe165125e2007-07-24 09:28:11 +02001331 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001332 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001334 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001335 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
1337 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Jens Axboec07e2b42007-07-18 13:27:58 +02001339 if (bio_check_eod(bio, nr_sectors))
1340 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
1342 /*
1343 * Resolve the mapping until finished. (drivers are
1344 * still free to implement/resolve their own stacking
1345 * by explicitly returning 0)
1346 *
1347 * NOTE: we don't repeat the blk_size check for each new device.
1348 * Stacking drivers are expected to know what they are doing.
1349 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001350 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001351 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 do {
1353 char b[BDEVNAME_SIZE];
1354
1355 q = bdev_get_queue(bio->bi_bdev);
1356 if (!q) {
1357 printk(KERN_ERR
1358 "generic_make_request: Trying to access "
1359 "nonexistent block-device %s (%Lu)\n",
1360 bdevname(bio->bi_bdev, b),
1361 (long long) bio->bi_sector);
1362end_io:
Jens Axboe51fd77b2007-11-02 08:49:08 +01001363 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 break;
1365 }
1366
Jens Axboe4fa253f2007-07-18 13:13:10 +02001367 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001368 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 bdevname(bio->bi_bdev, b),
1370 bio_sectors(bio),
1371 q->max_hw_sectors);
1372 goto end_io;
1373 }
1374
Nick Pigginfde6ad22005-06-23 00:08:53 -07001375 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 goto end_io;
1377
Akinobu Mitac17bb492006-12-08 02:39:46 -08001378 if (should_fail_request(bio))
1379 goto end_io;
1380
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 /*
1382 * If this device has partitions, remap block n
1383 * of partition p to block n+start(p) of the disk.
1384 */
1385 blk_partition_remap(bio);
1386
NeilBrown5ddfe962006-10-30 22:07:21 -08001387 if (old_sector != -1)
Jens Axboe4fa253f2007-07-18 13:13:10 +02001388 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08001389 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001390
1391 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
1392
NeilBrown5ddfe962006-10-30 22:07:21 -08001393 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001394 old_dev = bio->bi_bdev->bd_dev;
1395
Jens Axboec07e2b42007-07-18 13:27:58 +02001396 if (bio_check_eod(bio, nr_sectors))
1397 goto end_io;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001398 if (bio_empty_barrier(bio) && !q->prepare_flush_fn) {
1399 err = -EOPNOTSUPP;
1400 goto end_io;
1401 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001402
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 ret = q->make_request_fn(q, bio);
1404 } while (ret);
1405}
1406
Neil Brownd89d8792007-05-01 09:53:42 +02001407/*
1408 * We only want one ->make_request_fn to be active at a time,
1409 * else stack usage with stacked devices could be a problem.
1410 * So use current->bio_{list,tail} to keep a list of requests
1411 * submited by a make_request_fn function.
1412 * current->bio_tail is also used as a flag to say if
1413 * generic_make_request is currently active in this task or not.
1414 * If it is NULL, then no make_request is active. If it is non-NULL,
1415 * then a make_request is active, and new requests should be added
1416 * at the tail
1417 */
1418void generic_make_request(struct bio *bio)
1419{
1420 if (current->bio_tail) {
1421 /* make_request is active */
1422 *(current->bio_tail) = bio;
1423 bio->bi_next = NULL;
1424 current->bio_tail = &bio->bi_next;
1425 return;
1426 }
1427 /* following loop may be a bit non-obvious, and so deserves some
1428 * explanation.
1429 * Before entering the loop, bio->bi_next is NULL (as all callers
1430 * ensure that) so we have a list with a single bio.
1431 * We pretend that we have just taken it off a longer list, so
1432 * we assign bio_list to the next (which is NULL) and bio_tail
1433 * to &bio_list, thus initialising the bio_list of new bios to be
1434 * added. __generic_make_request may indeed add some more bios
1435 * through a recursive call to generic_make_request. If it
1436 * did, we find a non-NULL value in bio_list and re-enter the loop
1437 * from the top. In this case we really did just take the bio
1438 * of the top of the list (no pretending) and so fixup bio_list and
1439 * bio_tail or bi_next, and call into __generic_make_request again.
