blob: 4ce953f1b3909f38f213adcc0a39251a34875330 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080029#include <linux/fault-inject.h>
Li Zefan55782132009-06-09 13:43:05 +080030
31#define CREATE_TRACE_POINTS
32#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Jens Axboe8324aa92008-01-29 14:51:59 +010034#include "blk.h"
35
Ingo Molnar0bfc2452008-11-26 11:59:56 +010036EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020037EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Li Zefan55782132009-06-09 13:43:05 +080038EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010039
Jens Axboe165125e2007-07-24 09:28:11 +020040static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42/*
43 * For the allocated request tables
44 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010045static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47/*
48 * For queue allocation
49 */
Jens Axboe6728cb02008-01-31 13:03:55 +010050struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 * Controlling structure to kblockd
54 */
Jens Axboeff856ba2006-01-09 16:02:34 +010055static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Jens Axboe26b82562008-01-29 13:54:41 +010057static void drive_stat_acct(struct request *rq, int new_io)
58{
Jens Axboe28f13702008-05-07 10:15:46 +020059 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010060 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090061 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010062
Jens Axboec2553b52009-04-24 08:10:11 +020063 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010064 return;
65
Tejun Heo074a7ac2008-08-25 19:56:14 +090066 cpu = part_stat_lock();
Jens Axboef253b862010-10-24 22:06:02 +020067 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Tejun Heoc9959052008-08-25 19:47:21 +090068
Jens Axboef253b862010-10-24 22:06:02 +020069 if (!new_io)
Tejun Heo074a7ac2008-08-25 19:56:14 +090070 part_stat_inc(cpu, part, merges[rw]);
Jens Axboef253b862010-10-24 22:06:02 +020071 else {
Tejun Heo074a7ac2008-08-25 19:56:14 +090072 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020073 part_inc_in_flight(part, rw);
Jens Axboe26b82562008-01-29 13:54:41 +010074 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020075
Tejun Heo074a7ac2008-08-25 19:56:14 +090076 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010077}
78
Jens Axboe8324aa92008-01-29 14:51:59 +010079void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080{
81 int nr;
82
83 nr = q->nr_requests - (q->nr_requests / 8) + 1;
84 if (nr > q->nr_requests)
85 nr = q->nr_requests;
86 q->nr_congestion_on = nr;
87
88 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
89 if (nr < 1)
90 nr = 1;
91 q->nr_congestion_off = nr;
92}
93
Linus Torvalds1da177e2005-04-16 15:20:36 -070094/**
95 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
96 * @bdev: device
97 *
98 * Locates the passed device's request queue and returns the address of its
99 * backing_dev_info
100 *
101 * Will return NULL if the request queue cannot be located.
102 */
103struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
104{
105 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200106 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
108 if (q)
109 ret = &q->backing_dev_info;
110 return ret;
111}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112EXPORT_SYMBOL(blk_get_backing_dev_info);
113
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200114void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200116 memset(rq, 0, sizeof(*rq));
117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700119 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200120 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100121 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900122 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200123 INIT_HLIST_NODE(&rq->hash);
124 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200125 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800126 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100127 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100128 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900129 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700130 set_start_time_ns(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200132EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
NeilBrown5bb23a62007-09-27 12:46:13 +0200134static void req_bio_endio(struct request *rq, struct bio *bio,
135 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100136{
Jens Axboe165125e2007-07-24 09:28:11 +0200137 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100138
Tejun Heodd4c1332010-09-03 11:56:16 +0200139 if (&q->flush_rq != rq) {
NeilBrown5bb23a62007-09-27 12:46:13 +0200140 if (error)
141 clear_bit(BIO_UPTODATE, &bio->bi_flags);
142 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
143 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100144
NeilBrown5bb23a62007-09-27 12:46:13 +0200145 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100146 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700147 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200148 nbytes = bio->bi_size;
149 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100150
Keith Mannthey08bafc02008-11-25 10:24:35 +0100151 if (unlikely(rq->cmd_flags & REQ_QUIET))
152 set_bit(BIO_QUIET, &bio->bi_flags);
153
NeilBrown5bb23a62007-09-27 12:46:13 +0200154 bio->bi_size -= nbytes;
155 bio->bi_sector += (nbytes >> 9);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200156
157 if (bio_integrity(bio))
158 bio_integrity_advance(bio, nbytes);
159
NeilBrown5bb23a62007-09-27 12:46:13 +0200160 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200161 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200162 } else {
NeilBrown5bb23a62007-09-27 12:46:13 +0200163 /*
Tejun Heodd4c1332010-09-03 11:56:16 +0200164 * Okay, this is the sequenced flush request in
165 * progress, just record the error;
NeilBrown5bb23a62007-09-27 12:46:13 +0200166 */
Tejun Heodd4c1332010-09-03 11:56:16 +0200167 if (error && !q->flush_err)
168 q->flush_err = error;
NeilBrown5bb23a62007-09-27 12:46:13 +0200169 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100170}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172void blk_dump_rq_flags(struct request *rq, char *msg)
173{
174 int bit;
175
Jens Axboe6728cb02008-01-31 13:03:55 +0100176 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200177 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
178 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
Tejun Heo83096eb2009-05-07 22:24:39 +0900180 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
181 (unsigned long long)blk_rq_pos(rq),
182 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900183 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900184 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200186 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100187 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200188 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 printk("%02x ", rq->cmd[bit]);
190 printk("\n");
191 }
192}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193EXPORT_SYMBOL(blk_dump_rq_flags);
194
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195/*
196 * "plug" the device if there are no outstanding requests: this will
197 * force the transfer to start only after we have put all the requests
198 * on the list.
199 *
200 * This is called with interrupts off and no requests on the queue and
201 * with the queue lock held.
202 */
Jens Axboe165125e2007-07-24 09:28:11 +0200203void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
205 WARN_ON(!irqs_disabled());
206
207 /*
208 * don't plug a stopped queue, it must be paired with blk_start_queue()
209 * which will restart the queueing
210 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200211 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 return;
213
Jens Axboee48ec692008-07-03 13:18:54 +0200214 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100216 trace_block_plug(q);
Jens Axboe2056a782006-03-23 20:00:26 +0100217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219EXPORT_SYMBOL(blk_plug_device);
220
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200221/**
222 * blk_plug_device_unlocked - plug a device without queue lock held
223 * @q: The &struct request_queue to plug
224 *
225 * Description:
226 * Like @blk_plug_device(), but grabs the queue lock and disables
227 * interrupts.
228 **/
229void blk_plug_device_unlocked(struct request_queue *q)
230{
231 unsigned long flags;
232
233 spin_lock_irqsave(q->queue_lock, flags);
234 blk_plug_device(q);
235 spin_unlock_irqrestore(q->queue_lock, flags);
236}
237EXPORT_SYMBOL(blk_plug_device_unlocked);
238
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239/*
240 * remove the queue from the plugged list, if present. called with
241 * queue lock held and interrupts disabled.
242 */
Jens Axboe165125e2007-07-24 09:28:11 +0200243int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
245 WARN_ON(!irqs_disabled());
246
Jens Axboee48ec692008-07-03 13:18:54 +0200247 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 return 0;
249
250 del_timer(&q->unplug_timer);
251 return 1;
252}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253EXPORT_SYMBOL(blk_remove_plug);
254
255/*
256 * remove the plug and let it rip..
257 */
Jens Axboe165125e2007-07-24 09:28:11 +0200258void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200260 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 return;
Jens Axboea31a9732008-10-17 13:58:29 +0200262 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 return;
264
Jens Axboe22e2c502005-06-27 10:55:12 +0200265 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
268/**
269 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200270 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 *
272 * Description:
273 * Linux uses plugging to build bigger requests queues before letting
274 * the device have at them. If a queue is plugged, the I/O scheduler
275 * is still adding and merging requests on the queue. Once the queue
276 * gets unplugged, the request_fn defined for the queue is invoked and
277 * transfers started.
278 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200279void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200281 if (blk_queue_plugged(q)) {
282 spin_lock_irq(q->queue_lock);
283 __generic_unplug_device(q);
284 spin_unlock_irq(q->queue_lock);
285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286}
287EXPORT_SYMBOL(generic_unplug_device);
288
289static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
290 struct page *page)
291{
Jens Axboe165125e2007-07-24 09:28:11 +0200292 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500294 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
Jens Axboe86db1e22008-01-29 14:53:40 +0100297void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
Jens Axboe165125e2007-07-24 09:28:11 +0200299 struct request_queue *q =
300 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100302 trace_block_unplug_io(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 q->unplug_fn(q);
304}
305
Jens Axboe86db1e22008-01-29 14:53:40 +0100306void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
Jens Axboe165125e2007-07-24 09:28:11 +0200308 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100310 trace_block_unplug_timer(q);
Jens Axboe18887ad2008-07-28 13:08:45 +0200311 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312}
313
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500314void blk_unplug(struct request_queue *q)
315{
316 /*
317 * devices don't necessarily have an ->unplug_fn defined
318 */
319 if (q->unplug_fn) {
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100320 trace_block_unplug_io(q);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500321 q->unplug_fn(q);
322 }
323}
324EXPORT_SYMBOL(blk_unplug);
325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326/**
327 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200328 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 *
330 * Description:
331 * blk_start_queue() will clear the stop flag on the queue, and call
332 * the request_fn for the queue if it was in a stopped state when
333 * entered. Also see blk_stop_queue(). Queue lock must be held.
334 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200335void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200337 WARN_ON(!irqs_disabled());
338
Nick Piggin75ad23b2008-04-29 14:48:33 +0200339 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Tejun Heoa538cd02009-04-23 11:05:17 +0900340 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342EXPORT_SYMBOL(blk_start_queue);
343
344/**
345 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200346 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 *
348 * Description:
349 * The Linux block layer assumes that a block driver will consume all
350 * entries on the request queue when the request_fn strategy is called.
