blob: 90f22cc30799cf8bf71e25c4cf68718c3fbf0cf3 [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>
Jens Axboe73c10102011-03-08 13:19:51 +010030#include <linux/list_sort.h>
Li Zefan55782132009-06-09 13:43:05 +080031
32#define CREATE_TRACE_POINTS
33#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Jens Axboe8324aa92008-01-29 14:51:59 +010035#include "blk.h"
36
Mike Snitzerd07335e2010-11-16 12:52:38 +010037EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020038EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Li Zefan55782132009-06-09 13:43:05 +080039EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010040
Jens Axboe165125e2007-07-24 09:28:11 +020041static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43/*
44 * For the allocated request tables
45 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010046static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48/*
49 * For queue allocation
50 */
Jens Axboe6728cb02008-01-31 13:03:55 +010051struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 * Controlling structure to kblockd
55 */
Jens Axboeff856ba2006-01-09 16:02:34 +010056static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Jens Axboe26b82562008-01-29 13:54:41 +010058static void drive_stat_acct(struct request *rq, int new_io)
59{
Jens Axboe28f13702008-05-07 10:15:46 +020060 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010061 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090062 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010063
Jens Axboec2553b52009-04-24 08:10:11 +020064 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010065 return;
66
Tejun Heo074a7ac2008-08-25 19:56:14 +090067 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +090068
Jerome Marchand09e099d2011-01-05 16:57:38 +010069 if (!new_io) {
70 part = rq->part;
Tejun Heo074a7ac2008-08-25 19:56:14 +090071 part_stat_inc(cpu, part, merges[rw]);
Jerome Marchand09e099d2011-01-05 16:57:38 +010072 } else {
73 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Jens Axboe6c23a962011-01-07 08:43:37 +010074 if (!hd_struct_try_get(part)) {
Jerome Marchand09e099d2011-01-05 16:57:38 +010075 /*
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
78 *
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
82 */
83 part = &rq->rq_disk->part0;
Jens Axboe6c23a962011-01-07 08:43:37 +010084 hd_struct_get(part);
Jerome Marchand09e099d2011-01-05 16:57:38 +010085 }
Tejun Heo074a7ac2008-08-25 19:56:14 +090086 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020087 part_inc_in_flight(part, rw);
Jerome Marchand09e099d2011-01-05 16:57:38 +010088 rq->part = part;
Jens Axboe26b82562008-01-29 13:54:41 +010089 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020090
Tejun Heo074a7ac2008-08-25 19:56:14 +090091 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010092}
93
Jens Axboe8324aa92008-01-29 14:51:59 +010094void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095{
96 int nr;
97
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
100 nr = q->nr_requests;
101 q->nr_congestion_on = nr;
102
103 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
104 if (nr < 1)
105 nr = 1;
106 q->nr_congestion_off = nr;
107}
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109/**
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
111 * @bdev: device
112 *
113 * Locates the passed device's request queue and returns the address of its
114 * backing_dev_info
115 *
116 * Will return NULL if the request queue cannot be located.
117 */
118struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
119{
120 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200121 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
123 if (q)
124 ret = &q->backing_dev_info;
125 return ret;
126}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127EXPORT_SYMBOL(blk_get_backing_dev_info);
128
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200129void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200131 memset(rq, 0, sizeof(*rq));
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700134 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200135 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100136 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900137 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200138 INIT_HLIST_NODE(&rq->hash);
139 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200140 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800141 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100142 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100143 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900144 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700145 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100146 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200148EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
NeilBrown5bb23a62007-09-27 12:46:13 +0200150static void req_bio_endio(struct request *rq, struct bio *bio,
151 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100152{
Tejun Heo143a87f2011-01-25 12:43:52 +0100153 if (error)
154 clear_bit(BIO_UPTODATE, &bio->bi_flags);
155 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
156 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100157
Tejun Heo143a87f2011-01-25 12:43:52 +0100158 if (unlikely(nbytes > bio->bi_size)) {
159 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
160 __func__, nbytes, bio->bi_size);
161 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +0200162 }
Tejun Heo143a87f2011-01-25 12:43:52 +0100163
164 if (unlikely(rq->cmd_flags & REQ_QUIET))
165 set_bit(BIO_QUIET, &bio->bi_flags);
166
167 bio->bi_size -= nbytes;
168 bio->bi_sector += (nbytes >> 9);
169
170 if (bio_integrity(bio))
171 bio_integrity_advance(bio, nbytes);
172
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
175 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100176}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178void blk_dump_rq_flags(struct request *rq, char *msg)
179{
180 int bit;
181
Jens Axboe6728cb02008-01-31 13:03:55 +0100182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
184 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Tejun Heo83096eb2009-05-07 22:24:39 +0900186 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq),
188 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900189 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900190 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200192 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100193 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200194 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 printk("%02x ", rq->cmd[bit]);
196 printk("\n");
197 }
198}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199EXPORT_SYMBOL(blk_dump_rq_flags);
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201/*
Jens Axboe7eaceac2011-03-10 08:52:07 +0100202 * Make sure that plugs that were pending when this function was entered,
203 * are now complete and requests pushed to the queue.
204*/
205static inline void queue_sync_plugs(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 /*
Jens Axboe7eaceac2011-03-10 08:52:07 +0100208 * If the current process is plugged and has barriers submitted,
209 * we will livelock if we don't unplug first.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 */
Jens Axboe7eaceac2011-03-10 08:52:07 +0100211 blk_flush_plug(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500214static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200215{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500216 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200217
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500218 q = container_of(work, struct request_queue, delay_work.work);
219 spin_lock_irq(q->queue_lock);
Jens Axboe4c63f562011-03-10 08:58:35 +0100220 __blk_run_queue(q, false);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500221 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
224/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500225 * blk_delay_queue - restart queueing after defined interval
226 * @q: The &struct request_queue in question
227 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 *
229 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500230 * Sometimes queueing needs to be postponed for a little while, to allow
231 * resources to come back. This function will make sure that queueing is
232 * restarted around the specified time.
233 */
234void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500236 schedule_delayed_work(&q->delay_work, msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500238EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240/**
241 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200242 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 *
244 * Description:
245 * blk_start_queue() will clear the stop flag on the queue, and call
246 * the request_fn for the queue if it was in a stopped state when
247 * entered. Also see blk_stop_queue(). Queue lock must be held.
248 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200249void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200251 WARN_ON(!irqs_disabled());
252
Nick Piggin75ad23b2008-04-29 14:48:33 +0200253 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Tejun Heo1654e742011-03-02 08:48:05 -0500254 __blk_run_queue(q, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256EXPORT_SYMBOL(blk_start_queue);
257
258/**
259 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200260 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 *
262 * Description:
263 * The Linux block layer assumes that a block driver will consume all
264 * entries on the request queue when the request_fn strategy is called.
265 * Often this will not happen, because of hardware limitations (queue
266 * depth settings). If a device driver gets a 'queue full' response,
267 * or if it simply chooses not to queue more I/O at one point, it can
268 * call this function to prevent the request_fn from being called until
269 * the driver has signalled it's ready to go again. This happens by calling
270 * blk_start_queue() to restart queue operations. Queue lock must be held.
271 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200272void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
Jens Axboead3d9d72011-03-25 16:58:59 +0100274 __cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200275 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277EXPORT_SYMBOL(blk_stop_queue);
278
279/**
280 * blk_sync_queue - cancel any pending callbacks on a queue
281 * @q: the queue
282 *
283 * Description:
284 * The block layer may perform asynchronous callback activity
285 * on a queue, such as calling the unplug function after a timeout.
286 * A block device may call blk_sync_queue to ensure that any
287 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200288 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 * that its ->make_request_fn will not re-add plugging prior to calling
290 * this function.
291 *
Vivek Goyalda527772011-03-02 19:05:33 -0500292 * This function does not cancel any asynchronous activity arising
293 * out of elevator or throttling code. That would require elevaotor_exit()
294 * and blk_throtl_exit() to be called with queue lock initialized.
295 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 */
297void blk_sync_queue(struct request_queue *q)
298{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100299 del_timer_sync(&q->timeout);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500300 cancel_delayed_work_sync(&q->delay_work);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100301 queue_sync_plugs(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302}
303EXPORT_SYMBOL(blk_sync_queue);
304
305/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200306 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 * @q: The queue to run
Tejun Heo1654e742011-03-02 08:48:05 -0500308 * @force_kblockd: Don't run @q->request_fn directly. Use kblockd.
Jens Axboe80a4b582008-10-14 09:51:06 +0200309 *
310 * Description:
311 * See @blk_run_queue. This variant must be called with the queue lock
312 * held and interrupts disabled.
313 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 */
Tejun Heo1654e742011-03-02 08:48:05 -0500315void __blk_run_queue(struct request_queue *q, bool force_kblockd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
Tejun Heoa538cd02009-04-23 11:05:17 +0900317 if (unlikely(blk_queue_stopped(q)))
318 return;
319
Jens Axboedac07ec2006-05-11 08:20:16 +0200320 /*
321 * Only recurse once to avoid overrunning the stack, let the unplug
322 * handling reinvoke the handler shortly if we already got there.
323 */
Tejun Heo1654e742011-03-02 08:48:05 -0500324 if (!force_kblockd && !queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
Tejun Heoa538cd02009-04-23 11:05:17 +0900325 q->request_fn(q);
326 queue_flag_clear(QUEUE_FLAG_REENTER, q);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100327 } else
328 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200329}
330EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200331
Nick Piggin75ad23b2008-04-29 14:48:33 +0200332/**
333 * blk_run_queue - run a single device queue
334 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200335 *
336 * Description:
337 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900338 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200339 */
340void blk_run_queue(struct request_queue *q)
341{
342 unsigned long flags;
343
344 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo1654e742011-03-02 08:48:05 -0500345 __blk_run_queue(q, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 spin_unlock_irqrestore(q->queue_lock, flags);
347}
348EXPORT_SYMBOL(blk_run_queue);
349
Jens Axboe165125e2007-07-24 09:28:11 +0200350void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500351{
352 kobject_put(&q->kobj);
353}
Al Viro483f4af2006-03-18 18:34:37 -0500354
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500355/*
356 * Note: If a driver supplied the queue lock, it should not zap that lock
357 * unexpectedly as some queue cleanup components like elevator_exit() and
358 * blk_throtl_exit() need queue lock.
