blob: 15de223c7f9371a9da852825ea8857789d94ae70 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd42006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
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>
Tejun Heoe3c78ca2011-10-19 14:32:38 +020031#include <linux/delay.h>
Li Zefan55782132009-06-09 13:43:05 +080032
33#define CREATE_TRACE_POINTS
34#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Jens Axboe8324aa92008-01-29 14:51:59 +010036#include "blk.h"
37
Mike Snitzerd07335e2010-11-16 12:52:38 +010038EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020039EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Li Zefan55782132009-06-09 13:43:05 +080040EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
43 * For the allocated request tables
44 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010045static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47/*
48 * For queue allocation
49 */
Jens Axboe6728cb02008-01-31 13:03:55 +010050struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 * Controlling structure to kblockd
54 */
Jens Axboeff856ba2006-01-09 16:02:34 +010055static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Jens Axboe26b82562008-01-29 13:54:41 +010057static void drive_stat_acct(struct request *rq, int new_io)
58{
Jens Axboe28f13702008-05-07 10:15:46 +020059 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010060 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090061 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010062
Jens Axboec2553b52009-04-24 08:10:11 +020063 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010064 return;
65
Tejun Heo074a7ac2008-08-25 19:56:14 +090066 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +090067
Jerome Marchand09e099d2011-01-05 16:57:38 +010068 if (!new_io) {
69 part = rq->part;
Tejun Heo074a7ac2008-08-25 19:56:14 +090070 part_stat_inc(cpu, part, merges[rw]);
Jerome Marchand09e099d2011-01-05 16:57:38 +010071 } else {
72 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Jens Axboe6c23a962011-01-07 08:43:37 +010073 if (!hd_struct_try_get(part)) {
Jerome Marchand09e099d2011-01-05 16:57:38 +010074 /*
75 * The partition is already being removed,
76 * the request will be accounted on the disk only
77 *
78 * We take a reference on disk->part0 although that
79 * partition will never be deleted, so we can treat
80 * it as any other partition.
81 */
82 part = &rq->rq_disk->part0;
Jens Axboe6c23a962011-01-07 08:43:37 +010083 hd_struct_get(part);
Jerome Marchand09e099d2011-01-05 16:57:38 +010084 }
Tejun Heo074a7ac2008-08-25 19:56:14 +090085 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020086 part_inc_in_flight(part, rw);
Jerome Marchand09e099d2011-01-05 16:57:38 +010087 rq->part = part;
Jens Axboe26b82562008-01-29 13:54:41 +010088 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020089
Tejun Heo074a7ac2008-08-25 19:56:14 +090090 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010091}
92
Jens Axboe8324aa92008-01-29 14:51:59 +010093void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094{
95 int nr;
96
97 nr = q->nr_requests - (q->nr_requests / 8) + 1;
98 if (nr > q->nr_requests)
99 nr = q->nr_requests;
100 q->nr_congestion_on = nr;
101
102 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
103 if (nr < 1)
104 nr = 1;
105 q->nr_congestion_off = nr;
106}
107
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108/**
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
110 * @bdev: device
111 *
112 * Locates the passed device's request queue and returns the address of its
113 * backing_dev_info
114 *
115 * Will return NULL if the request queue cannot be located.
116 */
117struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
118{
119 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200120 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
122 if (q)
123 ret = &q->backing_dev_info;
124 return ret;
125}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126EXPORT_SYMBOL(blk_get_backing_dev_info);
127
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200128void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200130 memset(rq, 0, sizeof(*rq));
131
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700133 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200134 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100135 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900136 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200137 INIT_HLIST_NODE(&rq->hash);
138 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200139 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800140 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100141 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100142 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900143 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700144 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100145 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200147EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
NeilBrown5bb23a62007-09-27 12:46:13 +0200149static void req_bio_endio(struct request *rq, struct bio *bio,
150 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100151{
Tejun Heo143a87f2011-01-25 12:43:52 +0100152 if (error)
153 clear_bit(BIO_UPTODATE, &bio->bi_flags);
154 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
155 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100156
Tejun Heo143a87f2011-01-25 12:43:52 +0100157 if (unlikely(nbytes > bio->bi_size)) {
158 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
159 __func__, nbytes, bio->bi_size);
160 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +0200161 }
Tejun Heo143a87f2011-01-25 12:43:52 +0100162
163 if (unlikely(rq->cmd_flags & REQ_QUIET))
164 set_bit(BIO_QUIET, &bio->bi_flags);
165
166 bio->bi_size -= nbytes;
167 bio->bi_sector += (nbytes >> 9);
168
169 if (bio_integrity(bio))
170 bio_integrity_advance(bio, nbytes);
171
172 /* don't actually finish bio if it's part of flush sequence */
173 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
174 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100175}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177void blk_dump_rq_flags(struct request *rq, char *msg)
178{
179 int bit;
180
Jens Axboe6728cb02008-01-31 13:03:55 +0100181 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200182 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
183 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Tejun Heo83096eb2009-05-07 22:24:39 +0900185 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
186 (unsigned long long)blk_rq_pos(rq),
187 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900188 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900189 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200191 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100192 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200193 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 printk("%02x ", rq->cmd[bit]);
195 printk("\n");
196 }
197}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198EXPORT_SYMBOL(blk_dump_rq_flags);
199
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500200static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200201{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500202 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200203
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500204 q = container_of(work, struct request_queue, delay_work.work);
205 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200206 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500207 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
210/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500211 * blk_delay_queue - restart queueing after defined interval
212 * @q: The &struct request_queue in question
213 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 *
215 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500216 * Sometimes queueing needs to be postponed for a little while, to allow
217 * resources to come back. This function will make sure that queueing is
218 * restarted around the specified time.
219 */
220void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221{
Jens Axboe4521cc42011-04-18 11:36:39 +0200222 queue_delayed_work(kblockd_workqueue, &q->delay_work,
223 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500225EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500226
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227/**
228 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200229 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 *
231 * Description:
232 * blk_start_queue() will clear the stop flag on the queue, and call
233 * the request_fn for the queue if it was in a stopped state when
234 * entered. Also see blk_stop_queue(). Queue lock must be held.
235 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200236void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200238 WARN_ON(!irqs_disabled());
239
Nick Piggin75ad23b2008-04-29 14:48:33 +0200240 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200241 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243EXPORT_SYMBOL(blk_start_queue);
244
245/**
246 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200247 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 *
249 * Description:
250 * The Linux block layer assumes that a block driver will consume all
251 * entries on the request queue when the request_fn strategy is called.
252 * Often this will not happen, because of hardware limitations (queue
253 * depth settings). If a device driver gets a 'queue full' response,
254 * or if it simply chooses not to queue more I/O at one point, it can
255 * call this function to prevent the request_fn from being called until
256 * the driver has signalled it's ready to go again. This happens by calling
257 * blk_start_queue() to restart queue operations. Queue lock must be held.
258 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200259void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
Jens Axboead3d9d72011-03-25 16:58:59 +0100261 __cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200262 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263}
264EXPORT_SYMBOL(blk_stop_queue);
265
266/**
267 * blk_sync_queue - cancel any pending callbacks on a queue
268 * @q: the queue
269 *
270 * Description:
271 * The block layer may perform asynchronous callback activity
272 * on a queue, such as calling the unplug function after a timeout.
273 * A block device may call blk_sync_queue to ensure that any
274 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200275 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 * that its ->make_request_fn will not re-add plugging prior to calling
277 * this function.
278 *
Vivek Goyalda527772011-03-02 19:05:33 -0500279 * This function does not cancel any asynchronous activity arising
280 * out of elevator or throttling code. That would require elevaotor_exit()
281 * and blk_throtl_exit() to be called with queue lock initialized.
282 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 */
284void blk_sync_queue(struct request_queue *q)
285{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100286 del_timer_sync(&q->timeout);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500287 cancel_delayed_work_sync(&q->delay_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289EXPORT_SYMBOL(blk_sync_queue);
290
291/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200292 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200294 *
295 * Description:
296 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200297 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200299void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300{
Tejun Heoa538cd02009-04-23 11:05:17 +0900301 if (unlikely(blk_queue_stopped(q)))
302 return;
303
Jens Axboec21e6be2011-04-19 13:32:46 +0200304 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200305}
306EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200307
Nick Piggin75ad23b2008-04-29 14:48:33 +0200308/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200309 * blk_run_queue_async - run a single device queue in workqueue context
310 * @q: The queue to run
311 *
312 * Description:
313 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
314 * of us.
315 */
316void blk_run_queue_async(struct request_queue *q)
317{
Shaohua Li3ec717b2011-05-18 11:22:43 +0200318 if (likely(!blk_queue_stopped(q))) {
319 __cancel_delayed_work(&q->delay_work);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200320 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Shaohua Li3ec717b2011-05-18 11:22:43 +0200321 }
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200322}
Jens Axboec21e6be2011-04-19 13:32:46 +0200323EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200324
325/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200326 * blk_run_queue - run a single device queue
327 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200328 *
329 * Description:
330 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900331 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200332 */
333void blk_run_queue(struct request_queue *q)
334{
335 unsigned long flags;
336
337 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200338 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 spin_unlock_irqrestore(q->queue_lock, flags);
340}
341EXPORT_SYMBOL(blk_run_queue);
342
Jens Axboe165125e2007-07-24 09:28:11 +0200343void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500344{
345 kobject_put(&q->kobj);
346}
Jens Axboed86e0e82011-05-27 07:44:43 +0200347EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500348
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200349/**
350 * blk_drain_queue - drain requests from request_queue
351 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200352 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200353 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200354 * Drain requests from @q. If @drain_all is set, all requests are drained.
355 * If not, only ELVPRIV requests are drained. The caller is responsible
356 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200357 */
Tejun Heoc9a929d2011-10-19 14:42:16 +0200358void blk_drain_queue(struct request_queue *q, bool drain_all)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200359{
360 while (true) {
361 int nr_rqs;
362
363 spin_lock_irq(q->queue_lock);
364
365 elv_drain_elevator(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200366 if (drain_all)
367 blk_throtl_drain(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200368
Tejun Heo4eabc942011-12-15 20:03:04 +0100369 /*
370 * This function might be called on a queue which failed
371 * driver init after queue creation. Some drivers
372 * (e.g. fd) get unhappy in such cases. Kick queue iff
373 * dispatch queue has something on it.