351 * Often this will not happen, because of hardware limitations (queue
352 * depth settings). If a device driver gets a 'queue full' response,
353 * or if it simply chooses not to queue more I/O at one point, it can
354 * call this function to prevent the request_fn from being called until
355 * the driver has signalled it's ready to go again. This happens by calling
356 * blk_start_queue() to restart queue operations. Queue lock must be held.
357 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200358void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
360 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200361 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362}
363EXPORT_SYMBOL(blk_stop_queue);
364
365/**
366 * blk_sync_queue - cancel any pending callbacks on a queue
367 * @q: the queue
368 *
369 * Description:
370 * The block layer may perform asynchronous callback activity
371 * on a queue, such as calling the unplug function after a timeout.
372 * A block device may call blk_sync_queue to ensure that any
373 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200374 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * that its ->make_request_fn will not re-add plugging prior to calling
376 * this function.
377 *
378 */
379void blk_sync_queue(struct request_queue *q)
380{
381 del_timer_sync(&q->unplug_timer);
Jens Axboe70ed28b2008-11-19 14:38:39 +0100382 del_timer_sync(&q->timeout);
Cheng Renquan64d01dc2008-12-03 12:41:39 +0100383 cancel_work_sync(&q->unplug_work);
Vivek Goyale43473b2010-09-15 17:06:35 -0400384 throtl_shutdown_timer_wq(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385}
386EXPORT_SYMBOL(blk_sync_queue);
387
388/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200389 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200391 *
392 * Description:
393 * See @blk_run_queue. This variant must be called with the queue lock
394 * held and interrupts disabled.
395 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200397void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200400
Tejun Heoa538cd02009-04-23 11:05:17 +0900401 if (unlikely(blk_queue_stopped(q)))
402 return;
403
404 if (elv_queue_empty(q))
405 return;
406
Jens Axboedac07ec2006-05-11 08:20:16 +0200407 /*
408 * Only recurse once to avoid overrunning the stack, let the unplug
409 * handling reinvoke the handler shortly if we already got there.
410 */
Tejun Heoa538cd02009-04-23 11:05:17 +0900411 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
412 q->request_fn(q);
413 queue_flag_clear(QUEUE_FLAG_REENTER, q);
414 } else {
415 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
416 kblockd_schedule_work(q, &q->unplug_work);
417 }
Nick Piggin75ad23b2008-04-29 14:48:33 +0200418}
419EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200420
Nick Piggin75ad23b2008-04-29 14:48:33 +0200421/**
422 * blk_run_queue - run a single device queue
423 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200424 *
425 * Description:
426 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900427 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200428 */
429void blk_run_queue(struct request_queue *q)
430{
431 unsigned long flags;
432
433 spin_lock_irqsave(q->queue_lock, flags);
434 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 spin_unlock_irqrestore(q->queue_lock, flags);
436}
437EXPORT_SYMBOL(blk_run_queue);
438
Jens Axboe165125e2007-07-24 09:28:11 +0200439void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500440{
441 kobject_put(&q->kobj);
442}
Al Viro483f4af2006-03-18 18:34:37 -0500443
Jens Axboe6728cb02008-01-31 13:03:55 +0100444void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500445{
Jens Axboee3335de92008-09-18 09:22:54 -0700446 /*
447 * We know we have process context here, so we can be a little
448 * cautious and ensure that pending block actions on this device
449 * are done before moving on. Going into this function, we should
450 * not have processes doing IO to this device.
451 */
452 blk_sync_queue(q);
453
Matthew Garrett31373d02010-04-06 14:25:14 +0200454 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500455 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200456 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500457 mutex_unlock(&q->sysfs_lock);
458
459 if (q->elevator)
460 elevator_exit(q->elevator);
461
462 blk_put_queue(q);
463}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464EXPORT_SYMBOL(blk_cleanup_queue);
465
Jens Axboe165125e2007-07-24 09:28:11 +0200466static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
468 struct request_list *rl = &q->rq;
469
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400470 if (unlikely(rl->rq_pool))
471 return 0;
472
Jens Axboe1faa16d2009-04-06 14:48:01 +0200473 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
474 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200475 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200476 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
477 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Christoph Lameter19460892005-06-23 00:08:19 -0700479 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
480 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 if (!rl->rq_pool)
483 return -ENOMEM;
484
485 return 0;
486}
487
Jens Axboe165125e2007-07-24 09:28:11 +0200488struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
Christoph Lameter19460892005-06-23 00:08:19 -0700490 return blk_alloc_queue_node(gfp_mask, -1);
491}
492EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Jens Axboe165125e2007-07-24 09:28:11 +0200494struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700495{
Jens Axboe165125e2007-07-24 09:28:11 +0200496 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700497 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700498
Jens Axboe8324aa92008-01-29 14:51:59 +0100499 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700500 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 if (!q)
502 return NULL;
503
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700504 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
505 q->backing_dev_info.unplug_io_data = q;
Jens Axboe0989a022009-06-12 14:42:56 +0200506 q->backing_dev_info.ra_pages =
507 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
508 q->backing_dev_info.state = 0;
509 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200510 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200511
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700512 err = bdi_init(&q->backing_dev_info);
513 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100514 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700515 return NULL;
516 }
517
Vivek Goyale43473b2010-09-15 17:06:35 -0400518 if (blk_throtl_init(q)) {
519 kmem_cache_free(blk_requestq_cachep, q);
520 return NULL;
521 }
522
Matthew Garrett31373d02010-04-06 14:25:14 +0200523 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
524 laptop_mode_timer_fn, (unsigned long) q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 init_timer(&q->unplug_timer);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700526 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
527 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heodd4c1332010-09-03 11:56:16 +0200528 INIT_LIST_HEAD(&q->pending_flushes);
Peter Zijlstra713ada92008-10-16 13:44:57 +0200529 INIT_WORK(&q->unplug_work, blk_unplug_work);
Al Viro483f4af2006-03-18 18:34:37 -0500530
Jens Axboe8324aa92008-01-29 14:51:59 +0100531 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
Al Viro483f4af2006-03-18 18:34:37 -0500533 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700534 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 return q;
537}
Christoph Lameter19460892005-06-23 00:08:19 -0700538EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
540/**
541 * blk_init_queue - prepare a request queue for use with a block device
542 * @rfn: The function to be called to process requests that have been
543 * placed on the queue.
544 * @lock: Request queue spin lock
545 *
546 * Description:
547 * If a block device wishes to use the standard request handling procedures,
548 * which sorts requests and coalesces adjacent requests, then it must
549 * call blk_init_queue(). The function @rfn will be called when there
550 * are requests on the queue that need to be processed. If the device
551 * supports plugging, then @rfn may not be called immediately when requests
552 * are available on the queue, but may be called at some time later instead.
553 * Plugged queues are generally unplugged when a buffer belonging to one
554 * of the requests on the queue is needed, or due to memory pressure.
555 *
556 * @rfn is not required, or even expected, to remove all requests off the
557 * queue, but only as many as it can handle at a time. If it does leave
558 * requests on the queue, it is responsible for arranging that the requests
559 * get dealt with eventually.
560 *
561 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200562 * request queue; this lock will be taken also from interrupt context, so irq
563 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200565 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 * it didn't succeed.
567 *
568 * Note:
569 * blk_init_queue() must be paired with a blk_cleanup_queue() call
570 * when the block device is deactivated (such as at module unload).
571 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700572
Jens Axboe165125e2007-07-24 09:28:11 +0200573struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
Christoph Lameter19460892005-06-23 00:08:19 -0700575 return blk_init_queue_node(rfn, lock, -1);
576}
577EXPORT_SYMBOL(blk_init_queue);
578
Jens Axboe165125e2007-07-24 09:28:11 +0200579struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700580blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
581{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600582 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600584 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
585 if (!uninit_q)
586 return NULL;
587
588 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
589 if (!q)
590 blk_cleanup_queue(uninit_q);
591
592 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200593}
594EXPORT_SYMBOL(blk_init_queue_node);
595
596struct request_queue *
597blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
598 spinlock_t *lock)
599{
600 return blk_init_allocated_queue_node(q, rfn, lock, -1);
601}
602EXPORT_SYMBOL(blk_init_allocated_queue);
603
604struct request_queue *
605blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
606 spinlock_t *lock, int node_id)
607{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 if (!q)
609 return NULL;
610
Christoph Lameter19460892005-06-23 00:08:19 -0700611 q->node = node_id;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600612 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500613 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900617 q->unprep_rq_fn = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 q->unplug_fn = generic_unplug_device;
Jens Axboebc58ba92009-01-23 10:54:44 +0100619 q->queue_flags = QUEUE_FLAG_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 q->queue_lock = lock;
621
Jens Axboef3b144a2009-03-06 08:48:33 +0100622 /*
623 * This also sets hw/phys segments, boundary and size
624 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Alan Stern44ec9542007-02-20 11:01:57 -0500627 q->sg_reserved_size = INT_MAX;
628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 /*
630 * all done
631 */
632 if (!elevator_init(q, NULL)) {
633 blk_queue_congestion_threshold(q);
634 return q;
635 }
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 return NULL;
638}
Mike Snitzer01effb02010-05-11 08:57:42 +0200639EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
Jens Axboe165125e2007-07-24 09:28:11 +0200641int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700643 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500644 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 return 0;
646 }
647
648 return 1;
649}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Jens Axboe165125e2007-07-24 09:28:11 +0200651static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200653 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200654 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 mempool_free(rq, q->rq.rq_pool);
656}
657
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200658static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200659blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
661 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
662
663 if (!rq)
664 return NULL;
665
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200666 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200667
Jerome Marchand42dad762009-04-22 14:01:49 +0200668 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
Tejun Heocb98fc82005-10-28 08:29:39 +0200670 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200671 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200672 mempool_free(rq, q->rq.rq_pool);
673 return NULL;
674 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200675 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200676 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Tejun Heocb98fc82005-10-28 08:29:39 +0200678 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679}
680
681/*
682 * ioc_batching returns true if the ioc is a valid batching request and
683 * should be given priority access to a request.
