blob: ea70e6c80cd34f0c2b563ceabcc7189027fc5182 [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
369 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200370
371 if (drain_all)
372 nr_rqs = q->rq.count[0] + q->rq.count[1];
373 else
374 nr_rqs = q->rq.elvpriv;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200375
376 spin_unlock_irq(q->queue_lock);
377
378 if (!nr_rqs)
379 break;
380 msleep(10);
381 }
382}
383
Tejun Heoc9a929d2011-10-19 14:42:16 +0200384/**
385 * blk_cleanup_queue - shutdown a request queue
386 * @q: request queue to shutdown
387 *
388 * Mark @q DEAD, drain all pending requests, destroy and put it. All
389 * future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500390 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100391void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500392{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200393 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de2008-09-18 09:22:54 -0700394
Tejun Heoc9a929d2011-10-19 14:42:16 +0200395 /* mark @q DEAD, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500396 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200397 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200398
399 spin_lock_irq(lock);
400 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
401 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
402 queue_flag_set(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500403
Hannes Reinecke777eb1b2011-09-28 08:07:01 -0600404 if (q->queue_lock != &q->__queue_lock)
405 q->queue_lock = &q->__queue_lock;
Vivek Goyalda527772011-03-02 19:05:33 -0500406
Tejun Heoc9a929d2011-10-19 14:42:16 +0200407 spin_unlock_irq(lock);
408 mutex_unlock(&q->sysfs_lock);
409
Tejun Heo6dd9ad72011-11-03 18:52:11 +0100410 /*
411 * Drain all requests queued before DEAD marking. The caller might
412 * be trying to tear down @q before its elevator is initialized, in
413 * which case we don't want to call into draining.
414 */
415 if (q->elevator)
416 blk_drain_queue(q, true);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200417
418 /* @q won't process any more request, flush async actions */
419 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
420 blk_sync_queue(q);
421
422 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500423 blk_put_queue(q);
424}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425EXPORT_SYMBOL(blk_cleanup_queue);
426
Jens Axboe165125e2007-07-24 09:28:11 +0200427static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
429 struct request_list *rl = &q->rq;
430
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400431 if (unlikely(rl->rq_pool))
432 return 0;
433
Jens Axboe1faa16d2009-04-06 14:48:01 +0200434 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
435 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200436 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200437 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
438 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
Christoph Lameter19460892005-06-23 00:08:19 -0700440 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
441 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
443 if (!rl->rq_pool)
444 return -ENOMEM;
445
446 return 0;
447}
448
Jens Axboe165125e2007-07-24 09:28:11 +0200449struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450{
Christoph Lameter19460892005-06-23 00:08:19 -0700451 return blk_alloc_queue_node(gfp_mask, -1);
452}
453EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
Jens Axboe165125e2007-07-24 09:28:11 +0200455struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700456{
Jens Axboe165125e2007-07-24 09:28:11 +0200457 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700458 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700459
Jens Axboe8324aa92008-01-29 14:51:59 +0100460 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700461 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 if (!q)
463 return NULL;
464
Jens Axboe0989a022009-06-12 14:42:56 +0200465 q->backing_dev_info.ra_pages =
466 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
467 q->backing_dev_info.state = 0;
468 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200469 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200470
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700471 err = bdi_init(&q->backing_dev_info);
472 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100473 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700474 return NULL;
475 }
476
Vivek Goyale43473b2010-09-15 17:06:35 -0400477 if (blk_throtl_init(q)) {
478 kmem_cache_free(blk_requestq_cachep, q);
479 return NULL;
480 }
481
Matthew Garrett31373d02010-04-06 14:25:14 +0200482 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
483 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700484 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
485 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoae1b1532011-01-25 12:43:54 +0100486 INIT_LIST_HEAD(&q->flush_queue[0]);
487 INIT_LIST_HEAD(&q->flush_queue[1]);
488 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500489 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500490
Jens Axboe8324aa92008-01-29 14:51:59 +0100491 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Al Viro483f4af2006-03-18 18:34:37 -0500493 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700494 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500495
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500496 /*
497 * By default initialize queue_lock to internal lock and driver can
498 * override it later if need be.
499 */
500 q->queue_lock = &q->__queue_lock;
501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 return q;
503}
Christoph Lameter19460892005-06-23 00:08:19 -0700504EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
506/**
507 * blk_init_queue - prepare a request queue for use with a block device
508 * @rfn: The function to be called to process requests that have been
509 * placed on the queue.
510 * @lock: Request queue spin lock
511 *
512 * Description:
513 * If a block device wishes to use the standard request handling procedures,
514 * which sorts requests and coalesces adjacent requests, then it must
515 * call blk_init_queue(). The function @rfn will be called when there
516 * are requests on the queue that need to be processed. If the device
517 * supports plugging, then @rfn may not be called immediately when requests
518 * are available on the queue, but may be called at some time later instead.
519 * Plugged queues are generally unplugged when a buffer belonging to one
520 * of the requests on the queue is needed, or due to memory pressure.
521 *
522 * @rfn is not required, or even expected, to remove all requests off the
523 * queue, but only as many as it can handle at a time. If it does leave
524 * requests on the queue, it is responsible for arranging that the requests
525 * get dealt with eventually.
526 *
527 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200528 * request queue; this lock will be taken also from interrupt context, so irq
529 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200531 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 * it didn't succeed.
533 *
534 * Note:
535 * blk_init_queue() must be paired with a blk_cleanup_queue() call
536 * when the block device is deactivated (such as at module unload).
537 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700538
Jens Axboe165125e2007-07-24 09:28:11 +0200539struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540{
Christoph Lameter19460892005-06-23 00:08:19 -0700541 return blk_init_queue_node(rfn, lock, -1);
542}
543EXPORT_SYMBOL(blk_init_queue);
544
Jens Axboe165125e2007-07-24 09:28:11 +0200545struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700546blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
547{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600548 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600550 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
551 if (!uninit_q)
552 return NULL;
553
554 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
555 if (!q)
556 blk_cleanup_queue(uninit_q);
557
558 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200559}
560EXPORT_SYMBOL(blk_init_queue_node);
561
562struct request_queue *
563blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
564 spinlock_t *lock)
565{
566 return blk_init_allocated_queue_node(q, rfn, lock, -1);
567}
568EXPORT_SYMBOL(blk_init_allocated_queue);
569
570struct request_queue *
571blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
572 spinlock_t *lock, int node_id)
573{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 if (!q)
575 return NULL;
576
Christoph Lameter19460892005-06-23 00:08:19 -0700577 q->node = node_id;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600578 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500579 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900583 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100584 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500585
586 /* Override internal queue lock with supplied lock pointer */
587 if (lock)
588 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Jens Axboef3b144a2009-03-06 08:48:33 +0100590 /*
591 * This also sets hw/phys segments, boundary and size
592 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200593 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Alan Stern44ec9542007-02-20 11:01:57 -0500595 q->sg_reserved_size = INT_MAX;
596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 /*
598 * all done
599 */
600 if (!elevator_init(q, NULL)) {
601 blk_queue_congestion_threshold(q);
602 return q;
603 }
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 return NULL;
606}
Mike Snitzer01effb02010-05-11 08:57:42 +0200607EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
Jens Axboe165125e2007-07-24 09:28:11 +0200609int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700611 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500612 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 return 0;
614 }
615
616 return 1;
617}
Jens Axboed86e0e82011-05-27 07:44:43 +0200618EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Jens Axboe165125e2007-07-24 09:28:11 +0200620static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200622 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200623 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 mempool_free(rq, q->rq.rq_pool);
625}
626
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200627static struct request *
Tejun Heo75eb6c32011-10-19 14:31:22 +0200628blk_alloc_request(struct request_queue *q, unsigned int flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
630 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
631
632 if (!rq)
633 return NULL;
634
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200635 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200636
Jerome Marchand42dad762009-04-22 14:01:49 +0200637 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
Tejun Heo75eb6c32011-10-19 14:31:22 +0200639 if ((flags & REQ_ELVPRIV) &&
640 unlikely(elv_set_request(q, rq, gfp_mask))) {
641 mempool_free(rq, q->rq.rq_pool);
642 return NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Tejun Heocb98fc82005-10-28 08:29:39 +0200645 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
648/*
649 * ioc_batching returns true if the ioc is a valid batching request and
650 * should be given priority access to a request.
651 */
Jens Axboe165125e2007-07-24 09:28:11 +0200652static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653{
654 if (!ioc)
655 return 0;
656
657 /*
658 * Make sure the process is able to allocate at least 1 request
659 * even if the batch times out, otherwise we could theoretically
660 * lose wakeups.
661 */
662 return ioc->nr_batch_requests == q->nr_batching ||
663 (ioc->nr_batch_requests > 0
664 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
665}
666
667/*
668 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
669 * will cause the process to be a "batcher" on all queues in the system. This
670 * is the behaviour we want though - once it gets a wakeup it should be given
671 * a nice run.
