blob: 82a45898ba76ffee794c8c2d819bb5c132e51c03 [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
Jens Axboe7eaceac2011-03-10 08:52:07 +0100201/*
202 * 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)
206{
207 /*
208 * If the current process is plugged and has barriers submitted,
209 * we will livelock if we don't unplug first.
210 */
211 blk_flush_plug(current);
212}
213
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500214static void blk_delay_work(struct work_struct *work)
215{
216 struct request_queue *q;
217
218 q = container_of(work, struct request_queue, delay_work.work);
219 spin_lock_irq(q->queue_lock);
Jens Axboe73c10102011-03-08 13:19:51 +0100220 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500221 spin_unlock_irq(q->queue_lock);
222}
223
224/**
225 * blk_delay_queue - restart queueing after defined interval
226 * @q: The &struct request_queue in question
227 * @msecs: Delay in msecs
228 *
229 * Description:
230 * 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)
235{
236 schedule_delayed_work(&q->delay_work, msecs_to_jiffies(msecs));
237}
238EXPORT_SYMBOL(blk_delay_queue);
239
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 Heoa538cd02009-04-23 11:05:17 +0900254 __blk_run_queue(q);
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 Axboe3cca6dc2011-03-02 11:08:00 -0500274 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 *
292 */
293void blk_sync_queue(struct request_queue *q)
294{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100295 del_timer_sync(&q->timeout);
Vivek Goyale43473b2010-09-15 17:06:35 -0400296 throtl_shutdown_timer_wq(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500297 cancel_delayed_work_sync(&q->delay_work);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100298 queue_sync_plugs(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300EXPORT_SYMBOL(blk_sync_queue);
301
302/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200303 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200305 *
306 * Description:
307 * See @blk_run_queue. This variant must be called with the queue lock
308 * held and interrupts disabled.
309 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200311void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Tejun Heoa538cd02009-04-23 11:05:17 +0900313 if (unlikely(blk_queue_stopped(q)))
314 return;
315
Jens Axboedac07ec2006-05-11 08:20:16 +0200316 /*
317 * Only recurse once to avoid overrunning the stack, let the unplug
318 * handling reinvoke the handler shortly if we already got there.
319 */
Tejun Heoa538cd02009-04-23 11:05:17 +0900320 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
321 q->request_fn(q);
322 queue_flag_clear(QUEUE_FLAG_REENTER, q);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100323 } else
324 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200325}
326EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200327
Nick Piggin75ad23b2008-04-29 14:48:33 +0200328/**
329 * blk_run_queue - run a single device queue
330 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200331 *
332 * Description:
333 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900334 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200335 */
336void blk_run_queue(struct request_queue *q)
337{
338 unsigned long flags;
339
340 spin_lock_irqsave(q->queue_lock, flags);
341 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 spin_unlock_irqrestore(q->queue_lock, flags);
343}
344EXPORT_SYMBOL(blk_run_queue);
345
Jens Axboe165125e2007-07-24 09:28:11 +0200346void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500347{
348 kobject_put(&q->kobj);
349}
Al Viro483f4af2006-03-18 18:34:37 -0500350
Jens Axboe6728cb02008-01-31 13:03:55 +0100351void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500352{
Jens Axboee3335de92008-09-18 09:22:54 -0700353 /*
354 * We know we have process context here, so we can be a little
355 * cautious and ensure that pending block actions on this device
356 * are done before moving on. Going into this function, we should
357 * not have processes doing IO to this device.
358 */
359 blk_sync_queue(q);
360
Matthew Garrett31373d02010-04-06 14:25:14 +0200361 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500362 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200363 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500364 mutex_unlock(&q->sysfs_lock);
365
366 if (q->elevator)
367 elevator_exit(q->elevator);
368
369 blk_put_queue(q);
370}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371EXPORT_SYMBOL(blk_cleanup_queue);
372
Jens Axboe165125e2007-07-24 09:28:11 +0200373static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
375 struct request_list *rl = &q->rq;
376
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400377 if (unlikely(rl->rq_pool))
378 return 0;
379
Jens Axboe1faa16d2009-04-06 14:48:01 +0200380 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
381 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200382 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200383 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
384 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Christoph Lameter19460892005-06-23 00:08:19 -0700386 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
387 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389 if (!rl->rq_pool)
390 return -ENOMEM;
391
392 return 0;
393}
394
Jens Axboe165125e2007-07-24 09:28:11 +0200395struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
Christoph Lameter19460892005-06-23 00:08:19 -0700397 return blk_alloc_queue_node(gfp_mask, -1);
398}
399EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Jens Axboe165125e2007-07-24 09:28:11 +0200401struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700402{
Jens Axboe165125e2007-07-24 09:28:11 +0200403 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700404 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700405
Jens Axboe8324aa92008-01-29 14:51:59 +0100406 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700407 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 if (!q)
409 return NULL;
410
Jens Axboe0989a022009-06-12 14:42:56 +0200411 q->backing_dev_info.ra_pages =
412 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
413 q->backing_dev_info.state = 0;
414 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200415 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200416
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700417 err = bdi_init(&q->backing_dev_info);
418 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100419 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700420 return NULL;
421 }
422
Vivek Goyale43473b2010-09-15 17:06:35 -0400423 if (blk_throtl_init(q)) {
424 kmem_cache_free(blk_requestq_cachep, q);
425 return NULL;
426 }
427
Matthew Garrett31373d02010-04-06 14:25:14 +0200428 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
429 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700430 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
431 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoae1b1532011-01-25 12:43:54 +0100432 INIT_LIST_HEAD(&q->flush_queue[0]);
433 INIT_LIST_HEAD(&q->flush_queue[1]);
434 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500435 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500436
Jens Axboe8324aa92008-01-29 14:51:59 +0100437 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Al Viro483f4af2006-03-18 18:34:37 -0500439 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700440 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500441
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 return q;
443}
Christoph Lameter19460892005-06-23 00:08:19 -0700444EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446/**
447 * blk_init_queue - prepare a request queue for use with a block device
448 * @rfn: The function to be called to process requests that have been
449 * placed on the queue.
450 * @lock: Request queue spin lock
451 *
452 * Description:
453 * If a block device wishes to use the standard request handling procedures,
454 * which sorts requests and coalesces adjacent requests, then it must
455 * call blk_init_queue(). The function @rfn will be called when there
456 * are requests on the queue that need to be processed. If the device
457 * supports plugging, then @rfn may not be called immediately when requests
458 * are available on the queue, but may be called at some time later instead.
459 * Plugged queues are generally unplugged when a buffer belonging to one
460 * of the requests on the queue is needed, or due to memory pressure.
461 *
462 * @rfn is not required, or even expected, to remove all requests off the
463 * queue, but only as many as it can handle at a time. If it does leave
464 * requests on the queue, it is responsible for arranging that the requests
465 * get dealt with eventually.
466 *
467 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200468 * request queue; this lock will be taken also from interrupt context, so irq
469 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200471 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 * it didn't succeed.
473 *
474 * Note:
475 * blk_init_queue() must be paired with a blk_cleanup_queue() call
476 * when the block device is deactivated (such as at module unload).
477 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700478
Jens Axboe165125e2007-07-24 09:28:11 +0200479struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
Christoph Lameter19460892005-06-23 00:08:19 -0700481 return blk_init_queue_node(rfn, lock, -1);
482}
483EXPORT_SYMBOL(blk_init_queue);
484
Jens Axboe165125e2007-07-24 09:28:11 +0200485struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700486blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
487{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600488 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600490 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
491 if (!uninit_q)
492 return NULL;
493
494 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
495 if (!q)
496 blk_cleanup_queue(uninit_q);
497
498 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200499}
500EXPORT_SYMBOL(blk_init_queue_node);
501
502struct request_queue *
503blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
504 spinlock_t *lock)
505{
506 return blk_init_allocated_queue_node(q, rfn, lock, -1);
507}
508EXPORT_SYMBOL(blk_init_allocated_queue);
509
510struct request_queue *
511blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
512 spinlock_t *lock, int node_id)
513{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 if (!q)
515 return NULL;
516
Christoph Lameter19460892005-06-23 00:08:19 -0700517 q->node = node_id;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600518 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500519 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
521 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900523 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100524 q->queue_flags = QUEUE_FLAG_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 q->queue_lock = lock;
526
Jens Axboef3b144a2009-03-06 08:48:33 +0100527 /*
528 * This also sets hw/phys segments, boundary and size
529 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
Alan Stern44ec9542007-02-20 11:01:57 -0500532 q->sg_reserved_size = INT_MAX;
533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 /*
535 * all done
536 */
537 if (!elevator_init(q, NULL)) {
538 blk_queue_congestion_threshold(q);
539 return q;
540 }
541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 return NULL;
543}
Mike Snitzer01effb02010-05-11 08:57:42 +0200544EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Jens Axboe165125e2007-07-24 09:28:11 +0200546int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700548 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500549 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 return 0;
551 }
552
553 return 1;
554}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Jens Axboe165125e2007-07-24 09:28:11 +0200556static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557{
Jens Axboe73c10102011-03-08 13:19:51 +0100558 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
559
Jens Axboe4aff5e22006-08-10 08:44:47 +0200560 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200561 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 mempool_free(rq, q->rq.rq_pool);
563}
564
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200565static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200566blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
568 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
569
570 if (!rq)
571 return NULL;
572
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200573 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200574
Jerome Marchand42dad762009-04-22 14:01:49 +0200575 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Tejun Heocb98fc82005-10-28 08:29:39 +0200577 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200578 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200579 mempool_free(rq, q->rq.rq_pool);
580 return NULL;
581 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200582 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Tejun Heocb98fc82005-10-28 08:29:39 +0200585 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586}
587
588/*
589 * ioc_batching returns true if the ioc is a valid batching request and
590 * should be given priority access to a request.