359 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100360void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500361{
Jens Axboee3335de92008-09-18 09:22:54 -0700362 /*
363 * We know we have process context here, so we can be a little
364 * cautious and ensure that pending block actions on this device
365 * are done before moving on. Going into this function, we should
366 * not have processes doing IO to this device.
367 */
368 blk_sync_queue(q);
369
Matthew Garrett31373d02010-04-06 14:25:14 +0200370 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500371 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200372 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500373 mutex_unlock(&q->sysfs_lock);
374
375 if (q->elevator)
376 elevator_exit(q->elevator);
377
Vivek Goyalda527772011-03-02 19:05:33 -0500378 blk_throtl_exit(q);
379
Al Viro483f4af2006-03-18 18:34:37 -0500380 blk_put_queue(q);
381}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382EXPORT_SYMBOL(blk_cleanup_queue);
383
Jens Axboe165125e2007-07-24 09:28:11 +0200384static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385{
386 struct request_list *rl = &q->rq;
387
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400388 if (unlikely(rl->rq_pool))
389 return 0;
390
Jens Axboe1faa16d2009-04-06 14:48:01 +0200391 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
392 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200393 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200394 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
395 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
Christoph Lameter19460892005-06-23 00:08:19 -0700397 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
398 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
400 if (!rl->rq_pool)
401 return -ENOMEM;
402
403 return 0;
404}
405
Jens Axboe165125e2007-07-24 09:28:11 +0200406struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
Christoph Lameter19460892005-06-23 00:08:19 -0700408 return blk_alloc_queue_node(gfp_mask, -1);
409}
410EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Jens Axboe165125e2007-07-24 09:28:11 +0200412struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700413{
Jens Axboe165125e2007-07-24 09:28:11 +0200414 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700415 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700416
Jens Axboe8324aa92008-01-29 14:51:59 +0100417 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700418 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 if (!q)
420 return NULL;
421
Jens Axboe0989a022009-06-12 14:42:56 +0200422 q->backing_dev_info.ra_pages =
423 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
424 q->backing_dev_info.state = 0;
425 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200426 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200427
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700428 err = bdi_init(&q->backing_dev_info);
429 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100430 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700431 return NULL;
432 }
433
Vivek Goyale43473b2010-09-15 17:06:35 -0400434 if (blk_throtl_init(q)) {
435 kmem_cache_free(blk_requestq_cachep, q);
436 return NULL;
437 }
438
Matthew Garrett31373d02010-04-06 14:25:14 +0200439 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
440 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700441 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
442 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoae1b1532011-01-25 12:43:54 +0100443 INIT_LIST_HEAD(&q->flush_queue[0]);
444 INIT_LIST_HEAD(&q->flush_queue[1]);
445 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500446 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500447
Jens Axboe8324aa92008-01-29 14:51:59 +0100448 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
Al Viro483f4af2006-03-18 18:34:37 -0500450 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700451 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500452
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500453 /*
454 * By default initialize queue_lock to internal lock and driver can
455 * override it later if need be.
456 */
457 q->queue_lock = &q->__queue_lock;
458
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 return q;
460}
Christoph Lameter19460892005-06-23 00:08:19 -0700461EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463/**
464 * blk_init_queue - prepare a request queue for use with a block device
465 * @rfn: The function to be called to process requests that have been
466 * placed on the queue.
467 * @lock: Request queue spin lock
468 *
469 * Description:
470 * If a block device wishes to use the standard request handling procedures,
471 * which sorts requests and coalesces adjacent requests, then it must
472 * call blk_init_queue(). The function @rfn will be called when there
473 * are requests on the queue that need to be processed. If the device
474 * supports plugging, then @rfn may not be called immediately when requests
475 * are available on the queue, but may be called at some time later instead.
476 * Plugged queues are generally unplugged when a buffer belonging to one
477 * of the requests on the queue is needed, or due to memory pressure.
478 *
479 * @rfn is not required, or even expected, to remove all requests off the
480 * queue, but only as many as it can handle at a time. If it does leave
481 * requests on the queue, it is responsible for arranging that the requests
482 * get dealt with eventually.
483 *
484 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200485 * request queue; this lock will be taken also from interrupt context, so irq
486 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200488 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 * it didn't succeed.
490 *
491 * Note:
492 * blk_init_queue() must be paired with a blk_cleanup_queue() call
493 * when the block device is deactivated (such as at module unload).
494 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700495
Jens Axboe165125e2007-07-24 09:28:11 +0200496struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497{
Christoph Lameter19460892005-06-23 00:08:19 -0700498 return blk_init_queue_node(rfn, lock, -1);
499}
500EXPORT_SYMBOL(blk_init_queue);
501
Jens Axboe165125e2007-07-24 09:28:11 +0200502struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700503blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
504{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600505 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600507 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
508 if (!uninit_q)
509 return NULL;
510
511 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
512 if (!q)
513 blk_cleanup_queue(uninit_q);
514
515 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200516}
517EXPORT_SYMBOL(blk_init_queue_node);
518
519struct request_queue *
520blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
521 spinlock_t *lock)
522{
523 return blk_init_allocated_queue_node(q, rfn, lock, -1);
524}
525EXPORT_SYMBOL(blk_init_allocated_queue);
526
527struct request_queue *
528blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
529 spinlock_t *lock, int node_id)
530{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 if (!q)
532 return NULL;
533
Christoph Lameter19460892005-06-23 00:08:19 -0700534 q->node = node_id;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600535 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500536 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
538 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900540 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100541 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500542
543 /* Override internal queue lock with supplied lock pointer */
544 if (lock)
545 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
Jens Axboef3b144a2009-03-06 08:48:33 +0100547 /*
548 * This also sets hw/phys segments, boundary and size
549 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Alan Stern44ec9542007-02-20 11:01:57 -0500552 q->sg_reserved_size = INT_MAX;
553
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 /*
555 * all done
556 */
557 if (!elevator_init(q, NULL)) {
558 blk_queue_congestion_threshold(q);
559 return q;
560 }
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 return NULL;
563}
Mike Snitzer01effb02010-05-11 08:57:42 +0200564EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Jens Axboe165125e2007-07-24 09:28:11 +0200566int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700568 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500569 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 return 0;
571 }
572
573 return 1;
574}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
Jens Axboe165125e2007-07-24 09:28:11 +0200576static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
Jens Axboe73c10102011-03-08 13:19:51 +0100578 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
579
Jens Axboe4aff5e22006-08-10 08:44:47 +0200580 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200581 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 mempool_free(rq, q->rq.rq_pool);
583}
584
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200585static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200586blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587{
588 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
589
590 if (!rq)
591 return NULL;
592
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200593 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200594
Jerome Marchand42dad762009-04-22 14:01:49 +0200595 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
Tejun Heocb98fc82005-10-28 08:29:39 +0200597 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200598 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200599 mempool_free(rq, q->rq.rq_pool);
600 return NULL;
601 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200602 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200603 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
Tejun Heocb98fc82005-10-28 08:29:39 +0200605 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606}
607
608/*
609 * ioc_batching returns true if the ioc is a valid batching request and
610 * should be given priority access to a request.
611 */
Jens Axboe165125e2007-07-24 09:28:11 +0200612static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613{
614 if (!ioc)
615 return 0;
616
617 /*
618 * Make sure the process is able to allocate at least 1 request
619 * even if the batch times out, otherwise we could theoretically
620 * lose wakeups.
621 */
622 return ioc->nr_batch_requests == q->nr_batching ||
623 (ioc->nr_batch_requests > 0
624 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
625}
626
627/*
628 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
629 * will cause the process to be a "batcher" on all queues in the system. This
630 * is the behaviour we want though - once it gets a wakeup it should be given
631 * a nice run.
632 */
Jens Axboe165125e2007-07-24 09:28:11 +0200633static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
635 if (!ioc || ioc_batching(q, ioc))
636 return;
637
638 ioc->nr_batch_requests = q->nr_batching;
639 ioc->last_waited = jiffies;
640}
641
Jens Axboe1faa16d2009-04-06 14:48:01 +0200642static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
644 struct request_list *rl = &q->rq;
645
Jens Axboe1faa16d2009-04-06 14:48:01 +0200646 if (rl->count[sync] < queue_congestion_off_threshold(q))
647 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Jens Axboe1faa16d2009-04-06 14:48:01 +0200649 if (rl->count[sync] + 1 <= q->nr_requests) {
650 if (waitqueue_active(&rl->wait[sync]))
651 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Jens Axboe1faa16d2009-04-06 14:48:01 +0200653 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 }
655}
656
657/*
658 * A request has just been released. Account for it, update the full and
659 * congestion status, wake up any waiters. Called under q->queue_lock.
660 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200661static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
663 struct request_list *rl = &q->rq;
664
Jens Axboe1faa16d2009-04-06 14:48:01 +0200665 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200666 if (priv)
667 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Jens Axboe1faa16d2009-04-06 14:48:01 +0200669 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Jens Axboe1faa16d2009-04-06 14:48:01 +0200671 if (unlikely(rl->starved[sync ^ 1]))
672 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673}
674
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100676 * Determine if elevator data should be initialized when allocating the
677 * request associated with @bio.
678 */
679static bool blk_rq_should_init_elevator(struct bio *bio)
680{
681 if (!bio)
682 return true;
683
684 /*
685 * Flush requests do not use the elevator so skip initialization.
686 * This allows a request to share the flush and elevator data.
687 */
688 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
689 return false;
690
691 return true;
692}
693
694/*
Nick Piggind6344532005-06-28 20:45:14 -0700695 * Get a free request, queue_lock must be held.
696 * Returns NULL on failure, with queue_lock held.