672 */
Jens Axboe165125e2007-07-24 09:28:11 +0200673static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
675 if (!ioc || ioc_batching(q, ioc))
676 return;
677
678 ioc->nr_batch_requests = q->nr_batching;
679 ioc->last_waited = jiffies;
680}
681
Jens Axboe1faa16d2009-04-06 14:48:01 +0200682static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
684 struct request_list *rl = &q->rq;
685
Jens Axboe1faa16d2009-04-06 14:48:01 +0200686 if (rl->count[sync] < queue_congestion_off_threshold(q))
687 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Jens Axboe1faa16d2009-04-06 14:48:01 +0200689 if (rl->count[sync] + 1 <= q->nr_requests) {
690 if (waitqueue_active(&rl->wait[sync]))
691 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Jens Axboe1faa16d2009-04-06 14:48:01 +0200693 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
695}
696
697/*
698 * A request has just been released. Account for it, update the full and
699 * congestion status, wake up any waiters. Called under q->queue_lock.
700 */
Tejun Heo75eb6c32011-10-19 14:31:22 +0200701static void freed_request(struct request_queue *q, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
703 struct request_list *rl = &q->rq;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200704 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
Jens Axboe1faa16d2009-04-06 14:48:01 +0200706 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200707 if (flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200708 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
Jens Axboe1faa16d2009-04-06 14:48:01 +0200710 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
Jens Axboe1faa16d2009-04-06 14:48:01 +0200712 if (unlikely(rl->starved[sync ^ 1]))
713 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714}
715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100717 * Determine if elevator data should be initialized when allocating the
718 * request associated with @bio.
719 */
720static bool blk_rq_should_init_elevator(struct bio *bio)
721{
722 if (!bio)
723 return true;
724
725 /*
726 * Flush requests do not use the elevator so skip initialization.
727 * This allows a request to share the flush and elevator data.
728 */
729 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
730 return false;
731
732 return true;
733}
734
Tejun Heoda8303c2011-10-19 14:33:05 +0200735/**
736 * get_request - get a free request
737 * @q: request_queue to allocate request from
738 * @rw_flags: RW and SYNC flags
739 * @bio: bio to allocate request for (can be %NULL)
740 * @gfp_mask: allocation mask
741 *
742 * Get a free request from @q. This function may fail under memory
743 * pressure or if @q is dead.
744 *
745 * Must be callled with @q->queue_lock held and,
746 * Returns %NULL on failure, with @q->queue_lock held.
747 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 */
Jens Axboe165125e2007-07-24 09:28:11 +0200749static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100750 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751{
752 struct request *rq = NULL;
753 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100754 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200755 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200756 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Tejun Heoda8303c2011-10-19 14:33:05 +0200758 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
759 return NULL;
760
Jens Axboe7749a8d2006-12-13 13:02:26 +0100761 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100762 if (may_queue == ELV_MQUEUE_NO)
763 goto rq_starved;
764
Jens Axboe1faa16d2009-04-06 14:48:01 +0200765 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
766 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200767 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100768 /*
769 * The queue will fill after this allocation, so set
770 * it as full, and mark this process as "batching".
771 * This process will be allowed to complete a batch of
772 * requests, others will be blocked.
773 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200774 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100775 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200776 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100777 } else {
778 if (may_queue != ELV_MQUEUE_MUST
779 && !ioc_batching(q, ioc)) {
780 /*
781 * The queue is full and the allocating
782 * process is not a "batcher", and not
783 * exempted by the IO scheduler
784 */
785 goto out;
786 }
787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200789 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 }
791
Jens Axboe082cf692005-06-28 16:35:11 +0200792 /*
793 * Only allow batching queuers to allocate up to 50% over the defined
794 * limit of requests, otherwise we could have thousands of requests
795 * allocated with any setting of ->nr_requests
796 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200797 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200798 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100799
Jens Axboe1faa16d2009-04-06 14:48:01 +0200800 rl->count[is_sync]++;
801 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200802
Tejun Heo75eb6c32011-10-19 14:31:22 +0200803 if (blk_rq_should_init_elevator(bio) &&
804 !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
805 rw_flags |= REQ_ELVPRIV;
806 rl->elvpriv++;
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100807 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200808
Jens Axboef253b862010-10-24 22:06:02 +0200809 if (blk_queue_io_stat(q))
810 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 spin_unlock_irq(q->queue_lock);
812
Tejun Heo75eb6c32011-10-19 14:31:22 +0200813 rq = blk_alloc_request(q, rw_flags, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100814 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 /*
816 * Allocation failed presumably due to memory. Undo anything
817 * we might have messed up.
818 *
819 * Allocating task should really be put onto the front of the
820 * wait queue, but this is pretty rare.
821 */
822 spin_lock_irq(q->queue_lock);
Tejun Heo75eb6c32011-10-19 14:31:22 +0200823 freed_request(q, rw_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
825 /*
826 * in the very unlikely event that allocation failed and no
827 * requests for this direction was pending, mark us starved
828 * so that freeing of a request in the other direction will
829 * notice us. another possible fix would be to split the
830 * rq mempool into READ and WRITE
831 */
832rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200833 if (unlikely(rl->count[is_sync] == 0))
834 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 goto out;
837 }
838
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100839 /*
840 * ioc may be NULL here, and ioc_batching will be false. That's
841 * OK, if the queue is under the request limit then requests need
842 * not count toward the nr_batch_requests limit. There will always
843 * be some limit enforced by BLK_BATCH_TIME.
844 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 if (ioc_batching(q, ioc))
846 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100847
Jens Axboe1faa16d2009-04-06 14:48:01 +0200848 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 return rq;
851}
852
Tejun Heoda8303c2011-10-19 14:33:05 +0200853/**
854 * get_request_wait - get a free request with retry
855 * @q: request_queue to allocate request from
856 * @rw_flags: RW and SYNC flags
857 * @bio: bio to allocate request for (can be %NULL)
Nick Piggind6344532005-06-28 20:45:14 -0700858 *
Tejun Heoda8303c2011-10-19 14:33:05 +0200859 * Get a free request from @q. This function keeps retrying under memory
860 * pressure and fails iff @q is dead.
861 *
862 * Must be callled with @q->queue_lock held and,
863 * Returns %NULL on failure, with @q->queue_lock held.
864 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 */
Jens Axboe165125e2007-07-24 09:28:11 +0200866static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200867 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200869 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 struct request *rq;
871
Jens Axboe7749a8d2006-12-13 13:02:26 +0100872 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700873 while (!rq) {
874 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200875 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 struct request_list *rl = &q->rq;
877
Tejun Heoda8303c2011-10-19 14:33:05 +0200878 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
879 return NULL;
880
Jens Axboe1faa16d2009-04-06 14:48:01 +0200881 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 TASK_UNINTERRUPTIBLE);
883
Jens Axboe1faa16d2009-04-06 14:48:01 +0200884 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200886 spin_unlock_irq(q->queue_lock);
887 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200889 /*
890 * After sleeping, we become a "batching" process and
891 * will be able to allocate at least one request, and
892 * up to a big batch of them for a small period time.
893 * See ioc_batching, ioc_set_batching
894 */
895 ioc = current_io_context(GFP_NOIO, q->node);
896 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100897
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200898 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200899 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200900
901 rq = get_request(q, rw_flags, bio, GFP_NOIO);
902 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
904 return rq;
905}
906
Jens Axboe165125e2007-07-24 09:28:11 +0200907struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908{
909 struct request *rq;
910
911 BUG_ON(rw != READ && rw != WRITE);
912
Nick Piggind6344532005-06-28 20:45:14 -0700913 spin_lock_irq(q->queue_lock);
Tejun Heoda8303c2011-10-19 14:33:05 +0200914 if (gfp_mask & __GFP_WAIT)
Jens Axboe22e2c502005-06-27 10:55:12 +0200915 rq = get_request_wait(q, rw, NULL);
Tejun Heoda8303c2011-10-19 14:33:05 +0200916 else
Jens Axboe22e2c502005-06-27 10:55:12 +0200917 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +0200918 if (!rq)
919 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -0700920 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
922 return rq;
923}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924EXPORT_SYMBOL(blk_get_request);
925
926/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300927 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700928 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300929 * @bio: The bio describing the memory mappings that will be submitted for IO.
930 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700931 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200932 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300933 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
934 * type commands. Where the struct request needs to be farther initialized by
935 * the caller. It is passed a &struct bio, which describes the memory info of
936 * the I/O transfer.
937 *
938 * The caller of blk_make_request must make sure that bi_io_vec
939 * are set to describe the memory buffers. That bio_data_dir() will return
940 * the needed direction of the request. (And all bio's in the passed bio-chain
941 * are properly set accordingly)
942 *
943 * If called under none-sleepable conditions, mapped bio buffers must not
944 * need bouncing, by calling the appropriate masked or flagged allocator,
945 * suitable for the target device. Otherwise the call to blk_queue_bounce will
946 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200947 *
948 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
949 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
950 * anything but the first bio in the chain. Otherwise you risk waiting for IO
951 * completion of a bio that hasn't been submitted yet, thus resulting in a
952 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
953 * of bio_alloc(), as that avoids the mempool deadlock.
