blob: 3a78b00edd71cfb7b667d50c9d3fe3265a5dba28 [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
Tejun Heoa73f7302011-12-14 00:33:37 +010042DEFINE_IDA(blk_queue_ida);
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/*
45 * For the allocated request tables
46 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010047static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49/*
50 * For queue allocation
51 */
Jens Axboe6728cb02008-01-31 13:03:55 +010052struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 * Controlling structure to kblockd
56 */
Jens Axboeff856ba2006-01-09 16:02:34 +010057static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Jens Axboe26b82562008-01-29 13:54:41 +010059static void drive_stat_acct(struct request *rq, int new_io)
60{
Jens Axboe28f13702008-05-07 10:15:46 +020061 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010062 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090063 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010064
Jens Axboec2553b52009-04-24 08:10:11 +020065 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010066 return;
67
Tejun Heo074a7ac2008-08-25 19:56:14 +090068 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +090069
Jerome Marchand09e099d2011-01-05 16:57:38 +010070 if (!new_io) {
71 part = rq->part;
Tejun Heo074a7ac2008-08-25 19:56:14 +090072 part_stat_inc(cpu, part, merges[rw]);
Jerome Marchand09e099d2011-01-05 16:57:38 +010073 } else {
74 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Jens Axboe6c23a962011-01-07 08:43:37 +010075 if (!hd_struct_try_get(part)) {
Jerome Marchand09e099d2011-01-05 16:57:38 +010076 /*
77 * The partition is already being removed,
78 * the request will be accounted on the disk only
79 *
80 * We take a reference on disk->part0 although that
81 * partition will never be deleted, so we can treat
82 * it as any other partition.
83 */
84 part = &rq->rq_disk->part0;
Jens Axboe6c23a962011-01-07 08:43:37 +010085 hd_struct_get(part);
Jerome Marchand09e099d2011-01-05 16:57:38 +010086 }
Tejun Heo074a7ac2008-08-25 19:56:14 +090087 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020088 part_inc_in_flight(part, rw);
Jerome Marchand09e099d2011-01-05 16:57:38 +010089 rq->part = part;
Jens Axboe26b82562008-01-29 13:54:41 +010090 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020091
Tejun Heo074a7ac2008-08-25 19:56:14 +090092 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010093}
94
Jens Axboe8324aa92008-01-29 14:51:59 +010095void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096{
97 int nr;
98
99 nr = q->nr_requests - (q->nr_requests / 8) + 1;
100 if (nr > q->nr_requests)
101 nr = q->nr_requests;
102 q->nr_congestion_on = nr;
103
104 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
105 if (nr < 1)
106 nr = 1;
107 q->nr_congestion_off = nr;
108}
109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110/**
111 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * @bdev: device
113 *
114 * Locates the passed device's request queue and returns the address of its
115 * backing_dev_info
116 *
117 * Will return NULL if the request queue cannot be located.
118 */
119struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
120{
121 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200122 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124 if (q)
125 ret = &q->backing_dev_info;
126 return ret;
127}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128EXPORT_SYMBOL(blk_get_backing_dev_info);
129
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200130void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200132 memset(rq, 0, sizeof(*rq));
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700135 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200136 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100137 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900138 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200139 INIT_HLIST_NODE(&rq->hash);
140 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200141 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800142 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100143 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100144 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900145 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700146 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100147 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200149EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
NeilBrown5bb23a62007-09-27 12:46:13 +0200151static void req_bio_endio(struct request *rq, struct bio *bio,
152 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100153{
Tejun Heo143a87f2011-01-25 12:43:52 +0100154 if (error)
155 clear_bit(BIO_UPTODATE, &bio->bi_flags);
156 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
157 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100158
Tejun Heo143a87f2011-01-25 12:43:52 +0100159 if (unlikely(nbytes > bio->bi_size)) {
160 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
161 __func__, nbytes, bio->bi_size);
162 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +0200163 }
Tejun Heo143a87f2011-01-25 12:43:52 +0100164
165 if (unlikely(rq->cmd_flags & REQ_QUIET))
166 set_bit(BIO_QUIET, &bio->bi_flags);
167
168 bio->bi_size -= nbytes;
169 bio->bi_sector += (nbytes >> 9);
170
171 if (bio_integrity(bio))
172 bio_integrity_advance(bio, nbytes);
173
174 /* don't actually finish bio if it's part of flush sequence */
175 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
176 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100177}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179void blk_dump_rq_flags(struct request *rq, char *msg)
180{
181 int bit;
182
Jens Axboe6728cb02008-01-31 13:03:55 +0100183 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200184 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
185 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Tejun Heo83096eb2009-05-07 22:24:39 +0900187 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
188 (unsigned long long)blk_rq_pos(rq),
189 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900190 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900191 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200193 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100194 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200195 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 printk("%02x ", rq->cmd[bit]);
197 printk("\n");
198 }
199}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200EXPORT_SYMBOL(blk_dump_rq_flags);
201
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500202static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200203{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500204 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200205
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500206 q = container_of(work, struct request_queue, delay_work.work);
207 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200208 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500209 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
212/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500213 * blk_delay_queue - restart queueing after defined interval
214 * @q: The &struct request_queue in question
215 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 *
217 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500218 * Sometimes queueing needs to be postponed for a little while, to allow
219 * resources to come back. This function will make sure that queueing is
220 * restarted around the specified time.
221 */
222void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223{
Jens Axboe4521cc42011-04-18 11:36:39 +0200224 queue_delayed_work(kblockd_workqueue, &q->delay_work,
225 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500227EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500228
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229/**
230 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200231 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 *
233 * Description:
234 * blk_start_queue() will clear the stop flag on the queue, and call
235 * the request_fn for the queue if it was in a stopped state when
236 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200238void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200240 WARN_ON(!irqs_disabled());
241
Nick Piggin75ad23b2008-04-29 14:48:33 +0200242 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200243 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245EXPORT_SYMBOL(blk_start_queue);
246
247/**
248 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200249 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 *
251 * Description:
252 * The Linux block layer assumes that a block driver will consume all
253 * entries on the request queue when the request_fn strategy is called.
254 * Often this will not happen, because of hardware limitations (queue
255 * depth settings). If a device driver gets a 'queue full' response,
256 * or if it simply chooses not to queue more I/O at one point, it can
257 * call this function to prevent the request_fn from being called until
258 * the driver has signalled it's ready to go again. This happens by calling
259 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200261void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
Jens Axboead3d9d72011-03-25 16:58:59 +0100263 __cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200264 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
266EXPORT_SYMBOL(blk_stop_queue);
267
268/**
269 * blk_sync_queue - cancel any pending callbacks on a queue
270 * @q: the queue
271 *
272 * Description:
273 * The block layer may perform asynchronous callback activity
274 * on a queue, such as calling the unplug function after a timeout.
275 * A block device may call blk_sync_queue to ensure that any
276 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200277 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 * that its ->make_request_fn will not re-add plugging prior to calling
279 * this function.
280 *
Vivek Goyalda527772011-03-02 19:05:33 -0500281 * This function does not cancel any asynchronous activity arising
282 * out of elevator or throttling code. That would require elevaotor_exit()
283 * and blk_throtl_exit() to be called with queue lock initialized.
284 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 */
286void blk_sync_queue(struct request_queue *q)
287{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100288 del_timer_sync(&q->timeout);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500289 cancel_delayed_work_sync(&q->delay_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290}
291EXPORT_SYMBOL(blk_sync_queue);
292
293/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200294 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200296 *
297 * Description:
298 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200299 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200301void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Tejun Heoa538cd02009-04-23 11:05:17 +0900303 if (unlikely(blk_queue_stopped(q)))
304 return;
305
Jens Axboec21e6be2011-04-19 13:32:46 +0200306 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200307}
308EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200309
Nick Piggin75ad23b2008-04-29 14:48:33 +0200310/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200311 * blk_run_queue_async - run a single device queue in workqueue context
312 * @q: The queue to run
313 *
314 * Description:
315 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
316 * of us.
317 */
318void blk_run_queue_async(struct request_queue *q)
319{
Shaohua Li3ec717b2011-05-18 11:22:43 +0200320 if (likely(!blk_queue_stopped(q))) {
321 __cancel_delayed_work(&q->delay_work);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200322 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Shaohua Li3ec717b2011-05-18 11:22:43 +0200323 }
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200324}
Jens Axboec21e6be2011-04-19 13:32:46 +0200325EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200326
327/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200328 * blk_run_queue - run a single device queue
329 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200330 *
331 * Description:
332 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900333 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200334 */
335void blk_run_queue(struct request_queue *q)
336{
337 unsigned long flags;
338
339 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200340 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 spin_unlock_irqrestore(q->queue_lock, flags);
342}
343EXPORT_SYMBOL(blk_run_queue);
344
Jens Axboe165125e2007-07-24 09:28:11 +0200345void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500346{
347 kobject_put(&q->kobj);
348}
Jens Axboed86e0e82011-05-27 07:44:43 +0200349EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500350
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200351/**
352 * blk_drain_queue - drain requests from request_queue
353 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200354 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200355 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200356 * Drain requests from @q. If @drain_all is set, all requests are drained.
357 * If not, only ELVPRIV requests are drained. The caller is responsible
358 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200359 */
Tejun Heoc9a929d2011-10-19 14:42:16 +0200360void blk_drain_queue(struct request_queue *q, bool drain_all)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200361{
362 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100363 bool drain = false;
364 int i;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200365
366 spin_lock_irq(q->queue_lock);
367
368 elv_drain_elevator(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200369 if (drain_all)
370 blk_throtl_drain(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200371
Tejun Heo4eabc942011-12-15 20:03:04 +0100372 /*
373 * This function might be called on a queue which failed
374 * driver init after queue creation. Some drivers
375 * (e.g. fd) get unhappy in such cases. Kick queue iff
376 * dispatch queue has something on it.
377 */
378 if (!list_empty(&q->queue_head))
379 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200380
Tejun Heo481a7d62011-12-14 00:33:37 +0100381 drain |= q->rq.elvpriv;
382
383 /*
384 * Unfortunately, requests are queued at and tracked from
385 * multiple places and there's no single counter which can
386 * be drained. Check all the queues and counters.