697 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 */
Jens Axboe165125e2007-07-24 09:28:11 +0200699static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100700 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
702 struct request *rq = NULL;
703 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100704 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200705 const bool is_sync = rw_is_sync(rw_flags) != 0;
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100706 int may_queue, priv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Jens Axboe7749a8d2006-12-13 13:02:26 +0100708 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100709 if (may_queue == ELV_MQUEUE_NO)
710 goto rq_starved;
711
Jens Axboe1faa16d2009-04-06 14:48:01 +0200712 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
713 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200714 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100715 /*
716 * The queue will fill after this allocation, so set
717 * it as full, and mark this process as "batching".
718 * This process will be allowed to complete a batch of
719 * requests, others will be blocked.
720 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200721 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100722 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200723 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100724 } else {
725 if (may_queue != ELV_MQUEUE_MUST
726 && !ioc_batching(q, ioc)) {
727 /*
728 * The queue is full and the allocating
729 * process is not a "batcher", and not
730 * exempted by the IO scheduler
731 */
732 goto out;
733 }
734 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200736 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 }
738
Jens Axboe082cf692005-06-28 16:35:11 +0200739 /*
740 * Only allow batching queuers to allocate up to 50% over the defined
741 * limit of requests, otherwise we could have thousands of requests
742 * allocated with any setting of ->nr_requests
743 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200744 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200745 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100746
Jens Axboe1faa16d2009-04-06 14:48:01 +0200747 rl->count[is_sync]++;
748 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200749
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100750 if (blk_rq_should_init_elevator(bio)) {
751 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
752 if (priv)
753 rl->elvpriv++;
754 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200755
Jens Axboef253b862010-10-24 22:06:02 +0200756 if (blk_queue_io_stat(q))
757 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 spin_unlock_irq(q->queue_lock);
759
Jens Axboe7749a8d2006-12-13 13:02:26 +0100760 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100761 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 /*
763 * Allocation failed presumably due to memory. Undo anything
764 * we might have messed up.
765 *
766 * Allocating task should really be put onto the front of the
767 * wait queue, but this is pretty rare.
768 */
769 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200770 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
772 /*
773 * in the very unlikely event that allocation failed and no
774 * requests for this direction was pending, mark us starved
775 * so that freeing of a request in the other direction will
776 * notice us. another possible fix would be to split the
777 * rq mempool into READ and WRITE
778 */
779rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200780 if (unlikely(rl->count[is_sync] == 0))
781 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 goto out;
784 }
785
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100786 /*
787 * ioc may be NULL here, and ioc_batching will be false. That's
788 * OK, if the queue is under the request limit then requests need
789 * not count toward the nr_batch_requests limit. There will always
790 * be some limit enforced by BLK_BATCH_TIME.
791 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if (ioc_batching(q, ioc))
793 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100794
Jens Axboe1faa16d2009-04-06 14:48:01 +0200795 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 return rq;
798}
799
800/*
Jens Axboe7eaceac2011-03-10 08:52:07 +0100801 * No available requests for this queue, wait for some requests to become
802 * available.
Nick Piggind6344532005-06-28 20:45:14 -0700803 *
804 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 */
Jens Axboe165125e2007-07-24 09:28:11 +0200806static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200807 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200809 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 struct request *rq;
811
Jens Axboe7749a8d2006-12-13 13:02:26 +0100812 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700813 while (!rq) {
814 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200815 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 struct request_list *rl = &q->rq;
817
Jens Axboe1faa16d2009-04-06 14:48:01 +0200818 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 TASK_UNINTERRUPTIBLE);
820
Jens Axboe1faa16d2009-04-06 14:48:01 +0200821 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200823 spin_unlock_irq(q->queue_lock);
824 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200826 /*
827 * After sleeping, we become a "batching" process and
828 * will be able to allocate at least one request, and
829 * up to a big batch of them for a small period time.
830 * See ioc_batching, ioc_set_batching
831 */
832 ioc = current_io_context(GFP_NOIO, q->node);
833 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100834
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200835 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200836 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200837
838 rq = get_request(q, rw_flags, bio, GFP_NOIO);
839 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841 return rq;
842}
843
Jens Axboe165125e2007-07-24 09:28:11 +0200844struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845{
846 struct request *rq;
847
848 BUG_ON(rw != READ && rw != WRITE);
849
Nick Piggind6344532005-06-28 20:45:14 -0700850 spin_lock_irq(q->queue_lock);
851 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200852 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700853 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200854 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700855 if (!rq)
856 spin_unlock_irq(q->queue_lock);
857 }
858 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
860 return rq;
861}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862EXPORT_SYMBOL(blk_get_request);
863
864/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300865 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700866 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300867 * @bio: The bio describing the memory mappings that will be submitted for IO.
868 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700869 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200870 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300871 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
872 * type commands. Where the struct request needs to be farther initialized by
873 * the caller. It is passed a &struct bio, which describes the memory info of
874 * the I/O transfer.
875 *
876 * The caller of blk_make_request must make sure that bi_io_vec
877 * are set to describe the memory buffers. That bio_data_dir() will return
878 * the needed direction of the request. (And all bio's in the passed bio-chain
879 * are properly set accordingly)
880 *
881 * If called under none-sleepable conditions, mapped bio buffers must not
882 * need bouncing, by calling the appropriate masked or flagged allocator,
883 * suitable for the target device. Otherwise the call to blk_queue_bounce will
884 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200885 *
886 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
887 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
888 * anything but the first bio in the chain. Otherwise you risk waiting for IO
889 * completion of a bio that hasn't been submitted yet, thus resulting in a
890 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
891 * of bio_alloc(), as that avoids the mempool deadlock.
892 * If possible a big IO should be split into smaller parts when allocation
893 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200894 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300895struct request *blk_make_request(struct request_queue *q, struct bio *bio,
896 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200897{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300898 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
899
900 if (unlikely(!rq))
901 return ERR_PTR(-ENOMEM);
902
903 for_each_bio(bio) {
904 struct bio *bounce_bio = bio;
905 int ret;
906
907 blk_queue_bounce(q, &bounce_bio);
908 ret = blk_rq_append_bio(q, rq, bounce_bio);
909 if (unlikely(ret)) {
910 blk_put_request(rq);
911 return ERR_PTR(ret);
912 }
913 }
914
915 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200916}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300917EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200918
919/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 * blk_requeue_request - put a request back on queue
921 * @q: request queue where request should be inserted
922 * @rq: request to be inserted
923 *
924 * Description:
925 * Drivers often keep queueing requests until the hardware cannot accept
926 * more, when that condition happens we need to put the request back
927 * on the queue. Must be called with queue lock held.
928 */
Jens Axboe165125e2007-07-24 09:28:11 +0200929void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700931 blk_delete_timer(rq);
932 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100933 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100934
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 if (blk_rq_tagged(rq))
936 blk_queue_end_tag(q, rq);
937
James Bottomleyba396a62009-05-27 14:17:08 +0200938 BUG_ON(blk_queued_rq(rq));
939
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 elv_requeue_request(q, rq);
941}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942EXPORT_SYMBOL(blk_requeue_request);
943
Jens Axboe73c10102011-03-08 13:19:51 +0100944static void add_acct_request(struct request_queue *q, struct request *rq,
945 int where)
946{
947 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100948 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +0100949}
950
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200952 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 * @q: request queue where request should be inserted
954 * @rq: request to be inserted
955 * @at_head: insert request at head or tail of queue
956 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 *
958 * Description:
959 * Many block devices need to execute commands asynchronously, so they don't
960 * block the whole kernel from preemption during request execution. This is
961 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200962 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
963 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 *
965 * We have the option of inserting the head or the tail of the queue.
966 * Typically we use the tail for new ioctls and so forth. We use the head
967 * of the queue for things like a QUEUE_FULL message from a device, or a
968 * host that is unable to accept a particular command.
969 */
Jens Axboe165125e2007-07-24 09:28:11 +0200970void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500971 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972{
Tejun Heo 867d1192005-04-24 02:06:05 -0500973 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 unsigned long flags;
975
976 /*
977 * tell I/O scheduler that this isn't a regular read/write (ie it
978 * must not attempt merges on this) and that it acts as a soft
979 * barrier
980 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200981 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
983 rq->special = data;
984
985 spin_lock_irqsave(q->queue_lock, flags);
986
987 /*
988 * If command is tagged, release the tag
989 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500990 if (blk_rq_tagged(rq))
991 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Jens Axboe73c10102011-03-08 13:19:51 +0100993 add_acct_request(q, rq, where);
Tejun Heo1654e742011-03-02 08:48:05 -0500994 __blk_run_queue(q, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 spin_unlock_irqrestore(q->queue_lock, flags);
996}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997EXPORT_SYMBOL(blk_insert_request);
998
Tejun Heo074a7ac2008-08-25 19:56:14 +0900999static void part_round_stats_single(int cpu, struct hd_struct *part,
1000 unsigned long now)
1001{
1002 if (now == part->stamp)
1003 return;
1004
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001005 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001006 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001007 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001008 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1009 }
1010 part->stamp = now;
1011}
1012
1013/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001014 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1015 * @cpu: cpu number for stats access
1016 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 *
1018 * The average IO queue length and utilisation statistics are maintained
1019 * by observing the current state of the queue length and the amount of
1020 * time it has been in this state for.
1021 *
1022 * Normally, that accounting is done on IO completion, but that can result
1023 * in more than a second's worth of IO being accounted for within any one
1024 * second, leading to >100% utilisation. To deal with that, we call this
1025 * function to do a round-off before returning the results when reading
1026 * /proc/diskstats. This accounts immediately for all queue usage up to
1027 * the current jiffies and restarts the counters again.