591 */
Jens Axboe165125e2007-07-24 09:28:11 +0200592static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593{
594 if (!ioc)
595 return 0;
596
597 /*
598 * Make sure the process is able to allocate at least 1 request
599 * even if the batch times out, otherwise we could theoretically
600 * lose wakeups.
601 */
602 return ioc->nr_batch_requests == q->nr_batching ||
603 (ioc->nr_batch_requests > 0
604 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
605}
606
607/*
608 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
609 * will cause the process to be a "batcher" on all queues in the system. This
610 * is the behaviour we want though - once it gets a wakeup it should be given
611 * a nice run.
612 */
Jens Axboe165125e2007-07-24 09:28:11 +0200613static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614{
615 if (!ioc || ioc_batching(q, ioc))
616 return;
617
618 ioc->nr_batch_requests = q->nr_batching;
619 ioc->last_waited = jiffies;
620}
621
Jens Axboe1faa16d2009-04-06 14:48:01 +0200622static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
624 struct request_list *rl = &q->rq;
625
Jens Axboe1faa16d2009-04-06 14:48:01 +0200626 if (rl->count[sync] < queue_congestion_off_threshold(q))
627 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Jens Axboe1faa16d2009-04-06 14:48:01 +0200629 if (rl->count[sync] + 1 <= q->nr_requests) {
630 if (waitqueue_active(&rl->wait[sync]))
631 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
Jens Axboe1faa16d2009-04-06 14:48:01 +0200633 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
635}
636
637/*
638 * A request has just been released. Account for it, update the full and
639 * congestion status, wake up any waiters. Called under q->queue_lock.
640 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200641static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
643 struct request_list *rl = &q->rq;
644
Jens Axboe1faa16d2009-04-06 14:48:01 +0200645 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200646 if (priv)
647 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Jens Axboe1faa16d2009-04-06 14:48:01 +0200649 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Jens Axboe1faa16d2009-04-06 14:48:01 +0200651 if (unlikely(rl->starved[sync ^ 1]))
652 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653}
654
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100656 * Determine if elevator data should be initialized when allocating the
657 * request associated with @bio.
658 */
659static bool blk_rq_should_init_elevator(struct bio *bio)
660{
661 if (!bio)
662 return true;
663
664 /*
665 * Flush requests do not use the elevator so skip initialization.
666 * This allows a request to share the flush and elevator data.
667 */
668 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
669 return false;
670
671 return true;
672}
673
674/*
Nick Piggind6344532005-06-28 20:45:14 -0700675 * Get a free request, queue_lock must be held.
676 * Returns NULL on failure, with queue_lock held.
677 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 */
Jens Axboe165125e2007-07-24 09:28:11 +0200679static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100680 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
682 struct request *rq = NULL;
683 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100684 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200685 const bool is_sync = rw_is_sync(rw_flags) != 0;
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100686 int may_queue, priv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
Jens Axboe7749a8d2006-12-13 13:02:26 +0100688 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100689 if (may_queue == ELV_MQUEUE_NO)
690 goto rq_starved;
691
Jens Axboe1faa16d2009-04-06 14:48:01 +0200692 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
693 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200694 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100695 /*
696 * The queue will fill after this allocation, so set
697 * it as full, and mark this process as "batching".
698 * This process will be allowed to complete a batch of
699 * requests, others will be blocked.
700 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200701 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100702 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200703 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100704 } else {
705 if (may_queue != ELV_MQUEUE_MUST
706 && !ioc_batching(q, ioc)) {
707 /*
708 * The queue is full and the allocating
709 * process is not a "batcher", and not
710 * exempted by the IO scheduler
711 */
712 goto out;
713 }
714 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200716 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 }
718
Jens Axboe082cf692005-06-28 16:35:11 +0200719 /*
720 * Only allow batching queuers to allocate up to 50% over the defined
721 * limit of requests, otherwise we could have thousands of requests
722 * allocated with any setting of ->nr_requests
723 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200724 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200725 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100726
Jens Axboe1faa16d2009-04-06 14:48:01 +0200727 rl->count[is_sync]++;
728 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200729
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100730 if (blk_rq_should_init_elevator(bio)) {
731 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
732 if (priv)
733 rl->elvpriv++;
734 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200735
Jens Axboef253b862010-10-24 22:06:02 +0200736 if (blk_queue_io_stat(q))
737 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 spin_unlock_irq(q->queue_lock);
739
Jens Axboe7749a8d2006-12-13 13:02:26 +0100740 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100741 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 /*
743 * Allocation failed presumably due to memory. Undo anything
744 * we might have messed up.
745 *
746 * Allocating task should really be put onto the front of the
747 * wait queue, but this is pretty rare.
748 */
749 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200750 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
752 /*
753 * in the very unlikely event that allocation failed and no
754 * requests for this direction was pending, mark us starved
755 * so that freeing of a request in the other direction will
756 * notice us. another possible fix would be to split the
757 * rq mempool into READ and WRITE
758 */
759rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200760 if (unlikely(rl->count[is_sync] == 0))
761 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 goto out;
764 }
765
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100766 /*
767 * ioc may be NULL here, and ioc_batching will be false. That's
768 * OK, if the queue is under the request limit then requests need
769 * not count toward the nr_batch_requests limit. There will always
770 * be some limit enforced by BLK_BATCH_TIME.
771 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 if (ioc_batching(q, ioc))
773 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100774
Jens Axboe1faa16d2009-04-06 14:48:01 +0200775 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 return rq;
778}
779
780/*
Jens Axboe7eaceac2011-03-10 08:52:07 +0100781 * No available requests for this queue, wait for some requests to become
782 * available.
Nick Piggind6344532005-06-28 20:45:14 -0700783 *
784 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 */
Jens Axboe165125e2007-07-24 09:28:11 +0200786static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200787 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200789 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 struct request *rq;
791
Jens Axboe7749a8d2006-12-13 13:02:26 +0100792 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700793 while (!rq) {
794 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200795 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 struct request_list *rl = &q->rq;
797
Jens Axboe1faa16d2009-04-06 14:48:01 +0200798 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 TASK_UNINTERRUPTIBLE);
800
Jens Axboe1faa16d2009-04-06 14:48:01 +0200801 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200803 spin_unlock_irq(q->queue_lock);
804 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200806 /*
807 * After sleeping, we become a "batching" process and
808 * will be able to allocate at least one request, and
809 * up to a big batch of them for a small period time.
810 * See ioc_batching, ioc_set_batching
811 */
812 ioc = current_io_context(GFP_NOIO, q->node);
813 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100814
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200815 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200816 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200817
818 rq = get_request(q, rw_flags, bio, GFP_NOIO);
819 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821 return rq;
822}
823
Jens Axboe165125e2007-07-24 09:28:11 +0200824struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825{
826 struct request *rq;
827
828 BUG_ON(rw != READ && rw != WRITE);
829
Nick Piggind6344532005-06-28 20:45:14 -0700830 spin_lock_irq(q->queue_lock);
831 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200832 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700833 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200834 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700835 if (!rq)
836 spin_unlock_irq(q->queue_lock);
837 }
838 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
840 return rq;
841}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842EXPORT_SYMBOL(blk_get_request);
843
844/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300845 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700846 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300847 * @bio: The bio describing the memory mappings that will be submitted for IO.
848 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700849 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200850 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300851 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
852 * type commands. Where the struct request needs to be farther initialized by
853 * the caller. It is passed a &struct bio, which describes the memory info of
854 * the I/O transfer.
855 *
856 * The caller of blk_make_request must make sure that bi_io_vec
857 * are set to describe the memory buffers. That bio_data_dir() will return
858 * the needed direction of the request. (And all bio's in the passed bio-chain
859 * are properly set accordingly)
860 *
861 * If called under none-sleepable conditions, mapped bio buffers must not
862 * need bouncing, by calling the appropriate masked or flagged allocator,
863 * suitable for the target device. Otherwise the call to blk_queue_bounce will
864 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200865 *
866 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
867 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
868 * anything but the first bio in the chain. Otherwise you risk waiting for IO
869 * completion of a bio that hasn't been submitted yet, thus resulting in a
870 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
871 * of bio_alloc(), as that avoids the mempool deadlock.