954 * If possible a big IO should be split into smaller parts when allocation
955 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200956 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300957struct request *blk_make_request(struct request_queue *q, struct bio *bio,
958 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200959{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300960 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
961
962 if (unlikely(!rq))
963 return ERR_PTR(-ENOMEM);
964
965 for_each_bio(bio) {
966 struct bio *bounce_bio = bio;
967 int ret;
968
969 blk_queue_bounce(q, &bounce_bio);
970 ret = blk_rq_append_bio(q, rq, bounce_bio);
971 if (unlikely(ret)) {
972 blk_put_request(rq);
973 return ERR_PTR(ret);
974 }
975 }
976
977 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200978}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300979EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200980
981/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 * blk_requeue_request - put a request back on queue
983 * @q: request queue where request should be inserted
984 * @rq: request to be inserted
985 *
986 * Description:
987 * Drivers often keep queueing requests until the hardware cannot accept
988 * more, when that condition happens we need to put the request back
989 * on the queue. Must be called with queue lock held.
990 */
Jens Axboe165125e2007-07-24 09:28:11 +0200991void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700993 blk_delete_timer(rq);
994 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100995 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 if (blk_rq_tagged(rq))
998 blk_queue_end_tag(q, rq);
999
James Bottomleyba396a62009-05-27 14:17:08 +02001000 BUG_ON(blk_queued_rq(rq));
1001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 elv_requeue_request(q, rq);
1003}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004EXPORT_SYMBOL(blk_requeue_request);
1005
Jens Axboe73c10102011-03-08 13:19:51 +01001006static void add_acct_request(struct request_queue *q, struct request *rq,
1007 int where)
1008{
1009 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001010 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001011}
1012
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001014 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 * @q: request queue where request should be inserted
1016 * @rq: request to be inserted
1017 * @at_head: insert request at head or tail of queue
1018 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 *
1020 * Description:
1021 * Many block devices need to execute commands asynchronously, so they don't
1022 * block the whole kernel from preemption during request execution. This is
1023 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +02001024 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1025 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 *
1027 * We have the option of inserting the head or the tail of the queue.
1028 * Typically we use the tail for new ioctls and so forth. We use the head
1029 * of the queue for things like a QUEUE_FULL message from a device, or a
1030 * host that is unable to accept a particular command.
1031 */
Jens Axboe165125e2007-07-24 09:28:11 +02001032void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05001033 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
Tejun Heo 867d1192005-04-24 02:06:05 -05001035 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 unsigned long flags;
1037
1038 /*
1039 * tell I/O scheduler that this isn't a regular read/write (ie it
1040 * must not attempt merges on this) and that it acts as a soft
1041 * barrier
1042 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02001043 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045 rq->special = data;
1046
1047 spin_lock_irqsave(q->queue_lock, flags);
1048
1049 /*
1050 * If command is tagged, release the tag
1051 */
Tejun Heo 867d1192005-04-24 02:06:05 -05001052 if (blk_rq_tagged(rq))
1053 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Jens Axboe73c10102011-03-08 13:19:51 +01001055 add_acct_request(q, rq, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001056 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 spin_unlock_irqrestore(q->queue_lock, flags);
1058}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059EXPORT_SYMBOL(blk_insert_request);
1060
Tejun Heo074a7ac2008-08-25 19:56:14 +09001061static void part_round_stats_single(int cpu, struct hd_struct *part,
1062 unsigned long now)
1063{
1064 if (now == part->stamp)
1065 return;
1066
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001067 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001068 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001069 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001070 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1071 }
1072 part->stamp = now;
1073}
1074
1075/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001076 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1077 * @cpu: cpu number for stats access
1078 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 *
1080 * The average IO queue length and utilisation statistics are maintained
1081 * by observing the current state of the queue length and the amount of
1082 * time it has been in this state for.
1083 *
1084 * Normally, that accounting is done on IO completion, but that can result
1085 * in more than a second's worth of IO being accounted for within any one
1086 * second, leading to >100% utilisation. To deal with that, we call this
1087 * function to do a round-off before returning the results when reading
1088 * /proc/diskstats. This accounts immediately for all queue usage up to
1089 * the current jiffies and restarts the counters again.
1090 */
Tejun Heoc9959052008-08-25 19:47:21 +09001091void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001092{
1093 unsigned long now = jiffies;
1094
Tejun Heo074a7ac2008-08-25 19:56:14 +09001095 if (part->partno)
1096 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1097 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001098}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001099EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101/*
1102 * queue lock must be held
1103 */
Jens Axboe165125e2007-07-24 09:28:11 +02001104void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 if (unlikely(!q))
1107 return;
1108 if (unlikely(--req->ref_count))
1109 return;
1110
Tejun Heo8922e162005-10-20 16:23:44 +02001111 elv_completed_request(q, req);
1112
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001113 /* this is a bio leak */
1114 WARN_ON(req->bio != NULL);
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 /*
1117 * Request may not have originated from ll_rw_blk. if not,
1118 * it didn't come out of our reserved rq pools
1119 */
Jens Axboe49171e52006-08-10 08:59:11 +02001120 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001121 unsigned int flags = req->cmd_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001124 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
1126 blk_free_request(q, req);
Tejun Heo75eb6c32011-10-19 14:31:22 +02001127 freed_request(q, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 }
1129}
Mike Christie6e39b692005-11-11 05:30:24 -06001130EXPORT_SYMBOL_GPL(__blk_put_request);
1131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132void blk_put_request(struct request *req)
1133{
Tejun Heo8922e162005-10-20 16:23:44 +02001134 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001135 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001137 spin_lock_irqsave(q->queue_lock, flags);
1138 __blk_put_request(q, req);
1139 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141EXPORT_SYMBOL(blk_put_request);
1142
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001143/**
1144 * blk_add_request_payload - add a payload to a request
1145 * @rq: request to update
1146 * @page: page backing the payload
1147 * @len: length of the payload.
1148 *
1149 * This allows to later add a payload to an already submitted request by
1150 * a block driver. The driver needs to take care of freeing the payload
1151 * itself.
1152 *
1153 * Note that this is a quite horrible hack and nothing but handling of
1154 * discard requests should ever use it.
1155 */
1156void blk_add_request_payload(struct request *rq, struct page *page,
1157 unsigned int len)
1158{
1159 struct bio *bio = rq->bio;
1160
1161 bio->bi_io_vec->bv_page = page;
1162 bio->bi_io_vec->bv_offset = 0;
1163 bio->bi_io_vec->bv_len = len;
1164
1165 bio->bi_size = len;
1166 bio->bi_vcnt = 1;
1167 bio->bi_phys_segments = 1;
1168
1169 rq->__data_len = rq->resid_len = len;
1170 rq->nr_phys_segments = 1;
1171 rq->buffer = bio_data(bio);
1172}
1173EXPORT_SYMBOL_GPL(blk_add_request_payload);
1174
Jens Axboe73c10102011-03-08 13:19:51 +01001175static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1176 struct bio *bio)
1177{
1178 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1179
Jens Axboe73c10102011-03-08 13:19:51 +01001180 if (!ll_back_merge_fn(q, req, bio))
1181 return false;
1182
1183 trace_block_bio_backmerge(q, bio);
1184
1185 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1186 blk_rq_set_mixed_merge(req);
1187
1188 req->biotail->bi_next = bio;
1189 req->biotail = bio;
1190 req->__data_len += bio->bi_size;
1191 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1192
1193 drive_stat_acct(req, 0);
Vivek Goyal95cf3dd2011-05-23 10:02:19 +02001194 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001195 return true;
1196}
1197
1198static bool bio_attempt_front_merge(struct request_queue *q,
1199 struct request *req, struct bio *bio)
1200{
1201 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001202
Jens Axboe73c10102011-03-08 13:19:51 +01001203 if (!ll_front_merge_fn(q, req, bio))
1204 return false;
1205
1206 trace_block_bio_frontmerge(q, bio);
1207
1208 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1209 blk_rq_set_mixed_merge(req);
1210
Jens Axboe73c10102011-03-08 13:19:51 +01001211 bio->bi_next = req->bio;
1212 req->bio = bio;
1213
1214 /*
1215 * may not be valid. if the low level driver said
1216 * it didn't need a bounce buffer then it better
1217 * not touch req->buffer either...
1218 */
1219 req->buffer = bio_data(bio);
1220 req->__sector = bio->bi_sector;
1221 req->__data_len += bio->bi_size;
1222 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1223
1224 drive_stat_acct(req, 0);
Vivek Goyal95cf3dd2011-05-23 10:02:19 +02001225 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001226 return true;
1227}
1228
Tejun Heobd87b582011-10-19 14:33:08 +02001229/**
1230 * attempt_plug_merge - try to merge with %current's plugged list
1231 * @q: request_queue new bio is being queued at
1232 * @bio: new bio being queued
1233 * @request_count: out parameter for number of traversed plugged requests
1234 *
1235 * Determine whether @bio being queued on @q can be merged with a request
1236 * on %current's plugged list. Returns %true if merge was successful,
1237 * otherwise %false.
1238 *
1239 * This function is called without @q->queue_lock; however, elevator is
1240 * accessed iff there already are requests on the plugged list which in
1241 * turn guarantees validity of the elevator.
1242 *
1243 * Note that, on successful merge, elevator operation
1244 * elevator_bio_merged_fn() will be called without queue lock. Elevator
1245 * must be ready for this.