1028 */
Tejun Heoc9959052008-08-25 19:47:21 +09001029void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001030{
1031 unsigned long now = jiffies;
1032
Tejun Heo074a7ac2008-08-25 19:56:14 +09001033 if (part->partno)
1034 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1035 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001036}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001037EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039/*
1040 * queue lock must be held
1041 */
Jens Axboe165125e2007-07-24 09:28:11 +02001042void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 if (unlikely(!q))
1045 return;
1046 if (unlikely(--req->ref_count))
1047 return;
1048
Tejun Heo8922e162005-10-20 16:23:44 +02001049 elv_completed_request(q, req);
1050
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001051 /* this is a bio leak */
1052 WARN_ON(req->bio != NULL);
1053
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 /*
1055 * Request may not have originated from ll_rw_blk. if not,
1056 * it didn't come out of our reserved rq pools
1057 */
Jens Axboe49171e52006-08-10 08:59:11 +02001058 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001059 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001060 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001063 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
1065 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001066 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 }
1068}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001069EXPORT_SYMBOL_GPL(__blk_put_request);
1070
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071void blk_put_request(struct request *req)
1072{
Tejun Heo8922e162005-10-20 16:23:44 +02001073 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001074 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001076 spin_lock_irqsave(q->queue_lock, flags);
1077 __blk_put_request(q, req);
1078 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080EXPORT_SYMBOL(blk_put_request);
1081
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001082/**
1083 * blk_add_request_payload - add a payload to a request
1084 * @rq: request to update
1085 * @page: page backing the payload
1086 * @len: length of the payload.
1087 *
1088 * This allows to later add a payload to an already submitted request by
1089 * a block driver. The driver needs to take care of freeing the payload
1090 * itself.
1091 *
1092 * Note that this is a quite horrible hack and nothing but handling of
1093 * discard requests should ever use it.
1094 */
1095void blk_add_request_payload(struct request *rq, struct page *page,
1096 unsigned int len)
1097{
1098 struct bio *bio = rq->bio;
1099
1100 bio->bi_io_vec->bv_page = page;
1101 bio->bi_io_vec->bv_offset = 0;
1102 bio->bi_io_vec->bv_len = len;
1103
1104 bio->bi_size = len;
1105 bio->bi_vcnt = 1;
1106 bio->bi_phys_segments = 1;
1107
1108 rq->__data_len = rq->resid_len = len;
1109 rq->nr_phys_segments = 1;
1110 rq->buffer = bio_data(bio);
1111}
1112EXPORT_SYMBOL_GPL(blk_add_request_payload);
1113
Jens Axboe73c10102011-03-08 13:19:51 +01001114static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1115 struct bio *bio)
1116{
1117 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1118
1119 /*
1120 * Debug stuff, kill later
1121 */
1122 if (!rq_mergeable(req)) {
1123 blk_dump_rq_flags(req, "back");
1124 return false;
1125 }
1126
1127 if (!ll_back_merge_fn(q, req, bio))
1128 return false;
1129
1130 trace_block_bio_backmerge(q, bio);
1131
1132 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1133 blk_rq_set_mixed_merge(req);
1134
1135 req->biotail->bi_next = bio;
1136 req->biotail = bio;
1137 req->__data_len += bio->bi_size;
1138 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1139
1140 drive_stat_acct(req, 0);
1141 return true;
1142}
1143
1144static bool bio_attempt_front_merge(struct request_queue *q,
1145 struct request *req, struct bio *bio)
1146{
1147 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1148 sector_t sector;
1149
1150 /*
1151 * Debug stuff, kill later
1152 */
1153 if (!rq_mergeable(req)) {
1154 blk_dump_rq_flags(req, "front");
1155 return false;
1156 }
1157
1158 if (!ll_front_merge_fn(q, req, bio))
1159 return false;
1160
1161 trace_block_bio_frontmerge(q, bio);
1162
1163 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1164 blk_rq_set_mixed_merge(req);
1165
1166 sector = bio->bi_sector;
1167
1168 bio->bi_next = req->bio;
1169 req->bio = bio;
1170
1171 /*
1172 * may not be valid. if the low level driver said
1173 * it didn't need a bounce buffer then it better
1174 * not touch req->buffer either...
1175 */
1176 req->buffer = bio_data(bio);
1177 req->__sector = bio->bi_sector;
1178 req->__data_len += bio->bi_size;
1179 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1180
1181 drive_stat_acct(req, 0);
1182 return true;
1183}
1184
1185/*
1186 * Attempts to merge with the plugged list in the current process. Returns
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001187 * true if merge was successful, otherwise false.
Jens Axboe73c10102011-03-08 13:19:51 +01001188 */
1189static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1190 struct bio *bio)
1191{
1192 struct blk_plug *plug;
1193 struct request *rq;
1194 bool ret = false;
1195
1196 plug = tsk->plug;
1197 if (!plug)
1198 goto out;
1199
1200 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1201 int el_ret;
1202
1203 if (rq->q != q)
1204 continue;
1205
1206 el_ret = elv_try_merge(rq, bio);
1207 if (el_ret == ELEVATOR_BACK_MERGE) {
1208 ret = bio_attempt_back_merge(q, rq, bio);
1209 if (ret)
1210 break;
1211 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1212 ret = bio_attempt_front_merge(q, rq, bio);
1213 if (ret)
1214 break;
1215 }
1216 }
1217out:
1218 return ret;
1219}
1220
Jens Axboe86db1e22008-01-29 14:53:40 +01001221void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001222{
Jens Axboec7c22e42008-09-13 20:26:01 +02001223 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001224 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001225
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001226 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1227 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001228 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001229
Tejun Heo52d9e672006-01-06 09:49:58 +01001230 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001231 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001232 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001233 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001234}
1235
Jens Axboe165125e2007-07-24 09:28:11 +02001236static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001238 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001239 struct blk_plug *plug;
1240 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1241 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 /*
1244 * low level driver can indicate that it wants pages above a
1245 * certain limit bounced to low memory (ie for highmem, or even
1246 * ISA dma in theory)
1247 */
1248 blk_queue_bounce(q, &bio);
1249
Tejun Heo4fed9472010-09-03 11:56:17 +02001250 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001251 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001252 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001253 goto get_rq;
1254 }
1255
Jens Axboe73c10102011-03-08 13:19:51 +01001256 /*
1257 * Check if we can merge with the plugged list before grabbing
1258 * any locks.
1259 */
1260 if (attempt_plug_merge(current, q, bio))
1261 goto out;
1262
1263 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
1265 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001266 if (el_ret == ELEVATOR_BACK_MERGE) {
1267 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1268 if (bio_attempt_back_merge(q, req, bio)) {
1269 if (!attempt_back_merge(q, req))
1270 elv_merged_request(q, req, el_ret);
1271 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001272 }
Jens Axboe73c10102011-03-08 13:19:51 +01001273 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1274 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1275 if (bio_attempt_front_merge(q, req, bio)) {
1276 if (!attempt_front_merge(q, req))
1277 elv_merged_request(q, req, el_ret);
1278 goto out_unlock;
1279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 }
1281
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001283 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001284 * This sync check and mask will be re-done in init_request_from_bio(),
1285 * but we need to set it earlier to expose the sync flag to the
1286 * rq allocator and io schedulers.
1287 */
1288 rw_flags = bio_data_dir(bio);
1289 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001290 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001291
1292 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001293 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001294 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001295 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001296 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001297
Nick Piggin450991b2005-06-28 20:45:13 -07001298 /*
1299 * After dropping the lock and possibly sleeping here, our request
1300 * may now be mergeable after it had proven unmergeable (above).
1301 * We don't worry about that case for efficiency. It won't happen
1302 * often, and the elevators are able to handle it.
1303 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001304 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
Jens Axboec7c22e42008-09-13 20:26:01 +02001306 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
Jens Axboe73c10102011-03-08 13:19:51 +01001307 bio_flagged(bio, BIO_CPU_AFFINE)) {
1308 req->cpu = blk_cpu_to_group(get_cpu());
1309 put_cpu();
1310 }
Tejun Heodd831002010-09-03 11:56:16 +02001311
Jens Axboe73c10102011-03-08 13:19:51 +01001312 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001313 if (plug) {
Jens Axboe73c10102011-03-08 13:19:51 +01001314 if (!plug->should_sort && !list_empty(&plug->list)) {
1315 struct request *__rq;
1316
1317 __rq = list_entry_rq(plug->list.prev);
1318 if (__rq->q != q)
1319 plug->should_sort = 1;
1320 }
1321 /*
1322 * Debug flag, kill later
1323 */
1324 req->cmd_flags |= REQ_ON_PLUG;
1325 list_add_tail(&req->queuelist, &plug->list);
1326 drive_stat_acct(req, 1);
1327 } else {
1328 spin_lock_irq(q->queue_lock);
1329 add_acct_request(q, req, where);
Jens Axboe4c63f562011-03-10 08:58:35 +01001330 __blk_run_queue(q, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001331out_unlock:
1332 spin_unlock_irq(q->queue_lock);
1333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336}
1337
1338/*
1339 * If bio->bi_dev is a partition, remap the location
1340 */
1341static inline void blk_partition_remap(struct bio *bio)
1342{
1343 struct block_device *bdev = bio->bi_bdev;
1344
Jens Axboebf2de6f2007-09-27 13:01:25 +02001345 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 bio->bi_sector += p->start_sect;
1349 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001350
Mike Snitzerd07335e2010-11-16 12:52:38 +01001351 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1352 bdev->bd_dev,
1353 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 }
1355}
1356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357static void handle_bad_sector(struct bio *bio)
1358{
1359 char b[BDEVNAME_SIZE];
1360
1361 printk(KERN_INFO "attempt to access beyond end of device\n");
1362 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1363 bdevname(bio->bi_bdev, b),
1364 bio->bi_rw,
1365 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001366 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
1368 set_bit(BIO_EOF, &bio->bi_flags);
1369}
1370
Akinobu Mitac17bb492006-12-08 02:39:46 -08001371#ifdef CONFIG_FAIL_MAKE_REQUEST
1372
1373static DECLARE_FAULT_ATTR(fail_make_request);
1374
1375static int __init setup_fail_make_request(char *str)
1376{
1377 return setup_fault_attr(&fail_make_request, str);
1378}
1379__setup("fail_make_request=", setup_fail_make_request);
1380
1381static int should_fail_request(struct bio *bio)
1382{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001383 struct hd_struct *part = bio->bi_bdev->bd_part;
1384
1385 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001386 return should_fail(&fail_make_request, bio->bi_size);
1387
1388 return 0;
1389}
1390
1391static int __init fail_make_request_debugfs(void)
1392{
1393 return init_fault_attr_dentries(&fail_make_request,
1394 "fail_make_request");
1395}
1396
1397late_initcall(fail_make_request_debugfs);
1398
1399#else /* CONFIG_FAIL_MAKE_REQUEST */
1400
1401static inline int should_fail_request(struct bio *bio)
1402{
1403 return 0;
1404}
1405
1406#endif /* CONFIG_FAIL_MAKE_REQUEST */
1407
Jens Axboec07e2b42007-07-18 13:27:58 +02001408/*
1409 * Check whether this bio extends beyond the end of the device.