387 */
388 if (drain_all) {
389 drain |= !list_empty(&q->queue_head);
390 for (i = 0; i < 2; i++) {
391 drain |= q->rq.count[i];
392 drain |= q->in_flight[i];
393 drain |= !list_empty(&q->flush_queue[i]);
394 }
395 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200396
397 spin_unlock_irq(q->queue_lock);
398
Tejun Heo481a7d62011-12-14 00:33:37 +0100399 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200400 break;
401 msleep(10);
402 }
403}
404
Tejun Heoc9a929d2011-10-19 14:42:16 +0200405/**
406 * blk_cleanup_queue - shutdown a request queue
407 * @q: request queue to shutdown
408 *
409 * Mark @q DEAD, drain all pending requests, destroy and put it. All
410 * future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500411 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100412void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500413{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200414 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de2008-09-18 09:22:54 -0700415
Tejun Heoc9a929d2011-10-19 14:42:16 +0200416 /* mark @q DEAD, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500417 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200418 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200419
420 spin_lock_irq(lock);
421 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
422 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
423 queue_flag_set(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500424
Hannes Reinecke777eb1b2011-09-28 08:07:01 -0600425 if (q->queue_lock != &q->__queue_lock)
426 q->queue_lock = &q->__queue_lock;
Vivek Goyalda527772011-03-02 19:05:33 -0500427
Tejun Heoc9a929d2011-10-19 14:42:16 +0200428 spin_unlock_irq(lock);
429 mutex_unlock(&q->sysfs_lock);
430
Tejun Heo6dd9ad72011-11-03 18:52:11 +0100431 /*
432 * Drain all requests queued before DEAD marking. The caller might
433 * be trying to tear down @q before its elevator is initialized, in
434 * which case we don't want to call into draining.
435 */
436 if (q->elevator)
437 blk_drain_queue(q, true);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200438
439 /* @q won't process any more request, flush async actions */
440 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
441 blk_sync_queue(q);
442
443 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500444 blk_put_queue(q);
445}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446EXPORT_SYMBOL(blk_cleanup_queue);
447
Jens Axboe165125e2007-07-24 09:28:11 +0200448static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449{
450 struct request_list *rl = &q->rq;
451
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400452 if (unlikely(rl->rq_pool))
453 return 0;
454
Jens Axboe1faa16d2009-04-06 14:48:01 +0200455 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
456 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200457 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200458 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
459 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
Christoph Lameter19460892005-06-23 00:08:19 -0700461 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
462 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
464 if (!rl->rq_pool)
465 return -ENOMEM;
466
467 return 0;
468}
469
Jens Axboe165125e2007-07-24 09:28:11 +0200470struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
Christoph Lameter19460892005-06-23 00:08:19 -0700472 return blk_alloc_queue_node(gfp_mask, -1);
473}
474EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Jens Axboe165125e2007-07-24 09:28:11 +0200476struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700477{
Jens Axboe165125e2007-07-24 09:28:11 +0200478 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700479 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700480
Jens Axboe8324aa92008-01-29 14:51:59 +0100481 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700482 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 if (!q)
484 return NULL;
485
Tejun Heoa73f7302011-12-14 00:33:37 +0100486 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
487 if (q->id < 0)
488 goto fail_q;
489
Jens Axboe0989a022009-06-12 14:42:56 +0200490 q->backing_dev_info.ra_pages =
491 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
492 q->backing_dev_info.state = 0;
493 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200494 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100495 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200496
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700497 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100498 if (err)
499 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700500
Tejun Heoa73f7302011-12-14 00:33:37 +0100501 if (blk_throtl_init(q))
502 goto fail_id;
Vivek Goyale43473b2010-09-15 17:06:35 -0400503
Matthew Garrett31373d02010-04-06 14:25:14 +0200504 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
505 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700506 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
507 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100508 INIT_LIST_HEAD(&q->icq_list);
Tejun Heoae1b1532011-01-25 12:43:54 +0100509 INIT_LIST_HEAD(&q->flush_queue[0]);
510 INIT_LIST_HEAD(&q->flush_queue[1]);
511 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500512 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500513
Jens Axboe8324aa92008-01-29 14:51:59 +0100514 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Al Viro483f4af2006-03-18 18:34:37 -0500516 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700517 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500518
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500519 /*
520 * By default initialize queue_lock to internal lock and driver can
521 * override it later if need be.
522 */
523 q->queue_lock = &q->__queue_lock;
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100526
527fail_id:
528 ida_simple_remove(&blk_queue_ida, q->id);
529fail_q:
530 kmem_cache_free(blk_requestq_cachep, q);
531 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532}
Christoph Lameter19460892005-06-23 00:08:19 -0700533EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535/**
536 * blk_init_queue - prepare a request queue for use with a block device
537 * @rfn: The function to be called to process requests that have been
538 * placed on the queue.
539 * @lock: Request queue spin lock
540 *
541 * Description:
542 * If a block device wishes to use the standard request handling procedures,
543 * which sorts requests and coalesces adjacent requests, then it must
544 * call blk_init_queue(). The function @rfn will be called when there
545 * are requests on the queue that need to be processed. If the device
546 * supports plugging, then @rfn may not be called immediately when requests
547 * are available on the queue, but may be called at some time later instead.
548 * Plugged queues are generally unplugged when a buffer belonging to one
549 * of the requests on the queue is needed, or due to memory pressure.
550 *
551 * @rfn is not required, or even expected, to remove all requests off the
552 * queue, but only as many as it can handle at a time. If it does leave
553 * requests on the queue, it is responsible for arranging that the requests
554 * get dealt with eventually.
555 *
556 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200557 * request queue; this lock will be taken also from interrupt context, so irq
558 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200560 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 * it didn't succeed.
562 *
563 * Note:
564 * blk_init_queue() must be paired with a blk_cleanup_queue() call
565 * when the block device is deactivated (such as at module unload).
566 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700567
Jens Axboe165125e2007-07-24 09:28:11 +0200568struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
Christoph Lameter19460892005-06-23 00:08:19 -0700570 return blk_init_queue_node(rfn, lock, -1);
571}
572EXPORT_SYMBOL(blk_init_queue);
573
Jens Axboe165125e2007-07-24 09:28:11 +0200574struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700575blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
576{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600577 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600579 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
580 if (!uninit_q)
581 return NULL;
582
Mike Snitzer51514122011-11-23 10:59:13 +0100583 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600584 if (!q)
585 blk_cleanup_queue(uninit_q);
586
587 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200588}
589EXPORT_SYMBOL(blk_init_queue_node);
590
591struct request_queue *
592blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
593 spinlock_t *lock)
594{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 if (!q)
596 return NULL;
597
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600598 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500599 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
601 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900603 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100604 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500605
606 /* Override internal queue lock with supplied lock pointer */
607 if (lock)
608 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Jens Axboef3b144a2009-03-06 08:48:33 +0100610 /*
611 * This also sets hw/phys segments, boundary and size
612 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200613 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Alan Stern44ec9542007-02-20 11:01:57 -0500615 q->sg_reserved_size = INT_MAX;
616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 /*
618 * all done
619 */
620 if (!elevator_init(q, NULL)) {
621 blk_queue_congestion_threshold(q);
622 return q;
623 }
624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 return NULL;
626}
Mike Snitzer51514122011-11-23 10:59:13 +0100627EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Tejun Heo09ac46c2011-12-14 00:33:38 +0100629bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Tejun Heo34f60552011-12-14 00:33:37 +0100631 if (likely(!blk_queue_dead(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100632 __blk_get_queue(q);
633 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
635
Tejun Heo09ac46c2011-12-14 00:33:38 +0100636 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
Jens Axboed86e0e82011-05-27 07:44:43 +0200638EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Jens Axboe165125e2007-07-24 09:28:11 +0200640static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100642 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200643 elv_put_request(q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100644 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100645 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100646 }
647
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 mempool_free(rq, q->rq.rq_pool);
649}
650
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200651static struct request *
Tejun Heof1f8cc92011-12-14 00:33:42 +0100652blk_alloc_request(struct request_queue *q, struct io_cq *icq,
653 unsigned int flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
655 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
656
657 if (!rq)
658 return NULL;
659
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200660 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200661
Jerome Marchand42dad762009-04-22 14:01:49 +0200662 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Tejun Heof1f8cc92011-12-14 00:33:42 +0100664 if (flags & REQ_ELVPRIV) {
665 rq->elv.icq = icq;
666 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
667 mempool_free(rq, q->rq.rq_pool);
668 return NULL;
669 }
670 /* @rq->elv.icq holds on to io_context until @rq is freed */
671 if (icq)
672 get_io_context(icq->ioc);
Tejun Heocb98fc82005-10-28 08:29:39 +0200673 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Tejun Heocb98fc82005-10-28 08:29:39 +0200675 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
678/*
679 * ioc_batching returns true if the ioc is a valid batching request and
680 * should be given priority access to a request.
681 */
Jens Axboe165125e2007-07-24 09:28:11 +0200682static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
684 if (!ioc)
685 return 0;
686
687 /*
688 * Make sure the process is able to allocate at least 1 request
689 * even if the batch times out, otherwise we could theoretically
690 * lose wakeups.
691 */
692 return ioc->nr_batch_requests == q->nr_batching ||
693 (ioc->nr_batch_requests > 0
694 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
695}
696
697/*
698 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
699 * will cause the process to be a "batcher" on all queues in the system. This
700 * is the behaviour we want though - once it gets a wakeup it should be given
701 * a nice run.