1440 *
1441 * The loop was structured like this to make only one call to
1442 * __generic_make_request (which is important as it is large and
1443 * inlined) and to keep the structure simple.
1444 */
1445 BUG_ON(bio->bi_next);
1446 do {
1447 current->bio_list = bio->bi_next;
1448 if (bio->bi_next == NULL)
1449 current->bio_tail = &current->bio_list;
1450 else
1451 bio->bi_next = NULL;
1452 __generic_make_request(bio);
1453 bio = current->bio_list;
1454 } while (bio);
1455 current->bio_tail = NULL; /* deactivate */
1456}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457EXPORT_SYMBOL(generic_make_request);
1458
1459/**
1460 * submit_bio: submit a bio to the block device layer for I/O
1461 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1462 * @bio: The &struct bio which describes the I/O
1463 *
1464 * submit_bio() is very similar in purpose to generic_make_request(), and
1465 * uses that function to do most of the work. Both are fairly rough
1466 * interfaces, @bio must be presetup and ready for I/O.
1467 *
1468 */
1469void submit_bio(int rw, struct bio *bio)
1470{
1471 int count = bio_sectors(bio);
1472
Jens Axboe22e2c502005-06-27 10:55:12 +02001473 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
Jens Axboebf2de6f2007-09-27 13:01:25 +02001475 /*
1476 * If it's a regular read/write or a barrier with data attached,
1477 * go through the normal accounting stuff before submission.
1478 */
1479 if (!bio_empty_barrier(bio)) {
1480
1481 BIO_BUG_ON(!bio->bi_size);
1482 BIO_BUG_ON(!bio->bi_io_vec);
1483
1484 if (rw & WRITE) {
1485 count_vm_events(PGPGOUT, count);
1486 } else {
1487 task_io_account_read(bio->bi_size);
1488 count_vm_events(PGPGIN, count);
1489 }
1490
1491 if (unlikely(block_dump)) {
1492 char b[BDEVNAME_SIZE];
1493 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001494 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001495 (rw & WRITE) ? "WRITE" : "READ",
1496 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001497 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001498 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 }
1500
1501 generic_make_request(bio);
1502}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503EXPORT_SYMBOL(submit_bio);
1504
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001505/**
1506 * __end_that_request_first - end I/O on a request
1507 * @req: the request being processed
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001508 * @error: 0 for success, < 0 for error
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001509 * @nr_bytes: number of bytes to complete
1510 *
1511 * Description:
1512 * Ends I/O on a number of bytes attached to @req, and sets it up
1513 * for the next range of segments (if any) in the cluster.
1514 *
1515 * Return:
1516 * 0 - we are done with this request, call end_that_request_last()
1517 * 1 - still buffers pending for this request
1518 **/
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001519static int __end_that_request_first(struct request *req, int error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 int nr_bytes)
1521{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001522 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 struct bio *bio;
1524
Jens Axboe2056a782006-03-23 20:00:26 +01001525 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
1526
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 * for a REQ_BLOCK_PC request, we want to carry any eventual
1529 * sense key with us all the way through
1530 */
1531 if (!blk_pc_request(req))
1532 req->errors = 0;
1533
Jens Axboe6728cb02008-01-31 13:03:55 +01001534 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1535 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 req->rq_disk ? req->rq_disk->disk_name : "?",
1537 (unsigned long long)req->sector);
1538 }
1539
Jens Axboed72d9042005-11-01 08:35:42 +01001540 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboe28f13702008-05-07 10:15:46 +02001541 struct hd_struct *part = get_part(req->rq_disk, req->sector);
Jens Axboea3623572005-11-01 09:26:16 +01001542 const int rw = rq_data_dir(req);
1543
Jens Axboe28f13702008-05-07 10:15:46 +02001544 all_stat_add(req->rq_disk, part, sectors[rw],
1545 nr_bytes >> 9, req->sector);
Jens Axboed72d9042005-11-01 08:35:42 +01001546 }
1547
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 total_bytes = bio_nbytes = 0;
1549 while ((bio = req->bio) != NULL) {
1550 int nbytes;
1551
Jens Axboebf2de6f2007-09-27 13:01:25 +02001552 /*
1553 * For an empty barrier request, the low level driver must
1554 * store a potential error location in ->sector. We pass
1555 * that back up in ->bi_sector.