1410 */
1411static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1412{
1413 sector_t maxsector;
1414
1415 if (!nr_sectors)
1416 return 0;
1417
1418 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001419 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001420 if (maxsector) {
1421 sector_t sector = bio->bi_sector;
1422
1423 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1424 /*
1425 * This may well happen - the kernel calls bread()
1426 * without checking the size of the device, e.g., when
1427 * mounting a device.
1428 */
1429 handle_bad_sector(bio);
1430 return 1;
1431 }
1432 }
1433
1434 return 0;
1435}
1436
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001438 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 * @bio: The bio describing the location in memory and on the device.
1440 *
1441 * generic_make_request() is used to make I/O requests of block
1442 * devices. It is passed a &struct bio, which describes the I/O that needs
1443 * to be done.
1444 *
1445 * generic_make_request() does not return any status. The
1446 * success/failure status of the request, along with notification of
1447 * completion, is delivered asynchronously through the bio->bi_end_io
1448 * function described (one day) else where.
1449 *
1450 * The caller of generic_make_request must make sure that bi_io_vec
1451 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1452 * set to describe the device address, and the
1453 * bi_end_io and optionally bi_private are set to describe how
1454 * completion notification should be signaled.
1455 *
1456 * generic_make_request and the drivers it calls may use bi_next if this
1457 * bio happens to be merged with someone else, and may change bi_dev and
1458 * bi_sector for remaps as it sees fit. So the values of these fields
1459 * should NOT be depended on after the call to generic_make_request.
1460 */
Neil Brownd89d8792007-05-01 09:53:42 +02001461static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462{
Jens Axboe165125e2007-07-24 09:28:11 +02001463 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001464 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001466 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001467 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
1469 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
Jens Axboec07e2b42007-07-18 13:27:58 +02001471 if (bio_check_eod(bio, nr_sectors))
1472 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
1474 /*
1475 * Resolve the mapping until finished. (drivers are
1476 * still free to implement/resolve their own stacking
1477 * by explicitly returning 0)
1478 *
1479 * NOTE: we don't repeat the blk_size check for each new device.
1480 * Stacking drivers are expected to know what they are doing.
1481 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001482 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001483 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 do {
1485 char b[BDEVNAME_SIZE];
1486
1487 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001488 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 printk(KERN_ERR
1490 "generic_make_request: Trying to access "
1491 "nonexistent block-device %s (%Lu)\n",
1492 bdevname(bio->bi_bdev, b),
1493 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001494 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 }
1496
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001497 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
Christoph Hellwig67efc922009-09-30 13:54:20 +02001498 nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001499 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001500 bdevname(bio->bi_bdev, b),
1501 bio_sectors(bio),
1502 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 goto end_io;
1504 }
1505
Nick Pigginfde6ad22005-06-23 00:08:53 -07001506 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 goto end_io;
1508
Akinobu Mitac17bb492006-12-08 02:39:46 -08001509 if (should_fail_request(bio))
1510 goto end_io;
1511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 /*
1513 * If this device has partitions, remap block n
1514 * of partition p to block n+start(p) of the disk.
1515 */
1516 blk_partition_remap(bio);
1517
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001518 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1519 goto end_io;
1520
NeilBrown5ddfe962006-10-30 22:07:21 -08001521 if (old_sector != -1)
Mike Snitzerd07335e2010-11-16 12:52:38 +01001522 trace_block_bio_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001523
NeilBrown5ddfe962006-10-30 22:07:21 -08001524 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001525 old_dev = bio->bi_bdev->bd_dev;
1526
Jens Axboec07e2b42007-07-18 13:27:58 +02001527 if (bio_check_eod(bio, nr_sectors))
1528 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001529
Tejun Heo1e879012010-09-03 11:56:17 +02001530 /*
1531 * Filter flush bio's early so that make_request based
1532 * drivers without flush support don't have to worry
1533 * about them.
1534 */
1535 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1536 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1537 if (!nr_sectors) {
1538 err = 0;
1539 goto end_io;
1540 }
1541 }
1542
Adrian Hunter8d57a982010-08-11 14:17:49 -07001543 if ((bio->bi_rw & REQ_DISCARD) &&
1544 (!blk_queue_discard(q) ||
1545 ((bio->bi_rw & REQ_SECURE) &&
1546 !blk_queue_secdiscard(q)))) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001547 err = -EOPNOTSUPP;
1548 goto end_io;
1549 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001550
Vivek Goyale43473b2010-09-15 17:06:35 -04001551 blk_throtl_bio(q, &bio);
1552
1553 /*
1554 * If bio = NULL, bio has been throttled and will be submitted
1555 * later.
1556 */
1557 if (!bio)
1558 break;
1559
Minchan Kim01edede2009-09-08 21:56:38 +02001560 trace_block_bio_queue(q, bio);
1561
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 ret = q->make_request_fn(q, bio);
1563 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001564
1565 return;
1566
1567end_io:
1568 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569}
1570
Neil Brownd89d8792007-05-01 09:53:42 +02001571/*
1572 * We only want one ->make_request_fn to be active at a time,
1573 * else stack usage with stacked devices could be a problem.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001574 * So use current->bio_list to keep a list of requests
Neil Brownd89d8792007-05-01 09:53:42 +02001575 * submited by a make_request_fn function.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001576 * current->bio_list is also used as a flag to say if
Neil Brownd89d8792007-05-01 09:53:42 +02001577 * generic_make_request is currently active in this task or not.
1578 * If it is NULL, then no make_request is active. If it is non-NULL,
1579 * then a make_request is active, and new requests should be added
1580 * at the tail
1581 */
1582void generic_make_request(struct bio *bio)
1583{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001584 struct bio_list bio_list_on_stack;
1585
1586 if (current->bio_list) {
Neil Brownd89d8792007-05-01 09:53:42 +02001587 /* make_request is active */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001588 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001589 return;
1590 }
1591 /* following loop may be a bit non-obvious, and so deserves some
1592 * explanation.
1593 * Before entering the loop, bio->bi_next is NULL (as all callers
1594 * ensure that) so we have a list with a single bio.
1595 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001596 * we assign bio_list to a pointer to the bio_list_on_stack,
1597 * thus initialising the bio_list of new bios to be
Neil Brownd89d8792007-05-01 09:53:42 +02001598 * added. __generic_make_request may indeed add some more bios
1599 * through a recursive call to generic_make_request. If it
1600 * did, we find a non-NULL value in bio_list and re-enter the loop
1601 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001602 * of the top of the list (no pretending) and so remove it from
1603 * bio_list, and call into __generic_make_request again.
Neil Brownd89d8792007-05-01 09:53:42 +02001604 *
1605 * The loop was structured like this to make only one call to
1606 * __generic_make_request (which is important as it is large and
1607 * inlined) and to keep the structure simple.
1608 */
1609 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001610 bio_list_init(&bio_list_on_stack);
1611 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001612 do {
Neil Brownd89d8792007-05-01 09:53:42 +02001613 __generic_make_request(bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001614 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001615 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001616 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001617}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618EXPORT_SYMBOL(generic_make_request);
1619
1620/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001621 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1623 * @bio: The &struct bio which describes the I/O
1624 *
1625 * submit_bio() is very similar in purpose to generic_make_request(), and
1626 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001627 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 *
1629 */
1630void submit_bio(int rw, struct bio *bio)
1631{
1632 int count = bio_sectors(bio);
1633
Jens Axboe22e2c502005-06-27 10:55:12 +02001634 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
Jens Axboebf2de6f2007-09-27 13:01:25 +02001636 /*
1637 * If it's a regular read/write or a barrier with data attached,
1638 * go through the normal accounting stuff before submission.
1639 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001640 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001641 if (rw & WRITE) {
1642 count_vm_events(PGPGOUT, count);
1643 } else {
1644 task_io_account_read(bio->bi_size);
1645 count_vm_events(PGPGIN, count);
1646 }
1647
1648 if (unlikely(block_dump)) {
1649 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001650 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001651 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001652 (rw & WRITE) ? "WRITE" : "READ",
1653 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001654 bdevname(bio->bi_bdev, b),
1655 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 }
1658
1659 generic_make_request(bio);
1660}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661EXPORT_SYMBOL(submit_bio);
1662
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001663/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001664 * blk_rq_check_limits - Helper function to check a request for the queue limit
1665 * @q: the queue
1666 * @rq: the request being checked
1667 *
1668 * Description:
1669 * @rq may have been made based on weaker limitations of upper-level queues
1670 * in request stacking drivers, and it may violate the limitation of @q.
1671 * Since the block layer and the underlying device driver trust @rq
1672 * after it is inserted to @q, it should be checked against @q before
1673 * the insertion using this generic function.
1674 *
1675 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001676 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001677 * Request stacking drivers like request-based dm may change the queue
1678 * limits while requests are in the queue (e.g. dm's table swapping).
1679 * Such request stacking drivers should check those requests agaist
1680 * the new queue limits again when they dispatch those requests,
1681 * although such checkings are also done against the old queue limits
1682 * when submitting requests.
1683 */
1684int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1685{
ike Snitzer33839772010-08-08 12:11:33 -04001686 if (rq->cmd_flags & REQ_DISCARD)
1687 return 0;
1688
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001689 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1690 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001691 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1692 return -EIO;
1693 }
1694
1695 /*
1696 * queue's settings related to segment counting like q->bounce_pfn
1697 * may differ from that of other stacking queues.