702 */
Jens Axboe165125e2007-07-24 09:28:11 +0200703static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704{
705 if (!ioc || ioc_batching(q, ioc))
706 return;
707
708 ioc->nr_batch_requests = q->nr_batching;
709 ioc->last_waited = jiffies;
710}
711
Jens Axboe1faa16d2009-04-06 14:48:01 +0200712static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
714 struct request_list *rl = &q->rq;
715
Jens Axboe1faa16d2009-04-06 14:48:01 +0200716 if (rl->count[sync] < queue_congestion_off_threshold(q))
717 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Jens Axboe1faa16d2009-04-06 14:48:01 +0200719 if (rl->count[sync] + 1 <= q->nr_requests) {
720 if (waitqueue_active(&rl->wait[sync]))
721 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Jens Axboe1faa16d2009-04-06 14:48:01 +0200723 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 }
725}
726
727/*
728 * A request has just been released. Account for it, update the full and
729 * congestion status, wake up any waiters. Called under q->queue_lock.
730 */
Tejun Heo75eb6c32011-10-19 14:31:22 +0200731static void freed_request(struct request_queue *q, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
733 struct request_list *rl = &q->rq;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200734 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
Jens Axboe1faa16d2009-04-06 14:48:01 +0200736 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200737 if (flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200738 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Jens Axboe1faa16d2009-04-06 14:48:01 +0200740 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Jens Axboe1faa16d2009-04-06 14:48:01 +0200742 if (unlikely(rl->starved[sync ^ 1]))
743 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744}
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100747 * Determine if elevator data should be initialized when allocating the
748 * request associated with @bio.
749 */
750static bool blk_rq_should_init_elevator(struct bio *bio)
751{
752 if (!bio)
753 return true;
754
755 /*
756 * Flush requests do not use the elevator so skip initialization.
757 * This allows a request to share the flush and elevator data.
758 */
759 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
760 return false;
761
762 return true;
763}
764
Tejun Heoda8303c2011-10-19 14:33:05 +0200765/**
766 * get_request - get a free request
767 * @q: request_queue to allocate request from
768 * @rw_flags: RW and SYNC flags
769 * @bio: bio to allocate request for (can be %NULL)
770 * @gfp_mask: allocation mask
771 *
772 * Get a free request from @q. This function may fail under memory
773 * pressure or if @q is dead.
774 *
775 * Must be callled with @q->queue_lock held and,
776 * Returns %NULL on failure, with @q->queue_lock held.
777 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
Jens Axboe165125e2007-07-24 09:28:11 +0200779static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100780 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781{
782 struct request *rq = NULL;
783 struct request_list *rl = &q->rq;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100784 struct elevator_type *et;
Tejun Heof2dbd762011-12-14 00:33:40 +0100785 struct io_context *ioc;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100786 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200787 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heof2dbd762011-12-14 00:33:40 +0100788 bool retried = false;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200789 int may_queue;
Tejun Heof2dbd762011-12-14 00:33:40 +0100790retry:
Tejun Heof1f8cc92011-12-14 00:33:42 +0100791 et = q->elevator->type;
Tejun Heof2dbd762011-12-14 00:33:40 +0100792 ioc = current->io_context;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
Tejun Heo34f60552011-12-14 00:33:37 +0100794 if (unlikely(blk_queue_dead(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200795 return NULL;
796
Jens Axboe7749a8d2006-12-13 13:02:26 +0100797 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100798 if (may_queue == ELV_MQUEUE_NO)
799 goto rq_starved;
800
Jens Axboe1faa16d2009-04-06 14:48:01 +0200801 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
802 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100803 /*
804 * We want ioc to record batching state. If it's
805 * not already there, creating a new one requires
806 * dropping queue_lock, which in turn requires
807 * retesting conditions to avoid queue hang.
808 */
809 if (!ioc && !retried) {
810 spin_unlock_irq(q->queue_lock);
811 create_io_context(current, gfp_mask, q->node);
812 spin_lock_irq(q->queue_lock);
813 retried = true;
814 goto retry;
815 }
816
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100817 /*
818 * The queue will fill after this allocation, so set
819 * it as full, and mark this process as "batching".
820 * This process will be allowed to complete a batch of
821 * requests, others will be blocked.
822 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200823 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100824 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200825 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100826 } else {
827 if (may_queue != ELV_MQUEUE_MUST
828 && !ioc_batching(q, ioc)) {
829 /*
830 * The queue is full and the allocating
831 * process is not a "batcher", and not
832 * exempted by the IO scheduler
833 */
834 goto out;
835 }
836 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200838 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 }
840
Jens Axboe082cf692005-06-28 16:35:11 +0200841 /*
842 * Only allow batching queuers to allocate up to 50% over the defined
843 * limit of requests, otherwise we could have thousands of requests
844 * allocated with any setting of ->nr_requests
845 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200846 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200847 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100848
Jens Axboe1faa16d2009-04-06 14:48:01 +0200849 rl->count[is_sync]++;
850 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200851
Tejun Heof1f8cc92011-12-14 00:33:42 +0100852 /*
853 * Decide whether the new request will be managed by elevator. If
854 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
855 * prevent the current elevator from being destroyed until the new
856 * request is freed. This guarantees icq's won't be destroyed and
857 * makes creating new ones safe.
858 *
859 * Also, lookup icq while holding queue_lock. If it doesn't exist,
860 * it will be created after releasing queue_lock.
861 */
Tejun Heo75eb6c32011-10-19 14:31:22 +0200862 if (blk_rq_should_init_elevator(bio) &&
863 !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
864 rw_flags |= REQ_ELVPRIV;
865 rl->elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100866 if (et->icq_cache && ioc)
867 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100868 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200869
Jens Axboef253b862010-10-24 22:06:02 +0200870 if (blk_queue_io_stat(q))
871 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 spin_unlock_irq(q->queue_lock);
873
Tejun Heof1f8cc92011-12-14 00:33:42 +0100874 /* create icq if missing */
Shaohua Li05c30b92012-01-19 09:20:10 +0100875 if ((rw_flags & REQ_ELVPRIV) && unlikely(et->icq_cache && !icq)) {
Tejun Heof1f8cc92011-12-14 00:33:42 +0100876 icq = ioc_create_icq(q, gfp_mask);
Shaohua Li05c30b92012-01-19 09:20:10 +0100877 if (!icq)
878 goto fail_icq;
879 }
Tejun Heof1f8cc92011-12-14 00:33:42 +0100880
Shaohua Li05c30b92012-01-19 09:20:10 +0100881 rq = blk_alloc_request(q, icq, rw_flags, gfp_mask);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100882
Shaohua Li05c30b92012-01-19 09:20:10 +0100883fail_icq:
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100884 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 /*
886 * Allocation failed presumably due to memory. Undo anything
887 * we might have messed up.
888 *
889 * Allocating task should really be put onto the front of the
890 * wait queue, but this is pretty rare.
891 */
892 spin_lock_irq(q->queue_lock);
Tejun Heo75eb6c32011-10-19 14:31:22 +0200893 freed_request(q, rw_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
895 /*
896 * in the very unlikely event that allocation failed and no
897 * requests for this direction was pending, mark us starved
898 * so that freeing of a request in the other direction will
899 * notice us. another possible fix would be to split the
900 * rq mempool into READ and WRITE
901 */
902rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200903 if (unlikely(rl->count[is_sync] == 0))
904 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 goto out;
907 }
908
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100909 /*
910 * ioc may be NULL here, and ioc_batching will be false. That's
911 * OK, if the queue is under the request limit then requests need
912 * not count toward the nr_batch_requests limit. There will always
913 * be some limit enforced by BLK_BATCH_TIME.
914 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 if (ioc_batching(q, ioc))
916 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100917
Jens Axboe1faa16d2009-04-06 14:48:01 +0200918 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 return rq;
921}
922
Tejun Heoda8303c2011-10-19 14:33:05 +0200923/**
924 * get_request_wait - get a free request with retry
925 * @q: request_queue to allocate request from
926 * @rw_flags: RW and SYNC flags
927 * @bio: bio to allocate request for (can be %NULL)
Nick Piggind6344532005-06-28 20:45:14 -0700928 *
Tejun Heoda8303c2011-10-19 14:33:05 +0200929 * Get a free request from @q. This function keeps retrying under memory
930 * pressure and fails iff @q is dead.
931 *
932 * Must be callled with @q->queue_lock held and,
933 * Returns %NULL on failure, with @q->queue_lock held.
934 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 */
Jens Axboe165125e2007-07-24 09:28:11 +0200936static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200937 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200939 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 struct request *rq;
941
Jens Axboe7749a8d2006-12-13 13:02:26 +0100942 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700943 while (!rq) {
944 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 struct request_list *rl = &q->rq;
946
Tejun Heo34f60552011-12-14 00:33:37 +0100947 if (unlikely(blk_queue_dead(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200948 return NULL;
949
Jens Axboe1faa16d2009-04-06 14:48:01 +0200950 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 TASK_UNINTERRUPTIBLE);
952
Jens Axboe1faa16d2009-04-06 14:48:01 +0200953 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200955 spin_unlock_irq(q->queue_lock);
956 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200958 /*
959 * After sleeping, we become a "batching" process and
960 * will be able to allocate at least one request, and
961 * up to a big batch of them for a small period time.
962 * See ioc_batching, ioc_set_batching
963 */
Tejun Heof2dbd762011-12-14 00:33:40 +0100964 create_io_context(current, GFP_NOIO, q->node);
965 ioc_set_batching(q, current->io_context);
Jens Axboe2056a782006-03-23 20:00:26 +0100966
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200967 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200968 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200969
970 rq = get_request(q, rw_flags, bio, GFP_NOIO);
971 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 return rq;
974}
975
Jens Axboe165125e2007-07-24 09:28:11 +0200976struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977{
978 struct request *rq;
979
980 BUG_ON(rw != READ && rw != WRITE);
981
Nick Piggind6344532005-06-28 20:45:14 -0700982 spin_lock_irq(q->queue_lock);
Tejun Heoda8303c2011-10-19 14:33:05 +0200983 if (gfp_mask & __GFP_WAIT)
Jens Axboe22e2c502005-06-27 10:55:12 +0200984 rq = get_request_wait(q, rw, NULL);
Tejun Heoda8303c2011-10-19 14:33:05 +0200985 else
Jens Axboe22e2c502005-06-27 10:55:12 +0200986 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +0200987 if (!rq)
988 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -0700989 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
991 return rq;
992}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993EXPORT_SYMBOL(blk_get_request);
994
995/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300996 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700997 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300998 * @bio: The bio describing the memory mappings that will be submitted for IO.