684 */
Jens Axboe165125e2007-07-24 09:28:11 +0200685static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
687 if (!ioc)
688 return 0;
689
690 /*
691 * Make sure the process is able to allocate at least 1 request
692 * even if the batch times out, otherwise we could theoretically
693 * lose wakeups.
694 */
695 return ioc->nr_batch_requests == q->nr_batching ||
696 (ioc->nr_batch_requests > 0
697 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
698}
699
700/*
701 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
702 * will cause the process to be a "batcher" on all queues in the system. This
703 * is the behaviour we want though - once it gets a wakeup it should be given
704 * a nice run.
705 */
Jens Axboe165125e2007-07-24 09:28:11 +0200706static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
708 if (!ioc || ioc_batching(q, ioc))
709 return;
710
711 ioc->nr_batch_requests = q->nr_batching;
712 ioc->last_waited = jiffies;
713}
714
Jens Axboe1faa16d2009-04-06 14:48:01 +0200715static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716{
717 struct request_list *rl = &q->rq;
718
Jens Axboe1faa16d2009-04-06 14:48:01 +0200719 if (rl->count[sync] < queue_congestion_off_threshold(q))
720 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Jens Axboe1faa16d2009-04-06 14:48:01 +0200722 if (rl->count[sync] + 1 <= q->nr_requests) {
723 if (waitqueue_active(&rl->wait[sync]))
724 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Jens Axboe1faa16d2009-04-06 14:48:01 +0200726 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 }
728}
729
730/*
731 * A request has just been released. Account for it, update the full and
732 * congestion status, wake up any waiters. Called under q->queue_lock.
733 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200734static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
736 struct request_list *rl = &q->rq;
737
Jens Axboe1faa16d2009-04-06 14:48:01 +0200738 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200739 if (priv)
740 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Jens Axboe1faa16d2009-04-06 14:48:01 +0200742 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
Jens Axboe1faa16d2009-04-06 14:48:01 +0200744 if (unlikely(rl->starved[sync ^ 1]))
745 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
747
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748/*
Nick Piggind6344532005-06-28 20:45:14 -0700749 * Get a free request, queue_lock must be held.
750 * Returns NULL on failure, with queue_lock held.
751 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 */
Jens Axboe165125e2007-07-24 09:28:11 +0200753static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100754 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755{
756 struct request *rq = NULL;
757 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100758 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200759 const bool is_sync = rw_is_sync(rw_flags) != 0;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100760 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Jens Axboe7749a8d2006-12-13 13:02:26 +0100762 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100763 if (may_queue == ELV_MQUEUE_NO)
764 goto rq_starved;
765
Jens Axboe1faa16d2009-04-06 14:48:01 +0200766 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
767 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200768 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100769 /*
770 * The queue will fill after this allocation, so set
771 * it as full, and mark this process as "batching".
772 * This process will be allowed to complete a batch of
773 * requests, others will be blocked.
774 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200775 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100776 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200777 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100778 } else {
779 if (may_queue != ELV_MQUEUE_MUST
780 && !ioc_batching(q, ioc)) {
781 /*
782 * The queue is full and the allocating
783 * process is not a "batcher", and not
784 * exempted by the IO scheduler
785 */
786 goto out;
787 }
788 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200790 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
792
Jens Axboe082cf692005-06-28 16:35:11 +0200793 /*
794 * Only allow batching queuers to allocate up to 50% over the defined
795 * limit of requests, otherwise we could have thousands of requests
796 * allocated with any setting of ->nr_requests
797 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200798 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200799 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100800
Jens Axboe1faa16d2009-04-06 14:48:01 +0200801 rl->count[is_sync]++;
802 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200803
Jens Axboe64521d12005-10-28 08:30:39 +0200804 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Jens Axboef253b862010-10-24 22:06:02 +0200805 if (priv)
Tejun Heocb98fc82005-10-28 08:29:39 +0200806 rl->elvpriv++;
807
Jens Axboef253b862010-10-24 22:06:02 +0200808 if (blk_queue_io_stat(q))
809 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 spin_unlock_irq(q->queue_lock);
811
Jens Axboe7749a8d2006-12-13 13:02:26 +0100812 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100813 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 /*
815 * Allocation failed presumably due to memory. Undo anything
816 * we might have messed up.
817 *
818 * Allocating task should really be put onto the front of the
819 * wait queue, but this is pretty rare.
820 */
821 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200822 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
824 /*
825 * in the very unlikely event that allocation failed and no
826 * requests for this direction was pending, mark us starved
827 * so that freeing of a request in the other direction will
828 * notice us. another possible fix would be to split the
829 * rq mempool into READ and WRITE
830 */
831rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200832 if (unlikely(rl->count[is_sync] == 0))
833 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 goto out;
836 }
837
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100838 /*
839 * ioc may be NULL here, and ioc_batching will be false. That's
840 * OK, if the queue is under the request limit then requests need
841 * not count toward the nr_batch_requests limit. There will always
842 * be some limit enforced by BLK_BATCH_TIME.
843 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (ioc_batching(q, ioc))
845 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100846
Jens Axboe1faa16d2009-04-06 14:48:01 +0200847 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 return rq;
850}
851
852/*
853 * No available requests for this queue, unplug the device and wait for some
854 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700855 *
856 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 */
Jens Axboe165125e2007-07-24 09:28:11 +0200858static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200859 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200861 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 struct request *rq;
863
Jens Axboe7749a8d2006-12-13 13:02:26 +0100864 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700865 while (!rq) {
866 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200867 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 struct request_list *rl = &q->rq;
869
Jens Axboe1faa16d2009-04-06 14:48:01 +0200870 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 TASK_UNINTERRUPTIBLE);
872
Jens Axboe1faa16d2009-04-06 14:48:01 +0200873 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200875 __generic_unplug_device(q);
876 spin_unlock_irq(q->queue_lock);
877 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200879 /*
880 * After sleeping, we become a "batching" process and
881 * will be able to allocate at least one request, and
882 * up to a big batch of them for a small period time.
883 * See ioc_batching, ioc_set_batching
884 */
885 ioc = current_io_context(GFP_NOIO, q->node);
886 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100887
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200888 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200889 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200890
891 rq = get_request(q, rw_flags, bio, GFP_NOIO);
892 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893
894 return rq;
895}
896
Jens Axboe165125e2007-07-24 09:28:11 +0200897struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898{
899 struct request *rq;
900
901 BUG_ON(rw != READ && rw != WRITE);
902
Nick Piggind6344532005-06-28 20:45:14 -0700903 spin_lock_irq(q->queue_lock);
904 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200905 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700906 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200907 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700908 if (!rq)
909 spin_unlock_irq(q->queue_lock);
910 }
911 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
913 return rq;
914}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915EXPORT_SYMBOL(blk_get_request);
916
917/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300918 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700919 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300920 * @bio: The bio describing the memory mappings that will be submitted for IO.
921 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700922 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200923 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300924 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
925 * type commands. Where the struct request needs to be farther initialized by
926 * the caller. It is passed a &struct bio, which describes the memory info of
927 * the I/O transfer.
928 *
929 * The caller of blk_make_request must make sure that bi_io_vec
930 * are set to describe the memory buffers. That bio_data_dir() will return
931 * the needed direction of the request. (And all bio's in the passed bio-chain
932 * are properly set accordingly)
933 *
934 * If called under none-sleepable conditions, mapped bio buffers must not
935 * need bouncing, by calling the appropriate masked or flagged allocator,
936 * suitable for the target device. Otherwise the call to blk_queue_bounce will
937 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200938 *
939 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
940 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
941 * anything but the first bio in the chain. Otherwise you risk waiting for IO
942 * completion of a bio that hasn't been submitted yet, thus resulting in a
943 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
944 * of bio_alloc(), as that avoids the mempool deadlock.
945 * If possible a big IO should be split into smaller parts when allocation
946 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200947 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300948struct request *blk_make_request(struct request_queue *q, struct bio *bio,
949 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200950{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300951 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
952
953 if (unlikely(!rq))
954 return ERR_PTR(-ENOMEM);
955
956 for_each_bio(bio) {
957 struct bio *bounce_bio = bio;
958 int ret;
959
960 blk_queue_bounce(q, &bounce_bio);
961 ret = blk_rq_append_bio(q, rq, bounce_bio);
962 if (unlikely(ret)) {
963 blk_put_request(rq);
964 return ERR_PTR(ret);
965 }
966 }
967
968 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200969}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300970EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200971
972/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 * blk_requeue_request - put a request back on queue
974 * @q: request queue where request should be inserted
975 * @rq: request to be inserted
976 *
977 * Description:
978 * Drivers often keep queueing requests until the hardware cannot accept
979 * more, when that condition happens we need to put the request back
980 * on the queue. Must be called with queue lock held.
981 */
Jens Axboe165125e2007-07-24 09:28:11 +0200982void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700984 blk_delete_timer(rq);
985 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100986 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100987
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 if (blk_rq_tagged(rq))
989 blk_queue_end_tag(q, rq);
990
James Bottomleyba396a62009-05-27 14:17:08 +0200991 BUG_ON(blk_queued_rq(rq));
992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 elv_requeue_request(q, rq);
994}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995EXPORT_SYMBOL(blk_requeue_request);
996
997/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200998 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 * @q: request queue where request should be inserted
1000 * @rq: request to be inserted
1001 * @at_head: insert request at head or tail of queue
1002 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 *
1004 * Description:
1005 * Many block devices need to execute commands asynchronously, so they don't
1006 * block the whole kernel from preemption during request execution. This is
1007 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +02001008 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1009 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 *
1011 * We have the option of inserting the head or the tail of the queue.