Jens Axboe73c10102011-03-08 13:19:51 +01001246 */
Tejun Heobd87b582011-10-19 14:33:08 +02001247static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1248 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001249{
1250 struct blk_plug *plug;
1251 struct request *rq;
1252 bool ret = false;
1253
Tejun Heobd87b582011-10-19 14:33:08 +02001254 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001255 if (!plug)
1256 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001257 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001258
1259 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1260 int el_ret;
1261
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001262 (*request_count)++;
1263
Jens Axboe73c10102011-03-08 13:19:51 +01001264 if (rq->q != q)
1265 continue;
1266
1267 el_ret = elv_try_merge(rq, bio);
1268 if (el_ret == ELEVATOR_BACK_MERGE) {
1269 ret = bio_attempt_back_merge(q, rq, bio);
1270 if (ret)
1271 break;
1272 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1273 ret = bio_attempt_front_merge(q, rq, bio);
1274 if (ret)
1275 break;
1276 }
1277 }
1278out:
1279 return ret;
1280}
1281
Jens Axboe86db1e22008-01-29 14:53:40 +01001282void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001283{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001284 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001285
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001286 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1287 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001288 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001289
Tejun Heo52d9e672006-01-06 09:49:58 +01001290 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001291 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001292 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001293 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001294}
1295
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001296void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001298 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001299 struct blk_plug *plug;
1300 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1301 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001302 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 /*
1305 * low level driver can indicate that it wants pages above a
1306 * certain limit bounced to low memory (ie for highmem, or even
1307 * ISA dma in theory)
1308 */
1309 blk_queue_bounce(q, &bio);
1310
Tejun Heo4fed9472010-09-03 11:56:17 +02001311 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001312 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001313 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001314 goto get_rq;
1315 }
1316
Jens Axboe73c10102011-03-08 13:19:51 +01001317 /*
1318 * Check if we can merge with the plugged list before grabbing
1319 * any locks.
1320 */
Tejun Heobd87b582011-10-19 14:33:08 +02001321 if (attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001322 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001323
1324 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
1326 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001327 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001328 if (bio_attempt_back_merge(q, req, bio)) {
1329 if (!attempt_back_merge(q, req))
1330 elv_merged_request(q, req, el_ret);
1331 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001332 }
Jens Axboe73c10102011-03-08 13:19:51 +01001333 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001334 if (bio_attempt_front_merge(q, req, bio)) {
1335 if (!attempt_front_merge(q, req))
1336 elv_merged_request(q, req, el_ret);
1337 goto out_unlock;
1338 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 }
1340
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001342 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001343 * This sync check and mask will be re-done in init_request_from_bio(),
1344 * but we need to set it earlier to expose the sync flag to the
1345 * rq allocator and io schedulers.
1346 */
1347 rw_flags = bio_data_dir(bio);
1348 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001349 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001350
1351 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001352 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001353 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001354 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001355 req = get_request_wait(q, rw_flags, bio);
Tejun Heoda8303c2011-10-19 14:33:05 +02001356 if (unlikely(!req)) {
1357 bio_endio(bio, -ENODEV); /* @q is dead */
1358 goto out_unlock;
1359 }
Nick Piggind6344532005-06-28 20:45:14 -07001360
Nick Piggin450991b2005-06-28 20:45:13 -07001361 /*
1362 * After dropping the lock and possibly sleeping here, our request
1363 * may now be mergeable after it had proven unmergeable (above).
1364 * We don't worry about that case for efficiency. It won't happen
1365 * often, and the elevators are able to handle it.
1366 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001367 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Tao Ma9562ad92011-10-24 16:11:30 +02001369 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001370 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001371
Jens Axboe73c10102011-03-08 13:19:51 +01001372 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001373 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001374 /*
1375 * If this is the first request added after a plug, fire
1376 * of a plug trace. If others have been added before, check
1377 * if we have multiple devices in this plug. If so, make a
1378 * note to sort the list before dispatch.
1379 */
1380 if (list_empty(&plug->list))
1381 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001382 else {
1383 if (!plug->should_sort) {
1384 struct request *__rq;
Jens Axboe73c10102011-03-08 13:19:51 +01001385
Shaohua Li3540d5e2011-11-16 09:21:50 +01001386 __rq = list_entry_rq(plug->list.prev);
1387 if (__rq->q != q)
1388 plug->should_sort = 1;
1389 }
Shaohua Li019ceb72011-11-16 09:21:50 +01001390 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001391 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001392 trace_block_plug(q);
1393 }
Jens Axboe73c10102011-03-08 13:19:51 +01001394 }
Shaohua Lia6327162011-08-24 16:04:32 +02001395 list_add_tail(&req->queuelist, &plug->list);
1396 drive_stat_acct(req, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001397 } else {
1398 spin_lock_irq(q->queue_lock);
1399 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001400 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001401out_unlock:
1402 spin_unlock_irq(q->queue_lock);
1403 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404}
Jens Axboec20e8de2011-09-12 12:03:37 +02001405EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
1407/*
1408 * If bio->bi_dev is a partition, remap the location
1409 */
1410static inline void blk_partition_remap(struct bio *bio)
1411{
1412 struct block_device *bdev = bio->bi_bdev;
1413
Jens Axboebf2de6f2007-09-27 13:01:25 +02001414 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001416
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 bio->bi_sector += p->start_sect;
1418 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001419
Mike Snitzerd07335e2010-11-16 12:52:38 +01001420 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1421 bdev->bd_dev,
1422 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 }
1424}
1425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426static void handle_bad_sector(struct bio *bio)
1427{
1428 char b[BDEVNAME_SIZE];
1429
1430 printk(KERN_INFO "attempt to access beyond end of device\n");
1431 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1432 bdevname(bio->bi_bdev, b),
1433 bio->bi_rw,
1434 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001435 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
1437 set_bit(BIO_EOF, &bio->bi_flags);
1438}
1439
Akinobu Mitac17bb492006-12-08 02:39:46 -08001440#ifdef CONFIG_FAIL_MAKE_REQUEST
1441
1442static DECLARE_FAULT_ATTR(fail_make_request);
1443
1444static int __init setup_fail_make_request(char *str)
1445{
1446 return setup_fault_attr(&fail_make_request, str);
1447}
1448__setup("fail_make_request=", setup_fail_make_request);
1449
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001450static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001451{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001452 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001453}
1454
1455static int __init fail_make_request_debugfs(void)
1456{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001457 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1458 NULL, &fail_make_request);
1459
1460 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001461}
1462
1463late_initcall(fail_make_request_debugfs);
1464
1465#else /* CONFIG_FAIL_MAKE_REQUEST */
1466
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001467static inline bool should_fail_request(struct hd_struct *part,
1468 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001469{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001470 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001471}
1472
1473#endif /* CONFIG_FAIL_MAKE_REQUEST */
1474
Jens Axboec07e2b42007-07-18 13:27:58 +02001475/*
1476 * Check whether this bio extends beyond the end of the device.
1477 */
1478static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1479{
1480 sector_t maxsector;
1481
1482 if (!nr_sectors)
1483 return 0;
1484
1485 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001486 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001487 if (maxsector) {
1488 sector_t sector = bio->bi_sector;
1489
1490 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1491 /*
1492 * This may well happen - the kernel calls bread()
1493 * without checking the size of the device, e.g., when
1494 * mounting a device.
1495 */
1496 handle_bad_sector(bio);
1497 return 1;
1498 }
1499 }
1500
1501 return 0;
1502}
1503
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001504static noinline_for_stack bool
1505generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506{
Jens Axboe165125e2007-07-24 09:28:11 +02001507 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001508 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001509 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001510 char b[BDEVNAME_SIZE];
1511 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
1513 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Jens Axboec07e2b42007-07-18 13:27:58 +02001515 if (bio_check_eod(bio, nr_sectors))
1516 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001518 q = bdev_get_queue(bio->bi_bdev);
1519 if (unlikely(!q)) {
1520 printk(KERN_ERR
1521 "generic_make_request: Trying to access "
1522 "nonexistent block-device %s (%Lu)\n",
1523 bdevname(bio->bi_bdev, b),
1524 (long long) bio->bi_sector);
1525 goto end_io;
1526 }
1527
1528 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1529 nr_sectors > queue_max_hw_sectors(q))) {
1530 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1531 bdevname(bio->bi_bdev, b),
1532 bio_sectors(bio),
1533 queue_max_hw_sectors(q));
1534 goto end_io;
1535 }
1536
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001537 part = bio->bi_bdev->bd_part;
1538 if (should_fail_request(part, bio->bi_size) ||
1539 should_fail_request(&part_to_disk(part)->part0,
1540 bio->bi_size))
1541 goto end_io;
1542
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001544 * If this device has partitions, remap block n
1545 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001547 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001549 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1550 goto end_io;
1551
1552 if (bio_check_eod(bio, nr_sectors))
1553 goto end_io;
1554
1555 /*
1556 * Filter flush bio's early so that make_request based
1557 * drivers without flush support don't have to worry
1558 * about them.
1559 */
1560 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1561 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1562 if (!nr_sectors) {
1563 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001564 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001568 if ((bio->bi_rw & REQ_DISCARD) &&
1569 (!blk_queue_discard(q) ||
1570 ((bio->bi_rw & REQ_SECURE) &&
1571 !blk_queue_secdiscard(q)))) {
1572 err = -EOPNOTSUPP;
1573 goto end_io;
1574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
Tejun Heobc16a4f2011-10-19 14:33:01 +02001576 if (blk_throtl_bio(q, bio))
1577 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001578
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001579 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001580 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001581
1582end_io:
1583 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001584 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585}
1586
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001587/**
1588 * generic_make_request - hand a buffer to its device driver for I/O
1589 * @bio: The bio describing the location in memory and on the device.