999 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001000 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001001 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001002 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1003 * type commands. Where the struct request needs to be farther initialized by
1004 * the caller. It is passed a &struct bio, which describes the memory info of
1005 * the I/O transfer.
1006 *
1007 * The caller of blk_make_request must make sure that bi_io_vec
1008 * are set to describe the memory buffers. That bio_data_dir() will return
1009 * the needed direction of the request. (And all bio's in the passed bio-chain
1010 * are properly set accordingly)
1011 *
1012 * If called under none-sleepable conditions, mapped bio buffers must not
1013 * need bouncing, by calling the appropriate masked or flagged allocator,
1014 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1015 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001016 *
1017 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1018 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1019 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1020 * completion of a bio that hasn't been submitted yet, thus resulting in a
1021 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1022 * of bio_alloc(), as that avoids the mempool deadlock.
1023 * If possible a big IO should be split into smaller parts when allocation
1024 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001025 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001026struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1027 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001028{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001029 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1030
1031 if (unlikely(!rq))
1032 return ERR_PTR(-ENOMEM);
1033
1034 for_each_bio(bio) {
1035 struct bio *bounce_bio = bio;
1036 int ret;
1037
1038 blk_queue_bounce(q, &bounce_bio);
1039 ret = blk_rq_append_bio(q, rq, bounce_bio);
1040 if (unlikely(ret)) {
1041 blk_put_request(rq);
1042 return ERR_PTR(ret);
1043 }
1044 }
1045
1046 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001047}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001048EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001049
1050/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 * blk_requeue_request - put a request back on queue
1052 * @q: request queue where request should be inserted
1053 * @rq: request to be inserted
1054 *
1055 * Description:
1056 * Drivers often keep queueing requests until the hardware cannot accept
1057 * more, when that condition happens we need to put the request back
1058 * on the queue. Must be called with queue lock held.
1059 */
Jens Axboe165125e2007-07-24 09:28:11 +02001060void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001062 blk_delete_timer(rq);
1063 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001064 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001065
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 if (blk_rq_tagged(rq))
1067 blk_queue_end_tag(q, rq);
1068
James Bottomleyba396a62009-05-27 14:17:08 +02001069 BUG_ON(blk_queued_rq(rq));
1070
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 elv_requeue_request(q, rq);
1072}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073EXPORT_SYMBOL(blk_requeue_request);
1074
Jens Axboe73c10102011-03-08 13:19:51 +01001075static void add_acct_request(struct request_queue *q, struct request *rq,
1076 int where)
1077{
1078 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001079 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001080}
1081
Tejun Heo074a7ac2008-08-25 19:56:14 +09001082static void part_round_stats_single(int cpu, struct hd_struct *part,
1083 unsigned long now)
1084{
1085 if (now == part->stamp)
1086 return;
1087
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001088 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001089 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001090 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001091 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1092 }
1093 part->stamp = now;
1094}
1095
1096/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001097 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1098 * @cpu: cpu number for stats access
1099 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 *
1101 * The average IO queue length and utilisation statistics are maintained
1102 * by observing the current state of the queue length and the amount of
1103 * time it has been in this state for.
1104 *
1105 * Normally, that accounting is done on IO completion, but that can result
1106 * in more than a second's worth of IO being accounted for within any one
1107 * second, leading to >100% utilisation. To deal with that, we call this
1108 * function to do a round-off before returning the results when reading
1109 * /proc/diskstats. This accounts immediately for all queue usage up to
1110 * the current jiffies and restarts the counters again.
1111 */
Tejun Heoc9959052008-08-25 19:47:21 +09001112void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001113{
1114 unsigned long now = jiffies;
1115
Tejun Heo074a7ac2008-08-25 19:56:14 +09001116 if (part->partno)
1117 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1118 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001119}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001120EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001121
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122/*
1123 * queue lock must be held
1124 */
Jens Axboe165125e2007-07-24 09:28:11 +02001125void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 if (unlikely(!q))
1128 return;
1129 if (unlikely(--req->ref_count))
1130 return;
1131
Tejun Heo8922e162005-10-20 16:23:44 +02001132 elv_completed_request(q, req);
1133
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001134 /* this is a bio leak */
1135 WARN_ON(req->bio != NULL);
1136
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 /*
1138 * Request may not have originated from ll_rw_blk. if not,
1139 * it didn't come out of our reserved rq pools
1140 */
Jens Axboe49171e52006-08-10 08:59:11 +02001141 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001142 unsigned int flags = req->cmd_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001145 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
1147 blk_free_request(q, req);
Tejun Heo75eb6c32011-10-19 14:31:22 +02001148 freed_request(q, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 }
1150}
Mike Christie6e39b692005-11-11 05:30:24 -06001151EXPORT_SYMBOL_GPL(__blk_put_request);
1152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153void blk_put_request(struct request *req)
1154{
Tejun Heo8922e162005-10-20 16:23:44 +02001155 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001156 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001158 spin_lock_irqsave(q->queue_lock, flags);
1159 __blk_put_request(q, req);
1160 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162EXPORT_SYMBOL(blk_put_request);
1163
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001164/**
1165 * blk_add_request_payload - add a payload to a request
1166 * @rq: request to update
1167 * @page: page backing the payload
1168 * @len: length of the payload.
1169 *
1170 * This allows to later add a payload to an already submitted request by
1171 * a block driver. The driver needs to take care of freeing the payload
1172 * itself.
1173 *
1174 * Note that this is a quite horrible hack and nothing but handling of
1175 * discard requests should ever use it.
1176 */
1177void blk_add_request_payload(struct request *rq, struct page *page,
1178 unsigned int len)
1179{
1180 struct bio *bio = rq->bio;
1181
1182 bio->bi_io_vec->bv_page = page;
1183 bio->bi_io_vec->bv_offset = 0;
1184 bio->bi_io_vec->bv_len = len;
1185
1186 bio->bi_size = len;
1187 bio->bi_vcnt = 1;
1188 bio->bi_phys_segments = 1;
1189
1190 rq->__data_len = rq->resid_len = len;
1191 rq->nr_phys_segments = 1;
1192 rq->buffer = bio_data(bio);
1193}
1194EXPORT_SYMBOL_GPL(blk_add_request_payload);
1195
Jens Axboe73c10102011-03-08 13:19:51 +01001196static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1197 struct bio *bio)
1198{
1199 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1200
Jens Axboe73c10102011-03-08 13:19:51 +01001201 if (!ll_back_merge_fn(q, req, bio))
1202 return false;
1203
1204 trace_block_bio_backmerge(q, bio);
1205
1206 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1207 blk_rq_set_mixed_merge(req);
1208
1209 req->biotail->bi_next = bio;
1210 req->biotail = bio;
1211 req->__data_len += bio->bi_size;
1212 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1213
1214 drive_stat_acct(req, 0);
1215 return true;
1216}
1217
1218static bool bio_attempt_front_merge(struct request_queue *q,
1219 struct request *req, struct bio *bio)
1220{
1221 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001222
Jens Axboe73c10102011-03-08 13:19:51 +01001223 if (!ll_front_merge_fn(q, req, bio))
1224 return false;
1225
1226 trace_block_bio_frontmerge(q, bio);
1227
1228 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1229 blk_rq_set_mixed_merge(req);
1230
Jens Axboe73c10102011-03-08 13:19:51 +01001231 bio->bi_next = req->bio;
1232 req->bio = bio;
1233
1234 /*
1235 * may not be valid. if the low level driver said
1236 * it didn't need a bounce buffer then it better
1237 * not touch req->buffer either...
1238 */
1239 req->buffer = bio_data(bio);
1240 req->__sector = bio->bi_sector;
1241 req->__data_len += bio->bi_size;
1242 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1243
1244 drive_stat_acct(req, 0);
1245 return true;
1246}
1247
Tejun Heobd87b582011-10-19 14:33:08 +02001248/**
1249 * attempt_plug_merge - try to merge with %current's plugged list
1250 * @q: request_queue new bio is being queued at
1251 * @bio: new bio being queued
1252 * @request_count: out parameter for number of traversed plugged requests
1253 *
1254 * Determine whether @bio being queued on @q can be merged with a request
1255 * on %current's plugged list. Returns %true if merge was successful,
1256 * otherwise %false.
1257 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001258 * Plugging coalesces IOs from the same issuer for the same purpose without
1259 * going through @q->queue_lock. As such it's more of an issuing mechanism
1260 * than scheduling, and the request, while may have elvpriv data, is not
1261 * added on the elevator at this point. In addition, we don't have
1262 * reliable access to the elevator outside queue lock. Only check basic
1263 * merging parameters without querying the elevator.
Jens Axboe73c10102011-03-08 13:19:51 +01001264 */
Tejun Heobd87b582011-10-19 14:33:08 +02001265static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1266 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001267{
1268 struct blk_plug *plug;
1269 struct request *rq;
1270 bool ret = false;
1271
Tejun Heobd87b582011-10-19 14:33:08 +02001272 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001273 if (!plug)
1274 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001275 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001276
1277 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1278 int el_ret;
1279
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001280 (*request_count)++;
1281
Tejun Heo07c2bd32012-02-08 09:19:42 +01001282 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001283 continue;
1284
Tejun Heo050c8ea2012-02-08 09:19:38 +01001285 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001286 if (el_ret == ELEVATOR_BACK_MERGE) {
1287 ret = bio_attempt_back_merge(q, rq, bio);
1288 if (ret)
1289 break;
1290 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1291 ret = bio_attempt_front_merge(q, rq, bio);
1292 if (ret)
1293 break;
1294 }
1295 }
1296out:
1297 return ret;
1298}
1299
Jens Axboe86db1e22008-01-29 14:53:40 +01001300void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001301{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001302 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001303
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001304 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1305 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001306 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001307
Tejun Heo52d9e672006-01-06 09:49:58 +01001308 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001309 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001310 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001311 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001312}
1313
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001314void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001316 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001317 struct blk_plug *plug;
1318 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1319 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001320 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 /*
1323 * low level driver can indicate that it wants pages above a
1324 * certain limit bounced to low memory (ie for highmem, or even
1325 * ISA dma in theory)
1326 */
1327 blk_queue_bounce(q, &bio);
1328
Tejun Heo4fed9472010-09-03 11:56:17 +02001329 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001330 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001331 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001332 goto get_rq;
1333 }
1334
Jens Axboe73c10102011-03-08 13:19:51 +01001335 /*
1336 * Check if we can merge with the plugged list before grabbing
1337 * any locks.