1012 * Typically we use the tail for new ioctls and so forth. We use the head
1013 * of the queue for things like a QUEUE_FULL message from a device, or a
1014 * host that is unable to accept a particular command.
1015 */
Jens Axboe165125e2007-07-24 09:28:11 +02001016void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05001017 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018{
Tejun Heo 867d1192005-04-24 02:06:05 -05001019 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 unsigned long flags;
1021
1022 /*
1023 * tell I/O scheduler that this isn't a regular read/write (ie it
1024 * must not attempt merges on this) and that it acts as a soft
1025 * barrier
1026 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02001027 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029 rq->special = data;
1030
1031 spin_lock_irqsave(q->queue_lock, flags);
1032
1033 /*
1034 * If command is tagged, release the tag
1035 */
Tejun Heo 867d1192005-04-24 02:06:05 -05001036 if (blk_rq_tagged(rq))
1037 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Jerome Marchandb238b3d2007-10-23 15:05:46 +02001039 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -05001040 __elv_add_request(q, rq, where, 0);
Tejun Heoa7f55792009-04-23 11:05:17 +09001041 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 spin_unlock_irqrestore(q->queue_lock, flags);
1043}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044EXPORT_SYMBOL(blk_insert_request);
1045
Tejun Heo074a7ac2008-08-25 19:56:14 +09001046static void part_round_stats_single(int cpu, struct hd_struct *part,
1047 unsigned long now)
1048{
1049 if (now == part->stamp)
1050 return;
1051
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001052 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001053 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001054 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001055 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1056 }
1057 part->stamp = now;
1058}
1059
1060/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001061 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1062 * @cpu: cpu number for stats access
1063 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 *
1065 * The average IO queue length and utilisation statistics are maintained
1066 * by observing the current state of the queue length and the amount of
1067 * time it has been in this state for.
1068 *
1069 * Normally, that accounting is done on IO completion, but that can result
1070 * in more than a second's worth of IO being accounted for within any one
1071 * second, leading to >100% utilisation. To deal with that, we call this
1072 * function to do a round-off before returning the results when reading
1073 * /proc/diskstats. This accounts immediately for all queue usage up to
1074 * the current jiffies and restarts the counters again.
1075 */
Tejun Heoc9959052008-08-25 19:47:21 +09001076void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001077{
1078 unsigned long now = jiffies;
1079
Tejun Heo074a7ac2008-08-25 19:56:14 +09001080 if (part->partno)
1081 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1082 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001083}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001084EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001085
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086/*
1087 * queue lock must be held
1088 */
Jens Axboe165125e2007-07-24 09:28:11 +02001089void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (unlikely(!q))
1092 return;
1093 if (unlikely(--req->ref_count))
1094 return;
1095
Tejun Heo8922e162005-10-20 16:23:44 +02001096 elv_completed_request(q, req);
1097
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001098 /* this is a bio leak */
1099 WARN_ON(req->bio != NULL);
1100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 /*
1102 * Request may not have originated from ll_rw_blk. if not,
1103 * it didn't come out of our reserved rq pools
1104 */
Jens Axboe49171e52006-08-10 08:59:11 +02001105 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001106 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001107 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001110 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
1112 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001113 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 }
1115}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001116EXPORT_SYMBOL_GPL(__blk_put_request);
1117
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118void blk_put_request(struct request *req)
1119{
Tejun Heo8922e162005-10-20 16:23:44 +02001120 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001121 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001123 spin_lock_irqsave(q->queue_lock, flags);
1124 __blk_put_request(q, req);
1125 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127EXPORT_SYMBOL(blk_put_request);
1128
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001129/**
1130 * blk_add_request_payload - add a payload to a request
1131 * @rq: request to update
1132 * @page: page backing the payload
1133 * @len: length of the payload.
1134 *
1135 * This allows to later add a payload to an already submitted request by
1136 * a block driver. The driver needs to take care of freeing the payload
1137 * itself.
1138 *
1139 * Note that this is a quite horrible hack and nothing but handling of
1140 * discard requests should ever use it.
1141 */
1142void blk_add_request_payload(struct request *rq, struct page *page,
1143 unsigned int len)
1144{
1145 struct bio *bio = rq->bio;
1146
1147 bio->bi_io_vec->bv_page = page;
1148 bio->bi_io_vec->bv_offset = 0;
1149 bio->bi_io_vec->bv_len = len;
1150
1151 bio->bi_size = len;
1152 bio->bi_vcnt = 1;
1153 bio->bi_phys_segments = 1;
1154
1155 rq->__data_len = rq->resid_len = len;
1156 rq->nr_phys_segments = 1;
1157 rq->buffer = bio_data(bio);
1158}
1159EXPORT_SYMBOL_GPL(blk_add_request_payload);
1160
Jens Axboe86db1e22008-01-29 14:53:40 +01001161void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001162{
Jens Axboec7c22e42008-09-13 20:26:01 +02001163 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001164 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001165
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001166 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1167 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001168 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001169
Tejun Heo52d9e672006-01-06 09:49:58 +01001170 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001171 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001172 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001173 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001174}
1175
Jens Axboe644b2d92009-04-06 14:48:06 +02001176/*
1177 * Only disabling plugging for non-rotational devices if it does tagging
1178 * as well, otherwise we do need the proper merging
1179 */
1180static inline bool queue_should_plug(struct request_queue *q)
1181{
Jens Axboe79da06442010-02-23 08:40:43 +01001182 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
Jens Axboe644b2d92009-04-06 14:48:06 +02001183}
1184
Jens Axboe165125e2007-07-24 09:28:11 +02001185static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186{
Nick Piggin450991b2005-06-28 20:45:13 -07001187 struct request *req;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001188 int el_ret;
1189 unsigned int bytes = bio->bi_size;
Jens Axboe51da90f2006-07-18 04:14:45 +02001190 const unsigned short prio = bio_prio(bio);
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001191 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1192 const bool unplug = !!(bio->bi_rw & REQ_UNPLUG);
1193 const unsigned long ff = bio->bi_rw & REQ_FAILFAST_MASK;
Tejun Heo28e7d182010-09-03 11:56:16 +02001194 int where = ELEVATOR_INSERT_SORT;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001195 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 /*
1198 * low level driver can indicate that it wants pages above a
1199 * certain limit bounced to low memory (ie for highmem, or even
1200 * ISA dma in theory)
1201 */
1202 blk_queue_bounce(q, &bio);
1203
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 spin_lock_irq(q->queue_lock);
1205
Tejun Heo4fed9472010-09-03 11:56:17 +02001206 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Tejun Heo28e7d182010-09-03 11:56:16 +02001207 where = ELEVATOR_INSERT_FRONT;
1208 goto get_rq;
1209 }
1210
1211 if (elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 goto get_rq;
1213
1214 el_ret = elv_merge(q, &req, bio);
1215 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001216 case ELEVATOR_BACK_MERGE:
1217 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
Jens Axboe6728cb02008-01-31 13:03:55 +01001219 if (!ll_back_merge_fn(q, req, bio))
1220 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001222 trace_block_bio_backmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001223
Tejun Heo80a761f2009-07-03 17:48:17 +09001224 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1225 blk_rq_set_mixed_merge(req);
1226
Jens Axboe6728cb02008-01-31 13:03:55 +01001227 req->biotail->bi_next = bio;
1228 req->biotail = bio;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001229 req->__data_len += bytes;
Jens Axboe6728cb02008-01-31 13:03:55 +01001230 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001231 if (!blk_rq_cpu_valid(req))
1232 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001233 drive_stat_acct(req, 0);
Divyesh Shah812d4022010-04-08 21:14:23 -07001234 elv_bio_merged(q, req, bio);
Jens Axboe6728cb02008-01-31 13:03:55 +01001235 if (!attempt_back_merge(q, req))
1236 elv_merged_request(q, req, el_ret);
1237 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Jens Axboe6728cb02008-01-31 13:03:55 +01001239 case ELEVATOR_FRONT_MERGE:
1240 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Jens Axboe6728cb02008-01-31 13:03:55 +01001242 if (!ll_front_merge_fn(q, req, bio))
1243 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001245 trace_block_bio_frontmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001246
Tejun Heo80a761f2009-07-03 17:48:17 +09001247 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1248 blk_rq_set_mixed_merge(req);
1249 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1250 req->cmd_flags |= ff;
1251 }
1252
Jens Axboe6728cb02008-01-31 13:03:55 +01001253 bio->bi_next = req->bio;
1254 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
Jens Axboe6728cb02008-01-31 13:03:55 +01001256 /*
1257 * may not be valid. if the low level driver said
1258 * it didn't need a bounce buffer then it better
1259 * not touch req->buffer either...
1260 */
1261 req->buffer = bio_data(bio);
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001262 req->__sector = bio->bi_sector;
1263 req->__data_len += bytes;
Jens Axboe6728cb02008-01-31 13:03:55 +01001264 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001265 if (!blk_rq_cpu_valid(req))
1266 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001267 drive_stat_acct(req, 0);
Divyesh Shah812d4022010-04-08 21:14:23 -07001268 elv_bio_merged(q, req, bio);
Jens Axboe6728cb02008-01-31 13:03:55 +01001269 if (!attempt_front_merge(q, req))
1270 elv_merged_request(q, req, el_ret);
1271 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
Jens Axboe6728cb02008-01-31 13:03:55 +01001273 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1274 default:
1275 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 }
1277
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001279 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001280 * This sync check and mask will be re-done in init_request_from_bio(),
1281 * but we need to set it earlier to expose the sync flag to the
1282 * rq allocator and io schedulers.
1283 */
1284 rw_flags = bio_data_dir(bio);
1285 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001286 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001287
1288 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001289 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001290 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001291 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001292 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001293
Nick Piggin450991b2005-06-28 20:45:13 -07001294 /*
1295 * After dropping the lock and possibly sleeping here, our request
1296 * may now be mergeable after it had proven unmergeable (above).