1590 *
1591 * generic_make_request() is used to make I/O requests of block
1592 * devices. It is passed a &struct bio, which describes the I/O that needs
1593 * to be done.
1594 *
1595 * generic_make_request() does not return any status. The
1596 * success/failure status of the request, along with notification of
1597 * completion, is delivered asynchronously through the bio->bi_end_io
1598 * function described (one day) else where.
1599 *
1600 * The caller of generic_make_request must make sure that bi_io_vec
1601 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1602 * set to describe the device address, and the
1603 * bi_end_io and optionally bi_private are set to describe how
1604 * completion notification should be signaled.
1605 *
1606 * generic_make_request and the drivers it calls may use bi_next if this
1607 * bio happens to be merged with someone else, and may resubmit the bio to
1608 * a lower device by calling into generic_make_request recursively, which
1609 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001610 */
1611void generic_make_request(struct bio *bio)
1612{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001613 struct bio_list bio_list_on_stack;
1614
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001615 if (!generic_make_request_checks(bio))
1616 return;
1617
1618 /*
1619 * We only want one ->make_request_fn to be active at a time, else
1620 * stack usage with stacked devices could be a problem. So use
1621 * current->bio_list to keep a list of requests submited by a
1622 * make_request_fn function. current->bio_list is also used as a
1623 * flag to say if generic_make_request is currently active in this
1624 * task or not. If it is NULL, then no make_request is active. If
1625 * it is non-NULL, then a make_request is active, and new requests
1626 * should be added at the tail
1627 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001628 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001629 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001630 return;
1631 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001632
Neil Brownd89d8792007-05-01 09:53:42 +02001633 /* following loop may be a bit non-obvious, and so deserves some
1634 * explanation.
1635 * Before entering the loop, bio->bi_next is NULL (as all callers
1636 * ensure that) so we have a list with a single bio.
1637 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001638 * we assign bio_list to a pointer to the bio_list_on_stack,
1639 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001640 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001641 * through a recursive call to generic_make_request. If it
1642 * did, we find a non-NULL value in bio_list and re-enter the loop
1643 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001644 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001645 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001646 */
1647 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001648 bio_list_init(&bio_list_on_stack);
1649 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001650 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001651 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1652
1653 q->make_request_fn(q, bio);
1654
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001655 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001656 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001657 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001658}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659EXPORT_SYMBOL(generic_make_request);
1660
1661/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001662 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1664 * @bio: The &struct bio which describes the I/O
1665 *
1666 * submit_bio() is very similar in purpose to generic_make_request(), and
1667 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001668 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 *
1670 */
1671void submit_bio(int rw, struct bio *bio)
1672{
1673 int count = bio_sectors(bio);
1674
Jens Axboe22e2c502005-06-27 10:55:12 +02001675 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Jens Axboebf2de6f2007-09-27 13:01:25 +02001677 /*
1678 * If it's a regular read/write or a barrier with data attached,
1679 * go through the normal accounting stuff before submission.
1680 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001681 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001682 if (rw & WRITE) {
1683 count_vm_events(PGPGOUT, count);
1684 } else {
1685 task_io_account_read(bio->bi_size);
1686 count_vm_events(PGPGIN, count);
1687 }
1688
1689 if (unlikely(block_dump)) {
1690 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001691 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001692 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001693 (rw & WRITE) ? "WRITE" : "READ",
1694 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001695 bdevname(bio->bi_bdev, b),
1696 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 }
1699
1700 generic_make_request(bio);
1701}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702EXPORT_SYMBOL(submit_bio);
1703
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001704/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001705 * blk_rq_check_limits - Helper function to check a request for the queue limit
1706 * @q: the queue
1707 * @rq: the request being checked
1708 *
1709 * Description:
1710 * @rq may have been made based on weaker limitations of upper-level queues
1711 * in request stacking drivers, and it may violate the limitation of @q.
1712 * Since the block layer and the underlying device driver trust @rq
1713 * after it is inserted to @q, it should be checked against @q before
1714 * the insertion using this generic function.
1715 *
1716 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001717 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001718 * Request stacking drivers like request-based dm may change the queue
1719 * limits while requests are in the queue (e.g. dm's table swapping).
1720 * Such request stacking drivers should check those requests agaist
1721 * the new queue limits again when they dispatch those requests,
1722 * although such checkings are also done against the old queue limits
1723 * when submitting requests.
1724 */
1725int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1726{
ike Snitzer33839772010-08-08 12:11:33 -04001727 if (rq->cmd_flags & REQ_DISCARD)
1728 return 0;
1729
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001730 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1731 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001732 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1733 return -EIO;
1734 }
1735
1736 /*
1737 * queue's settings related to segment counting like q->bounce_pfn
1738 * may differ from that of other stacking queues.
1739 * Recalculate it to check the request correctly on this queue's
1740 * limitation.
1741 */
1742 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001743 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001744 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1745 return -EIO;
1746 }
1747
1748 return 0;
1749}
1750EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1751
1752/**
1753 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1754 * @q: the queue to submit the request
1755 * @rq: the request being queued
1756 */
1757int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1758{
1759 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001760 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001761
1762 if (blk_rq_check_limits(q, rq))
1763 return -EIO;
1764
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001765 if (rq->rq_disk &&
1766 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001767 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001768
1769 spin_lock_irqsave(q->queue_lock, flags);
1770
1771 /*
1772 * Submitting request must be dequeued before calling this function
1773 * because it will be linked to another request_queue
1774 */
1775 BUG_ON(blk_queued_rq(rq));
1776
Jeff Moyer4853aba2011-08-15 21:37:25 +02001777 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1778 where = ELEVATOR_INSERT_FLUSH;
1779
1780 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001781 if (where == ELEVATOR_INSERT_FLUSH)
1782 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001783 spin_unlock_irqrestore(q->queue_lock, flags);
1784
1785 return 0;
1786}
1787EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1788
Tejun Heo80a761f2009-07-03 17:48:17 +09001789/**
1790 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1791 * @rq: request to examine
1792 *
1793 * Description:
1794 * A request could be merge of IOs which require different failure
1795 * handling. This function determines the number of bytes which
1796 * can be failed from the beginning of the request without
1797 * crossing into area which need to be retried further.
1798 *
1799 * Return:
1800 * The number of bytes to fail.
1801 *
1802 * Context:
1803 * queue_lock must be held.
1804 */
1805unsigned int blk_rq_err_bytes(const struct request *rq)
1806{
1807 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1808 unsigned int bytes = 0;
1809 struct bio *bio;
1810
1811 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1812 return blk_rq_bytes(rq);
1813
1814 /*
1815 * Currently the only 'mixing' which can happen is between
1816 * different fastfail types. We can safely fail portions
1817 * which have all the failfast bits that the first one has -
1818 * the ones which are at least as eager to fail as the first
1819 * one.
1820 */
1821 for (bio = rq->bio; bio; bio = bio->bi_next) {
1822 if ((bio->bi_rw & ff) != ff)
1823 break;
1824 bytes += bio->bi_size;
1825 }
1826
1827 /* this could lead to infinite loop */
1828 BUG_ON(blk_rq_bytes(rq) && !bytes);
1829 return bytes;
1830}
1831EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1832
Jens Axboebc58ba92009-01-23 10:54:44 +01001833static void blk_account_io_completion(struct request *req, unsigned int bytes)
1834{
Jens Axboec2553b52009-04-24 08:10:11 +02001835 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001836 const int rw = rq_data_dir(req);
1837 struct hd_struct *part;
1838 int cpu;
1839
1840 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001841 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001842 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1843 part_stat_unlock();
1844 }
1845}
1846
1847static void blk_account_io_done(struct request *req)
1848{
Jens Axboebc58ba92009-01-23 10:54:44 +01001849 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001850 * Account IO completion. flush_rq isn't accounted as a
1851 * normal IO on queueing nor completion. Accounting the
1852 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001853 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001854 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001855 unsigned long duration = jiffies - req->start_time;
1856 const int rw = rq_data_dir(req);
1857 struct hd_struct *part;
1858 int cpu;
1859
1860 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001861 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001862
1863 part_stat_inc(cpu, part, ios[rw]);
1864 part_stat_add(cpu, part, ticks[rw], duration);
1865 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001866 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001867
Jens Axboe6c23a962011-01-07 08:43:37 +01001868 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001869 part_stat_unlock();
1870 }
1871}
1872
Tejun Heo53a08802008-12-03 12:41:26 +01001873/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001874 * blk_peek_request - peek at the top of a request queue
1875 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001876 *
1877 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001878 * Return the request at the top of @q. The returned request
1879 * should be started using blk_start_request() before LLD starts
1880 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001881 *
1882 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001883 * Pointer to the request at the top of @q if available. Null
1884 * otherwise.
1885 *
1886 * Context:
1887 * queue_lock must be held.