1338 */
Tejun Heobd87b582011-10-19 14:33:08 +02001339 if (attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001340 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001341
1342 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
1344 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001345 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001346 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001347 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001348 if (!attempt_back_merge(q, req))
1349 elv_merged_request(q, req, el_ret);
1350 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001351 }
Jens Axboe73c10102011-03-08 13:19:51 +01001352 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001353 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001354 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001355 if (!attempt_front_merge(q, req))
1356 elv_merged_request(q, req, el_ret);
1357 goto out_unlock;
1358 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 }
1360
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001362 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001363 * This sync check and mask will be re-done in init_request_from_bio(),
1364 * but we need to set it earlier to expose the sync flag to the
1365 * rq allocator and io schedulers.
1366 */
1367 rw_flags = bio_data_dir(bio);
1368 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001369 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001370
1371 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001372 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001373 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001374 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001375 req = get_request_wait(q, rw_flags, bio);
Tejun Heoda8303c2011-10-19 14:33:05 +02001376 if (unlikely(!req)) {
1377 bio_endio(bio, -ENODEV); /* @q is dead */
1378 goto out_unlock;
1379 }
Nick Piggind6344532005-06-28 20:45:14 -07001380
Nick Piggin450991b2005-06-28 20:45:13 -07001381 /*
1382 * After dropping the lock and possibly sleeping here, our request
1383 * may now be mergeable after it had proven unmergeable (above).
1384 * We don't worry about that case for efficiency. It won't happen
1385 * often, and the elevators are able to handle it.
1386 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001387 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
Tao Ma9562ad92011-10-24 16:11:30 +02001389 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001390 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001391
Jens Axboe73c10102011-03-08 13:19:51 +01001392 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001393 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001394 /*
1395 * If this is the first request added after a plug, fire
1396 * of a plug trace. If others have been added before, check
1397 * if we have multiple devices in this plug. If so, make a
1398 * note to sort the list before dispatch.
1399 */
1400 if (list_empty(&plug->list))
1401 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001402 else {
1403 if (!plug->should_sort) {
1404 struct request *__rq;
Jens Axboe73c10102011-03-08 13:19:51 +01001405
Shaohua Li3540d5e2011-11-16 09:21:50 +01001406 __rq = list_entry_rq(plug->list.prev);
1407 if (__rq->q != q)
1408 plug->should_sort = 1;
1409 }
Shaohua Li019ceb72011-11-16 09:21:50 +01001410 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001411 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001412 trace_block_plug(q);
1413 }
Jens Axboe73c10102011-03-08 13:19:51 +01001414 }
Shaohua Lia6327162011-08-24 16:04:32 +02001415 list_add_tail(&req->queuelist, &plug->list);
1416 drive_stat_acct(req, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001417 } else {
1418 spin_lock_irq(q->queue_lock);
1419 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001420 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001421out_unlock:
1422 spin_unlock_irq(q->queue_lock);
1423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424}
Jens Axboec20e8de2011-09-12 12:03:37 +02001425EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
1427/*
1428 * If bio->bi_dev is a partition, remap the location
1429 */
1430static inline void blk_partition_remap(struct bio *bio)
1431{
1432 struct block_device *bdev = bio->bi_bdev;
1433
Jens Axboebf2de6f2007-09-27 13:01:25 +02001434 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001436
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 bio->bi_sector += p->start_sect;
1438 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001439
Mike Snitzerd07335e2010-11-16 12:52:38 +01001440 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1441 bdev->bd_dev,
1442 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 }
1444}
1445
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446static void handle_bad_sector(struct bio *bio)
1447{
1448 char b[BDEVNAME_SIZE];
1449
1450 printk(KERN_INFO "attempt to access beyond end of device\n");
1451 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1452 bdevname(bio->bi_bdev, b),
1453 bio->bi_rw,
1454 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001455 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
1457 set_bit(BIO_EOF, &bio->bi_flags);
1458}
1459
Akinobu Mitac17bb492006-12-08 02:39:46 -08001460#ifdef CONFIG_FAIL_MAKE_REQUEST
1461
1462static DECLARE_FAULT_ATTR(fail_make_request);
1463
1464static int __init setup_fail_make_request(char *str)
1465{
1466 return setup_fault_attr(&fail_make_request, str);
1467}
1468__setup("fail_make_request=", setup_fail_make_request);
1469
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001470static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001471{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001472 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001473}
1474
1475static int __init fail_make_request_debugfs(void)
1476{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001477 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1478 NULL, &fail_make_request);
1479
1480 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001481}
1482
1483late_initcall(fail_make_request_debugfs);
1484
1485#else /* CONFIG_FAIL_MAKE_REQUEST */
1486
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001487static inline bool should_fail_request(struct hd_struct *part,
1488 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001489{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001490 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001491}
1492
1493#endif /* CONFIG_FAIL_MAKE_REQUEST */
1494
Jens Axboec07e2b42007-07-18 13:27:58 +02001495/*
1496 * Check whether this bio extends beyond the end of the device.
1497 */
1498static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1499{
1500 sector_t maxsector;
1501
1502 if (!nr_sectors)
1503 return 0;
1504
1505 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001506 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001507 if (maxsector) {
1508 sector_t sector = bio->bi_sector;
1509
1510 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1511 /*
1512 * This may well happen - the kernel calls bread()
1513 * without checking the size of the device, e.g., when
1514 * mounting a device.
1515 */
1516 handle_bad_sector(bio);
1517 return 1;
1518 }
1519 }
1520
1521 return 0;
1522}
1523
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001524static noinline_for_stack bool
1525generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526{
Jens Axboe165125e2007-07-24 09:28:11 +02001527 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001528 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001529 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001530 char b[BDEVNAME_SIZE];
1531 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
1533 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
Jens Axboec07e2b42007-07-18 13:27:58 +02001535 if (bio_check_eod(bio, nr_sectors))
1536 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001538 q = bdev_get_queue(bio->bi_bdev);
1539 if (unlikely(!q)) {
1540 printk(KERN_ERR
1541 "generic_make_request: Trying to access "
1542 "nonexistent block-device %s (%Lu)\n",
1543 bdevname(bio->bi_bdev, b),
1544 (long long) bio->bi_sector);
1545 goto end_io;
1546 }
1547
1548 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1549 nr_sectors > queue_max_hw_sectors(q))) {
1550 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1551 bdevname(bio->bi_bdev, b),
1552 bio_sectors(bio),
1553 queue_max_hw_sectors(q));
1554 goto end_io;
1555 }
1556
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001557 part = bio->bi_bdev->bd_part;
1558 if (should_fail_request(part, bio->bi_size) ||
1559 should_fail_request(&part_to_disk(part)->part0,
1560 bio->bi_size))
1561 goto end_io;
1562
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001564 * If this device has partitions, remap block n
1565 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001567 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001569 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1570 goto end_io;
1571
1572 if (bio_check_eod(bio, nr_sectors))
1573 goto end_io;
1574
1575 /*
1576 * Filter flush bio's early so that make_request based
1577 * drivers without flush support don't have to worry
1578 * about them.
1579 */
1580 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1581 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1582 if (!nr_sectors) {
1583 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001584 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001588 if ((bio->bi_rw & REQ_DISCARD) &&
1589 (!blk_queue_discard(q) ||
1590 ((bio->bi_rw & REQ_SECURE) &&
1591 !blk_queue_secdiscard(q)))) {
1592 err = -EOPNOTSUPP;
1593 goto end_io;
1594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
Tejun Heobc16a4f2011-10-19 14:33:01 +02001596 if (blk_throtl_bio(q, bio))
1597 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001598
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001599 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001600 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001601
1602end_io:
1603 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001604 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605}
1606
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001607/**
1608 * generic_make_request - hand a buffer to its device driver for I/O
1609 * @bio: The bio describing the location in memory and on the device.
1610 *
1611 * generic_make_request() is used to make I/O requests of block
1612 * devices. It is passed a &struct bio, which describes the I/O that needs
1613 * to be done.
1614 *
1615 * generic_make_request() does not return any status. The
1616 * success/failure status of the request, along with notification of
1617 * completion, is delivered asynchronously through the bio->bi_end_io
1618 * function described (one day) else where.
1619 *
1620 * The caller of generic_make_request must make sure that bi_io_vec
1621 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1622 * set to describe the device address, and the
1623 * bi_end_io and optionally bi_private are set to describe how
1624 * completion notification should be signaled.
1625 *
1626 * generic_make_request and the drivers it calls may use bi_next if this
1627 * bio happens to be merged with someone else, and may resubmit the bio to
1628 * a lower device by calling into generic_make_request recursively, which
1629 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001630 */
1631void generic_make_request(struct bio *bio)
1632{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001633 struct bio_list bio_list_on_stack;
1634
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001635 if (!generic_make_request_checks(bio))
1636 return;
1637
1638 /*
1639 * We only want one ->make_request_fn to be active at a time, else
1640 * stack usage with stacked devices could be a problem. So use
1641 * current->bio_list to keep a list of requests submited by a
1642 * make_request_fn function. current->bio_list is also used as a
1643 * flag to say if generic_make_request is currently active in this
1644 * task or not. If it is NULL, then no make_request is active. If
1645 * it is non-NULL, then a make_request is active, and new requests
1646 * should be added at the tail
1647 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001648 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001649 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001650 return;
1651 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001652
Neil Brownd89d8792007-05-01 09:53:42 +02001653 /* following loop may be a bit non-obvious, and so deserves some
1654 * explanation.