1297 * We don't worry about that case for efficiency. It won't happen
1298 * often, and the elevators are able to handle it.
1299 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001300 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Nick Piggin450991b2005-06-28 20:45:13 -07001302 spin_lock_irq(q->queue_lock);
Jens Axboec7c22e42008-09-13 20:26:01 +02001303 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1304 bio_flagged(bio, BIO_CPU_AFFINE))
1305 req->cpu = blk_cpu_to_group(smp_processor_id());
Jens Axboe644b2d92009-04-06 14:48:06 +02001306 if (queue_should_plug(q) && elv_queue_empty(q))
Nick Piggin450991b2005-06-28 20:45:13 -07001307 blk_plug_device(q);
Tejun Heodd831002010-09-03 11:56:16 +02001308
1309 /* insert the request into the elevator */
1310 drive_stat_acct(req, 1);
Tejun Heo28e7d182010-09-03 11:56:16 +02001311 __elv_add_request(q, req, where, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312out:
Jens Axboe644b2d92009-04-06 14:48:06 +02001313 if (unplug || !queue_should_plug(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 __generic_unplug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 spin_unlock_irq(q->queue_lock);
1316 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317}
1318
1319/*
1320 * If bio->bi_dev is a partition, remap the location
1321 */
1322static inline void blk_partition_remap(struct bio *bio)
1323{
1324 struct block_device *bdev = bio->bi_bdev;
1325
Jens Axboebf2de6f2007-09-27 13:01:25 +02001326 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001328
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 bio->bi_sector += p->start_sect;
1330 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001331
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001332 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
Alan D. Brunelle22a7c312009-05-04 16:35:08 -04001333 bdev->bd_dev,
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001334 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 }
1336}
1337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338static void handle_bad_sector(struct bio *bio)
1339{
1340 char b[BDEVNAME_SIZE];
1341
1342 printk(KERN_INFO "attempt to access beyond end of device\n");
1343 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1344 bdevname(bio->bi_bdev, b),
1345 bio->bi_rw,
1346 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001347 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
1349 set_bit(BIO_EOF, &bio->bi_flags);
1350}
1351
Akinobu Mitac17bb492006-12-08 02:39:46 -08001352#ifdef CONFIG_FAIL_MAKE_REQUEST
1353
1354static DECLARE_FAULT_ATTR(fail_make_request);
1355
1356static int __init setup_fail_make_request(char *str)
1357{
1358 return setup_fault_attr(&fail_make_request, str);
1359}
1360__setup("fail_make_request=", setup_fail_make_request);
1361
1362static int should_fail_request(struct bio *bio)
1363{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001364 struct hd_struct *part = bio->bi_bdev->bd_part;
1365
1366 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001367 return should_fail(&fail_make_request, bio->bi_size);
1368
1369 return 0;
1370}
1371
1372static int __init fail_make_request_debugfs(void)
1373{
1374 return init_fault_attr_dentries(&fail_make_request,
1375 "fail_make_request");
1376}
1377
1378late_initcall(fail_make_request_debugfs);
1379
1380#else /* CONFIG_FAIL_MAKE_REQUEST */
1381
1382static inline int should_fail_request(struct bio *bio)
1383{
1384 return 0;
1385}
1386
1387#endif /* CONFIG_FAIL_MAKE_REQUEST */
1388
Jens Axboec07e2b42007-07-18 13:27:58 +02001389/*
1390 * Check whether this bio extends beyond the end of the device.
1391 */
1392static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1393{
1394 sector_t maxsector;
1395
1396 if (!nr_sectors)
1397 return 0;
1398
1399 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001400 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001401 if (maxsector) {
1402 sector_t sector = bio->bi_sector;
1403
1404 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1405 /*
1406 * This may well happen - the kernel calls bread()
1407 * without checking the size of the device, e.g., when
1408 * mounting a device.
1409 */
1410 handle_bad_sector(bio);
1411 return 1;
1412 }
1413 }
1414
1415 return 0;
1416}
1417
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001419 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 * @bio: The bio describing the location in memory and on the device.
1421 *
1422 * generic_make_request() is used to make I/O requests of block
1423 * devices. It is passed a &struct bio, which describes the I/O that needs
1424 * to be done.
1425 *
1426 * generic_make_request() does not return any status. The
1427 * success/failure status of the request, along with notification of
1428 * completion, is delivered asynchronously through the bio->bi_end_io
1429 * function described (one day) else where.
1430 *
1431 * The caller of generic_make_request must make sure that bi_io_vec
1432 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1433 * set to describe the device address, and the
1434 * bi_end_io and optionally bi_private are set to describe how
1435 * completion notification should be signaled.
1436 *
1437 * generic_make_request and the drivers it calls may use bi_next if this
1438 * bio happens to be merged with someone else, and may change bi_dev and
1439 * bi_sector for remaps as it sees fit. So the values of these fields
1440 * should NOT be depended on after the call to generic_make_request.
1441 */
Neil Brownd89d8792007-05-01 09:53:42 +02001442static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443{
Jens Axboe165125e2007-07-24 09:28:11 +02001444 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001445 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001447 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001448 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
1450 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
Jens Axboec07e2b42007-07-18 13:27:58 +02001452 if (bio_check_eod(bio, nr_sectors))
1453 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
1455 /*
1456 * Resolve the mapping until finished. (drivers are
1457 * still free to implement/resolve their own stacking
1458 * by explicitly returning 0)
1459 *
1460 * NOTE: we don't repeat the blk_size check for each new device.
1461 * Stacking drivers are expected to know what they are doing.
1462 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001463 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001464 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 do {
1466 char b[BDEVNAME_SIZE];
1467
1468 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001469 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 printk(KERN_ERR
1471 "generic_make_request: Trying to access "
1472 "nonexistent block-device %s (%Lu)\n",
1473 bdevname(bio->bi_bdev, b),
1474 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001475 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 }
1477
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001478 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
Christoph Hellwig67efc922009-09-30 13:54:20 +02001479 nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001480 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001481 bdevname(bio->bi_bdev, b),
1482 bio_sectors(bio),
1483 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 goto end_io;
1485 }
1486
Nick Pigginfde6ad22005-06-23 00:08:53 -07001487 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 goto end_io;
1489
Akinobu Mitac17bb492006-12-08 02:39:46 -08001490 if (should_fail_request(bio))
1491 goto end_io;
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 /*
1494 * If this device has partitions, remap block n
1495 * of partition p to block n+start(p) of the disk.
1496 */
1497 blk_partition_remap(bio);
1498
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001499 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1500 goto end_io;
1501
NeilBrown5ddfe962006-10-30 22:07:21 -08001502 if (old_sector != -1)
Alan D. Brunelle22a7c312009-05-04 16:35:08 -04001503 trace_block_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001504
NeilBrown5ddfe962006-10-30 22:07:21 -08001505 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001506 old_dev = bio->bi_bdev->bd_dev;
1507
Jens Axboec07e2b42007-07-18 13:27:58 +02001508 if (bio_check_eod(bio, nr_sectors))
1509 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001510
Tejun Heo1e879012010-09-03 11:56:17 +02001511 /*
1512 * Filter flush bio's early so that make_request based
1513 * drivers without flush support don't have to worry
1514 * about them.
1515 */
1516 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1517 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1518 if (!nr_sectors) {
1519 err = 0;
1520 goto end_io;
1521 }
1522 }
1523
Adrian Hunter8d57a982010-08-11 14:17:49 -07001524 if ((bio->bi_rw & REQ_DISCARD) &&
1525 (!blk_queue_discard(q) ||
1526 ((bio->bi_rw & REQ_SECURE) &&
1527 !blk_queue_secdiscard(q)))) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001528 err = -EOPNOTSUPP;
1529 goto end_io;
1530 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001531
Vivek Goyale43473b2010-09-15 17:06:35 -04001532 blk_throtl_bio(q, &bio);
1533
1534 /*
1535 * If bio = NULL, bio has been throttled and will be submitted
1536 * later.
1537 */
1538 if (!bio)
1539 break;
1540
Minchan Kim01edede2009-09-08 21:56:38 +02001541 trace_block_bio_queue(q, bio);
1542
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 ret = q->make_request_fn(q, bio);
1544 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001545
1546 return;
1547
1548end_io:
1549 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550}
1551
Neil Brownd89d8792007-05-01 09:53:42 +02001552/*
1553 * We only want one ->make_request_fn to be active at a time,
1554 * else stack usage with stacked devices could be a problem.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001555 * So use current->bio_list to keep a list of requests
Neil Brownd89d8792007-05-01 09:53:42 +02001556 * submited by a make_request_fn function.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001557 * current->bio_list is also used as a flag to say if
Neil Brownd89d8792007-05-01 09:53:42 +02001558 * generic_make_request is currently active in this task or not.
1559 * If it is NULL, then no make_request is active. If it is non-NULL,
1560 * then a make_request is active, and new requests should be added
1561 * at the tail
1562 */
1563void generic_make_request(struct bio *bio)
1564{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001565 struct bio_list bio_list_on_stack;
1566
1567 if (current->bio_list) {
Neil Brownd89d8792007-05-01 09:53:42 +02001568 /* make_request is active */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001569 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001570 return;
1571 }
1572 /* following loop may be a bit non-obvious, and so deserves some
1573 * explanation.
1574 * Before entering the loop, bio->bi_next is NULL (as all callers
1575 * ensure that) so we have a list with a single bio.
1576 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001577 * we assign bio_list to a pointer to the bio_list_on_stack,
1578 * thus initialising the bio_list of new bios to be
Neil Brownd89d8792007-05-01 09:53:42 +02001579 * added. __generic_make_request may indeed add some more bios
1580 * through a recursive call to generic_make_request. If it
1581 * did, we find a non-NULL value in bio_list and re-enter the loop
1582 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001583 * of the top of the list (no pretending) and so remove it from
1584 * bio_list, and call into __generic_make_request again.