1556 */
1557 if (blk_empty_barrier(req))
1558 bio->bi_sector = req->sector;
1559
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 if (nr_bytes >= bio->bi_size) {
1561 req->bio = bio->bi_next;
1562 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02001563 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 next_idx = 0;
1565 bio_nbytes = 0;
1566 } else {
1567 int idx = bio->bi_idx + next_idx;
1568
1569 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1570 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01001571 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -07001572 __func__, bio->bi_idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 break;
1574 }
1575
1576 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1577 BIO_BUG_ON(nbytes > bio->bi_size);
1578
1579 /*
1580 * not a complete bvec done
1581 */
1582 if (unlikely(nbytes > nr_bytes)) {
1583 bio_nbytes += nr_bytes;
1584 total_bytes += nr_bytes;
1585 break;
1586 }
1587
1588 /*
1589 * advance to the next vector
1590 */
1591 next_idx++;
1592 bio_nbytes += nbytes;
1593 }
1594
1595 total_bytes += nbytes;
1596 nr_bytes -= nbytes;
1597
Jens Axboe6728cb02008-01-31 13:03:55 +01001598 bio = req->bio;
1599 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 /*
1601 * end more in this run, or just return 'not-done'
1602 */
1603 if (unlikely(nr_bytes <= 0))
1604 break;
1605 }
1606 }
1607
1608 /*
1609 * completely done
1610 */
1611 if (!req->bio)
1612 return 0;
1613
1614 /*
1615 * if the request wasn't completed, update state
1616 */
1617 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02001618 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 bio->bi_idx += next_idx;
1620 bio_iovec(bio)->bv_offset += nr_bytes;
1621 bio_iovec(bio)->bv_len -= nr_bytes;
1622 }
1623
1624 blk_recalc_rq_sectors(req, total_bytes >> 9);
1625 blk_recalc_rq_segments(req);
1626 return 1;
1627}
1628
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629/*
Jens Axboeff856ba2006-01-09 16:02:34 +01001630 * splice the completion data to a local structure and hand off to
1631 * process_completion_queue() to complete the requests
1632 */
1633static void blk_done_softirq(struct softirq_action *h)
1634{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07001635 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01001636
1637 local_irq_disable();
1638 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07001639 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01001640 local_irq_enable();
1641
1642 while (!list_empty(&local_list)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001643 struct request *rq;
Jens Axboeff856ba2006-01-09 16:02:34 +01001644
Jens Axboe6728cb02008-01-31 13:03:55 +01001645 rq = list_entry(local_list.next, struct request, donelist);
Jens Axboeff856ba2006-01-09 16:02:34 +01001646 list_del_init(&rq->donelist);
1647 rq->q->softirq_done_fn(rq);
1648 }
1649}
1650
Jens Axboe6728cb02008-01-31 13:03:55 +01001651static int __cpuinit blk_cpu_notify(struct notifier_block *self,
1652 unsigned long action, void *hcpu)
Jens Axboeff856ba2006-01-09 16:02:34 +01001653{
1654 /*
1655 * If a CPU goes away, splice its entries to the current CPU
1656 * and trigger a run of the softirq
1657 */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001658 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Jens Axboeff856ba2006-01-09 16:02:34 +01001659 int cpu = (unsigned long) hcpu;
1660
1661 local_irq_disable();
1662 list_splice_init(&per_cpu(blk_cpu_done, cpu),
1663 &__get_cpu_var(blk_cpu_done));
1664 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1665 local_irq_enable();
1666 }
1667
1668 return NOTIFY_OK;
1669}
1670
1671
Satyam Sharmadb47d472007-08-23 09:29:40 +02001672static struct notifier_block blk_cpu_notifier __cpuinitdata = {
Jens Axboeff856ba2006-01-09 16:02:34 +01001673 .notifier_call = blk_cpu_notify,
1674};
1675
Jens Axboeff856ba2006-01-09 16:02:34 +01001676/**
1677 * blk_complete_request - end I/O on a request
1678 * @req: the request being processed
1679 *
1680 * Description:
1681 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02001682 * unless the driver actually implements this in its completion callback
Jens Axboe4fa253f2007-07-18 13:13:10 +02001683 * through requeueing. The actual completion happens out-of-order,
Jens Axboeff856ba2006-01-09 16:02:34 +01001684 * through a softirq handler. The user must have registered a completion
1685 * callback through blk_queue_softirq_done().