1655 * Before entering the loop, bio->bi_next is NULL (as all callers
1656 * ensure that) so we have a list with a single bio.
1657 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001658 * we assign bio_list to a pointer to the bio_list_on_stack,
1659 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001660 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001661 * through a recursive call to generic_make_request. If it
1662 * did, we find a non-NULL value in bio_list and re-enter the loop
1663 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001664 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001665 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001666 */
1667 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001668 bio_list_init(&bio_list_on_stack);
1669 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001670 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001671 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1672
1673 q->make_request_fn(q, bio);
1674
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001675 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001676 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001677 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001678}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679EXPORT_SYMBOL(generic_make_request);
1680
1681/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001682 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1684 * @bio: The &struct bio which describes the I/O
1685 *
1686 * submit_bio() is very similar in purpose to generic_make_request(), and
1687 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001688 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 *
1690 */
1691void submit_bio(int rw, struct bio *bio)
1692{
1693 int count = bio_sectors(bio);
1694
Jens Axboe22e2c502005-06-27 10:55:12 +02001695 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
Jens Axboebf2de6f2007-09-27 13:01:25 +02001697 /*
1698 * If it's a regular read/write or a barrier with data attached,
1699 * go through the normal accounting stuff before submission.
1700 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001701 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001702 if (rw & WRITE) {
1703 count_vm_events(PGPGOUT, count);
1704 } else {
1705 task_io_account_read(bio->bi_size);
1706 count_vm_events(PGPGIN, count);
1707 }
1708
1709 if (unlikely(block_dump)) {
1710 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001711 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001712 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001713 (rw & WRITE) ? "WRITE" : "READ",
1714 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001715 bdevname(bio->bi_bdev, b),
1716 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001717 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 }
1719
1720 generic_make_request(bio);
1721}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722EXPORT_SYMBOL(submit_bio);
1723
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001724/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001725 * blk_rq_check_limits - Helper function to check a request for the queue limit
1726 * @q: the queue
1727 * @rq: the request being checked
1728 *
1729 * Description:
1730 * @rq may have been made based on weaker limitations of upper-level queues
1731 * in request stacking drivers, and it may violate the limitation of @q.
1732 * Since the block layer and the underlying device driver trust @rq
1733 * after it is inserted to @q, it should be checked against @q before
1734 * the insertion using this generic function.
1735 *
1736 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001737 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001738 * Request stacking drivers like request-based dm may change the queue
1739 * limits while requests are in the queue (e.g. dm's table swapping).
1740 * Such request stacking drivers should check those requests agaist
1741 * the new queue limits again when they dispatch those requests,
1742 * although such checkings are also done against the old queue limits
1743 * when submitting requests.
1744 */
1745int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1746{
ike Snitzer33839772010-08-08 12:11:33 -04001747 if (rq->cmd_flags & REQ_DISCARD)
1748 return 0;
1749
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001750 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1751 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001752 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1753 return -EIO;
1754 }
1755
1756 /*
1757 * queue's settings related to segment counting like q->bounce_pfn
1758 * may differ from that of other stacking queues.
1759 * Recalculate it to check the request correctly on this queue's
1760 * limitation.
1761 */
1762 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001763 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001764 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1765 return -EIO;
1766 }
1767
1768 return 0;
1769}
1770EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1771
1772/**
1773 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1774 * @q: the queue to submit the request
1775 * @rq: the request being queued
1776 */
1777int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1778{
1779 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001780 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001781
1782 if (blk_rq_check_limits(q, rq))
1783 return -EIO;
1784
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001785 if (rq->rq_disk &&
1786 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001787 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001788
1789 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo8ba61432011-12-14 00:33:37 +01001790 if (unlikely(blk_queue_dead(q))) {
1791 spin_unlock_irqrestore(q->queue_lock, flags);
1792 return -ENODEV;
1793 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001794
1795 /*
1796 * Submitting request must be dequeued before calling this function
1797 * because it will be linked to another request_queue
1798 */
1799 BUG_ON(blk_queued_rq(rq));
1800
Jeff Moyer4853aba2011-08-15 21:37:25 +02001801 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1802 where = ELEVATOR_INSERT_FLUSH;
1803
1804 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001805 if (where == ELEVATOR_INSERT_FLUSH)
1806 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001807 spin_unlock_irqrestore(q->queue_lock, flags);
1808
1809 return 0;
1810}
1811EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1812
Tejun Heo80a761f2009-07-03 17:48:17 +09001813/**
1814 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1815 * @rq: request to examine
1816 *
1817 * Description:
1818 * A request could be merge of IOs which require different failure
1819 * handling. This function determines the number of bytes which
1820 * can be failed from the beginning of the request without
1821 * crossing into area which need to be retried further.
1822 *
1823 * Return:
1824 * The number of bytes to fail.
1825 *
1826 * Context:
1827 * queue_lock must be held.
1828 */
1829unsigned int blk_rq_err_bytes(const struct request *rq)
1830{
1831 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1832 unsigned int bytes = 0;
1833 struct bio *bio;
1834
1835 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1836 return blk_rq_bytes(rq);
1837
1838 /*
1839 * Currently the only 'mixing' which can happen is between
1840 * different fastfail types. We can safely fail portions
1841 * which have all the failfast bits that the first one has -
1842 * the ones which are at least as eager to fail as the first
1843 * one.
1844 */
1845 for (bio = rq->bio; bio; bio = bio->bi_next) {
1846 if ((bio->bi_rw & ff) != ff)
1847 break;
1848 bytes += bio->bi_size;
1849 }
1850
1851 /* this could lead to infinite loop */
1852 BUG_ON(blk_rq_bytes(rq) && !bytes);
1853 return bytes;
1854}
1855EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1856
Jens Axboebc58ba92009-01-23 10:54:44 +01001857static void blk_account_io_completion(struct request *req, unsigned int bytes)
1858{
Jens Axboec2553b52009-04-24 08:10:11 +02001859 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001860 const int rw = rq_data_dir(req);
1861 struct hd_struct *part;
1862 int cpu;
1863
1864 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001865 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001866 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1867 part_stat_unlock();
1868 }
1869}
1870
1871static void blk_account_io_done(struct request *req)
1872{
Jens Axboebc58ba92009-01-23 10:54:44 +01001873 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001874 * Account IO completion. flush_rq isn't accounted as a
1875 * normal IO on queueing nor completion. Accounting the
1876 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001877 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001878 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001879 unsigned long duration = jiffies - req->start_time;
1880 const int rw = rq_data_dir(req);
1881 struct hd_struct *part;
1882 int cpu;
1883
1884 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001885 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001886
1887 part_stat_inc(cpu, part, ios[rw]);
1888 part_stat_add(cpu, part, ticks[rw], duration);
1889 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001890 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001891
Jens Axboe6c23a962011-01-07 08:43:37 +01001892 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001893 part_stat_unlock();
1894 }
1895}
1896
Tejun Heo53a08802008-12-03 12:41:26 +01001897/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001898 * blk_peek_request - peek at the top of a request queue
1899 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001900 *
1901 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001902 * Return the request at the top of @q. The returned request
1903 * should be started using blk_start_request() before LLD starts
1904 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001905 *
1906 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001907 * Pointer to the request at the top of @q if available. Null
1908 * otherwise.
1909 *
1910 * Context:
1911 * queue_lock must be held.
1912 */
1913struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001914{
1915 struct request *rq;
1916 int ret;
1917
1918 while ((rq = __elv_next_request(q)) != NULL) {
1919 if (!(rq->cmd_flags & REQ_STARTED)) {
1920 /*
1921 * This is the first time the device driver
1922 * sees this request (possibly after
1923 * requeueing). Notify IO scheduler.
1924 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001925 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001926 elv_activate_rq(q, rq);
1927
1928 /*
1929 * just mark as started even if we don't start
1930 * it, a request that has been delayed should
1931 * not be passed by new incoming requests
1932 */
1933 rq->cmd_flags |= REQ_STARTED;
1934 trace_block_rq_issue(q, rq);
1935 }
1936
1937 if (!q->boundary_rq || q->boundary_rq == rq) {
1938 q->end_sector = rq_end_sector(rq);
1939 q->boundary_rq = NULL;
1940 }
1941
1942 if (rq->cmd_flags & REQ_DONTPREP)
1943 break;
1944
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001945 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001946 /*
1947 * make sure space for the drain appears we
1948 * know we can do this because max_hw_segments
1949 * has been adjusted to be one fewer than the
1950 * device can handle
1951 */
1952 rq->nr_phys_segments++;
1953 }
1954
1955 if (!q->prep_rq_fn)
1956 break;
1957
1958 ret = q->prep_rq_fn(q, rq);
1959 if (ret == BLKPREP_OK) {
1960 break;
1961 } else if (ret == BLKPREP_DEFER) {
1962 /*
1963 * the request may have been (partially) prepped.
1964 * we need to keep this request in the front to
1965 * avoid resource deadlock. REQ_STARTED will
1966 * prevent other fs requests from passing this one.
1967 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001968 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001969 !(rq->cmd_flags & REQ_DONTPREP)) {
1970 /*
1971 * remove the space for the drain we added
1972 * so that we don't add it again
1973 */
1974 --rq->nr_phys_segments;
1975 }
1976
1977 rq = NULL;
1978 break;
1979 } else if (ret == BLKPREP_KILL) {
1980 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001981 /*
1982 * Mark this request as started so we don't trigger
1983 * any debug logic in the end I/O path.
1984 */
1985 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001986 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001987 } else {
1988 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1989 break;
1990 }
1991 }
1992
1993 return rq;
1994}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001995EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001996
Tejun Heo9934c8c2009-05-08 11:54:16 +09001997void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001998{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001999 struct request_queue *q = rq->q;
2000
Tejun Heo158dbda2009-04-23 11:05:18 +09002001 BUG_ON(list_empty(&rq->queuelist));
2002 BUG_ON(ELV_ON_HASH(rq));
2003
2004 list_del_init(&rq->queuelist);
2005
2006 /*
2007 * the time frame between a request being removed from the lists
2008 * and to it is freed is accounted as io that is in progress at
2009 * the driver side.