1698 * Recalculate it to check the request correctly on this queue's
1699 * limitation.
1700 */
1701 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001702 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001703 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1704 return -EIO;
1705 }
1706
1707 return 0;
1708}
1709EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1710
1711/**
1712 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1713 * @q: the queue to submit the request
1714 * @rq: the request being queued
1715 */
1716int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1717{
1718 unsigned long flags;
1719
1720 if (blk_rq_check_limits(q, rq))
1721 return -EIO;
1722
1723#ifdef CONFIG_FAIL_MAKE_REQUEST
1724 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1725 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1726 return -EIO;
1727#endif
1728
1729 spin_lock_irqsave(q->queue_lock, flags);
1730
1731 /*
1732 * Submitting request must be dequeued before calling this function
1733 * because it will be linked to another request_queue
1734 */
1735 BUG_ON(blk_queued_rq(rq));
1736
Jens Axboe73c10102011-03-08 13:19:51 +01001737 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001738 spin_unlock_irqrestore(q->queue_lock, flags);
1739
1740 return 0;
1741}
1742EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1743
Tejun Heo80a761f2009-07-03 17:48:17 +09001744/**
1745 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1746 * @rq: request to examine
1747 *
1748 * Description:
1749 * A request could be merge of IOs which require different failure
1750 * handling. This function determines the number of bytes which
1751 * can be failed from the beginning of the request without
1752 * crossing into area which need to be retried further.
1753 *
1754 * Return:
1755 * The number of bytes to fail.
1756 *
1757 * Context:
1758 * queue_lock must be held.
1759 */
1760unsigned int blk_rq_err_bytes(const struct request *rq)
1761{
1762 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1763 unsigned int bytes = 0;
1764 struct bio *bio;
1765
1766 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1767 return blk_rq_bytes(rq);
1768
1769 /*
1770 * Currently the only 'mixing' which can happen is between
1771 * different fastfail types. We can safely fail portions
1772 * which have all the failfast bits that the first one has -
1773 * the ones which are at least as eager to fail as the first
1774 * one.
1775 */
1776 for (bio = rq->bio; bio; bio = bio->bi_next) {
1777 if ((bio->bi_rw & ff) != ff)
1778 break;
1779 bytes += bio->bi_size;
1780 }
1781
1782 /* this could lead to infinite loop */
1783 BUG_ON(blk_rq_bytes(rq) && !bytes);
1784 return bytes;
1785}
1786EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1787
Jens Axboebc58ba92009-01-23 10:54:44 +01001788static void blk_account_io_completion(struct request *req, unsigned int bytes)
1789{
Jens Axboec2553b52009-04-24 08:10:11 +02001790 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001791 const int rw = rq_data_dir(req);
1792 struct hd_struct *part;
1793 int cpu;
1794
1795 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001796 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001797 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1798 part_stat_unlock();
1799 }
1800}
1801
1802static void blk_account_io_done(struct request *req)
1803{
Jens Axboebc58ba92009-01-23 10:54:44 +01001804 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001805 * Account IO completion. flush_rq isn't accounted as a
1806 * normal IO on queueing nor completion. Accounting the
1807 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001808 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001809 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001810 unsigned long duration = jiffies - req->start_time;
1811 const int rw = rq_data_dir(req);
1812 struct hd_struct *part;
1813 int cpu;
1814
1815 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001816 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001817
1818 part_stat_inc(cpu, part, ios[rw]);
1819 part_stat_add(cpu, part, ticks[rw], duration);
1820 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001821 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001822
Jens Axboe6c23a962011-01-07 08:43:37 +01001823 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001824 part_stat_unlock();
1825 }
1826}
1827
Tejun Heo53a08802008-12-03 12:41:26 +01001828/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001829 * blk_peek_request - peek at the top of a request queue
1830 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001831 *
1832 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001833 * Return the request at the top of @q. The returned request
1834 * should be started using blk_start_request() before LLD starts
1835 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001836 *
1837 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001838 * Pointer to the request at the top of @q if available. Null
1839 * otherwise.
1840 *
1841 * Context:
1842 * queue_lock must be held.
1843 */
1844struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001845{
1846 struct request *rq;
1847 int ret;
1848
1849 while ((rq = __elv_next_request(q)) != NULL) {
1850 if (!(rq->cmd_flags & REQ_STARTED)) {
1851 /*
1852 * This is the first time the device driver
1853 * sees this request (possibly after
1854 * requeueing). Notify IO scheduler.
1855 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001856 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001857 elv_activate_rq(q, rq);
1858
1859 /*
1860 * just mark as started even if we don't start
1861 * it, a request that has been delayed should
1862 * not be passed by new incoming requests
1863 */
1864 rq->cmd_flags |= REQ_STARTED;
1865 trace_block_rq_issue(q, rq);
1866 }
1867
1868 if (!q->boundary_rq || q->boundary_rq == rq) {
1869 q->end_sector = rq_end_sector(rq);
1870 q->boundary_rq = NULL;
1871 }
1872
1873 if (rq->cmd_flags & REQ_DONTPREP)
1874 break;
1875
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001876 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001877 /*
1878 * make sure space for the drain appears we
1879 * know we can do this because max_hw_segments
1880 * has been adjusted to be one fewer than the
1881 * device can handle
1882 */
1883 rq->nr_phys_segments++;
1884 }
1885
1886 if (!q->prep_rq_fn)
1887 break;
1888
1889 ret = q->prep_rq_fn(q, rq);
1890 if (ret == BLKPREP_OK) {
1891 break;
1892 } else if (ret == BLKPREP_DEFER) {
1893 /*
1894 * the request may have been (partially) prepped.
1895 * we need to keep this request in the front to
1896 * avoid resource deadlock. REQ_STARTED will
1897 * prevent other fs requests from passing this one.
1898 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001899 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001900 !(rq->cmd_flags & REQ_DONTPREP)) {
1901 /*
1902 * remove the space for the drain we added
1903 * so that we don't add it again
1904 */
1905 --rq->nr_phys_segments;
1906 }
1907
1908 rq = NULL;
1909 break;
1910 } else if (ret == BLKPREP_KILL) {
1911 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001912 /*
1913 * Mark this request as started so we don't trigger
1914 * any debug logic in the end I/O path.
1915 */
1916 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001917 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001918 } else {
1919 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1920 break;
1921 }
1922 }
1923
1924 return rq;
1925}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001926EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001927
Tejun Heo9934c8c2009-05-08 11:54:16 +09001928void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001929{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001930 struct request_queue *q = rq->q;
1931
Tejun Heo158dbda2009-04-23 11:05:18 +09001932 BUG_ON(list_empty(&rq->queuelist));
1933 BUG_ON(ELV_ON_HASH(rq));
1934
1935 list_del_init(&rq->queuelist);
1936
1937 /*
1938 * the time frame between a request being removed from the lists
1939 * and to it is freed is accounted as io that is in progress at
1940 * the driver side.
1941 */
Divyesh Shah91952912010-04-01 15:01:41 -07001942 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001943 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001944 set_io_start_time_ns(rq);
1945 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001946}
1947
Tejun Heo5efccd12009-04-23 11:05:18 +09001948/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001949 * blk_start_request - start request processing on the driver
1950 * @req: request to dequeue
1951 *
1952 * Description:
1953 * Dequeue @req and start timeout timer on it. This hands off the
1954 * request to the driver.
1955 *
1956 * Block internal functions which don't want to start timer should
1957 * call blk_dequeue_request().
1958 *
1959 * Context:
1960 * queue_lock must be held.
1961 */
1962void blk_start_request(struct request *req)
1963{
1964 blk_dequeue_request(req);
1965
1966 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001967 * We are now handing the request to the hardware, initialize
1968 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001969 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001970 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001971 if (unlikely(blk_bidi_rq(req)))
1972 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1973
Tejun Heo9934c8c2009-05-08 11:54:16 +09001974 blk_add_timer(req);
1975}
1976EXPORT_SYMBOL(blk_start_request);
1977
1978/**
1979 * blk_fetch_request - fetch a request from a request queue
1980 * @q: request queue to fetch a request from
1981 *
1982 * Description:
1983 * Return the request at the top of @q. The request is started on
1984 * return and LLD can start processing it immediately.
1985 *
1986 * Return:
1987 * Pointer to the request at the top of @q if available. Null
1988 * otherwise.
1989 *
1990 * Context:
1991 * queue_lock must be held.
1992 */
1993struct request *blk_fetch_request(struct request_queue *q)
1994{
1995 struct request *rq;
1996
1997 rq = blk_peek_request(q);
1998 if (rq)
1999 blk_start_request(rq);
2000 return rq;
2001}
2002EXPORT_SYMBOL(blk_fetch_request);
2003
2004/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002005 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002006 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002007 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002008 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002009 *
2010 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002011 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2012 * the request structure even if @req doesn't have leftover.
2013 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002014 *
2015 * This special helper function is only for request stacking drivers
2016 * (e.g. request-based dm) so that they can handle partial completion.
2017 * Actual device drivers should use blk_end_request instead.
2018 *
2019 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2020 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002021 *
2022 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002023 * %false - this request doesn't have any more data
2024 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002025 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002026bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002028 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 struct bio *bio;
2030
Tejun Heo2e60e022009-04-23 11:05:18 +09002031 if (!req->bio)
2032 return false;
2033
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002034 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002035
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002037 * For fs requests, rq is just carrier of independent bio's
2038 * and each partial completion should be handled separately.
2039 * Reset per-request error on each partial completion.