1888 */
1889struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001890{
1891 struct request *rq;
1892 int ret;
1893
1894 while ((rq = __elv_next_request(q)) != NULL) {
1895 if (!(rq->cmd_flags & REQ_STARTED)) {
1896 /*
1897 * This is the first time the device driver
1898 * sees this request (possibly after
1899 * requeueing). Notify IO scheduler.
1900 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001901 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001902 elv_activate_rq(q, rq);
1903
1904 /*
1905 * just mark as started even if we don't start
1906 * it, a request that has been delayed should
1907 * not be passed by new incoming requests
1908 */
1909 rq->cmd_flags |= REQ_STARTED;
1910 trace_block_rq_issue(q, rq);
1911 }
1912
1913 if (!q->boundary_rq || q->boundary_rq == rq) {
1914 q->end_sector = rq_end_sector(rq);
1915 q->boundary_rq = NULL;
1916 }
1917
1918 if (rq->cmd_flags & REQ_DONTPREP)
1919 break;
1920
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001921 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001922 /*
1923 * make sure space for the drain appears we
1924 * know we can do this because max_hw_segments
1925 * has been adjusted to be one fewer than the
1926 * device can handle
1927 */
1928 rq->nr_phys_segments++;
1929 }
1930
1931 if (!q->prep_rq_fn)
1932 break;
1933
1934 ret = q->prep_rq_fn(q, rq);
1935 if (ret == BLKPREP_OK) {
1936 break;
1937 } else if (ret == BLKPREP_DEFER) {
1938 /*
1939 * the request may have been (partially) prepped.
1940 * we need to keep this request in the front to
1941 * avoid resource deadlock. REQ_STARTED will
1942 * prevent other fs requests from passing this one.
1943 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001944 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001945 !(rq->cmd_flags & REQ_DONTPREP)) {
1946 /*
1947 * remove the space for the drain we added
1948 * so that we don't add it again
1949 */
1950 --rq->nr_phys_segments;
1951 }
1952
1953 rq = NULL;
1954 break;
1955 } else if (ret == BLKPREP_KILL) {
1956 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001957 /*
1958 * Mark this request as started so we don't trigger
1959 * any debug logic in the end I/O path.
1960 */
1961 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001962 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001963 } else {
1964 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1965 break;
1966 }
1967 }
1968
1969 return rq;
1970}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001971EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001972
Tejun Heo9934c8c2009-05-08 11:54:16 +09001973void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001974{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001975 struct request_queue *q = rq->q;
1976
Tejun Heo158dbda2009-04-23 11:05:18 +09001977 BUG_ON(list_empty(&rq->queuelist));
1978 BUG_ON(ELV_ON_HASH(rq));
1979
1980 list_del_init(&rq->queuelist);
1981
1982 /*
1983 * the time frame between a request being removed from the lists
1984 * and to it is freed is accounted as io that is in progress at
1985 * the driver side.
1986 */
Divyesh Shah91952912010-04-01 15:01:41 -07001987 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001988 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001989 set_io_start_time_ns(rq);
1990 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001991}
1992
Tejun Heo5efccd12009-04-23 11:05:18 +09001993/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001994 * blk_start_request - start request processing on the driver
1995 * @req: request to dequeue
1996 *
1997 * Description:
1998 * Dequeue @req and start timeout timer on it. This hands off the
1999 * request to the driver.
2000 *
2001 * Block internal functions which don't want to start timer should
2002 * call blk_dequeue_request().
2003 *
2004 * Context:
2005 * queue_lock must be held.
2006 */
2007void blk_start_request(struct request *req)
2008{
2009 blk_dequeue_request(req);
2010
2011 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002012 * We are now handing the request to the hardware, initialize
2013 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002014 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002015 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002016 if (unlikely(blk_bidi_rq(req)))
2017 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2018
Tejun Heo9934c8c2009-05-08 11:54:16 +09002019 blk_add_timer(req);
2020}
2021EXPORT_SYMBOL(blk_start_request);
2022
2023/**
2024 * blk_fetch_request - fetch a request from a request queue
2025 * @q: request queue to fetch a request from
2026 *
2027 * Description:
2028 * Return the request at the top of @q. The request is started on
2029 * return and LLD can start processing it immediately.
2030 *
2031 * Return:
2032 * Pointer to the request at the top of @q if available. Null
2033 * otherwise.
2034 *
2035 * Context:
2036 * queue_lock must be held.
2037 */
2038struct request *blk_fetch_request(struct request_queue *q)
2039{
2040 struct request *rq;
2041
2042 rq = blk_peek_request(q);
2043 if (rq)
2044 blk_start_request(rq);
2045 return rq;
2046}
2047EXPORT_SYMBOL(blk_fetch_request);
2048
2049/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002050 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002051 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002052 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002053 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002054 *
2055 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002056 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2057 * the request structure even if @req doesn't have leftover.
2058 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002059 *
2060 * This special helper function is only for request stacking drivers
2061 * (e.g. request-based dm) so that they can handle partial completion.
2062 * Actual device drivers should use blk_end_request instead.
2063 *
2064 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2065 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002066 *
2067 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002068 * %false - this request doesn't have any more data
2069 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002070 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002071bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002073 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 struct bio *bio;
2075
Tejun Heo2e60e022009-04-23 11:05:18 +09002076 if (!req->bio)
2077 return false;
2078
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002079 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002080
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002082 * For fs requests, rq is just carrier of independent bio's
2083 * and each partial completion should be handled separately.
2084 * Reset per-request error on each partial completion.
2085 *
2086 * TODO: tj: This is too subtle. It would be better to let
2087 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002089 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 req->errors = 0;
2091
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002092 if (error && req->cmd_type == REQ_TYPE_FS &&
2093 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002094 char *error_type;
2095
2096 switch (error) {
2097 case -ENOLINK:
2098 error_type = "recoverable transport";
2099 break;
2100 case -EREMOTEIO:
2101 error_type = "critical target";
2102 break;
2103 case -EBADE:
2104 error_type = "critical nexus";
2105 break;
2106 case -EIO:
2107 default:
2108 error_type = "I/O";
2109 break;
2110 }
2111 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2112 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2113 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 }
2115
Jens Axboebc58ba92009-01-23 10:54:44 +01002116 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002117
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 total_bytes = bio_nbytes = 0;
2119 while ((bio = req->bio) != NULL) {
2120 int nbytes;
2121
2122 if (nr_bytes >= bio->bi_size) {
2123 req->bio = bio->bi_next;
2124 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002125 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 next_idx = 0;
2127 bio_nbytes = 0;
2128 } else {
2129 int idx = bio->bi_idx + next_idx;
2130
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002131 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002133 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002134 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 break;
2136 }
2137
2138 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2139 BIO_BUG_ON(nbytes > bio->bi_size);
2140
2141 /*
2142 * not a complete bvec done
2143 */
2144 if (unlikely(nbytes > nr_bytes)) {
2145 bio_nbytes += nr_bytes;
2146 total_bytes += nr_bytes;
2147 break;
2148 }
2149
2150 /*
2151 * advance to the next vector
2152 */
2153 next_idx++;
2154 bio_nbytes += nbytes;
2155 }
2156
2157 total_bytes += nbytes;
2158 nr_bytes -= nbytes;
2159
Jens Axboe6728cb02008-01-31 13:03:55 +01002160 bio = req->bio;
2161 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 /*
2163 * end more in this run, or just return 'not-done'
2164 */
2165 if (unlikely(nr_bytes <= 0))
2166 break;
2167 }
2168 }
2169
2170 /*
2171 * completely done
2172 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002173 if (!req->bio) {
2174 /*
2175 * Reset counters so that the request stacking driver
2176 * can find how many bytes remain in the request
2177 * later.
2178 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002179 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002180 return false;
2181 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182
2183 /*
2184 * if the request wasn't completed, update state
2185 */
2186 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002187 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 bio->bi_idx += next_idx;
2189 bio_iovec(bio)->bv_offset += nr_bytes;
2190 bio_iovec(bio)->bv_len -= nr_bytes;
2191 }
2192
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002193 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002194 req->buffer = bio_data(req->bio);
2195
2196 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002197 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002198 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002199
Tejun Heo80a761f2009-07-03 17:48:17 +09002200 /* mixed attributes always follow the first bio */
2201 if (req->cmd_flags & REQ_MIXED_MERGE) {
2202 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2203 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2204 }
2205
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002206 /*
2207 * If total number of sectors is less than the first segment
2208 * size, something has gone terribly wrong.
2209 */
2210 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002211 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002212 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002213 }
2214
2215 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002217
Tejun Heo2e60e022009-04-23 11:05:18 +09002218 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219}
Tejun Heo2e60e022009-04-23 11:05:18 +09002220EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
Tejun Heo2e60e022009-04-23 11:05:18 +09002222static bool blk_update_bidi_request(struct request *rq, int error,
2223 unsigned int nr_bytes,
2224 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002225{
Tejun Heo2e60e022009-04-23 11:05:18 +09002226 if (blk_update_request(rq, error, nr_bytes))
2227 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002228
Tejun Heo2e60e022009-04-23 11:05:18 +09002229 /* Bidi request must be completed as a whole */
2230 if (unlikely(blk_bidi_rq(rq)) &&
2231 blk_update_request(rq->next_rq, error, bidi_bytes))
2232 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002233
Jens Axboee2e1a142010-06-09 10:42:09 +02002234 if (blk_queue_add_random(rq->q))
2235 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002236
2237 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238}
2239
James Bottomley28018c22010-07-01 19:49:17 +09002240/**
2241 * blk_unprep_request - unprepare a request
2242 * @req: the request
2243 *
2244 * This function makes a request ready for complete resubmission (or
2245 * completion). It happens only after all error handling is complete,
2246 * so represents the appropriate moment to deallocate any resources
2247 * that were allocated to the request in the prep_rq_fn. The queue
2248 * lock is held when calling this.