1686 **/
1687
1688void blk_complete_request(struct request *req)
1689{
1690 struct list_head *cpu_list;
1691 unsigned long flags;
1692
1693 BUG_ON(!req->q->softirq_done_fn);
Jens Axboe6728cb02008-01-31 13:03:55 +01001694
Jens Axboeff856ba2006-01-09 16:02:34 +01001695 local_irq_save(flags);
1696
1697 cpu_list = &__get_cpu_var(blk_cpu_done);
1698 list_add_tail(&req->donelist, cpu_list);
1699 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1700
1701 local_irq_restore(flags);
1702}
Jens Axboeff856ba2006-01-09 16:02:34 +01001703EXPORT_SYMBOL(blk_complete_request);
Jens Axboe6728cb02008-01-31 13:03:55 +01001704
Jens Axboeff856ba2006-01-09 16:02:34 +01001705/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 * queue lock must be held
1707 */
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001708static void end_that_request_last(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709{
1710 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01001711
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001712 if (blk_rq_tagged(req))
1713 blk_queue_end_tag(req->q, req);
1714
1715 if (blk_queued_rq(req))
1716 blkdev_dequeue_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718 if (unlikely(laptop_mode) && blk_fs_request(req))
1719 laptop_io_completion();
1720
Jens Axboefd0ff8a2006-05-23 11:23:49 +02001721 /*
1722 * Account IO completion. bar_rq isn't accounted as a normal
1723 * IO on queueing nor completion. Accounting the containing
1724 * request is enough.
1725 */
1726 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01001728 const int rw = rq_data_dir(req);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001729 struct hd_struct *part = get_part(disk, req->sector);
Jens Axboea3623572005-11-01 09:26:16 +01001730
Jens Axboe28f13702008-05-07 10:15:46 +02001731 __all_stat_inc(disk, part, ios[rw], req->sector);
1732 __all_stat_add(disk, part, ticks[rw], duration, req->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 disk_round_stats(disk);
1734 disk->in_flight--;
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001735 if (part) {
1736 part_round_stats(part);
1737 part->in_flight--;
1738 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 }
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001740
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01001742 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001743 else {
1744 if (blk_bidi_rq(req))
1745 __blk_put_request(req->next_rq->q, req->next_rq);
1746
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 }
1749}
1750
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751static inline void __end_request(struct request *rq, int uptodate,
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001752 unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001754 int error = 0;
1755
1756 if (uptodate <= 0)
1757 error = uptodate ? uptodate : -EIO;
1758
1759 __blk_end_request(rq, error, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760}
1761
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001762/**
1763 * blk_rq_bytes - Returns bytes left to complete in the entire request
Randy Dunlap5d87a052008-02-20 09:01:22 +01001764 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001765 **/
1766unsigned int blk_rq_bytes(struct request *rq)
Jens Axboea0cd1282007-09-21 10:41:07 +02001767{
1768 if (blk_fs_request(rq))
1769 return rq->hard_nr_sectors << 9;
1770
1771 return rq->data_len;
1772}
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001773EXPORT_SYMBOL_GPL(blk_rq_bytes);
1774
1775/**
1776 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
Randy Dunlap5d87a052008-02-20 09:01:22 +01001777 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001778 **/
1779unsigned int blk_rq_cur_bytes(struct request *rq)
1780{
1781 if (blk_fs_request(rq))
1782 return rq->current_nr_sectors << 9;
1783
1784 if (rq->bio)
1785 return rq->bio->bi_size;
1786
1787 return rq->data_len;
1788}
1789EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
Jens Axboea0cd1282007-09-21 10:41:07 +02001790
1791/**
1792 * end_queued_request - end all I/O on a queued request
1793 * @rq: the request being processed
1794 * @uptodate: error value or 0/1 uptodate flag
1795 *
1796 * Description:
1797 * Ends all I/O on a request, and removes it from the block layer queues.
1798 * Not suitable for normal IO completion, unless the driver still has
1799 * the request attached to the block layer.