872 * If possible a big IO should be split into smaller parts when allocation
873 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200874 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300875struct request *blk_make_request(struct request_queue *q, struct bio *bio,
876 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200877{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300878 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
879
880 if (unlikely(!rq))
881 return ERR_PTR(-ENOMEM);
882
883 for_each_bio(bio) {
884 struct bio *bounce_bio = bio;
885 int ret;
886
887 blk_queue_bounce(q, &bounce_bio);
888 ret = blk_rq_append_bio(q, rq, bounce_bio);
889 if (unlikely(ret)) {
890 blk_put_request(rq);
891 return ERR_PTR(ret);
892 }
893 }
894
895 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200896}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300897EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200898
899/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 * blk_requeue_request - put a request back on queue
901 * @q: request queue where request should be inserted
902 * @rq: request to be inserted
903 *
904 * Description:
905 * Drivers often keep queueing requests until the hardware cannot accept
906 * more, when that condition happens we need to put the request back
907 * on the queue. Must be called with queue lock held.
908 */
Jens Axboe165125e2007-07-24 09:28:11 +0200909void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700911 blk_delete_timer(rq);
912 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100913 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100914
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 if (blk_rq_tagged(rq))
916 blk_queue_end_tag(q, rq);
917
James Bottomleyba396a62009-05-27 14:17:08 +0200918 BUG_ON(blk_queued_rq(rq));
919
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 elv_requeue_request(q, rq);
921}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922EXPORT_SYMBOL(blk_requeue_request);
923
Jens Axboe73c10102011-03-08 13:19:51 +0100924static void add_acct_request(struct request_queue *q, struct request *rq,
925 int where)
926{
927 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100928 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +0100929}
930
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200932 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 * @q: request queue where request should be inserted
934 * @rq: request to be inserted
935 * @at_head: insert request at head or tail of queue
936 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 *
938 * Description:
939 * Many block devices need to execute commands asynchronously, so they don't
940 * block the whole kernel from preemption during request execution. This is
941 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200942 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
943 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 *
945 * We have the option of inserting the head or the tail of the queue.
946 * Typically we use the tail for new ioctls and so forth. We use the head
947 * of the queue for things like a QUEUE_FULL message from a device, or a
948 * host that is unable to accept a particular command.
949 */
Jens Axboe165125e2007-07-24 09:28:11 +0200950void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500951 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952{
Tejun Heo 867d1192005-04-24 02:06:05 -0500953 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 unsigned long flags;
955
956 /*
957 * tell I/O scheduler that this isn't a regular read/write (ie it
958 * must not attempt merges on this) and that it acts as a soft
959 * barrier
960 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200961 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
963 rq->special = data;
964
965 spin_lock_irqsave(q->queue_lock, flags);
966
967 /*
968 * If command is tagged, release the tag
969 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500970 if (blk_rq_tagged(rq))
971 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
Jens Axboe73c10102011-03-08 13:19:51 +0100973 add_acct_request(q, rq, where);
Tejun Heoa7f55792009-04-23 11:05:17 +0900974 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 spin_unlock_irqrestore(q->queue_lock, flags);
976}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977EXPORT_SYMBOL(blk_insert_request);
978
Tejun Heo074a7ac2008-08-25 19:56:14 +0900979static void part_round_stats_single(int cpu, struct hd_struct *part,
980 unsigned long now)
981{
982 if (now == part->stamp)
983 return;
984
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200985 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +0900986 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200987 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +0900988 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
989 }
990 part->stamp = now;
991}
992
993/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +0200994 * part_round_stats() - Round off the performance stats on a struct disk_stats.
995 * @cpu: cpu number for stats access
996 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 *
998 * The average IO queue length and utilisation statistics are maintained
999 * by observing the current state of the queue length and the amount of
1000 * time it has been in this state for.
1001 *
1002 * Normally, that accounting is done on IO completion, but that can result
1003 * in more than a second's worth of IO being accounted for within any one
1004 * second, leading to >100% utilisation. To deal with that, we call this
1005 * function to do a round-off before returning the results when reading
1006 * /proc/diskstats. This accounts immediately for all queue usage up to
1007 * the current jiffies and restarts the counters again.
1008 */
Tejun Heoc9959052008-08-25 19:47:21 +09001009void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001010{
1011 unsigned long now = jiffies;
1012
Tejun Heo074a7ac2008-08-25 19:56:14 +09001013 if (part->partno)
1014 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1015 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001016}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001017EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019/*
1020 * queue lock must be held
1021 */
Jens Axboe165125e2007-07-24 09:28:11 +02001022void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 if (unlikely(!q))
1025 return;
1026 if (unlikely(--req->ref_count))
1027 return;
1028
Tejun Heo8922e162005-10-20 16:23:44 +02001029 elv_completed_request(q, req);
1030
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001031 /* this is a bio leak */
1032 WARN_ON(req->bio != NULL);
1033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 /*
1035 * Request may not have originated from ll_rw_blk. if not,
1036 * it didn't come out of our reserved rq pools
1037 */
Jens Axboe49171e52006-08-10 08:59:11 +02001038 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001039 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001040 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001043 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001046 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
1048}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001049EXPORT_SYMBOL_GPL(__blk_put_request);
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051void blk_put_request(struct request *req)
1052{
Tejun Heo8922e162005-10-20 16:23:44 +02001053 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001054 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001056 spin_lock_irqsave(q->queue_lock, flags);
1057 __blk_put_request(q, req);
1058 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060EXPORT_SYMBOL(blk_put_request);
1061
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001062/**
1063 * blk_add_request_payload - add a payload to a request
1064 * @rq: request to update
1065 * @page: page backing the payload
1066 * @len: length of the payload.
1067 *
1068 * This allows to later add a payload to an already submitted request by
1069 * a block driver. The driver needs to take care of freeing the payload
1070 * itself.
1071 *
1072 * Note that this is a quite horrible hack and nothing but handling of
1073 * discard requests should ever use it.
1074 */
1075void blk_add_request_payload(struct request *rq, struct page *page,
1076 unsigned int len)
1077{
1078 struct bio *bio = rq->bio;
1079
1080 bio->bi_io_vec->bv_page = page;
1081 bio->bi_io_vec->bv_offset = 0;
1082 bio->bi_io_vec->bv_len = len;
1083
1084 bio->bi_size = len;
1085 bio->bi_vcnt = 1;
1086 bio->bi_phys_segments = 1;
1087
1088 rq->__data_len = rq->resid_len = len;
1089 rq->nr_phys_segments = 1;
1090 rq->buffer = bio_data(bio);
1091}
1092EXPORT_SYMBOL_GPL(blk_add_request_payload);
1093
Jens Axboe73c10102011-03-08 13:19:51 +01001094static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1095 struct bio *bio)
1096{
1097 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1098
1099 /*
1100 * Debug stuff, kill later
1101 */
1102 if (!rq_mergeable(req)) {
1103 blk_dump_rq_flags(req, "back");
1104 return false;
1105 }
1106
1107 if (!ll_back_merge_fn(q, req, bio))
1108 return false;
1109
1110 trace_block_bio_backmerge(q, bio);
1111
1112 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1113 blk_rq_set_mixed_merge(req);
1114
1115 req->biotail->bi_next = bio;
1116 req->biotail = bio;
1117 req->__data_len += bio->bi_size;
1118 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1119
1120 drive_stat_acct(req, 0);
1121 return true;
1122}
1123
1124static bool bio_attempt_front_merge(struct request_queue *q,
1125 struct request *req, struct bio *bio)
1126{
1127 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1128 sector_t sector;
1129
1130 /*
1131 * Debug stuff, kill later
1132 */
1133 if (!rq_mergeable(req)) {
1134 blk_dump_rq_flags(req, "front");
1135 return false;
1136 }
1137
1138 if (!ll_front_merge_fn(q, req, bio))
1139 return false;
1140
1141 trace_block_bio_frontmerge(q, bio);
1142
1143 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1144 blk_rq_set_mixed_merge(req);
1145
1146 sector = bio->bi_sector;
1147
1148 bio->bi_next = req->bio;
1149 req->bio = bio;
1150
1151 /*
1152 * may not be valid. if the low level driver said
1153 * it didn't need a bounce buffer then it better
1154 * not touch req->buffer either...
1155 */
1156 req->buffer = bio_data(bio);
1157 req->__sector = bio->bi_sector;
1158 req->__data_len += bio->bi_size;
1159 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1160
1161 drive_stat_acct(req, 0);
1162 return true;
1163}
1164
1165/*
1166 * Attempts to merge with the plugged list in the current process. Returns
1167 * true if merge was succesful, otherwise false.