2010 */
Divyesh Shah91952912010-04-01 15:01:41 -07002011 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002012 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002013 set_io_start_time_ns(rq);
2014 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002015}
2016
Tejun Heo5efccd12009-04-23 11:05:18 +09002017/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002018 * blk_start_request - start request processing on the driver
2019 * @req: request to dequeue
2020 *
2021 * Description:
2022 * Dequeue @req and start timeout timer on it. This hands off the
2023 * request to the driver.
2024 *
2025 * Block internal functions which don't want to start timer should
2026 * call blk_dequeue_request().
2027 *
2028 * Context:
2029 * queue_lock must be held.
2030 */
2031void blk_start_request(struct request *req)
2032{
2033 blk_dequeue_request(req);
2034
2035 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002036 * We are now handing the request to the hardware, initialize
2037 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002038 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002039 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002040 if (unlikely(blk_bidi_rq(req)))
2041 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2042
Tejun Heo9934c8c2009-05-08 11:54:16 +09002043 blk_add_timer(req);
2044}
2045EXPORT_SYMBOL(blk_start_request);
2046
2047/**
2048 * blk_fetch_request - fetch a request from a request queue
2049 * @q: request queue to fetch a request from
2050 *
2051 * Description:
2052 * Return the request at the top of @q. The request is started on
2053 * return and LLD can start processing it immediately.
2054 *
2055 * Return:
2056 * Pointer to the request at the top of @q if available. Null
2057 * otherwise.
2058 *
2059 * Context:
2060 * queue_lock must be held.
2061 */
2062struct request *blk_fetch_request(struct request_queue *q)
2063{
2064 struct request *rq;
2065
2066 rq = blk_peek_request(q);
2067 if (rq)
2068 blk_start_request(rq);
2069 return rq;
2070}
2071EXPORT_SYMBOL(blk_fetch_request);
2072
2073/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002074 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002075 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002076 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002077 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002078 *
2079 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002080 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2081 * the request structure even if @req doesn't have leftover.
2082 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002083 *
2084 * This special helper function is only for request stacking drivers
2085 * (e.g. request-based dm) so that they can handle partial completion.
2086 * Actual device drivers should use blk_end_request instead.
2087 *
2088 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2089 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002090 *
2091 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002092 * %false - this request doesn't have any more data
2093 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002094 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002095bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002097 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 struct bio *bio;
2099
Tejun Heo2e60e022009-04-23 11:05:18 +09002100 if (!req->bio)
2101 return false;
2102
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002103 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002104
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002106 * For fs requests, rq is just carrier of independent bio's
2107 * and each partial completion should be handled separately.
2108 * Reset per-request error on each partial completion.
2109 *
2110 * TODO: tj: This is too subtle. It would be better to let
2111 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002113 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 req->errors = 0;
2115
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002116 if (error && req->cmd_type == REQ_TYPE_FS &&
2117 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002118 char *error_type;
2119
2120 switch (error) {
2121 case -ENOLINK:
2122 error_type = "recoverable transport";
2123 break;
2124 case -EREMOTEIO:
2125 error_type = "critical target";
2126 break;
2127 case -EBADE:
2128 error_type = "critical nexus";
2129 break;
2130 case -EIO:
2131 default:
2132 error_type = "I/O";
2133 break;
2134 }
2135 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2136 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2137 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 }
2139
Jens Axboebc58ba92009-01-23 10:54:44 +01002140 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002141
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 total_bytes = bio_nbytes = 0;
2143 while ((bio = req->bio) != NULL) {
2144 int nbytes;
2145
2146 if (nr_bytes >= bio->bi_size) {
2147 req->bio = bio->bi_next;
2148 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002149 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 next_idx = 0;
2151 bio_nbytes = 0;
2152 } else {
2153 int idx = bio->bi_idx + next_idx;
2154
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002155 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002157 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002158 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 break;
2160 }
2161
2162 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2163 BIO_BUG_ON(nbytes > bio->bi_size);
2164
2165 /*
2166 * not a complete bvec done
2167 */
2168 if (unlikely(nbytes > nr_bytes)) {
2169 bio_nbytes += nr_bytes;
2170 total_bytes += nr_bytes;
2171 break;
2172 }
2173
2174 /*
2175 * advance to the next vector
2176 */
2177 next_idx++;
2178 bio_nbytes += nbytes;
2179 }
2180
2181 total_bytes += nbytes;
2182 nr_bytes -= nbytes;
2183
Jens Axboe6728cb02008-01-31 13:03:55 +01002184 bio = req->bio;
2185 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 /*
2187 * end more in this run, or just return 'not-done'
2188 */
2189 if (unlikely(nr_bytes <= 0))
2190 break;
2191 }
2192 }
2193
2194 /*
2195 * completely done
2196 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002197 if (!req->bio) {
2198 /*
2199 * Reset counters so that the request stacking driver
2200 * can find how many bytes remain in the request
2201 * later.
2202 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002203 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002204 return false;
2205 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206
2207 /*
2208 * if the request wasn't completed, update state
2209 */
2210 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002211 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 bio->bi_idx += next_idx;
2213 bio_iovec(bio)->bv_offset += nr_bytes;
2214 bio_iovec(bio)->bv_len -= nr_bytes;
2215 }
2216
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002217 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002218 req->buffer = bio_data(req->bio);
2219
2220 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002221 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002222 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002223
Tejun Heo80a761f2009-07-03 17:48:17 +09002224 /* mixed attributes always follow the first bio */
2225 if (req->cmd_flags & REQ_MIXED_MERGE) {
2226 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2227 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2228 }
2229
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002230 /*
2231 * If total number of sectors is less than the first segment
2232 * size, something has gone terribly wrong.
2233 */
2234 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002235 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002236 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002237 }
2238
2239 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002241
Tejun Heo2e60e022009-04-23 11:05:18 +09002242 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243}
Tejun Heo2e60e022009-04-23 11:05:18 +09002244EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
Tejun Heo2e60e022009-04-23 11:05:18 +09002246static bool blk_update_bidi_request(struct request *rq, int error,
2247 unsigned int nr_bytes,
2248 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002249{
Tejun Heo2e60e022009-04-23 11:05:18 +09002250 if (blk_update_request(rq, error, nr_bytes))
2251 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002252
Tejun Heo2e60e022009-04-23 11:05:18 +09002253 /* Bidi request must be completed as a whole */
2254 if (unlikely(blk_bidi_rq(rq)) &&
2255 blk_update_request(rq->next_rq, error, bidi_bytes))
2256 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002257
Jens Axboee2e1a142010-06-09 10:42:09 +02002258 if (blk_queue_add_random(rq->q))
2259 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002260
2261 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262}
2263
James Bottomley28018c22010-07-01 19:49:17 +09002264/**
2265 * blk_unprep_request - unprepare a request
2266 * @req: the request
2267 *
2268 * This function makes a request ready for complete resubmission (or
2269 * completion). It happens only after all error handling is complete,
2270 * so represents the appropriate moment to deallocate any resources
2271 * that were allocated to the request in the prep_rq_fn. The queue
2272 * lock is held when calling this.
2273 */
2274void blk_unprep_request(struct request *req)
2275{
2276 struct request_queue *q = req->q;
2277
2278 req->cmd_flags &= ~REQ_DONTPREP;
2279 if (q->unprep_rq_fn)
2280 q->unprep_rq_fn(q, req);
2281}
2282EXPORT_SYMBOL_GPL(blk_unprep_request);
2283
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284/*
2285 * queue lock must be held
2286 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002287static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002289 if (blk_rq_tagged(req))
2290 blk_queue_end_tag(req->q, req);
2291
James Bottomleyba396a62009-05-27 14:17:08 +02002292 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002294 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002295 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
Mike Andersone78042e2008-10-30 02:16:20 -07002297 blk_delete_timer(req);
2298
James Bottomley28018c22010-07-01 19:49:17 +09002299 if (req->cmd_flags & REQ_DONTPREP)
2300 blk_unprep_request(req);
2301
2302
Jens Axboebc58ba92009-01-23 10:54:44 +01002303 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002304
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002306 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002307 else {
2308 if (blk_bidi_rq(req))
2309 __blk_put_request(req->next_rq->q, req->next_rq);
2310
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 }
2313}
2314
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002315/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002316 * blk_end_bidi_request - Complete a bidi request
2317 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002318 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002319 * @nr_bytes: number of bytes to complete @rq
2320 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002321 *
2322 * Description:
2323 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002324 * Drivers that supports bidi can safely call this member for any
2325 * type of request, bidi or uni. In the later case @bidi_bytes is
2326 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002327 *
2328 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002329 * %false - we are done with this request
2330 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002331 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002332static bool blk_end_bidi_request(struct request *rq, int error,
2333 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002334{
2335 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002336 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002337
Tejun Heo2e60e022009-04-23 11:05:18 +09002338 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2339 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002340
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002341 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002342 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002343 spin_unlock_irqrestore(q->queue_lock, flags);
2344
Tejun Heo2e60e022009-04-23 11:05:18 +09002345 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002346}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002347
2348/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002349 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2350 * @rq: the request to complete
2351 * @error: %0 for success, < %0 for error
2352 * @nr_bytes: number of bytes to complete @rq
2353 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002354 *
2355 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002356 * Identical to blk_end_bidi_request() except that queue lock is
2357 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002358 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002359 * Return:
2360 * %false - we are done with this request
2361 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002362 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002363bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002364 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002365{
Tejun Heo2e60e022009-04-23 11:05:18 +09002366 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2367 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002368
Tejun Heo2e60e022009-04-23 11:05:18 +09002369 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002370
Tejun Heo2e60e022009-04-23 11:05:18 +09002371 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002372}
2373
2374/**
2375 * blk_end_request - Helper function for drivers to complete the request.
2376 * @rq: the request being processed
2377 * @error: %0 for success, < %0 for error
2378 * @nr_bytes: number of bytes to complete
2379 *
2380 * Description:
2381 * Ends I/O on a number of bytes attached to @rq.
2382 * If @rq has leftover, sets it up for the next range of segments.