374 */
375 if (!list_empty(&q->queue_head))
376 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200377
378 if (drain_all)
379 nr_rqs = q->rq.count[0] + q->rq.count[1];
380 else
381 nr_rqs = q->rq.elvpriv;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200382
383 spin_unlock_irq(q->queue_lock);
384
385 if (!nr_rqs)
386 break;
387 msleep(10);
388 }
389}
390
Tejun Heoc9a929d2011-10-19 14:42:16 +0200391/**
392 * blk_cleanup_queue - shutdown a request queue
393 * @q: request queue to shutdown
394 *
395 * Mark @q DEAD, drain all pending requests, destroy and put it. All
396 * future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500397 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100398void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500399{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200400 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de2008-09-18 09:22:54 -0700401
Tejun Heoc9a929d2011-10-19 14:42:16 +0200402 /* mark @q DEAD, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500403 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200404 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200405
406 spin_lock_irq(lock);
407 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
408 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
409 queue_flag_set(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500410
Hannes Reinecke777eb1b2011-09-28 08:07:01 -0600411 if (q->queue_lock != &q->__queue_lock)
412 q->queue_lock = &q->__queue_lock;
Vivek Goyalda527772011-03-02 19:05:33 -0500413
Tejun Heoc9a929d2011-10-19 14:42:16 +0200414 spin_unlock_irq(lock);
415 mutex_unlock(&q->sysfs_lock);
416
Tejun Heo6dd9ad72011-11-03 18:52:11 +0100417 /*
418 * Drain all requests queued before DEAD marking. The caller might
419 * be trying to tear down @q before its elevator is initialized, in
420 * which case we don't want to call into draining.
421 */
422 if (q->elevator)
423 blk_drain_queue(q, true);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200424
425 /* @q won't process any more request, flush async actions */
426 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
427 blk_sync_queue(q);
428
429 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500430 blk_put_queue(q);
431}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432EXPORT_SYMBOL(blk_cleanup_queue);
433
Jens Axboe165125e2007-07-24 09:28:11 +0200434static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435{
436 struct request_list *rl = &q->rq;
437
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400438 if (unlikely(rl->rq_pool))
439 return 0;
440
Jens Axboe1faa16d2009-04-06 14:48:01 +0200441 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
442 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200443 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200444 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
445 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Christoph Lameter19460892005-06-23 00:08:19 -0700447 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
448 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
450 if (!rl->rq_pool)
451 return -ENOMEM;
452
453 return 0;
454}
455
Jens Axboe165125e2007-07-24 09:28:11 +0200456struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
Christoph Lameter19460892005-06-23 00:08:19 -0700458 return blk_alloc_queue_node(gfp_mask, -1);
459}
460EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Jens Axboe165125e2007-07-24 09:28:11 +0200462struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700463{
Jens Axboe165125e2007-07-24 09:28:11 +0200464 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700465 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700466
Jens Axboe8324aa92008-01-29 14:51:59 +0100467 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700468 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 if (!q)
470 return NULL;
471
Jens Axboe0989a022009-06-12 14:42:56 +0200472 q->backing_dev_info.ra_pages =
473 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
474 q->backing_dev_info.state = 0;
475 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200476 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100477 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200478
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700479 err = bdi_init(&q->backing_dev_info);
480 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100481 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700482 return NULL;
483 }
484
Vivek Goyale43473b2010-09-15 17:06:35 -0400485 if (blk_throtl_init(q)) {
486 kmem_cache_free(blk_requestq_cachep, q);
487 return NULL;
488 }
489
Matthew Garrett31373d02010-04-06 14:25:14 +0200490 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
491 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700492 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
493 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoae1b1532011-01-25 12:43:54 +0100494 INIT_LIST_HEAD(&q->flush_queue[0]);
495 INIT_LIST_HEAD(&q->flush_queue[1]);
496 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500497 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500498
Jens Axboe8324aa92008-01-29 14:51:59 +0100499 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Al Viro483f4af2006-03-18 18:34:37 -0500501 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700502 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500503
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500504 /*
505 * By default initialize queue_lock to internal lock and driver can
506 * override it later if need be.
507 */
508 q->queue_lock = &q->__queue_lock;
509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 return q;
511}
Christoph Lameter19460892005-06-23 00:08:19 -0700512EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
514/**
515 * blk_init_queue - prepare a request queue for use with a block device
516 * @rfn: The function to be called to process requests that have been
517 * placed on the queue.
518 * @lock: Request queue spin lock
519 *
520 * Description:
521 * If a block device wishes to use the standard request handling procedures,
522 * which sorts requests and coalesces adjacent requests, then it must
523 * call blk_init_queue(). The function @rfn will be called when there
524 * are requests on the queue that need to be processed. If the device
525 * supports plugging, then @rfn may not be called immediately when requests
526 * are available on the queue, but may be called at some time later instead.
527 * Plugged queues are generally unplugged when a buffer belonging to one
528 * of the requests on the queue is needed, or due to memory pressure.
529 *
530 * @rfn is not required, or even expected, to remove all requests off the
531 * queue, but only as many as it can handle at a time. If it does leave
532 * requests on the queue, it is responsible for arranging that the requests
533 * get dealt with eventually.
534 *
535 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200536 * request queue; this lock will be taken also from interrupt context, so irq
537 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200539 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 * it didn't succeed.
541 *
542 * Note:
543 * blk_init_queue() must be paired with a blk_cleanup_queue() call
544 * when the block device is deactivated (such as at module unload).
545 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700546
Jens Axboe165125e2007-07-24 09:28:11 +0200547struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
Christoph Lameter19460892005-06-23 00:08:19 -0700549 return blk_init_queue_node(rfn, lock, -1);
550}
551EXPORT_SYMBOL(blk_init_queue);
552
Jens Axboe165125e2007-07-24 09:28:11 +0200553struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700554blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
555{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600556 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600558 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
559 if (!uninit_q)
560 return NULL;
561
Mike Snitzer51514122011-11-23 10:59:13 +0100562 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600563 if (!q)
564 blk_cleanup_queue(uninit_q);
565
566 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200567}
568EXPORT_SYMBOL(blk_init_queue_node);
569
570struct request_queue *
571blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
572 spinlock_t *lock)
573{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 if (!q)
575 return NULL;
576
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600577 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500578 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900582 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100583 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500584
585 /* Override internal queue lock with supplied lock pointer */
586 if (lock)
587 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Jens Axboef3b144a2009-03-06 08:48:33 +0100589 /*
590 * This also sets hw/phys segments, boundary and size
591 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200592 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Alan Stern44ec9542007-02-20 11:01:57 -0500594 q->sg_reserved_size = INT_MAX;
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 /*
597 * all done
598 */
599 if (!elevator_init(q, NULL)) {
600 blk_queue_congestion_threshold(q);
601 return q;
602 }
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 return NULL;
605}
Mike Snitzer51514122011-11-23 10:59:13 +0100606EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
Jens Axboe165125e2007-07-24 09:28:11 +0200608int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700610 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500611 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 return 0;
613 }
614
615 return 1;
616}
Jens Axboed86e0e82011-05-27 07:44:43 +0200617EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Jens Axboe165125e2007-07-24 09:28:11 +0200619static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200621 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200622 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 mempool_free(rq, q->rq.rq_pool);
624}
625
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200626static struct request *
Tejun Heo75eb6c32011-10-19 14:31:22 +0200627blk_alloc_request(struct request_queue *q, unsigned int flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
629 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
630
631 if (!rq)
632 return NULL;
633
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200634 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200635
Jerome Marchand42dad762009-04-22 14:01:49 +0200636 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
Tejun Heo75eb6c32011-10-19 14:31:22 +0200638 if ((flags & REQ_ELVPRIV) &&
639 unlikely(elv_set_request(q, rq, gfp_mask))) {
640 mempool_free(rq, q->rq.rq_pool);
641 return NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200642 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Tejun Heocb98fc82005-10-28 08:29:39 +0200644 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
647/*
648 * ioc_batching returns true if the ioc is a valid batching request and
649 * should be given priority access to a request.
650 */
Jens Axboe165125e2007-07-24 09:28:11 +0200651static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
653 if (!ioc)
654 return 0;
655
656 /*
657 * Make sure the process is able to allocate at least 1 request
658 * even if the batch times out, otherwise we could theoretically
659 * lose wakeups.
660 */
661 return ioc->nr_batch_requests == q->nr_batching ||
662 (ioc->nr_batch_requests > 0
663 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
664}
665
666/*
667 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
668 * will cause the process to be a "batcher" on all queues in the system. This
669 * is the behaviour we want though - once it gets a wakeup it should be given
670 * a nice run.
671 */
Jens Axboe165125e2007-07-24 09:28:11 +0200672static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
674 if (!ioc || ioc_batching(q, ioc))
675 return;
676
677 ioc->nr_batch_requests = q->nr_batching;
678 ioc->last_waited = jiffies;
679}
680
Jens Axboe1faa16d2009-04-06 14:48:01 +0200681static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
683 struct request_list *rl = &q->rq;
684
Jens Axboe1faa16d2009-04-06 14:48:01 +0200685 if (rl->count[sync] < queue_congestion_off_threshold(q))
686 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
Jens Axboe1faa16d2009-04-06 14:48:01 +0200688 if (rl->count[sync] + 1 <= q->nr_requests) {
689 if (waitqueue_active(&rl->wait[sync]))
690 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
Jens Axboe1faa16d2009-04-06 14:48:01 +0200692 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 }
694}
695
696/*
697 * A request has just been released. Account for it, update the full and
698 * congestion status, wake up any waiters. Called under q->queue_lock.
699 */
Tejun Heo75eb6c32011-10-19 14:31:22 +0200700static void freed_request(struct request_queue *q, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
702 struct request_list *rl = &q->rq;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200703 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
Jens Axboe1faa16d2009-04-06 14:48:01 +0200705 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200706 if (flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200707 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
Jens Axboe1faa16d2009-04-06 14:48:01 +0200709 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
Jens Axboe1faa16d2009-04-06 14:48:01 +0200711 if (unlikely(rl->starved[sync ^ 1]))
712 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713}
714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100716 * Determine if elevator data should be initialized when allocating the
717 * request associated with @bio.
718 */
719static bool blk_rq_should_init_elevator(struct bio *bio)
720{
721 if (!bio)
722 return true;
723
724 /*
725 * Flush requests do not use the elevator so skip initialization.
726 * This allows a request to share the flush and elevator data.