1168 */
1169static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1170 struct bio *bio)
1171{
1172 struct blk_plug *plug;
1173 struct request *rq;
1174 bool ret = false;
1175
1176 plug = tsk->plug;
1177 if (!plug)
1178 goto out;
1179
1180 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1181 int el_ret;
1182
1183 if (rq->q != q)
1184 continue;
1185
1186 el_ret = elv_try_merge(rq, bio);
1187 if (el_ret == ELEVATOR_BACK_MERGE) {
1188 ret = bio_attempt_back_merge(q, rq, bio);
1189 if (ret)
1190 break;
1191 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1192 ret = bio_attempt_front_merge(q, rq, bio);
1193 if (ret)
1194 break;
1195 }
1196 }
1197out:
1198 return ret;
1199}
1200
Jens Axboe86db1e22008-01-29 14:53:40 +01001201void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001202{
Jens Axboec7c22e42008-09-13 20:26:01 +02001203 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001204 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001205
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001206 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1207 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001208 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001209
Tejun Heo52d9e672006-01-06 09:49:58 +01001210 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001211 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001212 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001213 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001214}
1215
Jens Axboe165125e2007-07-24 09:28:11 +02001216static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001218 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001219 struct blk_plug *plug;
1220 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1221 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 /*
1224 * low level driver can indicate that it wants pages above a
1225 * certain limit bounced to low memory (ie for highmem, or even
1226 * ISA dma in theory)
1227 */
1228 blk_queue_bounce(q, &bio);
1229
Tejun Heo4fed9472010-09-03 11:56:17 +02001230 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001231 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001232 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001233 goto get_rq;
1234 }
1235
Jens Axboe73c10102011-03-08 13:19:51 +01001236 /*
1237 * Check if we can merge with the plugged list before grabbing
1238 * any locks.
1239 */
1240 if (attempt_plug_merge(current, q, bio))
1241 goto out;
1242
1243 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
1245 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001246 if (el_ret == ELEVATOR_BACK_MERGE) {
1247 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1248 if (bio_attempt_back_merge(q, req, bio)) {
1249 if (!attempt_back_merge(q, req))
1250 elv_merged_request(q, req, el_ret);
1251 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001252 }
Jens Axboe73c10102011-03-08 13:19:51 +01001253 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1254 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1255 if (bio_attempt_front_merge(q, req, bio)) {
1256 if (!attempt_front_merge(q, req))
1257 elv_merged_request(q, req, el_ret);
1258 goto out_unlock;
1259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 }
1261
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001263 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001264 * This sync check and mask will be re-done in init_request_from_bio(),
1265 * but we need to set it earlier to expose the sync flag to the
1266 * rq allocator and io schedulers.
1267 */
1268 rw_flags = bio_data_dir(bio);
1269 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001270 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001271
1272 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001273 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001274 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001275 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001276 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001277
Nick Piggin450991b2005-06-28 20:45:13 -07001278 /*
1279 * After dropping the lock and possibly sleeping here, our request
1280 * may now be mergeable after it had proven unmergeable (above).
1281 * We don't worry about that case for efficiency. It won't happen
1282 * often, and the elevators are able to handle it.
1283 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001284 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
Jens Axboec7c22e42008-09-13 20:26:01 +02001286 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
Jens Axboe73c10102011-03-08 13:19:51 +01001287 bio_flagged(bio, BIO_CPU_AFFINE)) {
1288 req->cpu = blk_cpu_to_group(get_cpu());
1289 put_cpu();
1290 }
Tejun Heodd831002010-09-03 11:56:16 +02001291
Jens Axboe73c10102011-03-08 13:19:51 +01001292 plug = current->plug;
1293 if (plug && !sync) {
1294 if (!plug->should_sort && !list_empty(&plug->list)) {
1295 struct request *__rq;
1296
1297 __rq = list_entry_rq(plug->list.prev);
1298 if (__rq->q != q)
1299 plug->should_sort = 1;
1300 }
1301 /*
1302 * Debug flag, kill later
1303 */
1304 req->cmd_flags |= REQ_ON_PLUG;
1305 list_add_tail(&req->queuelist, &plug->list);
1306 drive_stat_acct(req, 1);
1307 } else {
1308 spin_lock_irq(q->queue_lock);
1309 add_acct_request(q, req, where);
1310 __blk_run_queue(q);
1311out_unlock:
1312 spin_unlock_irq(q->queue_lock);
1313 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316}
1317
1318/*
1319 * If bio->bi_dev is a partition, remap the location
1320 */
1321static inline void blk_partition_remap(struct bio *bio)
1322{
1323 struct block_device *bdev = bio->bi_bdev;
1324
Jens Axboebf2de6f2007-09-27 13:01:25 +02001325 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 bio->bi_sector += p->start_sect;
1329 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001330
Mike Snitzerd07335e2010-11-16 12:52:38 +01001331 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1332 bdev->bd_dev,
1333 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 }
1335}
1336
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337static void handle_bad_sector(struct bio *bio)
1338{
1339 char b[BDEVNAME_SIZE];
1340
1341 printk(KERN_INFO "attempt to access beyond end of device\n");
1342 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1343 bdevname(bio->bi_bdev, b),
1344 bio->bi_rw,
1345 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001346 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
1348 set_bit(BIO_EOF, &bio->bi_flags);
1349}
1350
Akinobu Mitac17bb492006-12-08 02:39:46 -08001351#ifdef CONFIG_FAIL_MAKE_REQUEST
1352
1353static DECLARE_FAULT_ATTR(fail_make_request);
1354
1355static int __init setup_fail_make_request(char *str)
1356{
1357 return setup_fault_attr(&fail_make_request, str);
1358}
1359__setup("fail_make_request=", setup_fail_make_request);
1360
1361static int should_fail_request(struct bio *bio)
1362{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001363 struct hd_struct *part = bio->bi_bdev->bd_part;
1364
1365 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001366 return should_fail(&fail_make_request, bio->bi_size);
1367
1368 return 0;
1369}
1370
1371static int __init fail_make_request_debugfs(void)
1372{
1373 return init_fault_attr_dentries(&fail_make_request,
1374 "fail_make_request");
1375}
1376
1377late_initcall(fail_make_request_debugfs);
1378
1379#else /* CONFIG_FAIL_MAKE_REQUEST */
1380
1381static inline int should_fail_request(struct bio *bio)
1382{
1383 return 0;
1384}
1385
1386#endif /* CONFIG_FAIL_MAKE_REQUEST */
1387
Jens Axboec07e2b42007-07-18 13:27:58 +02001388/*
1389 * Check whether this bio extends beyond the end of the device.
1390 */
1391static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1392{
1393 sector_t maxsector;
1394
1395 if (!nr_sectors)
1396 return 0;
1397
1398 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001399 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001400 if (maxsector) {
1401 sector_t sector = bio->bi_sector;
1402
1403 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1404 /*
1405 * This may well happen - the kernel calls bread()
1406 * without checking the size of the device, e.g., when
1407 * mounting a device.
1408 */
1409 handle_bad_sector(bio);
1410 return 1;
1411 }
1412 }
1413
1414 return 0;
1415}
1416
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001418 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 * @bio: The bio describing the location in memory and on the device.
1420 *
1421 * generic_make_request() is used to make I/O requests of block
1422 * devices. It is passed a &struct bio, which describes the I/O that needs
1423 * to be done.
1424 *
1425 * generic_make_request() does not return any status. The
1426 * success/failure status of the request, along with notification of
1427 * completion, is delivered asynchronously through the bio->bi_end_io
1428 * function described (one day) else where.
1429 *
1430 * The caller of generic_make_request must make sure that bi_io_vec
1431 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1432 * set to describe the device address, and the
1433 * bi_end_io and optionally bi_private are set to describe how
1434 * completion notification should be signaled.
1435 *
1436 * generic_make_request and the drivers it calls may use bi_next if this
1437 * bio happens to be merged with someone else, and may change bi_dev and
1438 * bi_sector for remaps as it sees fit. So the values of these fields
1439 * should NOT be depended on after the call to generic_make_request.
1440 */
Neil Brownd89d8792007-05-01 09:53:42 +02001441static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442{
Jens Axboe165125e2007-07-24 09:28:11 +02001443 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001444 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001446 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001447 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
1449 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
Jens Axboec07e2b42007-07-18 13:27:58 +02001451 if (bio_check_eod(bio, nr_sectors))
1452 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453
1454 /*
1455 * Resolve the mapping until finished. (drivers are
1456 * still free to implement/resolve their own stacking
1457 * by explicitly returning 0)
1458 *
1459 * NOTE: we don't repeat the blk_size check for each new device.
1460 * Stacking drivers are expected to know what they are doing.
1461 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001462 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001463 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 do {
1465 char b[BDEVNAME_SIZE];
1466
1467 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001468 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 printk(KERN_ERR
1470 "generic_make_request: Trying to access "
1471 "nonexistent block-device %s (%Lu)\n",
1472 bdevname(bio->bi_bdev, b),
1473 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001474 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 }
1476
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001477 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
Christoph Hellwig67efc922009-09-30 13:54:20 +02001478 nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001479 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001480 bdevname(bio->bi_bdev, b),
1481 bio_sectors(bio),
1482 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 goto end_io;
1484 }
1485
Nick Pigginfde6ad22005-06-23 00:08:53 -07001486 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 goto end_io;
1488
Akinobu Mitac17bb492006-12-08 02:39:46 -08001489 if (should_fail_request(bio))
1490 goto end_io;
1491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 /*
1493 * If this device has partitions, remap block n
1494 * of partition p to block n+start(p) of the disk.