1800 *
1801 **/
1802void end_queued_request(struct request *rq, int uptodate)
1803{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001804 __end_request(rq, uptodate, blk_rq_bytes(rq));
Jens Axboea0cd1282007-09-21 10:41:07 +02001805}
1806EXPORT_SYMBOL(end_queued_request);
1807
1808/**
1809 * end_dequeued_request - end all I/O on a dequeued request
1810 * @rq: the request being processed
1811 * @uptodate: error value or 0/1 uptodate flag
1812 *
1813 * Description:
1814 * Ends all I/O on a request. The request must already have been
1815 * dequeued using blkdev_dequeue_request(), as is normally the case
1816 * for most drivers.
1817 *
1818 **/
1819void end_dequeued_request(struct request *rq, int uptodate)
1820{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001821 __end_request(rq, uptodate, blk_rq_bytes(rq));
Jens Axboea0cd1282007-09-21 10:41:07 +02001822}
1823EXPORT_SYMBOL(end_dequeued_request);
1824
1825
1826/**
1827 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07001828 * @req: the request being processed
Jens Axboea0cd1282007-09-21 10:41:07 +02001829 * @uptodate: error value or 0/1 uptodate flag
1830 *
1831 * Description:
1832 * Ends I/O on the current segment of a request. If that is the only
1833 * remaining segment, the request is also completed and freed.
1834 *
1835 * This is a remnant of how older block drivers handled IO completions.
1836 * Modern drivers typically end IO on the full request in one go, unless
1837 * they have a residual value to account for. For that case this function
1838 * isn't really useful, unless the residual just happens to be the
1839 * full current segment. In other words, don't use this function in new
1840 * code. Either use end_request_completely(), or the
1841 * end_that_request_chunk() (along with end_that_request_last()) for
1842 * partial completions.
1843 *
1844 **/
1845void end_request(struct request *req, int uptodate)
1846{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001847 __end_request(req, uptodate, req->hard_cur_sectors << 9);
Jens Axboea0cd1282007-09-21 10:41:07 +02001848}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849EXPORT_SYMBOL(end_request);
1850
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001851/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001852 * blk_end_io - Generic end_io function to complete a request.
1853 * @rq: the request being processed
1854 * @error: 0 for success, < 0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001855 * @nr_bytes: number of bytes to complete @rq
1856 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001857 * @drv_callback: function called between completion of bios in the request
1858 * and completion of the request.
1859 * If the callback returns non 0, this helper returns without
1860 * completion of the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001861 *
1862 * Description:
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001863 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001864 * If @rq has leftover, sets it up for the next range of segments.
1865 *
1866 * Return:
1867 * 0 - we are done with this request
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001868 * 1 - this request is not freed yet, it still has pending buffers.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001869 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001870static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1871 unsigned int bidi_bytes,
1872 int (drv_callback)(struct request *))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001873{
1874 struct request_queue *q = rq->q;
1875 unsigned long flags = 0UL;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001876
1877 if (blk_fs_request(rq) || blk_pc_request(rq)) {
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001878 if (__end_that_request_first(rq, error, nr_bytes))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001879 return 1;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001880
1881 /* Bidi request must be completed as a whole */
1882 if (blk_bidi_rq(rq) &&
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001883 __end_that_request_first(rq->next_rq, error, bidi_bytes))
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001884 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001885 }
1886
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001887 /* Special feature for tricky drivers */
1888 if (drv_callback && drv_callback(rq))
1889 return 1;
1890
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001891 add_disk_randomness(rq->rq_disk);
1892
1893 spin_lock_irqsave(q->queue_lock, flags);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001894 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001895 spin_unlock_irqrestore(q->queue_lock, flags);
1896
1897 return 0;
1898}
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001899
1900/**
1901 * blk_end_request - Helper function for drivers to complete the request.
1902 * @rq: the request being processed
1903 * @error: 0 for success, < 0 for error
1904 * @nr_bytes: number of bytes to complete
1905 *
1906 * Description:
1907 * Ends I/O on a number of bytes attached to @rq.
1908 * If @rq has leftover, sets it up for the next range of segments.
1909 *
1910 * Return:
1911 * 0 - we are done with this request
1912 * 1 - still buffers pending for this request
1913 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001914int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001915{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001916 return blk_end_io(rq, error, nr_bytes, 0, NULL);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001917}
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001918EXPORT_SYMBOL_GPL(blk_end_request);
1919
1920/**
1921 * __blk_end_request - Helper function for drivers to complete the request.