727 */
728 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
729 return false;
730
731 return true;
732}
733
Tejun Heoda8303c2011-10-19 14:33:05 +0200734/**
735 * get_request - get a free request
736 * @q: request_queue to allocate request from
737 * @rw_flags: RW and SYNC flags
738 * @bio: bio to allocate request for (can be %NULL)
739 * @gfp_mask: allocation mask
740 *
741 * Get a free request from @q. This function may fail under memory
742 * pressure or if @q is dead.
743 *
744 * Must be callled with @q->queue_lock held and,
745 * Returns %NULL on failure, with @q->queue_lock held.
746 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 */
Jens Axboe165125e2007-07-24 09:28:11 +0200748static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100749 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
751 struct request *rq = NULL;
752 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100753 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200754 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200755 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
Tejun Heoda8303c2011-10-19 14:33:05 +0200757 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
758 return NULL;
759
Jens Axboe7749a8d2006-12-13 13:02:26 +0100760 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100761 if (may_queue == ELV_MQUEUE_NO)
762 goto rq_starved;
763
Jens Axboe1faa16d2009-04-06 14:48:01 +0200764 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
765 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200766 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100767 /*
768 * The queue will fill after this allocation, so set
769 * it as full, and mark this process as "batching".
770 * This process will be allowed to complete a batch of
771 * requests, others will be blocked.
772 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200773 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100774 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200775 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100776 } else {
777 if (may_queue != ELV_MQUEUE_MUST
778 && !ioc_batching(q, ioc)) {
779 /*
780 * The queue is full and the allocating
781 * process is not a "batcher", and not
782 * exempted by the IO scheduler
783 */
784 goto out;
785 }
786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200788 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 }
790
Jens Axboe082cf692005-06-28 16:35:11 +0200791 /*
792 * Only allow batching queuers to allocate up to 50% over the defined
793 * limit of requests, otherwise we could have thousands of requests
794 * allocated with any setting of ->nr_requests
795 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200796 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200797 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100798
Jens Axboe1faa16d2009-04-06 14:48:01 +0200799 rl->count[is_sync]++;
800 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200801
Tejun Heo75eb6c32011-10-19 14:31:22 +0200802 if (blk_rq_should_init_elevator(bio) &&
803 !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
804 rw_flags |= REQ_ELVPRIV;
805 rl->elvpriv++;
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100806 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200807
Jens Axboef253b862010-10-24 22:06:02 +0200808 if (blk_queue_io_stat(q))
809 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 spin_unlock_irq(q->queue_lock);
811
Tejun Heo75eb6c32011-10-19 14:31:22 +0200812 rq = blk_alloc_request(q, rw_flags, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100813 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 /*
815 * Allocation failed presumably due to memory. Undo anything
816 * we might have messed up.
817 *
818 * Allocating task should really be put onto the front of the
819 * wait queue, but this is pretty rare.
820 */
821 spin_lock_irq(q->queue_lock);
Tejun Heo75eb6c32011-10-19 14:31:22 +0200822 freed_request(q, rw_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
824 /*
825 * in the very unlikely event that allocation failed and no
826 * requests for this direction was pending, mark us starved
827 * so that freeing of a request in the other direction will
828 * notice us. another possible fix would be to split the
829 * rq mempool into READ and WRITE
830 */
831rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200832 if (unlikely(rl->count[is_sync] == 0))
833 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 goto out;
836 }
837
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100838 /*
839 * ioc may be NULL here, and ioc_batching will be false. That's
840 * OK, if the queue is under the request limit then requests need
841 * not count toward the nr_batch_requests limit. There will always
842 * be some limit enforced by BLK_BATCH_TIME.
843 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (ioc_batching(q, ioc))
845 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100846
Jens Axboe1faa16d2009-04-06 14:48:01 +0200847 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 return rq;
850}
851
Tejun Heoda8303c2011-10-19 14:33:05 +0200852/**
853 * get_request_wait - get a free request with retry
854 * @q: request_queue to allocate request from
855 * @rw_flags: RW and SYNC flags
856 * @bio: bio to allocate request for (can be %NULL)
Nick Piggind6344532005-06-28 20:45:14 -0700857 *
Tejun Heoda8303c2011-10-19 14:33:05 +0200858 * Get a free request from @q. This function keeps retrying under memory
859 * pressure and fails iff @q is dead.
860 *
861 * Must be callled with @q->queue_lock held and,
862 * Returns %NULL on failure, with @q->queue_lock held.
863 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 */
Jens Axboe165125e2007-07-24 09:28:11 +0200865static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200866 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200868 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 struct request *rq;
870
Jens Axboe7749a8d2006-12-13 13:02:26 +0100871 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700872 while (!rq) {
873 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200874 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 struct request_list *rl = &q->rq;
876
Tejun Heoda8303c2011-10-19 14:33:05 +0200877 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
878 return NULL;
879
Jens Axboe1faa16d2009-04-06 14:48:01 +0200880 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 TASK_UNINTERRUPTIBLE);
882
Jens Axboe1faa16d2009-04-06 14:48:01 +0200883 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200885 spin_unlock_irq(q->queue_lock);
886 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200888 /*
889 * After sleeping, we become a "batching" process and
890 * will be able to allocate at least one request, and
891 * up to a big batch of them for a small period time.
892 * See ioc_batching, ioc_set_batching
893 */
894 ioc = current_io_context(GFP_NOIO, q->node);
895 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100896
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200897 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200898 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200899
900 rq = get_request(q, rw_flags, bio, GFP_NOIO);
901 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
903 return rq;
904}
905
Jens Axboe165125e2007-07-24 09:28:11 +0200906struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
908 struct request *rq;
909
910 BUG_ON(rw != READ && rw != WRITE);
911
Nick Piggind6344532005-06-28 20:45:14 -0700912 spin_lock_irq(q->queue_lock);
Tejun Heoda8303c2011-10-19 14:33:05 +0200913 if (gfp_mask & __GFP_WAIT)
Jens Axboe22e2c502005-06-27 10:55:12 +0200914 rq = get_request_wait(q, rw, NULL);
Tejun Heoda8303c2011-10-19 14:33:05 +0200915 else
Jens Axboe22e2c502005-06-27 10:55:12 +0200916 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +0200917 if (!rq)
918 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -0700919 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 return rq;
922}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923EXPORT_SYMBOL(blk_get_request);
924
925/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300926 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700927 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300928 * @bio: The bio describing the memory mappings that will be submitted for IO.
929 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700930 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200931 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300932 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
933 * type commands. Where the struct request needs to be farther initialized by
934 * the caller. It is passed a &struct bio, which describes the memory info of
935 * the I/O transfer.
936 *
937 * The caller of blk_make_request must make sure that bi_io_vec
938 * are set to describe the memory buffers. That bio_data_dir() will return
939 * the needed direction of the request. (And all bio's in the passed bio-chain
940 * are properly set accordingly)
941 *
942 * If called under none-sleepable conditions, mapped bio buffers must not
943 * need bouncing, by calling the appropriate masked or flagged allocator,
944 * suitable for the target device. Otherwise the call to blk_queue_bounce will
945 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200946 *
947 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
948 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
949 * anything but the first bio in the chain. Otherwise you risk waiting for IO
950 * completion of a bio that hasn't been submitted yet, thus resulting in a
951 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
952 * of bio_alloc(), as that avoids the mempool deadlock.
953 * If possible a big IO should be split into smaller parts when allocation
954 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200955 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300956struct request *blk_make_request(struct request_queue *q, struct bio *bio,
957 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200958{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300959 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
960
961 if (unlikely(!rq))
962 return ERR_PTR(-ENOMEM);
963
964 for_each_bio(bio) {
965 struct bio *bounce_bio = bio;
966 int ret;
967
968 blk_queue_bounce(q, &bounce_bio);
969 ret = blk_rq_append_bio(q, rq, bounce_bio);
970 if (unlikely(ret)) {
971 blk_put_request(rq);
972 return ERR_PTR(ret);
973 }
974 }
975
976 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200977}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300978EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200979
980/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 * blk_requeue_request - put a request back on queue
982 * @q: request queue where request should be inserted
983 * @rq: request to be inserted
984 *
985 * Description:
986 * Drivers often keep queueing requests until the hardware cannot accept
987 * more, when that condition happens we need to put the request back
988 * on the queue. Must be called with queue lock held.
989 */
Jens Axboe165125e2007-07-24 09:28:11 +0200990void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700992 blk_delete_timer(rq);
993 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100994 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 if (blk_rq_tagged(rq))
997 blk_queue_end_tag(q, rq);
998
James Bottomleyba396a62009-05-27 14:17:08 +0200999 BUG_ON(blk_queued_rq(rq));
1000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 elv_requeue_request(q, rq);
1002}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003EXPORT_SYMBOL(blk_requeue_request);
1004
Jens Axboe73c10102011-03-08 13:19:51 +01001005static void add_acct_request(struct request_queue *q, struct request *rq,
1006 int where)
1007{
1008 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001009 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001010}
1011
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001013 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 * @q: request queue where request should be inserted
1015 * @rq: request to be inserted
1016 * @at_head: insert request at head or tail of queue
1017 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 *
1019 * Description:
1020 * Many block devices need to execute commands asynchronously, so they don't
1021 * block the whole kernel from preemption during request execution. This is
1022 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +02001023 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1024 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 *
1026 * We have the option of inserting the head or the tail of the queue.
1027 * Typically we use the tail for new ioctls and so forth. We use the head
1028 * of the queue for things like a QUEUE_FULL message from a device, or a
1029 * host that is unable to accept a particular command.
1030 */
Jens Axboe165125e2007-07-24 09:28:11 +02001031void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05001032 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
Tejun Heo 867d1192005-04-24 02:06:05 -05001034 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 unsigned long flags;
1036
1037 /*
1038 * tell I/O scheduler that this isn't a regular read/write (ie it
1039 * must not attempt merges on this) and that it acts as a soft
1040 * barrier
1041 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02001042 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 rq->special = data;
1045
1046 spin_lock_irqsave(q->queue_lock, flags);
1047
1048 /*
1049 * If command is tagged, release the tag
1050 */
Tejun Heo 867d1192005-04-24 02:06:05 -05001051 if (blk_rq_tagged(rq))
1052 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Jens Axboe73c10102011-03-08 13:19:51 +01001054 add_acct_request(q, rq, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001055 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 spin_unlock_irqrestore(q->queue_lock, flags);
1057}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058EXPORT_SYMBOL(blk_insert_request);
1059
Tejun Heo074a7ac2008-08-25 19:56:14 +09001060static void part_round_stats_single(int cpu, struct hd_struct *part,
1061 unsigned long now)
1062{
1063 if (now == part->stamp)
1064 return;
1065
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001066 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001067 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001068 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001069 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1070 }
1071 part->stamp = now;
1072}
1073
1074/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001075 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1076 * @cpu: cpu number for stats access
1077 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 *
1079 * The average IO queue length and utilisation statistics are maintained
1080 * by observing the current state of the queue length and the amount of
1081 * time it has been in this state for.