1495 */
1496 blk_partition_remap(bio);
1497
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001498 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1499 goto end_io;
1500
NeilBrown5ddfe962006-10-30 22:07:21 -08001501 if (old_sector != -1)
Mike Snitzerd07335e2010-11-16 12:52:38 +01001502 trace_block_bio_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001503
NeilBrown5ddfe962006-10-30 22:07:21 -08001504 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001505 old_dev = bio->bi_bdev->bd_dev;
1506
Jens Axboec07e2b42007-07-18 13:27:58 +02001507 if (bio_check_eod(bio, nr_sectors))
1508 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001509
Tejun Heo1e879012010-09-03 11:56:17 +02001510 /*
1511 * Filter flush bio's early so that make_request based
1512 * drivers without flush support don't have to worry
1513 * about them.
1514 */
1515 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1516 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1517 if (!nr_sectors) {
1518 err = 0;
1519 goto end_io;
1520 }
1521 }
1522
Adrian Hunter8d57a982010-08-11 14:17:49 -07001523 if ((bio->bi_rw & REQ_DISCARD) &&
1524 (!blk_queue_discard(q) ||
1525 ((bio->bi_rw & REQ_SECURE) &&
1526 !blk_queue_secdiscard(q)))) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001527 err = -EOPNOTSUPP;
1528 goto end_io;
1529 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001530
Vivek Goyale43473b2010-09-15 17:06:35 -04001531 blk_throtl_bio(q, &bio);
1532
1533 /*
1534 * If bio = NULL, bio has been throttled and will be submitted
1535 * later.
1536 */
1537 if (!bio)
1538 break;
1539
Minchan Kim01edede2009-09-08 21:56:38 +02001540 trace_block_bio_queue(q, bio);
1541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 ret = q->make_request_fn(q, bio);
1543 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001544
1545 return;
1546
1547end_io:
1548 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549}
1550
Neil Brownd89d8792007-05-01 09:53:42 +02001551/*
1552 * We only want one ->make_request_fn to be active at a time,
1553 * else stack usage with stacked devices could be a problem.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001554 * So use current->bio_list to keep a list of requests
Neil Brownd89d8792007-05-01 09:53:42 +02001555 * submited by a make_request_fn function.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001556 * current->bio_list is also used as a flag to say if
Neil Brownd89d8792007-05-01 09:53:42 +02001557 * generic_make_request is currently active in this task or not.
1558 * If it is NULL, then no make_request is active. If it is non-NULL,
1559 * then a make_request is active, and new requests should be added
1560 * at the tail
1561 */
1562void generic_make_request(struct bio *bio)
1563{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001564 struct bio_list bio_list_on_stack;
1565
1566 if (current->bio_list) {
Neil Brownd89d8792007-05-01 09:53:42 +02001567 /* make_request is active */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001568 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001569 return;
1570 }
1571 /* following loop may be a bit non-obvious, and so deserves some
1572 * explanation.
1573 * Before entering the loop, bio->bi_next is NULL (as all callers
1574 * ensure that) so we have a list with a single bio.
1575 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001576 * we assign bio_list to a pointer to the bio_list_on_stack,
1577 * thus initialising the bio_list of new bios to be
Neil Brownd89d8792007-05-01 09:53:42 +02001578 * added. __generic_make_request may indeed add some more bios
1579 * through a recursive call to generic_make_request. If it
1580 * did, we find a non-NULL value in bio_list and re-enter the loop
1581 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001582 * of the top of the list (no pretending) and so remove it from
1583 * bio_list, and call into __generic_make_request again.
Neil Brownd89d8792007-05-01 09:53:42 +02001584 *
1585 * The loop was structured like this to make only one call to
1586 * __generic_make_request (which is important as it is large and
1587 * inlined) and to keep the structure simple.
1588 */
1589 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001590 bio_list_init(&bio_list_on_stack);
1591 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001592 do {
Neil Brownd89d8792007-05-01 09:53:42 +02001593 __generic_make_request(bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001594 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001595 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001596 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001597}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598EXPORT_SYMBOL(generic_make_request);
1599
1600/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001601 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1603 * @bio: The &struct bio which describes the I/O
1604 *
1605 * submit_bio() is very similar in purpose to generic_make_request(), and
1606 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001607 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 *
1609 */
1610void submit_bio(int rw, struct bio *bio)
1611{
1612 int count = bio_sectors(bio);
1613
Jens Axboe22e2c502005-06-27 10:55:12 +02001614 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615
Jens Axboebf2de6f2007-09-27 13:01:25 +02001616 /*
1617 * If it's a regular read/write or a barrier with data attached,
1618 * go through the normal accounting stuff before submission.
1619 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001620 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001621 if (rw & WRITE) {
1622 count_vm_events(PGPGOUT, count);
1623 } else {
1624 task_io_account_read(bio->bi_size);
1625 count_vm_events(PGPGIN, count);
1626 }
1627
1628 if (unlikely(block_dump)) {
1629 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001630 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001631 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001632 (rw & WRITE) ? "WRITE" : "READ",
1633 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001634 bdevname(bio->bi_bdev, b),
1635 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 }
1638
1639 generic_make_request(bio);
1640}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641EXPORT_SYMBOL(submit_bio);
1642
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001643/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001644 * blk_rq_check_limits - Helper function to check a request for the queue limit
1645 * @q: the queue
1646 * @rq: the request being checked
1647 *
1648 * Description:
1649 * @rq may have been made based on weaker limitations of upper-level queues
1650 * in request stacking drivers, and it may violate the limitation of @q.
1651 * Since the block layer and the underlying device driver trust @rq
1652 * after it is inserted to @q, it should be checked against @q before
1653 * the insertion using this generic function.
1654 *
1655 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001656 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001657 * Request stacking drivers like request-based dm may change the queue
1658 * limits while requests are in the queue (e.g. dm's table swapping).
1659 * Such request stacking drivers should check those requests agaist
1660 * the new queue limits again when they dispatch those requests,
1661 * although such checkings are also done against the old queue limits
1662 * when submitting requests.
1663 */
1664int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1665{
ike Snitzer33839772010-08-08 12:11:33 -04001666 if (rq->cmd_flags & REQ_DISCARD)
1667 return 0;
1668
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001669 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1670 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001671 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1672 return -EIO;
1673 }
1674
1675 /*
1676 * queue's settings related to segment counting like q->bounce_pfn
1677 * may differ from that of other stacking queues.
1678 * Recalculate it to check the request correctly on this queue's
1679 * limitation.
1680 */
1681 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001682 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001683 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1684 return -EIO;
1685 }
1686
1687 return 0;
1688}
1689EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1690
1691/**
1692 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1693 * @q: the queue to submit the request
1694 * @rq: the request being queued
1695 */
1696int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1697{
1698 unsigned long flags;
1699
1700 if (blk_rq_check_limits(q, rq))
1701 return -EIO;
1702
1703#ifdef CONFIG_FAIL_MAKE_REQUEST
1704 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1705 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1706 return -EIO;
1707#endif
1708
1709 spin_lock_irqsave(q->queue_lock, flags);
1710
1711 /*
1712 * Submitting request must be dequeued before calling this function
1713 * because it will be linked to another request_queue
1714 */
1715 BUG_ON(blk_queued_rq(rq));
1716
Jens Axboe73c10102011-03-08 13:19:51 +01001717 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001718 spin_unlock_irqrestore(q->queue_lock, flags);
1719
1720 return 0;
1721}
1722EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1723
Tejun Heo80a761f2009-07-03 17:48:17 +09001724/**
1725 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1726 * @rq: request to examine
1727 *
1728 * Description:
1729 * A request could be merge of IOs which require different failure
1730 * handling. This function determines the number of bytes which
1731 * can be failed from the beginning of the request without
1732 * crossing into area which need to be retried further.
1733 *
1734 * Return:
1735 * The number of bytes to fail.
1736 *
1737 * Context:
1738 * queue_lock must be held.
1739 */
1740unsigned int blk_rq_err_bytes(const struct request *rq)
1741{
1742 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1743 unsigned int bytes = 0;
1744 struct bio *bio;
1745
1746 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1747 return blk_rq_bytes(rq);
1748
1749 /*
1750 * Currently the only 'mixing' which can happen is between
1751 * different fastfail types. We can safely fail portions
1752 * which have all the failfast bits that the first one has -
1753 * the ones which are at least as eager to fail as the first
1754 * one.