2040 *
2041 * TODO: tj: This is too subtle. It would be better to let
2042 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002044 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 req->errors = 0;
2046
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002047 if (error && req->cmd_type == REQ_TYPE_FS &&
2048 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002049 char *error_type;
2050
2051 switch (error) {
2052 case -ENOLINK:
2053 error_type = "recoverable transport";
2054 break;
2055 case -EREMOTEIO:
2056 error_type = "critical target";
2057 break;
2058 case -EBADE:
2059 error_type = "critical nexus";
2060 break;
2061 case -EIO:
2062 default:
2063 error_type = "I/O";
2064 break;
2065 }
2066 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2067 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2068 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 }
2070
Jens Axboebc58ba92009-01-23 10:54:44 +01002071 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002072
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 total_bytes = bio_nbytes = 0;
2074 while ((bio = req->bio) != NULL) {
2075 int nbytes;
2076
2077 if (nr_bytes >= bio->bi_size) {
2078 req->bio = bio->bi_next;
2079 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002080 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 next_idx = 0;
2082 bio_nbytes = 0;
2083 } else {
2084 int idx = bio->bi_idx + next_idx;
2085
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002086 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002088 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002089 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 break;
2091 }
2092
2093 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2094 BIO_BUG_ON(nbytes > bio->bi_size);
2095
2096 /*
2097 * not a complete bvec done
2098 */
2099 if (unlikely(nbytes > nr_bytes)) {
2100 bio_nbytes += nr_bytes;
2101 total_bytes += nr_bytes;
2102 break;
2103 }
2104
2105 /*
2106 * advance to the next vector
2107 */
2108 next_idx++;
2109 bio_nbytes += nbytes;
2110 }
2111
2112 total_bytes += nbytes;
2113 nr_bytes -= nbytes;
2114
Jens Axboe6728cb02008-01-31 13:03:55 +01002115 bio = req->bio;
2116 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 /*
2118 * end more in this run, or just return 'not-done'
2119 */
2120 if (unlikely(nr_bytes <= 0))
2121 break;
2122 }
2123 }
2124
2125 /*
2126 * completely done
2127 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002128 if (!req->bio) {
2129 /*
2130 * Reset counters so that the request stacking driver
2131 * can find how many bytes remain in the request
2132 * later.
2133 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002134 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002135 return false;
2136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137
2138 /*
2139 * if the request wasn't completed, update state
2140 */
2141 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002142 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 bio->bi_idx += next_idx;
2144 bio_iovec(bio)->bv_offset += nr_bytes;
2145 bio_iovec(bio)->bv_len -= nr_bytes;
2146 }
2147
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002148 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002149 req->buffer = bio_data(req->bio);
2150
2151 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002152 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002153 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002154
Tejun Heo80a761f2009-07-03 17:48:17 +09002155 /* mixed attributes always follow the first bio */
2156 if (req->cmd_flags & REQ_MIXED_MERGE) {
2157 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2158 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2159 }
2160
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002161 /*
2162 * If total number of sectors is less than the first segment
2163 * size, something has gone terribly wrong.
2164 */
2165 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002166 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002167 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002168 }
2169
2170 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002172
Tejun Heo2e60e022009-04-23 11:05:18 +09002173 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174}
Tejun Heo2e60e022009-04-23 11:05:18 +09002175EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176
Tejun Heo2e60e022009-04-23 11:05:18 +09002177static bool blk_update_bidi_request(struct request *rq, int error,
2178 unsigned int nr_bytes,
2179 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002180{
Tejun Heo2e60e022009-04-23 11:05:18 +09002181 if (blk_update_request(rq, error, nr_bytes))
2182 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002183
Tejun Heo2e60e022009-04-23 11:05:18 +09002184 /* Bidi request must be completed as a whole */
2185 if (unlikely(blk_bidi_rq(rq)) &&
2186 blk_update_request(rq->next_rq, error, bidi_bytes))
2187 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002188
Jens Axboee2e1a142010-06-09 10:42:09 +02002189 if (blk_queue_add_random(rq->q))
2190 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002191
2192 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193}
2194
James Bottomley28018c22010-07-01 19:49:17 +09002195/**
2196 * blk_unprep_request - unprepare a request
2197 * @req: the request
2198 *
2199 * This function makes a request ready for complete resubmission (or
2200 * completion). It happens only after all error handling is complete,
2201 * so represents the appropriate moment to deallocate any resources
2202 * that were allocated to the request in the prep_rq_fn. The queue
2203 * lock is held when calling this.
2204 */
2205void blk_unprep_request(struct request *req)
2206{
2207 struct request_queue *q = req->q;
2208
2209 req->cmd_flags &= ~REQ_DONTPREP;
2210 if (q->unprep_rq_fn)
2211 q->unprep_rq_fn(q, req);
2212}
2213EXPORT_SYMBOL_GPL(blk_unprep_request);
2214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215/*
2216 * queue lock must be held
2217 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002218static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002220 if (blk_rq_tagged(req))
2221 blk_queue_end_tag(req->q, req);
2222
James Bottomleyba396a62009-05-27 14:17:08 +02002223 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002225 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002226 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
Mike Andersone78042e2008-10-30 02:16:20 -07002228 blk_delete_timer(req);
2229
James Bottomley28018c22010-07-01 19:49:17 +09002230 if (req->cmd_flags & REQ_DONTPREP)
2231 blk_unprep_request(req);
2232
2233
Jens Axboebc58ba92009-01-23 10:54:44 +01002234 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002235
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002237 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002238 else {
2239 if (blk_bidi_rq(req))
2240 __blk_put_request(req->next_rq->q, req->next_rq);
2241
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 }
2244}
2245
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002246/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002247 * blk_end_bidi_request - Complete a bidi request
2248 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002249 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002250 * @nr_bytes: number of bytes to complete @rq
2251 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002252 *
2253 * Description:
2254 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002255 * Drivers that supports bidi can safely call this member for any
2256 * type of request, bidi or uni. In the later case @bidi_bytes is
2257 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002258 *
2259 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002260 * %false - we are done with this request
2261 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002262 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002263static bool blk_end_bidi_request(struct request *rq, int error,
2264 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002265{
2266 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002267 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002268
Tejun Heo2e60e022009-04-23 11:05:18 +09002269 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2270 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002271
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002272 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002273 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002274 spin_unlock_irqrestore(q->queue_lock, flags);
2275
Tejun Heo2e60e022009-04-23 11:05:18 +09002276 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002277}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002278
2279/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002280 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2281 * @rq: the request to complete
2282 * @error: %0 for success, < %0 for error
2283 * @nr_bytes: number of bytes to complete @rq
2284 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002285 *
2286 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002287 * Identical to blk_end_bidi_request() except that queue lock is
2288 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002289 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002290 * Return:
2291 * %false - we are done with this request
2292 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002293 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002294static bool __blk_end_bidi_request(struct request *rq, int error,
2295 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002296{
Tejun Heo2e60e022009-04-23 11:05:18 +09002297 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2298 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002299
Tejun Heo2e60e022009-04-23 11:05:18 +09002300 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002301
Tejun Heo2e60e022009-04-23 11:05:18 +09002302 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002303}
2304
2305/**
2306 * blk_end_request - Helper function for drivers to complete the request.
2307 * @rq: the request being processed
2308 * @error: %0 for success, < %0 for error
2309 * @nr_bytes: number of bytes to complete
2310 *
2311 * Description:
2312 * Ends I/O on a number of bytes attached to @rq.
2313 * If @rq has leftover, sets it up for the next range of segments.
2314 *
2315 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002316 * %false - we are done with this request
2317 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002318 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002319bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002320{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002321 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002322}
Jens Axboe56ad1742009-07-28 22:11:24 +02002323EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002324
2325/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002326 * blk_end_request_all - Helper function for drives to finish the request.
2327 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002328 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002329 *
2330 * Description:
2331 * Completely finish @rq.
2332 */
2333void blk_end_request_all(struct request *rq, int error)
2334{
2335 bool pending;
2336 unsigned int bidi_bytes = 0;
2337
2338 if (unlikely(blk_bidi_rq(rq)))
2339 bidi_bytes = blk_rq_bytes(rq->next_rq);
2340
2341 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2342 BUG_ON(pending);
2343}
Jens Axboe56ad1742009-07-28 22:11:24 +02002344EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002345
2346/**
2347 * blk_end_request_cur - Helper function to finish the current request chunk.
2348 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002349 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002350 *
2351 * Description:
2352 * Complete the current consecutively mapped chunk from @rq.
2353 *
2354 * Return:
2355 * %false - we are done with this request
2356 * %true - still buffers pending for this request
2357 */
2358bool blk_end_request_cur(struct request *rq, int error)
2359{
2360 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2361}
Jens Axboe56ad1742009-07-28 22:11:24 +02002362EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002363
2364/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002365 * blk_end_request_err - Finish a request till the next failure boundary.
2366 * @rq: the request to finish till the next failure boundary for
2367 * @error: must be negative errno
2368 *
2369 * Description:
2370 * Complete @rq till the next failure boundary.
2371 *
2372 * Return:
2373 * %false - we are done with this request
2374 * %true - still buffers pending for this request
2375 */
2376bool blk_end_request_err(struct request *rq, int error)
2377{
2378 WARN_ON(error >= 0);
2379 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2380}
2381EXPORT_SYMBOL_GPL(blk_end_request_err);
2382
2383/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002384 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002385 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002386 * @error: %0 for success, < %0 for error
2387 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002388 *
2389 * Description:
2390 * Must be called with queue lock held unlike blk_end_request().
2391 *
2392 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002393 * %false - we are done with this request
2394 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002395 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002396bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002397{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002398 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002399}
Jens Axboe56ad1742009-07-28 22:11:24 +02002400EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002401
2402/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002403 * __blk_end_request_all - Helper function for drives to finish the request.
2404 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002405 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002406 *
2407 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002408 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002409 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002410void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002411{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002412 bool pending;
2413 unsigned int bidi_bytes = 0;
2414
2415 if (unlikely(blk_bidi_rq(rq)))
2416 bidi_bytes = blk_rq_bytes(rq->next_rq);
2417
2418 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2419 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002420}
Jens Axboe56ad1742009-07-28 22:11:24 +02002421EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002422
2423/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002424 * __blk_end_request_cur - Helper function to finish the current request chunk.
2425 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002426 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002427 *
2428 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002429 * Complete the current consecutively mapped chunk from @rq. Must
2430 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002431 *
2432 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002433 * %false - we are done with this request
2434 * %true - still buffers pending for this request
2435 */
2436bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002437{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002438 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002439}
Jens Axboe56ad1742009-07-28 22:11:24 +02002440EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002441
Tejun Heo80a761f2009-07-03 17:48:17 +09002442/**
2443 * __blk_end_request_err - Finish a request till the next failure boundary.