1082 *
1083 * Normally, that accounting is done on IO completion, but that can result
1084 * in more than a second's worth of IO being accounted for within any one
1085 * second, leading to >100% utilisation. To deal with that, we call this
1086 * function to do a round-off before returning the results when reading
1087 * /proc/diskstats. This accounts immediately for all queue usage up to
1088 * the current jiffies and restarts the counters again.
1089 */
Tejun Heoc9959052008-08-25 19:47:21 +09001090void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001091{
1092 unsigned long now = jiffies;
1093
Tejun Heo074a7ac2008-08-25 19:56:14 +09001094 if (part->partno)
1095 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1096 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001097}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001098EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100/*
1101 * queue lock must be held
1102 */
Jens Axboe165125e2007-07-24 09:28:11 +02001103void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 if (unlikely(!q))
1106 return;
1107 if (unlikely(--req->ref_count))
1108 return;
1109
Tejun Heo8922e162005-10-20 16:23:44 +02001110 elv_completed_request(q, req);
1111
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001112 /* this is a bio leak */
1113 WARN_ON(req->bio != NULL);
1114
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 /*
1116 * Request may not have originated from ll_rw_blk. if not,
1117 * it didn't come out of our reserved rq pools
1118 */
Jens Axboe49171e52006-08-10 08:59:11 +02001119 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001120 unsigned int flags = req->cmd_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001123 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
1125 blk_free_request(q, req);
Tejun Heo75eb6c32011-10-19 14:31:22 +02001126 freed_request(q, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
1128}
Mike Christie6e39b692005-11-11 05:30:24 -06001129EXPORT_SYMBOL_GPL(__blk_put_request);
1130
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131void blk_put_request(struct request *req)
1132{
Tejun Heo8922e162005-10-20 16:23:44 +02001133 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001134 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001136 spin_lock_irqsave(q->queue_lock, flags);
1137 __blk_put_request(q, req);
1138 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140EXPORT_SYMBOL(blk_put_request);
1141
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001142/**
1143 * blk_add_request_payload - add a payload to a request
1144 * @rq: request to update
1145 * @page: page backing the payload
1146 * @len: length of the payload.
1147 *
1148 * This allows to later add a payload to an already submitted request by
1149 * a block driver. The driver needs to take care of freeing the payload
1150 * itself.
1151 *
1152 * Note that this is a quite horrible hack and nothing but handling of
1153 * discard requests should ever use it.
1154 */
1155void blk_add_request_payload(struct request *rq, struct page *page,
1156 unsigned int len)
1157{
1158 struct bio *bio = rq->bio;
1159
1160 bio->bi_io_vec->bv_page = page;
1161 bio->bi_io_vec->bv_offset = 0;
1162 bio->bi_io_vec->bv_len = len;
1163
1164 bio->bi_size = len;
1165 bio->bi_vcnt = 1;
1166 bio->bi_phys_segments = 1;
1167
1168 rq->__data_len = rq->resid_len = len;
1169 rq->nr_phys_segments = 1;
1170 rq->buffer = bio_data(bio);
1171}
1172EXPORT_SYMBOL_GPL(blk_add_request_payload);
1173
Jens Axboe73c10102011-03-08 13:19:51 +01001174static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1175 struct bio *bio)
1176{
1177 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1178
Jens Axboe73c10102011-03-08 13:19:51 +01001179 if (!ll_back_merge_fn(q, req, bio))
1180 return false;
1181
1182 trace_block_bio_backmerge(q, bio);
1183
1184 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1185 blk_rq_set_mixed_merge(req);
1186
1187 req->biotail->bi_next = bio;
1188 req->biotail = bio;
1189 req->__data_len += bio->bi_size;
1190 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1191
1192 drive_stat_acct(req, 0);
Vivek Goyal95cf3dd2011-05-23 10:02:19 +02001193 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001194 return true;
1195}
1196
1197static bool bio_attempt_front_merge(struct request_queue *q,
1198 struct request *req, struct bio *bio)
1199{
1200 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001201
Jens Axboe73c10102011-03-08 13:19:51 +01001202 if (!ll_front_merge_fn(q, req, bio))
1203 return false;
1204
1205 trace_block_bio_frontmerge(q, bio);
1206
1207 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1208 blk_rq_set_mixed_merge(req);
1209
Jens Axboe73c10102011-03-08 13:19:51 +01001210 bio->bi_next = req->bio;
1211 req->bio = bio;
1212
1213 /*
1214 * may not be valid. if the low level driver said
1215 * it didn't need a bounce buffer then it better
1216 * not touch req->buffer either...
1217 */
1218 req->buffer = bio_data(bio);
1219 req->__sector = bio->bi_sector;
1220 req->__data_len += bio->bi_size;
1221 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1222
1223 drive_stat_acct(req, 0);
Vivek Goyal95cf3dd2011-05-23 10:02:19 +02001224 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001225 return true;
1226}
1227
Tejun Heobd87b582011-10-19 14:33:08 +02001228/**
1229 * attempt_plug_merge - try to merge with %current's plugged list
1230 * @q: request_queue new bio is being queued at
1231 * @bio: new bio being queued
1232 * @request_count: out parameter for number of traversed plugged requests
1233 *
1234 * Determine whether @bio being queued on @q can be merged with a request
1235 * on %current's plugged list. Returns %true if merge was successful,
1236 * otherwise %false.
1237 *
1238 * This function is called without @q->queue_lock; however, elevator is
1239 * accessed iff there already are requests on the plugged list which in
1240 * turn guarantees validity of the elevator.
1241 *
1242 * Note that, on successful merge, elevator operation
1243 * elevator_bio_merged_fn() will be called without queue lock. Elevator
1244 * must be ready for this.
Jens Axboe73c10102011-03-08 13:19:51 +01001245 */
Tejun Heobd87b582011-10-19 14:33:08 +02001246static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1247 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001248{
1249 struct blk_plug *plug;
1250 struct request *rq;
1251 bool ret = false;
1252
Tejun Heobd87b582011-10-19 14:33:08 +02001253 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001254 if (!plug)
1255 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001256 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001257
1258 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1259 int el_ret;
1260
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001261 (*request_count)++;
1262
Jens Axboe73c10102011-03-08 13:19:51 +01001263 if (rq->q != q)
1264 continue;
1265
1266 el_ret = elv_try_merge(rq, bio);
1267 if (el_ret == ELEVATOR_BACK_MERGE) {
1268 ret = bio_attempt_back_merge(q, rq, bio);
1269 if (ret)
1270 break;
1271 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1272 ret = bio_attempt_front_merge(q, rq, bio);
1273 if (ret)
1274 break;
1275 }
1276 }
1277out:
1278 return ret;
1279}
1280
Jens Axboe86db1e22008-01-29 14:53:40 +01001281void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001282{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001283 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001284
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001285 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1286 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001287 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001288
Tejun Heo52d9e672006-01-06 09:49:58 +01001289 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001290 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001291 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001292 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001293}
1294
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001295void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001297 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001298 struct blk_plug *plug;
1299 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1300 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001301 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 /*
1304 * low level driver can indicate that it wants pages above a
1305 * certain limit bounced to low memory (ie for highmem, or even
1306 * ISA dma in theory)
1307 */
1308 blk_queue_bounce(q, &bio);
1309
Tejun Heo4fed9472010-09-03 11:56:17 +02001310 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001311 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001312 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001313 goto get_rq;
1314 }
1315
Jens Axboe73c10102011-03-08 13:19:51 +01001316 /*
1317 * Check if we can merge with the plugged list before grabbing
1318 * any locks.
1319 */
Tejun Heobd87b582011-10-19 14:33:08 +02001320 if (attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001321 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001322
1323 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001326 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001327 if (bio_attempt_back_merge(q, req, bio)) {
1328 if (!attempt_back_merge(q, req))
1329 elv_merged_request(q, req, el_ret);
1330 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001331 }
Jens Axboe73c10102011-03-08 13:19:51 +01001332 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001333 if (bio_attempt_front_merge(q, req, bio)) {
1334 if (!attempt_front_merge(q, req))
1335 elv_merged_request(q, req, el_ret);
1336 goto out_unlock;
1337 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 }
1339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001341 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001342 * This sync check and mask will be re-done in init_request_from_bio(),
1343 * but we need to set it earlier to expose the sync flag to the
1344 * rq allocator and io schedulers.
1345 */
1346 rw_flags = bio_data_dir(bio);
1347 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001348 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001349
1350 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001351 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001352 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001353 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001354 req = get_request_wait(q, rw_flags, bio);
Tejun Heoda8303c2011-10-19 14:33:05 +02001355 if (unlikely(!req)) {
1356 bio_endio(bio, -ENODEV); /* @q is dead */
1357 goto out_unlock;
1358 }
Nick Piggind6344532005-06-28 20:45:14 -07001359
Nick Piggin450991b2005-06-28 20:45:13 -07001360 /*
1361 * After dropping the lock and possibly sleeping here, our request
1362 * may now be mergeable after it had proven unmergeable (above).
1363 * We don't worry about that case for efficiency. It won't happen
1364 * often, and the elevators are able to handle it.
1365 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001366 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Tao Ma9562ad92011-10-24 16:11:30 +02001368 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001369 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001370
Jens Axboe73c10102011-03-08 13:19:51 +01001371 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001372 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001373 /*
1374 * If this is the first request added after a plug, fire
1375 * of a plug trace. If others have been added before, check
1376 * if we have multiple devices in this plug. If so, make a
1377 * note to sort the list before dispatch.