1755 */
1756 for (bio = rq->bio; bio; bio = bio->bi_next) {
1757 if ((bio->bi_rw & ff) != ff)
1758 break;
1759 bytes += bio->bi_size;
1760 }
1761
1762 /* this could lead to infinite loop */
1763 BUG_ON(blk_rq_bytes(rq) && !bytes);
1764 return bytes;
1765}
1766EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1767
Jens Axboebc58ba92009-01-23 10:54:44 +01001768static void blk_account_io_completion(struct request *req, unsigned int bytes)
1769{
Jens Axboec2553b52009-04-24 08:10:11 +02001770 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001771 const int rw = rq_data_dir(req);
1772 struct hd_struct *part;
1773 int cpu;
1774
1775 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001776 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001777 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1778 part_stat_unlock();
1779 }
1780}
1781
1782static void blk_account_io_done(struct request *req)
1783{
Jens Axboebc58ba92009-01-23 10:54:44 +01001784 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001785 * Account IO completion. flush_rq isn't accounted as a
1786 * normal IO on queueing nor completion. Accounting the
1787 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001788 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001789 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001790 unsigned long duration = jiffies - req->start_time;
1791 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
1798 part_stat_inc(cpu, part, ios[rw]);
1799 part_stat_add(cpu, part, ticks[rw], duration);
1800 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001801 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001802
Jens Axboe6c23a962011-01-07 08:43:37 +01001803 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001804 part_stat_unlock();
1805 }
1806}
1807
Tejun Heo53a08802008-12-03 12:41:26 +01001808/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001809 * blk_peek_request - peek at the top of a request queue
1810 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001811 *
1812 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001813 * Return the request at the top of @q. The returned request
1814 * should be started using blk_start_request() before LLD starts
1815 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001816 *
1817 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001818 * Pointer to the request at the top of @q if available. Null
1819 * otherwise.
1820 *
1821 * Context:
1822 * queue_lock must be held.
1823 */
1824struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001825{
1826 struct request *rq;
1827 int ret;
1828
1829 while ((rq = __elv_next_request(q)) != NULL) {
1830 if (!(rq->cmd_flags & REQ_STARTED)) {
1831 /*
1832 * This is the first time the device driver
1833 * sees this request (possibly after
1834 * requeueing). Notify IO scheduler.
1835 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001836 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001837 elv_activate_rq(q, rq);
1838
1839 /*
1840 * just mark as started even if we don't start
1841 * it, a request that has been delayed should
1842 * not be passed by new incoming requests
1843 */
1844 rq->cmd_flags |= REQ_STARTED;
1845 trace_block_rq_issue(q, rq);
1846 }
1847
1848 if (!q->boundary_rq || q->boundary_rq == rq) {
1849 q->end_sector = rq_end_sector(rq);
1850 q->boundary_rq = NULL;
1851 }
1852
1853 if (rq->cmd_flags & REQ_DONTPREP)
1854 break;
1855
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001856 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001857 /*
1858 * make sure space for the drain appears we
1859 * know we can do this because max_hw_segments
1860 * has been adjusted to be one fewer than the
1861 * device can handle
1862 */
1863 rq->nr_phys_segments++;
1864 }
1865
1866 if (!q->prep_rq_fn)
1867 break;
1868
1869 ret = q->prep_rq_fn(q, rq);
1870 if (ret == BLKPREP_OK) {
1871 break;
1872 } else if (ret == BLKPREP_DEFER) {
1873 /*
1874 * the request may have been (partially) prepped.
1875 * we need to keep this request in the front to
1876 * avoid resource deadlock. REQ_STARTED will
1877 * prevent other fs requests from passing this one.
1878 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001879 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001880 !(rq->cmd_flags & REQ_DONTPREP)) {
1881 /*
1882 * remove the space for the drain we added
1883 * so that we don't add it again
1884 */
1885 --rq->nr_phys_segments;
1886 }
1887
1888 rq = NULL;
1889 break;
1890 } else if (ret == BLKPREP_KILL) {
1891 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001892 /*
1893 * Mark this request as started so we don't trigger
1894 * any debug logic in the end I/O path.
1895 */
1896 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001897 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001898 } else {
1899 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1900 break;
1901 }
1902 }
1903
1904 return rq;
1905}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001906EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001907
Tejun Heo9934c8c2009-05-08 11:54:16 +09001908void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001909{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001910 struct request_queue *q = rq->q;
1911
Tejun Heo158dbda2009-04-23 11:05:18 +09001912 BUG_ON(list_empty(&rq->queuelist));
1913 BUG_ON(ELV_ON_HASH(rq));
1914
1915 list_del_init(&rq->queuelist);
1916
1917 /*
1918 * the time frame between a request being removed from the lists
1919 * and to it is freed is accounted as io that is in progress at
1920 * the driver side.
1921 */
Divyesh Shah91952912010-04-01 15:01:41 -07001922 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001923 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001924 set_io_start_time_ns(rq);
1925 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001926}
1927
Tejun Heo5efccd12009-04-23 11:05:18 +09001928/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001929 * blk_start_request - start request processing on the driver
1930 * @req: request to dequeue
1931 *
1932 * Description:
1933 * Dequeue @req and start timeout timer on it. This hands off the
1934 * request to the driver.
1935 *
1936 * Block internal functions which don't want to start timer should
1937 * call blk_dequeue_request().
1938 *
1939 * Context:
1940 * queue_lock must be held.
1941 */
1942void blk_start_request(struct request *req)
1943{
1944 blk_dequeue_request(req);
1945
1946 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001947 * We are now handing the request to the hardware, initialize
1948 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001949 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001950 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001951 if (unlikely(blk_bidi_rq(req)))
1952 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1953
Tejun Heo9934c8c2009-05-08 11:54:16 +09001954 blk_add_timer(req);
1955}
1956EXPORT_SYMBOL(blk_start_request);
1957
1958/**
1959 * blk_fetch_request - fetch a request from a request queue
1960 * @q: request queue to fetch a request from
1961 *
1962 * Description:
1963 * Return the request at the top of @q. The request is started on
1964 * return and LLD can start processing it immediately.
1965 *
1966 * Return:
1967 * Pointer to the request at the top of @q if available. Null
1968 * otherwise.
1969 *
1970 * Context:
1971 * queue_lock must be held.
1972 */
1973struct request *blk_fetch_request(struct request_queue *q)
1974{
1975 struct request *rq;
1976
1977 rq = blk_peek_request(q);
1978 if (rq)
1979 blk_start_request(rq);
1980 return rq;
1981}
1982EXPORT_SYMBOL(blk_fetch_request);
1983
1984/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001985 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001986 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001987 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001988 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001989 *
1990 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001991 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1992 * the request structure even if @req doesn't have leftover.
1993 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09001994 *
1995 * This special helper function is only for request stacking drivers
1996 * (e.g. request-based dm) so that they can handle partial completion.
1997 * Actual device drivers should use blk_end_request instead.
1998 *
1999 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2000 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002001 *
2002 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002003 * %false - this request doesn't have any more data
2004 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002005 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002006bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002008 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 struct bio *bio;
2010
Tejun Heo2e60e022009-04-23 11:05:18 +09002011 if (!req->bio)
2012 return false;
2013
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002014 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002017 * For fs requests, rq is just carrier of independent bio's
2018 * and each partial completion should be handled separately.
2019 * Reset per-request error on each partial completion.
2020 *
2021 * TODO: tj: This is too subtle. It would be better to let
2022 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002024 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 req->errors = 0;
2026
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002027 if (error && req->cmd_type == REQ_TYPE_FS &&
2028 !(req->cmd_flags & REQ_QUIET)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01002029 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 req->rq_disk ? req->rq_disk->disk_name : "?",
Tejun Heo83096eb2009-05-07 22:24:39 +09002031 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 }
2033
Jens Axboebc58ba92009-01-23 10:54:44 +01002034 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002035
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 total_bytes = bio_nbytes = 0;
2037 while ((bio = req->bio) != NULL) {
2038 int nbytes;
2039
2040 if (nr_bytes >= bio->bi_size) {
2041 req->bio = bio->bi_next;
2042 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002043 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 next_idx = 0;
2045 bio_nbytes = 0;
2046 } else {
2047 int idx = bio->bi_idx + next_idx;
2048
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002049 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002051 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002052 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 break;
2054 }
2055
2056 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2057 BIO_BUG_ON(nbytes > bio->bi_size);
2058
2059 /*
2060 * not a complete bvec done
2061 */
2062 if (unlikely(nbytes > nr_bytes)) {
2063 bio_nbytes += nr_bytes;
2064 total_bytes += nr_bytes;
2065 break;
2066 }
2067
2068 /*
2069 * advance to the next vector
2070 */
2071 next_idx++;
2072 bio_nbytes += nbytes;
2073 }
2074
2075 total_bytes += nbytes;
2076 nr_bytes -= nbytes;
2077
Jens Axboe6728cb02008-01-31 13:03:55 +01002078 bio = req->bio;
2079 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 /*
2081 * end more in this run, or just return 'not-done'
2082 */
2083 if (unlikely(nr_bytes <= 0))
2084 break;
2085 }
2086 }
2087
2088 /*
2089 * completely done
2090 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002091 if (!req->bio) {
2092 /*
2093 * Reset counters so that the request stacking driver
2094 * can find how many bytes remain in the request
2095 * later.