2444 * @rq: the request to finish till the next failure boundary for
2445 * @error: must be negative errno
2446 *
2447 * Description:
2448 * Complete @rq till the next failure boundary. Must be called
2449 * with queue lock held.
2450 *
2451 * Return:
2452 * %false - we are done with this request
2453 * %true - still buffers pending for this request
2454 */
2455bool __blk_end_request_err(struct request *rq, int error)
2456{
2457 WARN_ON(error >= 0);
2458 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2459}
2460EXPORT_SYMBOL_GPL(__blk_end_request_err);
2461
Jens Axboe86db1e22008-01-29 14:53:40 +01002462void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2463 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002465 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002466 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
David Woodhousefb2dce82008-08-05 18:01:53 +01002468 if (bio_has_data(bio)) {
2469 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002470 rq->buffer = bio_data(bio);
2471 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002472 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474
NeilBrown66846572007-08-16 13:31:28 +02002475 if (bio->bi_bdev)
2476 rq->rq_disk = bio->bi_bdev->bd_disk;
2477}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002479#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2480/**
2481 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2482 * @rq: the request to be flushed
2483 *
2484 * Description:
2485 * Flush all pages in @rq.
2486 */
2487void rq_flush_dcache_pages(struct request *rq)
2488{
2489 struct req_iterator iter;
2490 struct bio_vec *bvec;
2491
2492 rq_for_each_segment(bvec, rq, iter)
2493 flush_dcache_page(bvec->bv_page);
2494}
2495EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2496#endif
2497
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002498/**
2499 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2500 * @q : the queue of the device being checked
2501 *
2502 * Description:
2503 * Check if underlying low-level drivers of a device are busy.
2504 * If the drivers want to export their busy state, they must set own
2505 * exporting function using blk_queue_lld_busy() first.
2506 *
2507 * Basically, this function is used only by request stacking drivers
2508 * to stop dispatching requests to underlying devices when underlying
2509 * devices are busy. This behavior helps more I/O merging on the queue
2510 * of the request stacking driver and prevents I/O throughput regression
2511 * on burst I/O load.
2512 *
2513 * Return:
2514 * 0 - Not busy (The request stacking driver should dispatch request)
2515 * 1 - Busy (The request stacking driver should stop dispatching request)
2516 */
2517int blk_lld_busy(struct request_queue *q)
2518{
2519 if (q->lld_busy_fn)
2520 return q->lld_busy_fn(q);
2521
2522 return 0;
2523}
2524EXPORT_SYMBOL_GPL(blk_lld_busy);
2525
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002526/**
2527 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2528 * @rq: the clone request to be cleaned up
2529 *
2530 * Description:
2531 * Free all bios in @rq for a cloned request.
2532 */
2533void blk_rq_unprep_clone(struct request *rq)
2534{
2535 struct bio *bio;
2536
2537 while ((bio = rq->bio) != NULL) {
2538 rq->bio = bio->bi_next;
2539
2540 bio_put(bio);
2541 }
2542}
2543EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2544
2545/*
2546 * Copy attributes of the original request to the clone request.
2547 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2548 */
2549static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2550{
2551 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002552 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002553 dst->cmd_type = src->cmd_type;
2554 dst->__sector = blk_rq_pos(src);
2555 dst->__data_len = blk_rq_bytes(src);
2556 dst->nr_phys_segments = src->nr_phys_segments;
2557 dst->ioprio = src->ioprio;
2558 dst->extra_len = src->extra_len;
2559}
2560
2561/**
2562 * blk_rq_prep_clone - Helper function to setup clone request
2563 * @rq: the request to be setup
2564 * @rq_src: original request to be cloned
2565 * @bs: bio_set that bios for clone are allocated from
2566 * @gfp_mask: memory allocation mask for bio
2567 * @bio_ctr: setup function to be called for each clone bio.
2568 * Returns %0 for success, non %0 for failure.
2569 * @data: private data to be passed to @bio_ctr
2570 *
2571 * Description:
2572 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2573 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2574 * are not copied, and copying such parts is the caller's responsibility.
2575 * Also, pages which the original bios are pointing to are not copied
2576 * and the cloned bios just point same pages.
2577 * So cloned bios must be completed before original bios, which means
2578 * the caller must complete @rq before @rq_src.
2579 */
2580int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2581 struct bio_set *bs, gfp_t gfp_mask,
2582 int (*bio_ctr)(struct bio *, struct bio *, void *),
2583 void *data)
2584{
2585 struct bio *bio, *bio_src;
2586
2587 if (!bs)
2588 bs = fs_bio_set;
2589
2590 blk_rq_init(NULL, rq);
2591
2592 __rq_for_each_bio(bio_src, rq_src) {
2593 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2594 if (!bio)
2595 goto free_and_out;
2596
2597 __bio_clone(bio, bio_src);
2598
2599 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002600 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002601 goto free_and_out;
2602
2603 if (bio_ctr && bio_ctr(bio, bio_src, data))
2604 goto free_and_out;
2605
2606 if (rq->bio) {
2607 rq->biotail->bi_next = bio;
2608 rq->biotail = bio;
2609 } else
2610 rq->bio = rq->biotail = bio;
2611 }
2612
2613 __blk_rq_prep_clone(rq, rq_src);
2614
2615 return 0;
2616
2617free_and_out:
2618 if (bio)
2619 bio_free(bio, bs);
2620 blk_rq_unprep_clone(rq);
2621
2622 return -ENOMEM;
2623}
2624EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2625
Jens Axboe18887ad2008-07-28 13:08:45 +02002626int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627{
2628 return queue_work(kblockd_workqueue, work);
2629}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630EXPORT_SYMBOL(kblockd_schedule_work);
2631
Vivek Goyale43473b2010-09-15 17:06:35 -04002632int kblockd_schedule_delayed_work(struct request_queue *q,
2633 struct delayed_work *dwork, unsigned long delay)
2634{
2635 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2636}
2637EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2638
Jens Axboe73c10102011-03-08 13:19:51 +01002639#define PLUG_MAGIC 0x91827364
2640
2641void blk_start_plug(struct blk_plug *plug)
2642{
2643 struct task_struct *tsk = current;
2644
2645 plug->magic = PLUG_MAGIC;
2646 INIT_LIST_HEAD(&plug->list);
2647 plug->should_sort = 0;
2648
2649 /*
2650 * If this is a nested plug, don't actually assign it. It will be
2651 * flushed on its own.
2652 */
2653 if (!tsk->plug) {
2654 /*
2655 * Store ordering should not be needed here, since a potential
2656 * preempt will imply a full memory barrier
2657 */
2658 tsk->plug = plug;
2659 }
2660}
2661EXPORT_SYMBOL(blk_start_plug);
2662
2663static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2664{
2665 struct request *rqa = container_of(a, struct request, queuelist);
2666 struct request *rqb = container_of(b, struct request, queuelist);
2667
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002668 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002669}
2670
2671static void flush_plug_list(struct blk_plug *plug)
2672{
2673 struct request_queue *q;
2674 unsigned long flags;
2675 struct request *rq;
2676
2677 BUG_ON(plug->magic != PLUG_MAGIC);
2678
2679 if (list_empty(&plug->list))
2680 return;
2681
2682 if (plug->should_sort)
2683 list_sort(NULL, &plug->list, plug_rq_cmp);
2684
2685 q = NULL;
2686 local_irq_save(flags);
2687 while (!list_empty(&plug->list)) {
2688 rq = list_entry_rq(plug->list.next);
2689 list_del_init(&rq->queuelist);
2690 BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2691 BUG_ON(!rq->q);
2692 if (rq->q != q) {
2693 if (q) {
Jens Axboe4c63f562011-03-10 08:58:35 +01002694 __blk_run_queue(q, false);
Jens Axboe73c10102011-03-08 13:19:51 +01002695 spin_unlock(q->queue_lock);
2696 }
2697 q = rq->q;
2698 spin_lock(q->queue_lock);
2699 }
2700 rq->cmd_flags &= ~REQ_ON_PLUG;
2701
2702 /*
2703 * rq is already accounted, so use raw insert
2704 */
Jens Axboe401a18e2011-03-25 16:57:52 +01002705 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2706 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2707 else
2708 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe73c10102011-03-08 13:19:51 +01002709 }
2710
2711 if (q) {
Jens Axboe4c63f562011-03-10 08:58:35 +01002712 __blk_run_queue(q, false);
Jens Axboe73c10102011-03-08 13:19:51 +01002713 spin_unlock(q->queue_lock);
2714 }
2715
2716 BUG_ON(!list_empty(&plug->list));
2717 local_irq_restore(flags);
2718}
2719
2720static void __blk_finish_plug(struct task_struct *tsk, struct blk_plug *plug)
2721{
2722 flush_plug_list(plug);
2723
2724 if (plug == tsk->plug)
2725 tsk->plug = NULL;
2726}
2727
2728void blk_finish_plug(struct blk_plug *plug)
2729{
2730 if (plug)
2731 __blk_finish_plug(current, plug);
2732}
2733EXPORT_SYMBOL(blk_finish_plug);
2734
2735void __blk_flush_plug(struct task_struct *tsk, struct blk_plug *plug)
2736{
2737 __blk_finish_plug(tsk, plug);
2738 tsk->plug = plug;
2739}
2740EXPORT_SYMBOL(__blk_flush_plug);
2741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742int __init blk_dev_init(void)
2743{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002744 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2745 sizeof(((struct request *)0)->cmd_flags));
2746
Tejun Heo89b90be2011-01-03 15:01:47 +01002747 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2748 kblockd_workqueue = alloc_workqueue("kblockd",
2749 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 if (!kblockd_workqueue)
2751 panic("Failed to create kblockd\n");
2752
2753 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002754 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755
Jens Axboe8324aa92008-01-29 14:51:59 +01002756 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002757 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 return 0;
2760}