1378 */
1379 if (list_empty(&plug->list))
1380 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001381 else {
1382 if (!plug->should_sort) {
1383 struct request *__rq;
Jens Axboe73c10102011-03-08 13:19:51 +01001384
Shaohua Li3540d5e2011-11-16 09:21:50 +01001385 __rq = list_entry_rq(plug->list.prev);
1386 if (__rq->q != q)
1387 plug->should_sort = 1;
1388 }
Shaohua Li019ceb72011-11-16 09:21:50 +01001389 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001390 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001391 trace_block_plug(q);
1392 }
Jens Axboe73c10102011-03-08 13:19:51 +01001393 }
Shaohua Lia6327162011-08-24 16:04:32 +02001394 list_add_tail(&req->queuelist, &plug->list);
1395 drive_stat_acct(req, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001396 } else {
1397 spin_lock_irq(q->queue_lock);
1398 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001399 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001400out_unlock:
1401 spin_unlock_irq(q->queue_lock);
1402 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403}
Jens Axboec20e8de2011-09-12 12:03:37 +02001404EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
1406/*
1407 * If bio->bi_dev is a partition, remap the location
1408 */
1409static inline void blk_partition_remap(struct bio *bio)
1410{
1411 struct block_device *bdev = bio->bi_bdev;
1412
Jens Axboebf2de6f2007-09-27 13:01:25 +02001413 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 bio->bi_sector += p->start_sect;
1417 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001418
Mike Snitzerd07335e2010-11-16 12:52:38 +01001419 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1420 bdev->bd_dev,
1421 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 }
1423}
1424
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425static void handle_bad_sector(struct bio *bio)
1426{
1427 char b[BDEVNAME_SIZE];
1428
1429 printk(KERN_INFO "attempt to access beyond end of device\n");
1430 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1431 bdevname(bio->bi_bdev, b),
1432 bio->bi_rw,
1433 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001434 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436 set_bit(BIO_EOF, &bio->bi_flags);
1437}
1438
Akinobu Mitac17bb492006-12-08 02:39:46 -08001439#ifdef CONFIG_FAIL_MAKE_REQUEST
1440
1441static DECLARE_FAULT_ATTR(fail_make_request);
1442
1443static int __init setup_fail_make_request(char *str)
1444{
1445 return setup_fault_attr(&fail_make_request, str);
1446}
1447__setup("fail_make_request=", setup_fail_make_request);
1448
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001449static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001450{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001451 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001452}
1453
1454static int __init fail_make_request_debugfs(void)
1455{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001456 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1457 NULL, &fail_make_request);
1458
1459 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001460}
1461
1462late_initcall(fail_make_request_debugfs);
1463
1464#else /* CONFIG_FAIL_MAKE_REQUEST */
1465
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001466static inline bool should_fail_request(struct hd_struct *part,
1467 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001468{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001469 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001470}
1471
1472#endif /* CONFIG_FAIL_MAKE_REQUEST */
1473
Jens Axboec07e2b42007-07-18 13:27:58 +02001474/*
1475 * Check whether this bio extends beyond the end of the device.
1476 */
1477static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1478{
1479 sector_t maxsector;
1480
1481 if (!nr_sectors)
1482 return 0;
1483
1484 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001485 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001486 if (maxsector) {
1487 sector_t sector = bio->bi_sector;
1488
1489 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1490 /*
1491 * This may well happen - the kernel calls bread()
1492 * without checking the size of the device, e.g., when
1493 * mounting a device.
1494 */
1495 handle_bad_sector(bio);
1496 return 1;
1497 }
1498 }
1499
1500 return 0;
1501}
1502
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001503static noinline_for_stack bool
1504generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505{
Jens Axboe165125e2007-07-24 09:28:11 +02001506 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001507 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001508 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001509 char b[BDEVNAME_SIZE];
1510 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
1512 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Jens Axboec07e2b42007-07-18 13:27:58 +02001514 if (bio_check_eod(bio, nr_sectors))
1515 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001517 q = bdev_get_queue(bio->bi_bdev);
1518 if (unlikely(!q)) {
1519 printk(KERN_ERR
1520 "generic_make_request: Trying to access "
1521 "nonexistent block-device %s (%Lu)\n",
1522 bdevname(bio->bi_bdev, b),
1523 (long long) bio->bi_sector);
1524 goto end_io;
1525 }
1526
1527 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1528 nr_sectors > queue_max_hw_sectors(q))) {
1529 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1530 bdevname(bio->bi_bdev, b),
1531 bio_sectors(bio),
1532 queue_max_hw_sectors(q));
1533 goto end_io;
1534 }
1535
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001536 part = bio->bi_bdev->bd_part;
1537 if (should_fail_request(part, bio->bi_size) ||
1538 should_fail_request(&part_to_disk(part)->part0,
1539 bio->bi_size))
1540 goto end_io;
1541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001543 * If this device has partitions, remap block n
1544 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001546 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001548 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1549 goto end_io;
1550
1551 if (bio_check_eod(bio, nr_sectors))
1552 goto end_io;
1553
1554 /*
1555 * Filter flush bio's early so that make_request based
1556 * drivers without flush support don't have to worry
1557 * about them.
1558 */
1559 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1560 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1561 if (!nr_sectors) {
1562 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001563 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001565 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001567 if ((bio->bi_rw & REQ_DISCARD) &&
1568 (!blk_queue_discard(q) ||
1569 ((bio->bi_rw & REQ_SECURE) &&
1570 !blk_queue_secdiscard(q)))) {
1571 err = -EOPNOTSUPP;
1572 goto end_io;
1573 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
Tejun Heobc16a4f2011-10-19 14:33:01 +02001575 if (blk_throtl_bio(q, bio))
1576 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001577
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001578 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001579 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001580
1581end_io:
1582 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001583 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584}
1585
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001586/**
1587 * generic_make_request - hand a buffer to its device driver for I/O
1588 * @bio: The bio describing the location in memory and on the device.
1589 *
1590 * generic_make_request() is used to make I/O requests of block
1591 * devices. It is passed a &struct bio, which describes the I/O that needs
1592 * to be done.
1593 *
1594 * generic_make_request() does not return any status. The
1595 * success/failure status of the request, along with notification of
1596 * completion, is delivered asynchronously through the bio->bi_end_io
1597 * function described (one day) else where.
1598 *
1599 * The caller of generic_make_request must make sure that bi_io_vec
1600 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1601 * set to describe the device address, and the
1602 * bi_end_io and optionally bi_private are set to describe how
1603 * completion notification should be signaled.
1604 *
1605 * generic_make_request and the drivers it calls may use bi_next if this
1606 * bio happens to be merged with someone else, and may resubmit the bio to
1607 * a lower device by calling into generic_make_request recursively, which
1608 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001609 */
1610void generic_make_request(struct bio *bio)
1611{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001612 struct bio_list bio_list_on_stack;
1613
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001614 if (!generic_make_request_checks(bio))
1615 return;
1616
1617 /*
1618 * We only want one ->make_request_fn to be active at a time, else
1619 * stack usage with stacked devices could be a problem. So use
1620 * current->bio_list to keep a list of requests submited by a
1621 * make_request_fn function. current->bio_list is also used as a
1622 * flag to say if generic_make_request is currently active in this
1623 * task or not. If it is NULL, then no make_request is active. If
1624 * it is non-NULL, then a make_request is active, and new requests
1625 * should be added at the tail
1626 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001627 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001628 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001629 return;
1630 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001631
Neil Brownd89d8792007-05-01 09:53:42 +02001632 /* following loop may be a bit non-obvious, and so deserves some
1633 * explanation.
1634 * Before entering the loop, bio->bi_next is NULL (as all callers
1635 * ensure that) so we have a list with a single bio.
1636 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001637 * we assign bio_list to a pointer to the bio_list_on_stack,
1638 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001639 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001640 * through a recursive call to generic_make_request. If it
1641 * did, we find a non-NULL value in bio_list and re-enter the loop
1642 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001643 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001644 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001645 */
1646 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001647 bio_list_init(&bio_list_on_stack);
1648 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001649 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001650 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1651
1652 q->make_request_fn(q, bio);
1653
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001654 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001655 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001656 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001657}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658EXPORT_SYMBOL(generic_make_request);
1659
1660/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001661 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1663 * @bio: The &struct bio which describes the I/O
1664 *
1665 * submit_bio() is very similar in purpose to generic_make_request(), and
1666 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001667 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 *
1669 */
1670void submit_bio(int rw, struct bio *bio)
1671{
1672 int count = bio_sectors(bio);
1673
Jens Axboe22e2c502005-06-27 10:55:12 +02001674 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
Jens Axboebf2de6f2007-09-27 13:01:25 +02001676 /*
1677 * If it's a regular read/write or a barrier with data attached,
1678 * go through the normal accounting stuff before submission.
1679 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001680 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001681 if (rw & WRITE) {
1682 count_vm_events(PGPGOUT, count);
1683 } else {
1684 task_io_account_read(bio->bi_size);
1685 count_vm_events(PGPGIN, count);
1686 }
1687
1688 if (unlikely(block_dump)) {
1689 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001690 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001691 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001692 (rw & WRITE) ? "WRITE" : "READ",
1693 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001694 bdevname(bio->bi_bdev, b),
1695 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001696 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 }
1698
1699 generic_make_request(bio);
1700}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701EXPORT_SYMBOL(submit_bio);
1702
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001703/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001704 * blk_rq_check_limits - Helper function to check a request for the queue limit
1705 * @q: the queue
1706 * @rq: the request being checked
1707 *
1708 * Description:
1709 * @rq may have been made based on weaker limitations of upper-level queues
1710 * in request stacking drivers, and it may violate the limitation of @q.
1711 * Since the block layer and the underlying device driver trust @rq
1712 * after it is inserted to @q, it should be checked against @q before
1713 * the insertion using this generic function.
1714 *
1715 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001716 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001717 * Request stacking drivers like request-based dm may change the queue
1718 * limits while requests are in the queue (e.g. dm's table swapping).
1719 * Such request stacking drivers should check those requests agaist
1720 * the new queue limits again when they dispatch those requests,
1721 * although such checkings are also done against the old queue limits
1722 * when submitting requests.
1723 */
1724int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1725{
ike Snitzer33839772010-08-08 12:11:33 -04001726 if (rq->cmd_flags & REQ_DISCARD)
1727 return 0;
1728
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001729 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1730 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001731 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1732 return -EIO;
1733 }
1734
1735 /*
1736 * queue's settings related to segment counting like q->bounce_pfn
1737 * may differ from that of other stacking queues.
1738 * Recalculate it to check the request correctly on this queue's
1739 * limitation.