2096 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002097 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002098 return false;
2099 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
2101 /*
2102 * if the request wasn't completed, update state
2103 */
2104 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002105 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 bio->bi_idx += next_idx;
2107 bio_iovec(bio)->bv_offset += nr_bytes;
2108 bio_iovec(bio)->bv_len -= nr_bytes;
2109 }
2110
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002111 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002112 req->buffer = bio_data(req->bio);
2113
2114 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002115 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002116 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002117
Tejun Heo80a761f2009-07-03 17:48:17 +09002118 /* mixed attributes always follow the first bio */
2119 if (req->cmd_flags & REQ_MIXED_MERGE) {
2120 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2121 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2122 }
2123
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002124 /*
2125 * If total number of sectors is less than the first segment
2126 * size, something has gone terribly wrong.
2127 */
2128 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2129 printk(KERN_ERR "blk: request botched\n");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002130 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002131 }
2132
2133 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002135
Tejun Heo2e60e022009-04-23 11:05:18 +09002136 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137}
Tejun Heo2e60e022009-04-23 11:05:18 +09002138EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139
Tejun Heo2e60e022009-04-23 11:05:18 +09002140static bool blk_update_bidi_request(struct request *rq, int error,
2141 unsigned int nr_bytes,
2142 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002143{
Tejun Heo2e60e022009-04-23 11:05:18 +09002144 if (blk_update_request(rq, error, nr_bytes))
2145 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002146
Tejun Heo2e60e022009-04-23 11:05:18 +09002147 /* Bidi request must be completed as a whole */
2148 if (unlikely(blk_bidi_rq(rq)) &&
2149 blk_update_request(rq->next_rq, error, bidi_bytes))
2150 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002151
Jens Axboee2e1a142010-06-09 10:42:09 +02002152 if (blk_queue_add_random(rq->q))
2153 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002154
2155 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156}
2157
James Bottomley28018c22010-07-01 19:49:17 +09002158/**
2159 * blk_unprep_request - unprepare a request
2160 * @req: the request
2161 *
2162 * This function makes a request ready for complete resubmission (or
2163 * completion). It happens only after all error handling is complete,
2164 * so represents the appropriate moment to deallocate any resources
2165 * that were allocated to the request in the prep_rq_fn. The queue
2166 * lock is held when calling this.
2167 */
2168void blk_unprep_request(struct request *req)
2169{
2170 struct request_queue *q = req->q;
2171
2172 req->cmd_flags &= ~REQ_DONTPREP;
2173 if (q->unprep_rq_fn)
2174 q->unprep_rq_fn(q, req);
2175}
2176EXPORT_SYMBOL_GPL(blk_unprep_request);
2177
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178/*
2179 * queue lock must be held
2180 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002181static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002183 if (blk_rq_tagged(req))
2184 blk_queue_end_tag(req->q, req);
2185
James Bottomleyba396a62009-05-27 14:17:08 +02002186 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002188 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002189 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190
Mike Andersone78042e2008-10-30 02:16:20 -07002191 blk_delete_timer(req);
2192
James Bottomley28018c22010-07-01 19:49:17 +09002193 if (req->cmd_flags & REQ_DONTPREP)
2194 blk_unprep_request(req);
2195
2196
Jens Axboebc58ba92009-01-23 10:54:44 +01002197 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002198
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002200 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002201 else {
2202 if (blk_bidi_rq(req))
2203 __blk_put_request(req->next_rq->q, req->next_rq);
2204
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 }
2207}
2208
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002209/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002210 * blk_end_bidi_request - Complete a bidi request
2211 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002212 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002213 * @nr_bytes: number of bytes to complete @rq
2214 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002215 *
2216 * Description:
2217 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002218 * Drivers that supports bidi can safely call this member for any
2219 * type of request, bidi or uni. In the later case @bidi_bytes is
2220 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002221 *
2222 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002223 * %false - we are done with this request
2224 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002225 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002226static bool blk_end_bidi_request(struct request *rq, int error,
2227 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002228{
2229 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002230 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002231
Tejun Heo2e60e022009-04-23 11:05:18 +09002232 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2233 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002234
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002235 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002236 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002237 spin_unlock_irqrestore(q->queue_lock, flags);
2238
Tejun Heo2e60e022009-04-23 11:05:18 +09002239 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002240}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002241
2242/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002243 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2244 * @rq: the request to complete
2245 * @error: %0 for success, < %0 for error
2246 * @nr_bytes: number of bytes to complete @rq
2247 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002248 *
2249 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002250 * Identical to blk_end_bidi_request() except that queue lock is
2251 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002252 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002253 * Return:
2254 * %false - we are done with this request
2255 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002256 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002257static bool __blk_end_bidi_request(struct request *rq, int error,
2258 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002259{
Tejun Heo2e60e022009-04-23 11:05:18 +09002260 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2261 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002262
Tejun Heo2e60e022009-04-23 11:05:18 +09002263 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002264
Tejun Heo2e60e022009-04-23 11:05:18 +09002265 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002266}
2267
2268/**
2269 * blk_end_request - Helper function for drivers to complete the request.
2270 * @rq: the request being processed
2271 * @error: %0 for success, < %0 for error
2272 * @nr_bytes: number of bytes to complete
2273 *
2274 * Description:
2275 * Ends I/O on a number of bytes attached to @rq.
2276 * If @rq has leftover, sets it up for the next range of segments.
2277 *
2278 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002279 * %false - we are done with this request
2280 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002281 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002282bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002283{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002284 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002285}
Jens Axboe56ad1742009-07-28 22:11:24 +02002286EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002287
2288/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002289 * blk_end_request_all - Helper function for drives to finish the request.
2290 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002291 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002292 *
2293 * Description:
2294 * Completely finish @rq.
2295 */
2296void blk_end_request_all(struct request *rq, int error)
2297{
2298 bool pending;
2299 unsigned int bidi_bytes = 0;
2300
2301 if (unlikely(blk_bidi_rq(rq)))
2302 bidi_bytes = blk_rq_bytes(rq->next_rq);
2303
2304 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2305 BUG_ON(pending);
2306}
Jens Axboe56ad1742009-07-28 22:11:24 +02002307EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002308
2309/**
2310 * blk_end_request_cur - Helper function to finish the current request chunk.
2311 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002312 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002313 *
2314 * Description:
2315 * Complete the current consecutively mapped chunk from @rq.
2316 *
2317 * Return:
2318 * %false - we are done with this request
2319 * %true - still buffers pending for this request
2320 */
2321bool blk_end_request_cur(struct request *rq, int error)
2322{
2323 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2324}
Jens Axboe56ad1742009-07-28 22:11:24 +02002325EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002326
2327/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002328 * blk_end_request_err - Finish a request till the next failure boundary.
2329 * @rq: the request to finish till the next failure boundary for
2330 * @error: must be negative errno
2331 *
2332 * Description:
2333 * Complete @rq till the next failure boundary.
2334 *
2335 * Return:
2336 * %false - we are done with this request
2337 * %true - still buffers pending for this request
2338 */
2339bool blk_end_request_err(struct request *rq, int error)
2340{
2341 WARN_ON(error >= 0);
2342 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2343}
2344EXPORT_SYMBOL_GPL(blk_end_request_err);
2345
2346/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002347 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002348 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002349 * @error: %0 for success, < %0 for error
2350 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002351 *
2352 * Description:
2353 * Must be called with queue lock held unlike blk_end_request().
2354 *
2355 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002356 * %false - we are done with this request
2357 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002358 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002359bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002360{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002361 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002362}
Jens Axboe56ad1742009-07-28 22:11:24 +02002363EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002364
2365/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002366 * __blk_end_request_all - Helper function for drives to finish the request.
2367 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002368 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002369 *
2370 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002371 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002372 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002373void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002374{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002375 bool pending;
2376 unsigned int bidi_bytes = 0;
2377
2378 if (unlikely(blk_bidi_rq(rq)))
2379 bidi_bytes = blk_rq_bytes(rq->next_rq);
2380
2381 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2382 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002383}
Jens Axboe56ad1742009-07-28 22:11:24 +02002384EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002385
2386/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002387 * __blk_end_request_cur - Helper function to finish the current request chunk.
2388 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002389 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002390 *
2391 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002392 * Complete the current consecutively mapped chunk from @rq. Must
2393 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002394 *
2395 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002396 * %false - we are done with this request
2397 * %true - still buffers pending for this request
2398 */
2399bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002400{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002401 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002402}
Jens Axboe56ad1742009-07-28 22:11:24 +02002403EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002404
Tejun Heo80a761f2009-07-03 17:48:17 +09002405/**
2406 * __blk_end_request_err - Finish a request till the next failure boundary.
2407 * @rq: the request to finish till the next failure boundary for
2408 * @error: must be negative errno
2409 *
2410 * Description:
2411 * Complete @rq till the next failure boundary. Must be called
2412 * with queue lock held.