2249 */
2250void blk_unprep_request(struct request *req)
2251{
2252 struct request_queue *q = req->q;
2253
2254 req->cmd_flags &= ~REQ_DONTPREP;
2255 if (q->unprep_rq_fn)
2256 q->unprep_rq_fn(q, req);
2257}
2258EXPORT_SYMBOL_GPL(blk_unprep_request);
2259
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260/*
2261 * queue lock must be held
2262 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002263static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002265 if (blk_rq_tagged(req))
2266 blk_queue_end_tag(req->q, req);
2267
James Bottomleyba396a62009-05-27 14:17:08 +02002268 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002270 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002271 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
Mike Andersone78042e2008-10-30 02:16:20 -07002273 blk_delete_timer(req);
2274
James Bottomley28018c22010-07-01 19:49:17 +09002275 if (req->cmd_flags & REQ_DONTPREP)
2276 blk_unprep_request(req);
2277
2278
Jens Axboebc58ba92009-01-23 10:54:44 +01002279 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002280
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002282 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002283 else {
2284 if (blk_bidi_rq(req))
2285 __blk_put_request(req->next_rq->q, req->next_rq);
2286
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 }
2289}
2290
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002291/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002292 * blk_end_bidi_request - Complete a bidi request
2293 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002294 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002295 * @nr_bytes: number of bytes to complete @rq
2296 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002297 *
2298 * Description:
2299 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002300 * Drivers that supports bidi can safely call this member for any
2301 * type of request, bidi or uni. In the later case @bidi_bytes is
2302 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002303 *
2304 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002305 * %false - we are done with this request
2306 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002307 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002308static bool blk_end_bidi_request(struct request *rq, int error,
2309 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002310{
2311 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002312 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002313
Tejun Heo2e60e022009-04-23 11:05:18 +09002314 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2315 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002316
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002317 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002318 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002319 spin_unlock_irqrestore(q->queue_lock, flags);
2320
Tejun Heo2e60e022009-04-23 11:05:18 +09002321 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002322}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002323
2324/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002325 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2326 * @rq: the request to complete
2327 * @error: %0 for success, < %0 for error
2328 * @nr_bytes: number of bytes to complete @rq
2329 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002330 *
2331 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002332 * Identical to blk_end_bidi_request() except that queue lock is
2333 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002334 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002335 * Return:
2336 * %false - we are done with this request
2337 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002338 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002339bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002340 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002341{
Tejun Heo2e60e022009-04-23 11:05:18 +09002342 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2343 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002344
Tejun Heo2e60e022009-04-23 11:05:18 +09002345 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002346
Tejun Heo2e60e022009-04-23 11:05:18 +09002347 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002348}
2349
2350/**
2351 * blk_end_request - Helper function for drivers to complete the request.
2352 * @rq: the request being processed
2353 * @error: %0 for success, < %0 for error
2354 * @nr_bytes: number of bytes to complete
2355 *
2356 * Description:
2357 * Ends I/O on a number of bytes attached to @rq.
2358 * If @rq has leftover, sets it up for the next range of segments.
2359 *
2360 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002361 * %false - we are done with this request
2362 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002363 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002364bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002365{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002366 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002367}
Jens Axboe56ad1742009-07-28 22:11:24 +02002368EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002369
2370/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002371 * blk_end_request_all - Helper function for drives to finish the request.
2372 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002373 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002374 *
2375 * Description:
2376 * Completely finish @rq.
2377 */
2378void blk_end_request_all(struct request *rq, int error)
2379{
2380 bool pending;
2381 unsigned int bidi_bytes = 0;
2382
2383 if (unlikely(blk_bidi_rq(rq)))
2384 bidi_bytes = blk_rq_bytes(rq->next_rq);
2385
2386 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2387 BUG_ON(pending);
2388}
Jens Axboe56ad1742009-07-28 22:11:24 +02002389EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002390
2391/**
2392 * blk_end_request_cur - Helper function to finish the current request chunk.
2393 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002394 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002395 *
2396 * Description:
2397 * Complete the current consecutively mapped chunk from @rq.
2398 *
2399 * Return:
2400 * %false - we are done with this request
2401 * %true - still buffers pending for this request
2402 */
2403bool blk_end_request_cur(struct request *rq, int error)
2404{
2405 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2406}
Jens Axboe56ad1742009-07-28 22:11:24 +02002407EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002408
2409/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002410 * blk_end_request_err - Finish a request till the next failure boundary.
2411 * @rq: the request to finish till the next failure boundary for
2412 * @error: must be negative errno
2413 *
2414 * Description:
2415 * Complete @rq till the next failure boundary.
2416 *
2417 * Return:
2418 * %false - we are done with this request
2419 * %true - still buffers pending for this request
2420 */
2421bool blk_end_request_err(struct request *rq, int error)
2422{
2423 WARN_ON(error >= 0);
2424 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2425}
2426EXPORT_SYMBOL_GPL(blk_end_request_err);
2427
2428/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002429 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002430 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002431 * @error: %0 for success, < %0 for error
2432 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002433 *
2434 * Description:
2435 * Must be called with queue lock held unlike blk_end_request().
2436 *
2437 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002438 * %false - we are done with this request
2439 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002440 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002441bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002442{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002443 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002444}
Jens Axboe56ad1742009-07-28 22:11:24 +02002445EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002446
2447/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002448 * __blk_end_request_all - Helper function for drives to finish the request.
2449 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002450 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002451 *
2452 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002453 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002454 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002455void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002456{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002457 bool pending;
2458 unsigned int bidi_bytes = 0;
2459
2460 if (unlikely(blk_bidi_rq(rq)))
2461 bidi_bytes = blk_rq_bytes(rq->next_rq);
2462
2463 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2464 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002465}
Jens Axboe56ad1742009-07-28 22:11:24 +02002466EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002467
2468/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002469 * __blk_end_request_cur - Helper function to finish the current request chunk.
2470 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002471 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002472 *
2473 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002474 * Complete the current consecutively mapped chunk from @rq. Must
2475 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002476 *
2477 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002478 * %false - we are done with this request
2479 * %true - still buffers pending for this request
2480 */
2481bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002482{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002483 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002484}
Jens Axboe56ad1742009-07-28 22:11:24 +02002485EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002486
Tejun Heo80a761f2009-07-03 17:48:17 +09002487/**
2488 * __blk_end_request_err - Finish a request till the next failure boundary.
2489 * @rq: the request to finish till the next failure boundary for
2490 * @error: must be negative errno
2491 *
2492 * Description:
2493 * Complete @rq till the next failure boundary. Must be called
2494 * with queue lock held.
2495 *
2496 * Return:
2497 * %false - we are done with this request
2498 * %true - still buffers pending for this request
2499 */
2500bool __blk_end_request_err(struct request *rq, int error)
2501{
2502 WARN_ON(error >= 0);
2503 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2504}
2505EXPORT_SYMBOL_GPL(__blk_end_request_err);
2506
Jens Axboe86db1e22008-01-29 14:53:40 +01002507void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2508 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002510 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002511 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512
David Woodhousefb2dce82008-08-05 18:01:53 +01002513 if (bio_has_data(bio)) {
2514 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002515 rq->buffer = bio_data(bio);
2516 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002517 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
NeilBrown66846572007-08-16 13:31:28 +02002520 if (bio->bi_bdev)
2521 rq->rq_disk = bio->bi_bdev->bd_disk;
2522}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002524#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2525/**
2526 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2527 * @rq: the request to be flushed
2528 *
2529 * Description:
2530 * Flush all pages in @rq.
2531 */
2532void rq_flush_dcache_pages(struct request *rq)
2533{
2534 struct req_iterator iter;
2535 struct bio_vec *bvec;
2536
2537 rq_for_each_segment(bvec, rq, iter)
2538 flush_dcache_page(bvec->bv_page);
2539}
2540EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2541#endif
2542
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002543/**
2544 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2545 * @q : the queue of the device being checked
2546 *
2547 * Description:
2548 * Check if underlying low-level drivers of a device are busy.
2549 * If the drivers want to export their busy state, they must set own
2550 * exporting function using blk_queue_lld_busy() first.
2551 *
2552 * Basically, this function is used only by request stacking drivers
2553 * to stop dispatching requests to underlying devices when underlying
2554 * devices are busy. This behavior helps more I/O merging on the queue
2555 * of the request stacking driver and prevents I/O throughput regression
2556 * on burst I/O load.