1740 */
1741 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001742 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001743 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1744 return -EIO;
1745 }
1746
1747 return 0;
1748}
1749EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1750
1751/**
1752 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1753 * @q: the queue to submit the request
1754 * @rq: the request being queued
1755 */
1756int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1757{
1758 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001759 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001760
1761 if (blk_rq_check_limits(q, rq))
1762 return -EIO;
1763
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001764 if (rq->rq_disk &&
1765 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001766 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001767
1768 spin_lock_irqsave(q->queue_lock, flags);
1769
1770 /*
1771 * Submitting request must be dequeued before calling this function
1772 * because it will be linked to another request_queue
1773 */
1774 BUG_ON(blk_queued_rq(rq));
1775
Jeff Moyer4853aba2011-08-15 21:37:25 +02001776 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1777 where = ELEVATOR_INSERT_FLUSH;
1778
1779 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001780 if (where == ELEVATOR_INSERT_FLUSH)
1781 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001782 spin_unlock_irqrestore(q->queue_lock, flags);
1783
1784 return 0;
1785}
1786EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1787
Tejun Heo80a761f2009-07-03 17:48:17 +09001788/**
1789 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1790 * @rq: request to examine
1791 *
1792 * Description:
1793 * A request could be merge of IOs which require different failure
1794 * handling. This function determines the number of bytes which
1795 * can be failed from the beginning of the request without
1796 * crossing into area which need to be retried further.
1797 *
1798 * Return:
1799 * The number of bytes to fail.
1800 *
1801 * Context:
1802 * queue_lock must be held.
1803 */
1804unsigned int blk_rq_err_bytes(const struct request *rq)
1805{
1806 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1807 unsigned int bytes = 0;
1808 struct bio *bio;
1809
1810 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1811 return blk_rq_bytes(rq);
1812
1813 /*
1814 * Currently the only 'mixing' which can happen is between
1815 * different fastfail types. We can safely fail portions
1816 * which have all the failfast bits that the first one has -
1817 * the ones which are at least as eager to fail as the first
1818 * one.
1819 */
1820 for (bio = rq->bio; bio; bio = bio->bi_next) {
1821 if ((bio->bi_rw & ff) != ff)
1822 break;
1823 bytes += bio->bi_size;
1824 }
1825
1826 /* this could lead to infinite loop */
1827 BUG_ON(blk_rq_bytes(rq) && !bytes);
1828 return bytes;
1829}
1830EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1831
Jens Axboebc58ba92009-01-23 10:54:44 +01001832static void blk_account_io_completion(struct request *req, unsigned int bytes)
1833{
Jens Axboec2553b52009-04-24 08:10:11 +02001834 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001835 const int rw = rq_data_dir(req);
1836 struct hd_struct *part;
1837 int cpu;
1838
1839 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001840 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001841 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1842 part_stat_unlock();
1843 }
1844}
1845
1846static void blk_account_io_done(struct request *req)
1847{
Jens Axboebc58ba92009-01-23 10:54:44 +01001848 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001849 * Account IO completion. flush_rq isn't accounted as a
1850 * normal IO on queueing nor completion. Accounting the
1851 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001852 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001853 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001854 unsigned long duration = jiffies - req->start_time;
1855 const int rw = rq_data_dir(req);
1856 struct hd_struct *part;
1857 int cpu;
1858
1859 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001860 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001861
1862 part_stat_inc(cpu, part, ios[rw]);
1863 part_stat_add(cpu, part, ticks[rw], duration);
1864 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001865 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001866
Jens Axboe6c23a962011-01-07 08:43:37 +01001867 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001868 part_stat_unlock();
1869 }
1870}
1871
Tejun Heo53a08802008-12-03 12:41:26 +01001872/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001873 * blk_peek_request - peek at the top of a request queue
1874 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001875 *
1876 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001877 * Return the request at the top of @q. The returned request
1878 * should be started using blk_start_request() before LLD starts
1879 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001880 *
1881 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001882 * Pointer to the request at the top of @q if available. Null
1883 * otherwise.
1884 *
1885 * Context:
1886 * queue_lock must be held.
1887 */
1888struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001889{
1890 struct request *rq;
1891 int ret;
1892
1893 while ((rq = __elv_next_request(q)) != NULL) {
1894 if (!(rq->cmd_flags & REQ_STARTED)) {
1895 /*
1896 * This is the first time the device driver
1897 * sees this request (possibly after
1898 * requeueing). Notify IO scheduler.
1899 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001900 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001901 elv_activate_rq(q, rq);
1902
1903 /*
1904 * just mark as started even if we don't start
1905 * it, a request that has been delayed should
1906 * not be passed by new incoming requests
1907 */
1908 rq->cmd_flags |= REQ_STARTED;
1909 trace_block_rq_issue(q, rq);
1910 }
1911
1912 if (!q->boundary_rq || q->boundary_rq == rq) {
1913 q->end_sector = rq_end_sector(rq);
1914 q->boundary_rq = NULL;
1915 }
1916
1917 if (rq->cmd_flags & REQ_DONTPREP)
1918 break;
1919
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001920 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001921 /*
1922 * make sure space for the drain appears we
1923 * know we can do this because max_hw_segments
1924 * has been adjusted to be one fewer than the
1925 * device can handle
1926 */
1927 rq->nr_phys_segments++;
1928 }
1929
1930 if (!q->prep_rq_fn)
1931 break;
1932
1933 ret = q->prep_rq_fn(q, rq);
1934 if (ret == BLKPREP_OK) {
1935 break;
1936 } else if (ret == BLKPREP_DEFER) {
1937 /*
1938 * the request may have been (partially) prepped.
1939 * we need to keep this request in the front to
1940 * avoid resource deadlock. REQ_STARTED will
1941 * prevent other fs requests from passing this one.
1942 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001943 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001944 !(rq->cmd_flags & REQ_DONTPREP)) {
1945 /*
1946 * remove the space for the drain we added
1947 * so that we don't add it again
1948 */
1949 --rq->nr_phys_segments;
1950 }
1951
1952 rq = NULL;
1953 break;
1954 } else if (ret == BLKPREP_KILL) {
1955 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001956 /*
1957 * Mark this request as started so we don't trigger
1958 * any debug logic in the end I/O path.
1959 */
1960 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001961 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001962 } else {
1963 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1964 break;
1965 }
1966 }
1967
1968 return rq;
1969}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001970EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001971
Tejun Heo9934c8c2009-05-08 11:54:16 +09001972void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001973{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001974 struct request_queue *q = rq->q;
1975
Tejun Heo158dbda2009-04-23 11:05:18 +09001976 BUG_ON(list_empty(&rq->queuelist));
1977 BUG_ON(ELV_ON_HASH(rq));
1978
1979 list_del_init(&rq->queuelist);
1980
1981 /*
1982 * the time frame between a request being removed from the lists
1983 * and to it is freed is accounted as io that is in progress at
1984 * the driver side.
1985 */
Divyesh Shah91952912010-04-01 15:01:41 -07001986 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001987 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001988 set_io_start_time_ns(rq);
1989 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001990}
1991
Tejun Heo5efccd12009-04-23 11:05:18 +09001992/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001993 * blk_start_request - start request processing on the driver
1994 * @req: request to dequeue
1995 *
1996 * Description:
1997 * Dequeue @req and start timeout timer on it. This hands off the
1998 * request to the driver.
1999 *
2000 * Block internal functions which don't want to start timer should
2001 * call blk_dequeue_request().
2002 *
2003 * Context:
2004 * queue_lock must be held.
2005 */
2006void blk_start_request(struct request *req)
2007{
2008 blk_dequeue_request(req);
2009
2010 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002011 * We are now handing the request to the hardware, initialize
2012 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002013 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002014 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002015 if (unlikely(blk_bidi_rq(req)))
2016 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2017
Tejun Heo9934c8c2009-05-08 11:54:16 +09002018 blk_add_timer(req);
2019}
2020EXPORT_SYMBOL(blk_start_request);
2021
2022/**
2023 * blk_fetch_request - fetch a request from a request queue
2024 * @q: request queue to fetch a request from
2025 *
2026 * Description:
2027 * Return the request at the top of @q. The request is started on
2028 * return and LLD can start processing it immediately.
2029 *
2030 * Return:
2031 * Pointer to the request at the top of @q if available. Null
2032 * otherwise.
2033 *
2034 * Context:
2035 * queue_lock must be held.
2036 */
2037struct request *blk_fetch_request(struct request_queue *q)
2038{
2039 struct request *rq;
2040
2041 rq = blk_peek_request(q);
2042 if (rq)
2043 blk_start_request(rq);
2044 return rq;
2045}
2046EXPORT_SYMBOL(blk_fetch_request);
2047
2048/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002049 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002050 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002051 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002052 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002053 *
2054 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002055 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2056 * the request structure even if @req doesn't have leftover.
2057 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002058 *
2059 * This special helper function is only for request stacking drivers
2060 * (e.g. request-based dm) so that they can handle partial completion.
2061 * Actual device drivers should use blk_end_request instead.
2062 *
2063 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2064 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002065 *
2066 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002067 * %false - this request doesn't have any more data
2068 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002069 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002070bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002072 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 struct bio *bio;
2074
Tejun Heo2e60e022009-04-23 11:05:18 +09002075 if (!req->bio)
2076 return false;
2077
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002078 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002079
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002081 * For fs requests, rq is just carrier of independent bio's
2082 * and each partial completion should be handled separately.
2083 * Reset per-request error on each partial completion.
2084 *
2085 * TODO: tj: This is too subtle. It would be better to let
2086 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002088 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 req->errors = 0;
2090
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002091 if (error && req->cmd_type == REQ_TYPE_FS &&
2092 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002093 char *error_type;
2094
2095 switch (error) {
2096 case -ENOLINK:
2097 error_type = "recoverable transport";
2098 break;
2099 case -EREMOTEIO:
2100 error_type = "critical target";
2101 break;
2102 case -EBADE:
2103 error_type = "critical nexus";
2104 break;
2105 case -EIO:
2106 default:
2107 error_type = "I/O";
2108 break;
2109 }
2110 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2111 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2112 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 }
2114
Jens Axboebc58ba92009-01-23 10:54:44 +01002115 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002116
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 total_bytes = bio_nbytes = 0;
2118 while ((bio = req->bio) != NULL) {
2119 int nbytes;
2120
2121 if (nr_bytes >= bio->bi_size) {
2122 req->bio = bio->bi_next;
2123 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002124 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 next_idx = 0;
2126 bio_nbytes = 0;
2127 } else {
2128 int idx = bio->bi_idx + next_idx;
2129
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002130 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002132 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002133 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 break;
2135 }
2136
2137 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2138 BIO_BUG_ON(nbytes > bio->bi_size);
2139
2140 /*
2141 * not a complete bvec done
2142 */
2143 if (unlikely(nbytes > nr_bytes)) {
2144 bio_nbytes += nr_bytes;
2145 total_bytes += nr_bytes;
2146 break;
2147 }
2148
2149 /*
2150 * advance to the next vector
2151 */
2152 next_idx++;
2153 bio_nbytes += nbytes;
2154 }
2155
2156 total_bytes += nbytes;
2157 nr_bytes -= nbytes;
2158
Jens Axboe6728cb02008-01-31 13:03:55 +01002159 bio = req->bio;
2160 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 /*
2162 * end more in this run, or just return 'not-done'
2163 */
2164 if (unlikely(nr_bytes <= 0))
2165 break;
2166 }
2167 }
2168
2169 /*
2170 * completely done
2171 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002172 if (!req->bio) {
2173 /*
2174 * Reset counters so that the request stacking driver
2175 * can find how many bytes remain in the request
2176 * later.