Neil Brownd89d8792007-05-01 09:53:42 +02001585 *
1586 * The loop was structured like this to make only one call to
1587 * __generic_make_request (which is important as it is large and
1588 * inlined) and to keep the structure simple.
1589 */
1590 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001591 bio_list_init(&bio_list_on_stack);
1592 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001593 do {
Neil Brownd89d8792007-05-01 09:53:42 +02001594 __generic_make_request(bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001595 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001596 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001597 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001598}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599EXPORT_SYMBOL(generic_make_request);
1600
1601/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001602 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1604 * @bio: The &struct bio which describes the I/O
1605 *
1606 * submit_bio() is very similar in purpose to generic_make_request(), and
1607 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001608 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 *
1610 */
1611void submit_bio(int rw, struct bio *bio)
1612{
1613 int count = bio_sectors(bio);
1614
Jens Axboe22e2c502005-06-27 10:55:12 +02001615 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
Jens Axboebf2de6f2007-09-27 13:01:25 +02001617 /*
1618 * If it's a regular read/write or a barrier with data attached,
1619 * go through the normal accounting stuff before submission.
1620 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001621 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001622 if (rw & WRITE) {
1623 count_vm_events(PGPGOUT, count);
1624 } else {
1625 task_io_account_read(bio->bi_size);
1626 count_vm_events(PGPGIN, count);
1627 }
1628
1629 if (unlikely(block_dump)) {
1630 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001631 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001632 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001633 (rw & WRITE) ? "WRITE" : "READ",
1634 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001635 bdevname(bio->bi_bdev, b),
1636 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
1639
1640 generic_make_request(bio);
1641}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642EXPORT_SYMBOL(submit_bio);
1643
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001644/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001645 * blk_rq_check_limits - Helper function to check a request for the queue limit
1646 * @q: the queue
1647 * @rq: the request being checked
1648 *
1649 * Description:
1650 * @rq may have been made based on weaker limitations of upper-level queues
1651 * in request stacking drivers, and it may violate the limitation of @q.
1652 * Since the block layer and the underlying device driver trust @rq
1653 * after it is inserted to @q, it should be checked against @q before
1654 * the insertion using this generic function.
1655 *
1656 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001657 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001658 * Request stacking drivers like request-based dm may change the queue
1659 * limits while requests are in the queue (e.g. dm's table swapping).
1660 * Such request stacking drivers should check those requests agaist
1661 * the new queue limits again when they dispatch those requests,
1662 * although such checkings are also done against the old queue limits
1663 * when submitting requests.
1664 */
1665int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1666{
ike Snitzer33839772010-08-08 12:11:33 -04001667 if (rq->cmd_flags & REQ_DISCARD)
1668 return 0;
1669
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001670 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1671 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001672 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1673 return -EIO;
1674 }
1675
1676 /*
1677 * queue's settings related to segment counting like q->bounce_pfn
1678 * may differ from that of other stacking queues.
1679 * Recalculate it to check the request correctly on this queue's
1680 * limitation.
1681 */
1682 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001683 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001684 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1685 return -EIO;
1686 }
1687
1688 return 0;
1689}
1690EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1691
1692/**
1693 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1694 * @q: the queue to submit the request
1695 * @rq: the request being queued
1696 */
1697int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1698{
1699 unsigned long flags;
1700
1701 if (blk_rq_check_limits(q, rq))
1702 return -EIO;
1703
1704#ifdef CONFIG_FAIL_MAKE_REQUEST
1705 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1706 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1707 return -EIO;
1708#endif
1709
1710 spin_lock_irqsave(q->queue_lock, flags);
1711
1712 /*
1713 * Submitting request must be dequeued before calling this function
1714 * because it will be linked to another request_queue
1715 */
1716 BUG_ON(blk_queued_rq(rq));
1717
1718 drive_stat_acct(rq, 1);
1719 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1720
1721 spin_unlock_irqrestore(q->queue_lock, flags);
1722
1723 return 0;
1724}
1725EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1726
Tejun Heo80a761f2009-07-03 17:48:17 +09001727/**
1728 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1729 * @rq: request to examine
1730 *
1731 * Description:
1732 * A request could be merge of IOs which require different failure
1733 * handling. This function determines the number of bytes which
1734 * can be failed from the beginning of the request without
1735 * crossing into area which need to be retried further.
1736 *
1737 * Return:
1738 * The number of bytes to fail.
1739 *
1740 * Context:
1741 * queue_lock must be held.
1742 */
1743unsigned int blk_rq_err_bytes(const struct request *rq)
1744{
1745 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1746 unsigned int bytes = 0;
1747 struct bio *bio;
1748
1749 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1750 return blk_rq_bytes(rq);
1751
1752 /*
1753 * Currently the only 'mixing' which can happen is between
1754 * different fastfail types. We can safely fail portions
1755 * which have all the failfast bits that the first one has -
1756 * the ones which are at least as eager to fail as the first
1757 * one.
1758 */
1759 for (bio = rq->bio; bio; bio = bio->bi_next) {
1760 if ((bio->bi_rw & ff) != ff)
1761 break;
1762 bytes += bio->bi_size;
1763 }
1764
1765 /* this could lead to infinite loop */
1766 BUG_ON(blk_rq_bytes(rq) && !bytes);
1767 return bytes;
1768}
1769EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1770
Jens Axboebc58ba92009-01-23 10:54:44 +01001771static void blk_account_io_completion(struct request *req, unsigned int bytes)
1772{
Jens Axboec2553b52009-04-24 08:10:11 +02001773 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001774 const int rw = rq_data_dir(req);
1775 struct hd_struct *part;
1776 int cpu;
1777
1778 cpu = part_stat_lock();
Jens Axboef253b862010-10-24 22:06:02 +02001779 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001780 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1781 part_stat_unlock();
1782 }
1783}
1784
1785static void blk_account_io_done(struct request *req)
1786{
Jens Axboebc58ba92009-01-23 10:54:44 +01001787 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001788 * Account IO completion. flush_rq isn't accounted as a
1789 * normal IO on queueing nor completion. Accounting the
1790 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001791 */
Tejun Heodd4c1332010-09-03 11:56:16 +02001792 if (blk_do_io_stat(req) && req != &req->q->flush_rq) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001793 unsigned long duration = jiffies - req->start_time;
1794 const int rw = rq_data_dir(req);
1795 struct hd_struct *part;
1796 int cpu;
1797
1798 cpu = part_stat_lock();
Jens Axboef253b862010-10-24 22:06:02 +02001799 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001800
1801 part_stat_inc(cpu, part, ios[rw]);
1802 part_stat_add(cpu, part, ticks[rw], duration);
1803 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001804 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001805
1806 part_stat_unlock();
1807 }
1808}
1809
Tejun Heo53a08802008-12-03 12:41:26 +01001810/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001811 * blk_peek_request - peek at the top of a request queue
1812 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001813 *
1814 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001815 * Return the request at the top of @q. The returned request
1816 * should be started using blk_start_request() before LLD starts
1817 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001818 *
1819 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001820 * Pointer to the request at the top of @q if available. Null
1821 * otherwise.
1822 *
1823 * Context:
1824 * queue_lock must be held.
1825 */
1826struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001827{
1828 struct request *rq;
1829 int ret;
1830
1831 while ((rq = __elv_next_request(q)) != NULL) {
1832 if (!(rq->cmd_flags & REQ_STARTED)) {
1833 /*
1834 * This is the first time the device driver
1835 * sees this request (possibly after
1836 * requeueing). Notify IO scheduler.
1837 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001838 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001839 elv_activate_rq(q, rq);
1840
1841 /*
1842 * just mark as started even if we don't start
1843 * it, a request that has been delayed should
1844 * not be passed by new incoming requests
1845 */
1846 rq->cmd_flags |= REQ_STARTED;
1847 trace_block_rq_issue(q, rq);
1848 }
1849
1850 if (!q->boundary_rq || q->boundary_rq == rq) {
1851 q->end_sector = rq_end_sector(rq);
1852 q->boundary_rq = NULL;
1853 }
1854
1855 if (rq->cmd_flags & REQ_DONTPREP)
1856 break;
1857
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001858 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001859 /*
1860 * make sure space for the drain appears we
1861 * know we can do this because max_hw_segments
1862 * has been adjusted to be one fewer than the
1863 * device can handle
1864 */
1865 rq->nr_phys_segments++;
1866 }
1867
1868 if (!q->prep_rq_fn)
1869 break;
1870
1871 ret = q->prep_rq_fn(q, rq);
1872 if (ret == BLKPREP_OK) {
1873 break;
1874 } else if (ret == BLKPREP_DEFER) {
1875 /*
1876 * the request may have been (partially) prepped.
1877 * we need to keep this request in the front to
1878 * avoid resource deadlock. REQ_STARTED will
1879 * prevent other fs requests from passing this one.
1880 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001881 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001882 !(rq->cmd_flags & REQ_DONTPREP)) {
1883 /*
1884 * remove the space for the drain we added
1885 * so that we don't add it again
1886 */
1887 --rq->nr_phys_segments;
1888 }
1889
1890 rq = NULL;
1891 break;
1892 } else if (ret == BLKPREP_KILL) {
1893 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001894 /*
1895 * Mark this request as started so we don't trigger
1896 * any debug logic in the end I/O path.
1897 */
1898 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001899 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001900 } else {
1901 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1902 break;
1903 }
1904 }
1905
1906 return rq;
1907}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001908EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001909
Tejun Heo9934c8c2009-05-08 11:54:16 +09001910void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001911{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001912 struct request_queue *q = rq->q;
1913
Tejun Heo158dbda2009-04-23 11:05:18 +09001914 BUG_ON(list_empty(&rq->queuelist));
1915 BUG_ON(ELV_ON_HASH(rq));
1916
1917 list_del_init(&rq->queuelist);
1918
1919 /*
1920 * the time frame between a request being removed from the lists
1921 * and to it is freed is accounted as io that is in progress at
1922 * the driver side.