2557 *
2558 * Return:
2559 * 0 - Not busy (The request stacking driver should dispatch request)
2560 * 1 - Busy (The request stacking driver should stop dispatching request)
2561 */
2562int blk_lld_busy(struct request_queue *q)
2563{
2564 if (q->lld_busy_fn)
2565 return q->lld_busy_fn(q);
2566
2567 return 0;
2568}
2569EXPORT_SYMBOL_GPL(blk_lld_busy);
2570
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002571/**
2572 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2573 * @rq: the clone request to be cleaned up
2574 *
2575 * Description:
2576 * Free all bios in @rq for a cloned request.
2577 */
2578void blk_rq_unprep_clone(struct request *rq)
2579{
2580 struct bio *bio;
2581
2582 while ((bio = rq->bio) != NULL) {
2583 rq->bio = bio->bi_next;
2584
2585 bio_put(bio);
2586 }
2587}
2588EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2589
2590/*
2591 * Copy attributes of the original request to the clone request.
2592 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2593 */
2594static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2595{
2596 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002597 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002598 dst->cmd_type = src->cmd_type;
2599 dst->__sector = blk_rq_pos(src);
2600 dst->__data_len = blk_rq_bytes(src);
2601 dst->nr_phys_segments = src->nr_phys_segments;
2602 dst->ioprio = src->ioprio;
2603 dst->extra_len = src->extra_len;
2604}
2605
2606/**
2607 * blk_rq_prep_clone - Helper function to setup clone request
2608 * @rq: the request to be setup
2609 * @rq_src: original request to be cloned
2610 * @bs: bio_set that bios for clone are allocated from
2611 * @gfp_mask: memory allocation mask for bio
2612 * @bio_ctr: setup function to be called for each clone bio.
2613 * Returns %0 for success, non %0 for failure.
2614 * @data: private data to be passed to @bio_ctr
2615 *
2616 * Description:
2617 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2618 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2619 * are not copied, and copying such parts is the caller's responsibility.
2620 * Also, pages which the original bios are pointing to are not copied
2621 * and the cloned bios just point same pages.
2622 * So cloned bios must be completed before original bios, which means
2623 * the caller must complete @rq before @rq_src.
2624 */
2625int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2626 struct bio_set *bs, gfp_t gfp_mask,
2627 int (*bio_ctr)(struct bio *, struct bio *, void *),
2628 void *data)
2629{
2630 struct bio *bio, *bio_src;
2631
2632 if (!bs)
2633 bs = fs_bio_set;
2634
2635 blk_rq_init(NULL, rq);
2636
2637 __rq_for_each_bio(bio_src, rq_src) {
2638 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2639 if (!bio)
2640 goto free_and_out;
2641
2642 __bio_clone(bio, bio_src);
2643
2644 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002645 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002646 goto free_and_out;
2647
2648 if (bio_ctr && bio_ctr(bio, bio_src, data))
2649 goto free_and_out;
2650
2651 if (rq->bio) {
2652 rq->biotail->bi_next = bio;
2653 rq->biotail = bio;
2654 } else
2655 rq->bio = rq->biotail = bio;
2656 }
2657
2658 __blk_rq_prep_clone(rq, rq_src);
2659
2660 return 0;
2661
2662free_and_out:
2663 if (bio)
2664 bio_free(bio, bs);
2665 blk_rq_unprep_clone(rq);
2666
2667 return -ENOMEM;
2668}
2669EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2670
Jens Axboe18887ad2008-07-28 13:08:45 +02002671int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672{
2673 return queue_work(kblockd_workqueue, work);
2674}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675EXPORT_SYMBOL(kblockd_schedule_work);
2676
Vivek Goyale43473b2010-09-15 17:06:35 -04002677int kblockd_schedule_delayed_work(struct request_queue *q,
2678 struct delayed_work *dwork, unsigned long delay)
2679{
2680 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2681}
2682EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2683
Jens Axboe73c10102011-03-08 13:19:51 +01002684#define PLUG_MAGIC 0x91827364
2685
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002686/**
2687 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2688 * @plug: The &struct blk_plug that needs to be initialized
2689 *
2690 * Description:
2691 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2692 * pending I/O should the task end up blocking between blk_start_plug() and
2693 * blk_finish_plug(). This is important from a performance perspective, but
2694 * also ensures that we don't deadlock. For instance, if the task is blocking
2695 * for a memory allocation, memory reclaim could end up wanting to free a
2696 * page belonging to that request that is currently residing in our private
2697 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2698 * this kind of deadlock.
2699 */
Jens Axboe73c10102011-03-08 13:19:51 +01002700void blk_start_plug(struct blk_plug *plug)
2701{
2702 struct task_struct *tsk = current;
2703
2704 plug->magic = PLUG_MAGIC;
2705 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002706 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002707 plug->should_sort = 0;
2708
2709 /*
2710 * If this is a nested plug, don't actually assign it. It will be
2711 * flushed on its own.
2712 */
2713 if (!tsk->plug) {
2714 /*
2715 * Store ordering should not be needed here, since a potential
2716 * preempt will imply a full memory barrier
2717 */
2718 tsk->plug = plug;
2719 }
2720}
2721EXPORT_SYMBOL(blk_start_plug);
2722
2723static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2724{
2725 struct request *rqa = container_of(a, struct request, queuelist);
2726 struct request *rqb = container_of(b, struct request, queuelist);
2727
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002728 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002729}
2730
Jens Axboe49cac012011-04-16 13:51:05 +02002731/*
2732 * If 'from_schedule' is true, then postpone the dispatch of requests
2733 * until a safe kblockd context. We due this to avoid accidental big
2734 * additional stack usage in driver dispatch, in places where the originally
2735 * plugger did not intend it.
2736 */
Jens Axboef6603782011-04-15 15:49:07 +02002737static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002738 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002739 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002740{
Jens Axboe49cac012011-04-16 13:51:05 +02002741 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002742
2743 /*
2744 * If we are punting this to kblockd, then we can safely drop
2745 * the queue_lock before waking kblockd (which needs to take
2746 * this lock).
2747 */
2748 if (from_schedule) {
2749 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002750 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002751 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002752 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002753 spin_unlock(q->queue_lock);
2754 }
2755
Jens Axboe94b5eb22011-04-12 10:12:19 +02002756}
2757
NeilBrown048c9372011-04-18 09:52:22 +02002758static void flush_plug_callbacks(struct blk_plug *plug)
2759{
2760 LIST_HEAD(callbacks);
2761
2762 if (list_empty(&plug->cb_list))
2763 return;
2764
2765 list_splice_init(&plug->cb_list, &callbacks);
2766
2767 while (!list_empty(&callbacks)) {
2768 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2769 struct blk_plug_cb,
2770 list);
2771 list_del(&cb->list);
2772 cb->callback(cb);
2773 }
2774}
2775
Jens Axboe49cac012011-04-16 13:51:05 +02002776void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002777{
2778 struct request_queue *q;
2779 unsigned long flags;
2780 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002781 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002782 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002783
2784 BUG_ON(plug->magic != PLUG_MAGIC);
2785
NeilBrown048c9372011-04-18 09:52:22 +02002786 flush_plug_callbacks(plug);
Jens Axboe73c10102011-03-08 13:19:51 +01002787 if (list_empty(&plug->list))
2788 return;
2789
NeilBrown109b8122011-04-11 14:13:10 +02002790 list_splice_init(&plug->list, &list);
2791
2792 if (plug->should_sort) {
2793 list_sort(NULL, &list, plug_rq_cmp);
2794 plug->should_sort = 0;
2795 }
Jens Axboe73c10102011-03-08 13:19:51 +01002796
2797 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002798 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002799
2800 /*
2801 * Save and disable interrupts here, to avoid doing it for every
2802 * queue lock we have to take.
2803 */
Jens Axboe73c10102011-03-08 13:19:51 +01002804 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002805 while (!list_empty(&list)) {
2806 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002807 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002808 BUG_ON(!rq->q);
2809 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002810 /*
2811 * This drops the queue lock
2812 */
2813 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002814 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002815 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002816 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002817 spin_lock(q->queue_lock);
2818 }
Jens Axboe73c10102011-03-08 13:19:51 +01002819 /*
2820 * rq is already accounted, so use raw insert
2821 */
Jens Axboe401a18e2011-03-25 16:57:52 +01002822 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2823 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2824 else
2825 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002826
2827 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01002828 }
2829
Jens Axboe99e22592011-04-18 09:59:55 +02002830 /*
2831 * This drops the queue lock
2832 */
2833 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002834 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002835
Jens Axboe73c10102011-03-08 13:19:51 +01002836 local_irq_restore(flags);
2837}
Jens Axboe73c10102011-03-08 13:19:51 +01002838
2839void blk_finish_plug(struct blk_plug *plug)
2840{
Jens Axboef6603782011-04-15 15:49:07 +02002841 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02002842
2843 if (plug == current->plug)
2844 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01002845}
2846EXPORT_SYMBOL(blk_finish_plug);
2847
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848int __init blk_dev_init(void)
2849{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002850 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2851 sizeof(((struct request *)0)->cmd_flags));
2852
Tejun Heo89b90be2011-01-03 15:01:47 +01002853 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2854 kblockd_workqueue = alloc_workqueue("kblockd",
2855 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 if (!kblockd_workqueue)
2857 panic("Failed to create kblockd\n");
2858
2859 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002860 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861
Jens Axboe8324aa92008-01-29 14:51:59 +01002862 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002863 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 return 0;
2866}