2177 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002178 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002179 return false;
2180 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
2182 /*
2183 * if the request wasn't completed, update state
2184 */
2185 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002186 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 bio->bi_idx += next_idx;
2188 bio_iovec(bio)->bv_offset += nr_bytes;
2189 bio_iovec(bio)->bv_len -= nr_bytes;
2190 }
2191
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002192 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002193 req->buffer = bio_data(req->bio);
2194
2195 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002196 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002197 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002198
Tejun Heo80a761f2009-07-03 17:48:17 +09002199 /* mixed attributes always follow the first bio */
2200 if (req->cmd_flags & REQ_MIXED_MERGE) {
2201 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2202 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2203 }
2204
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002205 /*
2206 * If total number of sectors is less than the first segment
2207 * size, something has gone terribly wrong.
2208 */
2209 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002210 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002211 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002212 }
2213
2214 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002216
Tejun Heo2e60e022009-04-23 11:05:18 +09002217 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218}
Tejun Heo2e60e022009-04-23 11:05:18 +09002219EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
Tejun Heo2e60e022009-04-23 11:05:18 +09002221static bool blk_update_bidi_request(struct request *rq, int error,
2222 unsigned int nr_bytes,
2223 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002224{
Tejun Heo2e60e022009-04-23 11:05:18 +09002225 if (blk_update_request(rq, error, nr_bytes))
2226 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002227
Tejun Heo2e60e022009-04-23 11:05:18 +09002228 /* Bidi request must be completed as a whole */
2229 if (unlikely(blk_bidi_rq(rq)) &&
2230 blk_update_request(rq->next_rq, error, bidi_bytes))
2231 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002232
Jens Axboee2e1a142010-06-09 10:42:09 +02002233 if (blk_queue_add_random(rq->q))
2234 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002235
2236 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237}
2238
James Bottomley28018c22010-07-01 19:49:17 +09002239/**
2240 * blk_unprep_request - unprepare a request
2241 * @req: the request
2242 *
2243 * This function makes a request ready for complete resubmission (or
2244 * completion). It happens only after all error handling is complete,
2245 * so represents the appropriate moment to deallocate any resources
2246 * that were allocated to the request in the prep_rq_fn. The queue
2247 * lock is held when calling this.
2248 */
2249void blk_unprep_request(struct request *req)
2250{
2251 struct request_queue *q = req->q;
2252
2253 req->cmd_flags &= ~REQ_DONTPREP;
2254 if (q->unprep_rq_fn)
2255 q->unprep_rq_fn(q, req);
2256}
2257EXPORT_SYMBOL_GPL(blk_unprep_request);
2258
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259/*
2260 * queue lock must be held
2261 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002262static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002264 if (blk_rq_tagged(req))
2265 blk_queue_end_tag(req->q, req);
2266
James Bottomleyba396a62009-05-27 14:17:08 +02002267 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002269 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002270 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271
Mike Andersone78042e2008-10-30 02:16:20 -07002272 blk_delete_timer(req);
2273
James Bottomley28018c22010-07-01 19:49:17 +09002274 if (req->cmd_flags & REQ_DONTPREP)
2275 blk_unprep_request(req);
2276
2277
Jens Axboebc58ba92009-01-23 10:54:44 +01002278 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002279
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002281 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002282 else {
2283 if (blk_bidi_rq(req))
2284 __blk_put_request(req->next_rq->q, req->next_rq);
2285
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 }
2288}
2289
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002290/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002291 * blk_end_bidi_request - Complete a bidi request
2292 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002293 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002294 * @nr_bytes: number of bytes to complete @rq
2295 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002296 *
2297 * Description:
2298 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002299 * Drivers that supports bidi can safely call this member for any
2300 * type of request, bidi or uni. In the later case @bidi_bytes is
2301 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002302 *
2303 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002304 * %false - we are done with this request
2305 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002306 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002307static bool blk_end_bidi_request(struct request *rq, int error,
2308 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002309{
2310 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002311 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002312
Tejun Heo2e60e022009-04-23 11:05:18 +09002313 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2314 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002315
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002316 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002317 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002318 spin_unlock_irqrestore(q->queue_lock, flags);
2319
Tejun Heo2e60e022009-04-23 11:05:18 +09002320 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002321}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002322
2323/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002324 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2325 * @rq: the request to complete
2326 * @error: %0 for success, < %0 for error
2327 * @nr_bytes: number of bytes to complete @rq
2328 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002329 *
2330 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002331 * Identical to blk_end_bidi_request() except that queue lock is
2332 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002333 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002334 * Return:
2335 * %false - we are done with this request
2336 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002337 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002338bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002339 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002340{
Tejun Heo2e60e022009-04-23 11:05:18 +09002341 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2342 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002343
Tejun Heo2e60e022009-04-23 11:05:18 +09002344 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002345
Tejun Heo2e60e022009-04-23 11:05:18 +09002346 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002347}
2348
2349/**
2350 * blk_end_request - Helper function for drivers to complete the request.
2351 * @rq: the request being processed
2352 * @error: %0 for success, < %0 for error
2353 * @nr_bytes: number of bytes to complete
2354 *
2355 * Description:
2356 * Ends I/O on a number of bytes attached to @rq.
2357 * If @rq has leftover, sets it up for the next range of segments.
2358 *
2359 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002360 * %false - we are done with this request
2361 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002362 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002363bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002364{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002365 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002366}
Jens Axboe56ad1742009-07-28 22:11:24 +02002367EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002368
2369/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002370 * blk_end_request_all - Helper function for drives to finish the request.
2371 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002372 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002373 *
2374 * Description:
2375 * Completely finish @rq.
2376 */
2377void blk_end_request_all(struct request *rq, int error)
2378{
2379 bool pending;
2380 unsigned int bidi_bytes = 0;
2381
2382 if (unlikely(blk_bidi_rq(rq)))
2383 bidi_bytes = blk_rq_bytes(rq->next_rq);
2384
2385 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2386 BUG_ON(pending);
2387}
Jens Axboe56ad1742009-07-28 22:11:24 +02002388EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002389
2390/**
2391 * blk_end_request_cur - Helper function to finish the current request chunk.
2392 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002393 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002394 *
2395 * Description:
2396 * Complete the current consecutively mapped chunk from @rq.
2397 *
2398 * Return:
2399 * %false - we are done with this request
2400 * %true - still buffers pending for this request
2401 */
2402bool blk_end_request_cur(struct request *rq, int error)
2403{
2404 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2405}
Jens Axboe56ad1742009-07-28 22:11:24 +02002406EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002407
2408/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002409 * blk_end_request_err - Finish a request till the next failure boundary.
2410 * @rq: the request to finish till the next failure boundary for
2411 * @error: must be negative errno
2412 *
2413 * Description:
2414 * Complete @rq till the next failure boundary.
2415 *
2416 * Return:
2417 * %false - we are done with this request
2418 * %true - still buffers pending for this request
2419 */
2420bool blk_end_request_err(struct request *rq, int error)
2421{
2422 WARN_ON(error >= 0);
2423 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2424}
2425EXPORT_SYMBOL_GPL(blk_end_request_err);
2426
2427/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002428 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002429 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002430 * @error: %0 for success, < %0 for error
2431 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002432 *
2433 * Description:
2434 * Must be called with queue lock held unlike blk_end_request().
2435 *
2436 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002437 * %false - we are done with this request
2438 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002439 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002440bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002441{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002442 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002443}
Jens Axboe56ad1742009-07-28 22:11:24 +02002444EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002445
2446/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002447 * __blk_end_request_all - Helper function for drives to finish the request.
2448 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002449 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002450 *
2451 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002452 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002453 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002454void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002455{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002456 bool pending;
2457 unsigned int bidi_bytes = 0;
2458
2459 if (unlikely(blk_bidi_rq(rq)))
2460 bidi_bytes = blk_rq_bytes(rq->next_rq);
2461
2462 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2463 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002464}
Jens Axboe56ad1742009-07-28 22:11:24 +02002465EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002466
2467/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002468 * __blk_end_request_cur - Helper function to finish the current request chunk.
2469 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002470 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002471 *
2472 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002473 * Complete the current consecutively mapped chunk from @rq. Must
2474 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002475 *
2476 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002477 * %false - we are done with this request
2478 * %true - still buffers pending for this request
2479 */
2480bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002481{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002482 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002483}
Jens Axboe56ad1742009-07-28 22:11:24 +02002484EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002485
Tejun Heo80a761f2009-07-03 17:48:17 +09002486/**
2487 * __blk_end_request_err - Finish a request till the next failure boundary.
2488 * @rq: the request to finish till the next failure boundary for
2489 * @error: must be negative errno
2490 *
2491 * Description:
2492 * Complete @rq till the next failure boundary. Must be called
2493 * with queue lock held.
2494 *
2495 * Return:
2496 * %false - we are done with this request
2497 * %true - still buffers pending for this request
2498 */
2499bool __blk_end_request_err(struct request *rq, int error)
2500{
2501 WARN_ON(error >= 0);
2502 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2503}
2504EXPORT_SYMBOL_GPL(__blk_end_request_err);
2505
Jens Axboe86db1e22008-01-29 14:53:40 +01002506void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2507 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002509 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002510 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
David Woodhousefb2dce82008-08-05 18:01:53 +01002512 if (bio_has_data(bio)) {
2513 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002514 rq->buffer = bio_data(bio);
2515 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002516 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518
NeilBrown66846572007-08-16 13:31:28 +02002519 if (bio->bi_bdev)
2520 rq->rq_disk = bio->bi_bdev->bd_disk;
2521}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002523#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2524/**
2525 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2526 * @rq: the request to be flushed
2527 *
2528 * Description:
2529 * Flush all pages in @rq.