1923 */
Divyesh Shah91952912010-04-01 15:01:41 -07001924 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001925 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001926 set_io_start_time_ns(rq);
1927 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001928}
1929
Tejun Heo5efccd12009-04-23 11:05:18 +09001930/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001931 * blk_start_request - start request processing on the driver
1932 * @req: request to dequeue
1933 *
1934 * Description:
1935 * Dequeue @req and start timeout timer on it. This hands off the
1936 * request to the driver.
1937 *
1938 * Block internal functions which don't want to start timer should
1939 * call blk_dequeue_request().
1940 *
1941 * Context:
1942 * queue_lock must be held.
1943 */
1944void blk_start_request(struct request *req)
1945{
1946 blk_dequeue_request(req);
1947
1948 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001949 * We are now handing the request to the hardware, initialize
1950 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001951 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001952 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001953 if (unlikely(blk_bidi_rq(req)))
1954 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1955
Tejun Heo9934c8c2009-05-08 11:54:16 +09001956 blk_add_timer(req);
1957}
1958EXPORT_SYMBOL(blk_start_request);
1959
1960/**
1961 * blk_fetch_request - fetch a request from a request queue
1962 * @q: request queue to fetch a request from
1963 *
1964 * Description:
1965 * Return the request at the top of @q. The request is started on
1966 * return and LLD can start processing it immediately.
1967 *
1968 * Return:
1969 * Pointer to the request at the top of @q if available. Null
1970 * otherwise.
1971 *
1972 * Context:
1973 * queue_lock must be held.
1974 */
1975struct request *blk_fetch_request(struct request_queue *q)
1976{
1977 struct request *rq;
1978
1979 rq = blk_peek_request(q);
1980 if (rq)
1981 blk_start_request(rq);
1982 return rq;
1983}
1984EXPORT_SYMBOL(blk_fetch_request);
1985
1986/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001987 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001988 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001989 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001990 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001991 *
1992 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001993 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1994 * the request structure even if @req doesn't have leftover.
1995 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09001996 *
1997 * This special helper function is only for request stacking drivers
1998 * (e.g. request-based dm) so that they can handle partial completion.
1999 * Actual device drivers should use blk_end_request instead.
2000 *
2001 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2002 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002003 *
2004 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002005 * %false - this request doesn't have any more data
2006 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002007 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002008bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002010 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 struct bio *bio;
2012
Tejun Heo2e60e022009-04-23 11:05:18 +09002013 if (!req->bio)
2014 return false;
2015
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002016 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002017
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002019 * For fs requests, rq is just carrier of independent bio's
2020 * and each partial completion should be handled separately.
2021 * Reset per-request error on each partial completion.
2022 *
2023 * TODO: tj: This is too subtle. It would be better to let
2024 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002026 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 req->errors = 0;
2028
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002029 if (error && req->cmd_type == REQ_TYPE_FS &&
2030 !(req->cmd_flags & REQ_QUIET)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01002031 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 req->rq_disk ? req->rq_disk->disk_name : "?",
Tejun Heo83096eb2009-05-07 22:24:39 +09002033 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 }
2035
Jens Axboebc58ba92009-01-23 10:54:44 +01002036 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 total_bytes = bio_nbytes = 0;
2039 while ((bio = req->bio) != NULL) {
2040 int nbytes;
2041
2042 if (nr_bytes >= bio->bi_size) {
2043 req->bio = bio->bi_next;
2044 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002045 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 next_idx = 0;
2047 bio_nbytes = 0;
2048 } else {
2049 int idx = bio->bi_idx + next_idx;
2050
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002051 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002053 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002054 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 break;
2056 }
2057
2058 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2059 BIO_BUG_ON(nbytes > bio->bi_size);
2060
2061 /*
2062 * not a complete bvec done
2063 */
2064 if (unlikely(nbytes > nr_bytes)) {
2065 bio_nbytes += nr_bytes;
2066 total_bytes += nr_bytes;
2067 break;
2068 }
2069
2070 /*
2071 * advance to the next vector
2072 */
2073 next_idx++;
2074 bio_nbytes += nbytes;
2075 }
2076
2077 total_bytes += nbytes;
2078 nr_bytes -= nbytes;
2079
Jens Axboe6728cb02008-01-31 13:03:55 +01002080 bio = req->bio;
2081 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 /*
2083 * end more in this run, or just return 'not-done'
2084 */
2085 if (unlikely(nr_bytes <= 0))
2086 break;
2087 }
2088 }
2089
2090 /*
2091 * completely done
2092 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002093 if (!req->bio) {
2094 /*
2095 * Reset counters so that the request stacking driver
2096 * can find how many bytes remain in the request
2097 * later.
2098 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002099 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002100 return false;
2101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102
2103 /*
2104 * if the request wasn't completed, update state
2105 */
2106 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002107 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 bio->bi_idx += next_idx;
2109 bio_iovec(bio)->bv_offset += nr_bytes;
2110 bio_iovec(bio)->bv_len -= nr_bytes;
2111 }
2112
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002113 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002114 req->buffer = bio_data(req->bio);
2115
2116 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002117 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002118 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002119
Tejun Heo80a761f2009-07-03 17:48:17 +09002120 /* mixed attributes always follow the first bio */
2121 if (req->cmd_flags & REQ_MIXED_MERGE) {
2122 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2123 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2124 }
2125
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002126 /*
2127 * If total number of sectors is less than the first segment
2128 * size, something has gone terribly wrong.
2129 */
2130 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2131 printk(KERN_ERR "blk: request botched\n");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002132 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002133 }
2134
2135 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002137
Tejun Heo2e60e022009-04-23 11:05:18 +09002138 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139}
Tejun Heo2e60e022009-04-23 11:05:18 +09002140EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141
Tejun Heo2e60e022009-04-23 11:05:18 +09002142static bool blk_update_bidi_request(struct request *rq, int error,
2143 unsigned int nr_bytes,
2144 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002145{
Tejun Heo2e60e022009-04-23 11:05:18 +09002146 if (blk_update_request(rq, error, nr_bytes))
2147 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002148
Tejun Heo2e60e022009-04-23 11:05:18 +09002149 /* Bidi request must be completed as a whole */
2150 if (unlikely(blk_bidi_rq(rq)) &&
2151 blk_update_request(rq->next_rq, error, bidi_bytes))
2152 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002153
Jens Axboee2e1a142010-06-09 10:42:09 +02002154 if (blk_queue_add_random(rq->q))
2155 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002156
2157 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158}
2159
James Bottomley28018c22010-07-01 19:49:17 +09002160/**
2161 * blk_unprep_request - unprepare a request
2162 * @req: the request
2163 *
2164 * This function makes a request ready for complete resubmission (or
2165 * completion). It happens only after all error handling is complete,
2166 * so represents the appropriate moment to deallocate any resources
2167 * that were allocated to the request in the prep_rq_fn. The queue
2168 * lock is held when calling this.
2169 */
2170void blk_unprep_request(struct request *req)
2171{
2172 struct request_queue *q = req->q;
2173
2174 req->cmd_flags &= ~REQ_DONTPREP;
2175 if (q->unprep_rq_fn)
2176 q->unprep_rq_fn(q, req);
2177}
2178EXPORT_SYMBOL_GPL(blk_unprep_request);
2179
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180/*
2181 * queue lock must be held
2182 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002183static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002185 if (blk_rq_tagged(req))
2186 blk_queue_end_tag(req->q, req);
2187
James Bottomleyba396a62009-05-27 14:17:08 +02002188 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002190 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002191 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
Mike Andersone78042e2008-10-30 02:16:20 -07002193 blk_delete_timer(req);
2194
James Bottomley28018c22010-07-01 19:49:17 +09002195 if (req->cmd_flags & REQ_DONTPREP)
2196 blk_unprep_request(req);
2197
2198
Jens Axboebc58ba92009-01-23 10:54:44 +01002199 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002200
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002202 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002203 else {
2204 if (blk_bidi_rq(req))
2205 __blk_put_request(req->next_rq->q, req->next_rq);
2206
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 }
2209}
2210
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002211/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002212 * blk_end_bidi_request - Complete a bidi request
2213 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002214 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002215 * @nr_bytes: number of bytes to complete @rq
2216 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002217 *
2218 * Description:
2219 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002220 * Drivers that supports bidi can safely call this member for any
2221 * type of request, bidi or uni. In the later case @bidi_bytes is
2222 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002223 *
2224 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002225 * %false - we are done with this request
2226 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002227 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002228static bool blk_end_bidi_request(struct request *rq, int error,
2229 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002230{
2231 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002232 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002233
Tejun Heo2e60e022009-04-23 11:05:18 +09002234 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2235 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002236
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002237 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002238 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002239 spin_unlock_irqrestore(q->queue_lock, flags);
2240
Tejun Heo2e60e022009-04-23 11:05:18 +09002241 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002242}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002243
2244/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002245 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2246 * @rq: the request to complete
2247 * @error: %0 for success, < %0 for error
2248 * @nr_bytes: number of bytes to complete @rq
2249 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002250 *
2251 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002252 * Identical to blk_end_bidi_request() except that queue lock is
2253 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002254 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002255 * Return:
2256 * %false - we are done with this request
2257 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002258 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002259static bool __blk_end_bidi_request(struct request *rq, int error,
2260 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002261{
Tejun Heo2e60e022009-04-23 11:05:18 +09002262 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2263 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002264
Tejun Heo2e60e022009-04-23 11:05:18 +09002265 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002266
Tejun Heo2e60e022009-04-23 11:05:18 +09002267 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002268}
2269
2270/**
2271 * blk_end_request - Helper function for drivers to complete the request.