2413 *
2414 * Return:
2415 * %false - we are done with this request
2416 * %true - still buffers pending for this request
2417 */
2418bool __blk_end_request_err(struct request *rq, int error)
2419{
2420 WARN_ON(error >= 0);
2421 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2422}
2423EXPORT_SYMBOL_GPL(__blk_end_request_err);
2424
Jens Axboe86db1e22008-01-29 14:53:40 +01002425void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2426 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002428 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002429 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
David Woodhousefb2dce82008-08-05 18:01:53 +01002431 if (bio_has_data(bio)) {
2432 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002433 rq->buffer = bio_data(bio);
2434 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002435 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437
NeilBrown66846572007-08-16 13:31:28 +02002438 if (bio->bi_bdev)
2439 rq->rq_disk = bio->bi_bdev->bd_disk;
2440}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002442#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2443/**
2444 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2445 * @rq: the request to be flushed
2446 *
2447 * Description:
2448 * Flush all pages in @rq.
2449 */
2450void rq_flush_dcache_pages(struct request *rq)
2451{
2452 struct req_iterator iter;
2453 struct bio_vec *bvec;
2454
2455 rq_for_each_segment(bvec, rq, iter)
2456 flush_dcache_page(bvec->bv_page);
2457}
2458EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2459#endif
2460
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002461/**
2462 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2463 * @q : the queue of the device being checked
2464 *
2465 * Description:
2466 * Check if underlying low-level drivers of a device are busy.
2467 * If the drivers want to export their busy state, they must set own
2468 * exporting function using blk_queue_lld_busy() first.
2469 *
2470 * Basically, this function is used only by request stacking drivers
2471 * to stop dispatching requests to underlying devices when underlying
2472 * devices are busy. This behavior helps more I/O merging on the queue
2473 * of the request stacking driver and prevents I/O throughput regression
2474 * on burst I/O load.
2475 *
2476 * Return:
2477 * 0 - Not busy (The request stacking driver should dispatch request)
2478 * 1 - Busy (The request stacking driver should stop dispatching request)
2479 */
2480int blk_lld_busy(struct request_queue *q)
2481{
2482 if (q->lld_busy_fn)
2483 return q->lld_busy_fn(q);
2484
2485 return 0;
2486}
2487EXPORT_SYMBOL_GPL(blk_lld_busy);
2488
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002489/**
2490 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2491 * @rq: the clone request to be cleaned up
2492 *
2493 * Description:
2494 * Free all bios in @rq for a cloned request.
2495 */
2496void blk_rq_unprep_clone(struct request *rq)
2497{
2498 struct bio *bio;
2499
2500 while ((bio = rq->bio) != NULL) {
2501 rq->bio = bio->bi_next;
2502
2503 bio_put(bio);
2504 }
2505}
2506EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2507
2508/*
2509 * Copy attributes of the original request to the clone request.
2510 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2511 */
2512static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2513{
2514 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002515 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002516 dst->cmd_type = src->cmd_type;
2517 dst->__sector = blk_rq_pos(src);
2518 dst->__data_len = blk_rq_bytes(src);
2519 dst->nr_phys_segments = src->nr_phys_segments;
2520 dst->ioprio = src->ioprio;
2521 dst->extra_len = src->extra_len;
2522}
2523
2524/**
2525 * blk_rq_prep_clone - Helper function to setup clone request
2526 * @rq: the request to be setup
2527 * @rq_src: original request to be cloned
2528 * @bs: bio_set that bios for clone are allocated from
2529 * @gfp_mask: memory allocation mask for bio
2530 * @bio_ctr: setup function to be called for each clone bio.
2531 * Returns %0 for success, non %0 for failure.
2532 * @data: private data to be passed to @bio_ctr
2533 *
2534 * Description:
2535 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2536 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2537 * are not copied, and copying such parts is the caller's responsibility.
2538 * Also, pages which the original bios are pointing to are not copied
2539 * and the cloned bios just point same pages.
2540 * So cloned bios must be completed before original bios, which means
2541 * the caller must complete @rq before @rq_src.
2542 */
2543int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2544 struct bio_set *bs, gfp_t gfp_mask,
2545 int (*bio_ctr)(struct bio *, struct bio *, void *),
2546 void *data)
2547{
2548 struct bio *bio, *bio_src;
2549
2550 if (!bs)
2551 bs = fs_bio_set;
2552
2553 blk_rq_init(NULL, rq);
2554
2555 __rq_for_each_bio(bio_src, rq_src) {
2556 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2557 if (!bio)
2558 goto free_and_out;
2559
2560 __bio_clone(bio, bio_src);
2561
2562 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002563 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002564 goto free_and_out;
2565
2566 if (bio_ctr && bio_ctr(bio, bio_src, data))
2567 goto free_and_out;
2568
2569 if (rq->bio) {
2570 rq->biotail->bi_next = bio;
2571 rq->biotail = bio;
2572 } else
2573 rq->bio = rq->biotail = bio;
2574 }
2575
2576 __blk_rq_prep_clone(rq, rq_src);
2577
2578 return 0;
2579
2580free_and_out:
2581 if (bio)
2582 bio_free(bio, bs);
2583 blk_rq_unprep_clone(rq);
2584
2585 return -ENOMEM;
2586}
2587EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2588
Jens Axboe18887ad2008-07-28 13:08:45 +02002589int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590{
2591 return queue_work(kblockd_workqueue, work);
2592}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593EXPORT_SYMBOL(kblockd_schedule_work);
2594
Vivek Goyale43473b2010-09-15 17:06:35 -04002595int kblockd_schedule_delayed_work(struct request_queue *q,
2596 struct delayed_work *dwork, unsigned long delay)
2597{
2598 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2599}
2600EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2601
Jens Axboe73c10102011-03-08 13:19:51 +01002602#define PLUG_MAGIC 0x91827364
2603
2604void blk_start_plug(struct blk_plug *plug)
2605{
2606 struct task_struct *tsk = current;
2607
2608 plug->magic = PLUG_MAGIC;
2609 INIT_LIST_HEAD(&plug->list);
2610 plug->should_sort = 0;
2611
2612 /*
2613 * If this is a nested plug, don't actually assign it. It will be
2614 * flushed on its own.
2615 */
2616 if (!tsk->plug) {
2617 /*
2618 * Store ordering should not be needed here, since a potential
2619 * preempt will imply a full memory barrier
2620 */
2621 tsk->plug = plug;
2622 }
2623}
2624EXPORT_SYMBOL(blk_start_plug);
2625
2626static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2627{
2628 struct request *rqa = container_of(a, struct request, queuelist);
2629 struct request *rqb = container_of(b, struct request, queuelist);
2630
2631 return !(rqa->q == rqb->q);
2632}
2633
2634static void flush_plug_list(struct blk_plug *plug)
2635{
2636 struct request_queue *q;
2637 unsigned long flags;
2638 struct request *rq;
2639
2640 BUG_ON(plug->magic != PLUG_MAGIC);
2641
2642 if (list_empty(&plug->list))
2643 return;
2644
2645 if (plug->should_sort)
2646 list_sort(NULL, &plug->list, plug_rq_cmp);
2647
2648 q = NULL;
2649 local_irq_save(flags);
2650 while (!list_empty(&plug->list)) {
2651 rq = list_entry_rq(plug->list.next);
2652 list_del_init(&rq->queuelist);
2653 BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2654 BUG_ON(!rq->q);
2655 if (rq->q != q) {
2656 if (q) {
2657 __blk_run_queue(q);
2658 spin_unlock(q->queue_lock);
2659 }
2660 q = rq->q;
2661 spin_lock(q->queue_lock);
2662 }
2663 rq->cmd_flags &= ~REQ_ON_PLUG;
2664
2665 /*
2666 * rq is already accounted, so use raw insert
2667 */
Jens Axboe7eaceac2011-03-10 08:52:07 +01002668 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT);
Jens Axboe73c10102011-03-08 13:19:51 +01002669 }
2670
2671 if (q) {
2672 __blk_run_queue(q);
2673 spin_unlock(q->queue_lock);
2674 }
2675
2676 BUG_ON(!list_empty(&plug->list));
2677 local_irq_restore(flags);
2678}
2679
2680static void __blk_finish_plug(struct task_struct *tsk, struct blk_plug *plug)
2681{
2682 flush_plug_list(plug);
2683
2684 if (plug == tsk->plug)
2685 tsk->plug = NULL;
2686}
2687
2688void blk_finish_plug(struct blk_plug *plug)
2689{
2690 if (plug)
2691 __blk_finish_plug(current, plug);
2692}
2693EXPORT_SYMBOL(blk_finish_plug);
2694
2695void __blk_flush_plug(struct task_struct *tsk, struct blk_plug *plug)
2696{
2697 __blk_finish_plug(tsk, plug);
2698 tsk->plug = plug;
2699}
2700EXPORT_SYMBOL(__blk_flush_plug);
2701
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702int __init blk_dev_init(void)
2703{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002704 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2705 sizeof(((struct request *)0)->cmd_flags));
2706
Tejun Heo89b90be2011-01-03 15:01:47 +01002707 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2708 kblockd_workqueue = alloc_workqueue("kblockd",
2709 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 if (!kblockd_workqueue)
2711 panic("Failed to create kblockd\n");
2712
2713 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002714 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715
Jens Axboe8324aa92008-01-29 14:51:59 +01002716 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002717 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 return 0;
2720}