2530 */
2531void rq_flush_dcache_pages(struct request *rq)
2532{
2533 struct req_iterator iter;
2534 struct bio_vec *bvec;
2535
2536 rq_for_each_segment(bvec, rq, iter)
2537 flush_dcache_page(bvec->bv_page);
2538}
2539EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2540#endif
2541
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002542/**
2543 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2544 * @q : the queue of the device being checked
2545 *
2546 * Description:
2547 * Check if underlying low-level drivers of a device are busy.
2548 * If the drivers want to export their busy state, they must set own
2549 * exporting function using blk_queue_lld_busy() first.
2550 *
2551 * Basically, this function is used only by request stacking drivers
2552 * to stop dispatching requests to underlying devices when underlying
2553 * devices are busy. This behavior helps more I/O merging on the queue
2554 * of the request stacking driver and prevents I/O throughput regression
2555 * on burst I/O load.
2556 *
2557 * Return:
2558 * 0 - Not busy (The request stacking driver should dispatch request)
2559 * 1 - Busy (The request stacking driver should stop dispatching request)
2560 */
2561int blk_lld_busy(struct request_queue *q)
2562{
2563 if (q->lld_busy_fn)
2564 return q->lld_busy_fn(q);
2565
2566 return 0;
2567}
2568EXPORT_SYMBOL_GPL(blk_lld_busy);
2569
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002570/**
2571 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2572 * @rq: the clone request to be cleaned up
2573 *
2574 * Description:
2575 * Free all bios in @rq for a cloned request.
2576 */
2577void blk_rq_unprep_clone(struct request *rq)
2578{
2579 struct bio *bio;
2580
2581 while ((bio = rq->bio) != NULL) {
2582 rq->bio = bio->bi_next;
2583
2584 bio_put(bio);
2585 }
2586}
2587EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2588
2589/*
2590 * Copy attributes of the original request to the clone request.
2591 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2592 */
2593static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2594{
2595 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002596 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002597 dst->cmd_type = src->cmd_type;
2598 dst->__sector = blk_rq_pos(src);
2599 dst->__data_len = blk_rq_bytes(src);
2600 dst->nr_phys_segments = src->nr_phys_segments;
2601 dst->ioprio = src->ioprio;
2602 dst->extra_len = src->extra_len;
2603}
2604
2605/**
2606 * blk_rq_prep_clone - Helper function to setup clone request
2607 * @rq: the request to be setup
2608 * @rq_src: original request to be cloned
2609 * @bs: bio_set that bios for clone are allocated from
2610 * @gfp_mask: memory allocation mask for bio
2611 * @bio_ctr: setup function to be called for each clone bio.
2612 * Returns %0 for success, non %0 for failure.
2613 * @data: private data to be passed to @bio_ctr
2614 *
2615 * Description:
2616 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2617 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2618 * are not copied, and copying such parts is the caller's responsibility.
2619 * Also, pages which the original bios are pointing to are not copied
2620 * and the cloned bios just point same pages.
2621 * So cloned bios must be completed before original bios, which means
2622 * the caller must complete @rq before @rq_src.
2623 */
2624int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2625 struct bio_set *bs, gfp_t gfp_mask,
2626 int (*bio_ctr)(struct bio *, struct bio *, void *),
2627 void *data)
2628{
2629 struct bio *bio, *bio_src;
2630
2631 if (!bs)
2632 bs = fs_bio_set;
2633
2634 blk_rq_init(NULL, rq);
2635
2636 __rq_for_each_bio(bio_src, rq_src) {
2637 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2638 if (!bio)
2639 goto free_and_out;
2640
2641 __bio_clone(bio, bio_src);
2642
2643 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002644 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002645 goto free_and_out;
2646
2647 if (bio_ctr && bio_ctr(bio, bio_src, data))
2648 goto free_and_out;
2649
2650 if (rq->bio) {
2651 rq->biotail->bi_next = bio;
2652 rq->biotail = bio;
2653 } else
2654 rq->bio = rq->biotail = bio;
2655 }
2656
2657 __blk_rq_prep_clone(rq, rq_src);
2658
2659 return 0;
2660
2661free_and_out:
2662 if (bio)
2663 bio_free(bio, bs);
2664 blk_rq_unprep_clone(rq);
2665
2666 return -ENOMEM;
2667}
2668EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2669
Jens Axboe18887ad2008-07-28 13:08:45 +02002670int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671{
2672 return queue_work(kblockd_workqueue, work);
2673}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674EXPORT_SYMBOL(kblockd_schedule_work);
2675
Vivek Goyale43473b2010-09-15 17:06:35 -04002676int kblockd_schedule_delayed_work(struct request_queue *q,
2677 struct delayed_work *dwork, unsigned long delay)
2678{
2679 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2680}
2681EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2682
Jens Axboe73c10102011-03-08 13:19:51 +01002683#define PLUG_MAGIC 0x91827364
2684
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002685/**
2686 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2687 * @plug: The &struct blk_plug that needs to be initialized
2688 *
2689 * Description:
2690 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2691 * pending I/O should the task end up blocking between blk_start_plug() and
2692 * blk_finish_plug(). This is important from a performance perspective, but
2693 * also ensures that we don't deadlock. For instance, if the task is blocking
2694 * for a memory allocation, memory reclaim could end up wanting to free a
2695 * page belonging to that request that is currently residing in our private
2696 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2697 * this kind of deadlock.
2698 */
Jens Axboe73c10102011-03-08 13:19:51 +01002699void blk_start_plug(struct blk_plug *plug)
2700{
2701 struct task_struct *tsk = current;
2702
2703 plug->magic = PLUG_MAGIC;
2704 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002705 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002706 plug->should_sort = 0;
2707
2708 /*
2709 * If this is a nested plug, don't actually assign it. It will be
2710 * flushed on its own.
2711 */
2712 if (!tsk->plug) {
2713 /*
2714 * Store ordering should not be needed here, since a potential
2715 * preempt will imply a full memory barrier
2716 */
2717 tsk->plug = plug;
2718 }
2719}
2720EXPORT_SYMBOL(blk_start_plug);
2721
2722static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2723{
2724 struct request *rqa = container_of(a, struct request, queuelist);
2725 struct request *rqb = container_of(b, struct request, queuelist);
2726
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002727 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002728}
2729
Jens Axboe49cac012011-04-16 13:51:05 +02002730/*
2731 * If 'from_schedule' is true, then postpone the dispatch of requests
2732 * until a safe kblockd context. We due this to avoid accidental big
2733 * additional stack usage in driver dispatch, in places where the originally
2734 * plugger did not intend it.
2735 */
Jens Axboef6603782011-04-15 15:49:07 +02002736static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002737 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002738 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002739{
Jens Axboe49cac012011-04-16 13:51:05 +02002740 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002741
2742 /*
2743 * If we are punting this to kblockd, then we can safely drop
2744 * the queue_lock before waking kblockd (which needs to take
2745 * this lock).
2746 */
2747 if (from_schedule) {
2748 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002749 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002750 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002751 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002752 spin_unlock(q->queue_lock);
2753 }
2754
Jens Axboe94b5eb22011-04-12 10:12:19 +02002755}
2756
NeilBrown048c9372011-04-18 09:52:22 +02002757static void flush_plug_callbacks(struct blk_plug *plug)
2758{
2759 LIST_HEAD(callbacks);
2760
2761 if (list_empty(&plug->cb_list))
2762 return;
2763
2764 list_splice_init(&plug->cb_list, &callbacks);
2765
2766 while (!list_empty(&callbacks)) {
2767 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2768 struct blk_plug_cb,
2769 list);
2770 list_del(&cb->list);
2771 cb->callback(cb);
2772 }
2773}
2774
Jens Axboe49cac012011-04-16 13:51:05 +02002775void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002776{
2777 struct request_queue *q;
2778 unsigned long flags;
2779 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002780 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002781 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002782
2783 BUG_ON(plug->magic != PLUG_MAGIC);
2784
NeilBrown048c9372011-04-18 09:52:22 +02002785 flush_plug_callbacks(plug);
Jens Axboe73c10102011-03-08 13:19:51 +01002786 if (list_empty(&plug->list))
2787 return;
2788
NeilBrown109b8122011-04-11 14:13:10 +02002789 list_splice_init(&plug->list, &list);
2790
2791 if (plug->should_sort) {
2792 list_sort(NULL, &list, plug_rq_cmp);
2793 plug->should_sort = 0;
2794 }
Jens Axboe73c10102011-03-08 13:19:51 +01002795
2796 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002797 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002798
2799 /*
2800 * Save and disable interrupts here, to avoid doing it for every
2801 * queue lock we have to take.
2802 */
Jens Axboe73c10102011-03-08 13:19:51 +01002803 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002804 while (!list_empty(&list)) {
2805 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002806 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002807 BUG_ON(!rq->q);
2808 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002809 /*
2810 * This drops the queue lock
2811 */
2812 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002813 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002814 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002815 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002816 spin_lock(q->queue_lock);
2817 }
Jens Axboe73c10102011-03-08 13:19:51 +01002818 /*
2819 * rq is already accounted, so use raw insert
2820 */
Jens Axboe401a18e2011-03-25 16:57:52 +01002821 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2822 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2823 else
2824 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002825
2826 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01002827 }
2828
Jens Axboe99e22592011-04-18 09:59:55 +02002829 /*
2830 * This drops the queue lock
2831 */
2832 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002833 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002834
Jens Axboe73c10102011-03-08 13:19:51 +01002835 local_irq_restore(flags);
2836}
Jens Axboe73c10102011-03-08 13:19:51 +01002837
2838void blk_finish_plug(struct blk_plug *plug)
2839{
Jens Axboef6603782011-04-15 15:49:07 +02002840 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02002841
2842 if (plug == current->plug)
2843 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01002844}
2845EXPORT_SYMBOL(blk_finish_plug);
2846
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847int __init blk_dev_init(void)
2848{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002849 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2850 sizeof(((struct request *)0)->cmd_flags));
2851
Tejun Heo89b90be2011-01-03 15:01:47 +01002852 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2853 kblockd_workqueue = alloc_workqueue("kblockd",
2854 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 if (!kblockd_workqueue)
2856 panic("Failed to create kblockd\n");
2857
2858 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002859 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860
Jens Axboe8324aa92008-01-29 14:51:59 +01002861 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002862 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 return 0;
2865}