1922 * @rq: the request being processed
1923 * @error: 0 for success, < 0 for error
1924 * @nr_bytes: number of bytes to complete
1925 *
1926 * Description:
1927 * Must be called with queue lock held unlike blk_end_request().
1928 *
1929 * Return:
1930 * 0 - we are done with this request
1931 * 1 - still buffers pending for this request
1932 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001933int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001934{
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001935 if (blk_fs_request(rq) || blk_pc_request(rq)) {
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001936 if (__end_that_request_first(rq, error, nr_bytes))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001937 return 1;
1938 }
1939
1940 add_disk_randomness(rq->rq_disk);
1941
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001942 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001943
1944 return 0;
1945}
1946EXPORT_SYMBOL_GPL(__blk_end_request);
1947
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001948/**
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001949 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1950 * @rq: the bidi request being processed
1951 * @error: 0 for success, < 0 for error
1952 * @nr_bytes: number of bytes to complete @rq
1953 * @bidi_bytes: number of bytes to complete @rq->next_rq
1954 *
1955 * Description:
1956 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1957 *
1958 * Return:
1959 * 0 - we are done with this request
1960 * 1 - still buffers pending for this request
1961 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001962int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
1963 unsigned int bidi_bytes)
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001964{
1965 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
1966}
1967EXPORT_SYMBOL_GPL(blk_end_bidi_request);
1968
1969/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001970 * blk_end_request_callback - Special helper function for tricky drivers
1971 * @rq: the request being processed
1972 * @error: 0 for success, < 0 for error
1973 * @nr_bytes: number of bytes to complete
1974 * @drv_callback: function called between completion of bios in the request
1975 * and completion of the request.
1976 * If the callback returns non 0, this helper returns without
1977 * completion of the request.
1978 *
1979 * Description:
1980 * Ends I/O on a number of bytes attached to @rq.
1981 * If @rq has leftover, sets it up for the next range of segments.
1982 *
1983 * This special helper function is used only for existing tricky drivers.
1984 * (e.g. cdrom_newpc_intr() of ide-cd)
1985 * This interface will be removed when such drivers are rewritten.
1986 * Don't use this interface in other places anymore.
1987 *
1988 * Return:
1989 * 0 - we are done with this request
1990 * 1 - this request is not freed yet.
1991 * this request still has pending buffers or
1992 * the driver doesn't want to finish this request yet.
1993 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001994int blk_end_request_callback(struct request *rq, int error,
1995 unsigned int nr_bytes,
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001996 int (drv_callback)(struct request *))
1997{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001998 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001999}
2000EXPORT_SYMBOL_GPL(blk_end_request_callback);
2001
Jens Axboe86db1e22008-01-29 14:53:40 +01002002void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2003 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002005 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
2006 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007
2008 rq->nr_phys_segments = bio_phys_segments(q, bio);
2009 rq->nr_hw_segments = bio_hw_segments(q, bio);
2010 rq->current_nr_sectors = bio_cur_sectors(bio);
2011 rq->hard_cur_sectors = rq->current_nr_sectors;
2012 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
2013 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002014 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
2016 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017
NeilBrown66846572007-08-16 13:31:28 +02002018 if (bio->bi_bdev)
2019 rq->rq_disk = bio->bi_bdev->bd_disk;
2020}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
2022int kblockd_schedule_work(struct work_struct *work)
2023{
2024 return queue_work(kblockd_workqueue, work);
2025}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026EXPORT_SYMBOL(kblockd_schedule_work);
2027
Andrew Morton19a75d82007-05-09 02:33:56 -07002028void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07002030 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031}
Andrew Morton19a75d82007-05-09 02:33:56 -07002032EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
2034int __init blk_dev_init(void)
2035{
Jens Axboeff856ba2006-01-09 16:02:34 +01002036 int i;
2037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 kblockd_workqueue = create_workqueue("kblockd");
2039 if (!kblockd_workqueue)
2040 panic("Failed to create kblockd\n");
2041
2042 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002043 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
Jens Axboe8324aa92008-01-29 14:51:59 +01002045 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002046 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002048 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01002049 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
2050
2051 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002052 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01002053
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 return 0;
2055}
2056