2272 * @rq: the request being processed
2273 * @error: %0 for success, < %0 for error
2274 * @nr_bytes: number of bytes to complete
2275 *
2276 * Description:
2277 * Ends I/O on a number of bytes attached to @rq.
2278 * If @rq has leftover, sets it up for the next range of segments.
2279 *
2280 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002281 * %false - we are done with this request
2282 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002283 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002284bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002285{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002286 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002287}
Jens Axboe56ad1742009-07-28 22:11:24 +02002288EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002289
2290/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002291 * blk_end_request_all - Helper function for drives to finish the request.
2292 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002293 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002294 *
2295 * Description:
2296 * Completely finish @rq.
2297 */
2298void blk_end_request_all(struct request *rq, int error)
2299{
2300 bool pending;
2301 unsigned int bidi_bytes = 0;
2302
2303 if (unlikely(blk_bidi_rq(rq)))
2304 bidi_bytes = blk_rq_bytes(rq->next_rq);
2305
2306 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2307 BUG_ON(pending);
2308}
Jens Axboe56ad1742009-07-28 22:11:24 +02002309EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002310
2311/**
2312 * blk_end_request_cur - Helper function to finish the current request chunk.
2313 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002314 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002315 *
2316 * Description:
2317 * Complete the current consecutively mapped chunk from @rq.
2318 *
2319 * Return:
2320 * %false - we are done with this request
2321 * %true - still buffers pending for this request
2322 */
2323bool blk_end_request_cur(struct request *rq, int error)
2324{
2325 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2326}
Jens Axboe56ad1742009-07-28 22:11:24 +02002327EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002328
2329/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002330 * blk_end_request_err - Finish a request till the next failure boundary.
2331 * @rq: the request to finish till the next failure boundary for
2332 * @error: must be negative errno
2333 *
2334 * Description:
2335 * Complete @rq till the next failure boundary.
2336 *
2337 * Return:
2338 * %false - we are done with this request
2339 * %true - still buffers pending for this request
2340 */
2341bool blk_end_request_err(struct request *rq, int error)
2342{
2343 WARN_ON(error >= 0);
2344 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2345}
2346EXPORT_SYMBOL_GPL(blk_end_request_err);
2347
2348/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002349 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002350 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002351 * @error: %0 for success, < %0 for error
2352 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002353 *
2354 * Description:
2355 * Must be called with queue lock held unlike blk_end_request().
2356 *
2357 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002358 * %false - we are done with this request
2359 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002360 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002361bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002362{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002363 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002364}
Jens Axboe56ad1742009-07-28 22:11:24 +02002365EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002366
2367/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002368 * __blk_end_request_all - Helper function for drives to finish the request.
2369 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002370 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002371 *
2372 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002373 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002374 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002375void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002376{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002377 bool pending;
2378 unsigned int bidi_bytes = 0;
2379
2380 if (unlikely(blk_bidi_rq(rq)))
2381 bidi_bytes = blk_rq_bytes(rq->next_rq);
2382
2383 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2384 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002385}
Jens Axboe56ad1742009-07-28 22:11:24 +02002386EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002387
2388/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002389 * __blk_end_request_cur - Helper function to finish the current request chunk.
2390 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002391 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002392 *
2393 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002394 * Complete the current consecutively mapped chunk from @rq. Must
2395 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002396 *
2397 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002398 * %false - we are done with this request
2399 * %true - still buffers pending for this request
2400 */
2401bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002402{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002403 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002404}
Jens Axboe56ad1742009-07-28 22:11:24 +02002405EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002406
Tejun Heo80a761f2009-07-03 17:48:17 +09002407/**
2408 * __blk_end_request_err - Finish a request till the next failure boundary.
2409 * @rq: the request to finish till the next failure boundary for
2410 * @error: must be negative errno
2411 *
2412 * Description:
2413 * Complete @rq till the next failure boundary. Must be called
2414 * with queue lock held.
2415 *
2416 * Return:
2417 * %false - we are done with this request
2418 * %true - still buffers pending for this request
2419 */
2420bool __blk_end_request_err(struct request *rq, int error)
2421{
2422 WARN_ON(error >= 0);
2423 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2424}
2425EXPORT_SYMBOL_GPL(__blk_end_request_err);
2426
Jens Axboe86db1e22008-01-29 14:53:40 +01002427void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2428 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002430 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002431 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432
David Woodhousefb2dce82008-08-05 18:01:53 +01002433 if (bio_has_data(bio)) {
2434 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002435 rq->buffer = bio_data(bio);
2436 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002437 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439
NeilBrown66846572007-08-16 13:31:28 +02002440 if (bio->bi_bdev)
2441 rq->rq_disk = bio->bi_bdev->bd_disk;
2442}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002444#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2445/**
2446 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2447 * @rq: the request to be flushed
2448 *
2449 * Description:
2450 * Flush all pages in @rq.
2451 */
2452void rq_flush_dcache_pages(struct request *rq)
2453{
2454 struct req_iterator iter;
2455 struct bio_vec *bvec;
2456
2457 rq_for_each_segment(bvec, rq, iter)
2458 flush_dcache_page(bvec->bv_page);
2459}
2460EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2461#endif
2462
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002463/**
2464 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2465 * @q : the queue of the device being checked
2466 *
2467 * Description:
2468 * Check if underlying low-level drivers of a device are busy.
2469 * If the drivers want to export their busy state, they must set own
2470 * exporting function using blk_queue_lld_busy() first.
2471 *
2472 * Basically, this function is used only by request stacking drivers
2473 * to stop dispatching requests to underlying devices when underlying
2474 * devices are busy. This behavior helps more I/O merging on the queue
2475 * of the request stacking driver and prevents I/O throughput regression
2476 * on burst I/O load.
2477 *
2478 * Return:
2479 * 0 - Not busy (The request stacking driver should dispatch request)
2480 * 1 - Busy (The request stacking driver should stop dispatching request)
2481 */
2482int blk_lld_busy(struct request_queue *q)
2483{
2484 if (q->lld_busy_fn)
2485 return q->lld_busy_fn(q);
2486
2487 return 0;
2488}
2489EXPORT_SYMBOL_GPL(blk_lld_busy);
2490
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002491/**
2492 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2493 * @rq: the clone request to be cleaned up
2494 *
2495 * Description:
2496 * Free all bios in @rq for a cloned request.
2497 */
2498void blk_rq_unprep_clone(struct request *rq)
2499{
2500 struct bio *bio;
2501
2502 while ((bio = rq->bio) != NULL) {
2503 rq->bio = bio->bi_next;
2504
2505 bio_put(bio);
2506 }
2507}
2508EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2509
2510/*
2511 * Copy attributes of the original request to the clone request.
2512 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2513 */
2514static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2515{
2516 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002517 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002518 dst->cmd_type = src->cmd_type;
2519 dst->__sector = blk_rq_pos(src);
2520 dst->__data_len = blk_rq_bytes(src);
2521 dst->nr_phys_segments = src->nr_phys_segments;
2522 dst->ioprio = src->ioprio;
2523 dst->extra_len = src->extra_len;
2524}
2525
2526/**
2527 * blk_rq_prep_clone - Helper function to setup clone request
2528 * @rq: the request to be setup
2529 * @rq_src: original request to be cloned
2530 * @bs: bio_set that bios for clone are allocated from
2531 * @gfp_mask: memory allocation mask for bio
2532 * @bio_ctr: setup function to be called for each clone bio.
2533 * Returns %0 for success, non %0 for failure.
2534 * @data: private data to be passed to @bio_ctr
2535 *
2536 * Description:
2537 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2538 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2539 * are not copied, and copying such parts is the caller's responsibility.
2540 * Also, pages which the original bios are pointing to are not copied
2541 * and the cloned bios just point same pages.
2542 * So cloned bios must be completed before original bios, which means
2543 * the caller must complete @rq before @rq_src.
2544 */
2545int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2546 struct bio_set *bs, gfp_t gfp_mask,
2547 int (*bio_ctr)(struct bio *, struct bio *, void *),
2548 void *data)
2549{
2550 struct bio *bio, *bio_src;
2551
2552 if (!bs)
2553 bs = fs_bio_set;
2554
2555 blk_rq_init(NULL, rq);
2556
2557 __rq_for_each_bio(bio_src, rq_src) {
2558 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2559 if (!bio)
2560 goto free_and_out;
2561
2562 __bio_clone(bio, bio_src);
2563
2564 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002565 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002566 goto free_and_out;
2567
2568 if (bio_ctr && bio_ctr(bio, bio_src, data))
2569 goto free_and_out;
2570
2571 if (rq->bio) {
2572 rq->biotail->bi_next = bio;
2573 rq->biotail = bio;
2574 } else
2575 rq->bio = rq->biotail = bio;
2576 }
2577
2578 __blk_rq_prep_clone(rq, rq_src);
2579
2580 return 0;
2581
2582free_and_out:
2583 if (bio)
2584 bio_free(bio, bs);
2585 blk_rq_unprep_clone(rq);
2586
2587 return -ENOMEM;
2588}
2589EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2590
Jens Axboe18887ad2008-07-28 13:08:45 +02002591int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592{
2593 return queue_work(kblockd_workqueue, work);
2594}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595EXPORT_SYMBOL(kblockd_schedule_work);
2596
Vivek Goyale43473b2010-09-15 17:06:35 -04002597int kblockd_schedule_delayed_work(struct request_queue *q,
2598 struct delayed_work *dwork, unsigned long delay)
2599{
2600 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2601}
2602EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2603
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604int __init blk_dev_init(void)
2605{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002606 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2607 sizeof(((struct request *)0)->cmd_flags));
2608
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 kblockd_workqueue = create_workqueue("kblockd");
2610 if (!kblockd_workqueue)
2611 panic("Failed to create kblockd\n");
2612
2613 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002614 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615
Jens Axboe8324aa92008-01-29 14:51:59 +01002616 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002617 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 return 0;
2620}