2383 *
2384 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002385 * %false - we are done with this request
2386 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002387 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002388bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002389{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002390 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002391}
Jens Axboe56ad1742009-07-28 22:11:24 +02002392EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002393
2394/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002395 * blk_end_request_all - Helper function for drives to finish the request.
2396 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002397 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002398 *
2399 * Description:
2400 * Completely finish @rq.
2401 */
2402void blk_end_request_all(struct request *rq, int error)
2403{
2404 bool pending;
2405 unsigned int bidi_bytes = 0;
2406
2407 if (unlikely(blk_bidi_rq(rq)))
2408 bidi_bytes = blk_rq_bytes(rq->next_rq);
2409
2410 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2411 BUG_ON(pending);
2412}
Jens Axboe56ad1742009-07-28 22:11:24 +02002413EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002414
2415/**
2416 * blk_end_request_cur - Helper function to finish the current request chunk.
2417 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002418 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002419 *
2420 * Description:
2421 * Complete the current consecutively mapped chunk from @rq.
2422 *
2423 * Return:
2424 * %false - we are done with this request
2425 * %true - still buffers pending for this request
2426 */
2427bool blk_end_request_cur(struct request *rq, int error)
2428{
2429 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2430}
Jens Axboe56ad1742009-07-28 22:11:24 +02002431EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002432
2433/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002434 * blk_end_request_err - Finish a request till the next failure boundary.
2435 * @rq: the request to finish till the next failure boundary for
2436 * @error: must be negative errno
2437 *
2438 * Description:
2439 * Complete @rq till the next failure boundary.
2440 *
2441 * Return:
2442 * %false - we are done with this request
2443 * %true - still buffers pending for this request
2444 */
2445bool blk_end_request_err(struct request *rq, int error)
2446{
2447 WARN_ON(error >= 0);
2448 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2449}
2450EXPORT_SYMBOL_GPL(blk_end_request_err);
2451
2452/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002453 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002454 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002455 * @error: %0 for success, < %0 for error
2456 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002457 *
2458 * Description:
2459 * Must be called with queue lock held unlike blk_end_request().
2460 *
2461 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002462 * %false - we are done with this request
2463 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002464 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002465bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002466{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002467 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002468}
Jens Axboe56ad1742009-07-28 22:11:24 +02002469EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002470
2471/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002472 * __blk_end_request_all - Helper function for drives to finish the request.
2473 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002474 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002475 *
2476 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002477 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002478 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002479void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002480{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002481 bool pending;
2482 unsigned int bidi_bytes = 0;
2483
2484 if (unlikely(blk_bidi_rq(rq)))
2485 bidi_bytes = blk_rq_bytes(rq->next_rq);
2486
2487 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2488 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002489}
Jens Axboe56ad1742009-07-28 22:11:24 +02002490EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002491
2492/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002493 * __blk_end_request_cur - Helper function to finish the current request chunk.
2494 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002495 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002496 *
2497 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002498 * Complete the current consecutively mapped chunk from @rq. Must
2499 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002500 *
2501 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002502 * %false - we are done with this request
2503 * %true - still buffers pending for this request
2504 */
2505bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002506{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002507 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002508}
Jens Axboe56ad1742009-07-28 22:11:24 +02002509EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002510
Tejun Heo80a761f2009-07-03 17:48:17 +09002511/**
2512 * __blk_end_request_err - Finish a request till the next failure boundary.
2513 * @rq: the request to finish till the next failure boundary for
2514 * @error: must be negative errno
2515 *
2516 * Description:
2517 * Complete @rq till the next failure boundary. Must be called
2518 * with queue lock held.
2519 *
2520 * Return:
2521 * %false - we are done with this request
2522 * %true - still buffers pending for this request
2523 */
2524bool __blk_end_request_err(struct request *rq, int error)
2525{
2526 WARN_ON(error >= 0);
2527 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2528}
2529EXPORT_SYMBOL_GPL(__blk_end_request_err);
2530
Jens Axboe86db1e22008-01-29 14:53:40 +01002531void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2532 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002534 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002535 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
David Woodhousefb2dce82008-08-05 18:01:53 +01002537 if (bio_has_data(bio)) {
2538 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002539 rq->buffer = bio_data(bio);
2540 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002541 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
NeilBrown66846572007-08-16 13:31:28 +02002544 if (bio->bi_bdev)
2545 rq->rq_disk = bio->bi_bdev->bd_disk;
2546}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002548#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2549/**
2550 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2551 * @rq: the request to be flushed
2552 *
2553 * Description:
2554 * Flush all pages in @rq.
2555 */
2556void rq_flush_dcache_pages(struct request *rq)
2557{
2558 struct req_iterator iter;
2559 struct bio_vec *bvec;
2560
2561 rq_for_each_segment(bvec, rq, iter)
2562 flush_dcache_page(bvec->bv_page);
2563}
2564EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2565#endif
2566
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002567/**
2568 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2569 * @q : the queue of the device being checked
2570 *
2571 * Description:
2572 * Check if underlying low-level drivers of a device are busy.
2573 * If the drivers want to export their busy state, they must set own
2574 * exporting function using blk_queue_lld_busy() first.
2575 *
2576 * Basically, this function is used only by request stacking drivers
2577 * to stop dispatching requests to underlying devices when underlying
2578 * devices are busy. This behavior helps more I/O merging on the queue
2579 * of the request stacking driver and prevents I/O throughput regression
2580 * on burst I/O load.
2581 *
2582 * Return:
2583 * 0 - Not busy (The request stacking driver should dispatch request)
2584 * 1 - Busy (The request stacking driver should stop dispatching request)
2585 */
2586int blk_lld_busy(struct request_queue *q)
2587{
2588 if (q->lld_busy_fn)
2589 return q->lld_busy_fn(q);
2590
2591 return 0;
2592}
2593EXPORT_SYMBOL_GPL(blk_lld_busy);
2594
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002595/**
2596 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2597 * @rq: the clone request to be cleaned up
2598 *
2599 * Description:
2600 * Free all bios in @rq for a cloned request.
2601 */
2602void blk_rq_unprep_clone(struct request *rq)
2603{
2604 struct bio *bio;
2605
2606 while ((bio = rq->bio) != NULL) {
2607 rq->bio = bio->bi_next;
2608
2609 bio_put(bio);
2610 }
2611}
2612EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2613
2614/*
2615 * Copy attributes of the original request to the clone request.
2616 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2617 */
2618static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2619{
2620 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002621 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002622 dst->cmd_type = src->cmd_type;
2623 dst->__sector = blk_rq_pos(src);
2624 dst->__data_len = blk_rq_bytes(src);
2625 dst->nr_phys_segments = src->nr_phys_segments;
2626 dst->ioprio = src->ioprio;
2627 dst->extra_len = src->extra_len;
2628}
2629
2630/**
2631 * blk_rq_prep_clone - Helper function to setup clone request
2632 * @rq: the request to be setup
2633 * @rq_src: original request to be cloned
2634 * @bs: bio_set that bios for clone are allocated from
2635 * @gfp_mask: memory allocation mask for bio
2636 * @bio_ctr: setup function to be called for each clone bio.
2637 * Returns %0 for success, non %0 for failure.
2638 * @data: private data to be passed to @bio_ctr
2639 *
2640 * Description:
2641 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2642 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2643 * are not copied, and copying such parts is the caller's responsibility.
2644 * Also, pages which the original bios are pointing to are not copied
2645 * and the cloned bios just point same pages.
2646 * So cloned bios must be completed before original bios, which means
2647 * the caller must complete @rq before @rq_src.
2648 */
2649int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2650 struct bio_set *bs, gfp_t gfp_mask,
2651 int (*bio_ctr)(struct bio *, struct bio *, void *),
2652 void *data)
2653{
2654 struct bio *bio, *bio_src;
2655
2656 if (!bs)
2657 bs = fs_bio_set;
2658
2659 blk_rq_init(NULL, rq);
2660
2661 __rq_for_each_bio(bio_src, rq_src) {
2662 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2663 if (!bio)
2664 goto free_and_out;
2665
2666 __bio_clone(bio, bio_src);
2667
2668 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002669 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002670 goto free_and_out;
2671
2672 if (bio_ctr && bio_ctr(bio, bio_src, data))
2673 goto free_and_out;
2674
2675 if (rq->bio) {
2676 rq->biotail->bi_next = bio;
2677 rq->biotail = bio;
2678 } else
2679 rq->bio = rq->biotail = bio;
2680 }
2681
2682 __blk_rq_prep_clone(rq, rq_src);
2683
2684 return 0;
2685
2686free_and_out:
2687 if (bio)
2688 bio_free(bio, bs);
2689 blk_rq_unprep_clone(rq);
2690
2691 return -ENOMEM;
2692}
2693EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2694
Jens Axboe18887ad2008-07-28 13:08:45 +02002695int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696{
2697 return queue_work(kblockd_workqueue, work);
2698}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699EXPORT_SYMBOL(kblockd_schedule_work);
2700
Vivek Goyale43473b2010-09-15 17:06:35 -04002701int kblockd_schedule_delayed_work(struct request_queue *q,
2702 struct delayed_work *dwork, unsigned long delay)
2703{
2704 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2705}
2706EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2707
Jens Axboe73c10102011-03-08 13:19:51 +01002708#define PLUG_MAGIC 0x91827364
2709
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002710/**
2711 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2712 * @plug: The &struct blk_plug that needs to be initialized
2713 *
2714 * Description:
2715 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2716 * pending I/O should the task end up blocking between blk_start_plug() and
2717 * blk_finish_plug(). This is important from a performance perspective, but
2718 * also ensures that we don't deadlock. For instance, if the task is blocking
2719 * for a memory allocation, memory reclaim could end up wanting to free a
2720 * page belonging to that request that is currently residing in our private
2721 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2722 * this kind of deadlock.
2723 */
Jens Axboe73c10102011-03-08 13:19:51 +01002724void blk_start_plug(struct blk_plug *plug)
2725{
2726 struct task_struct *tsk = current;
2727
2728 plug->magic = PLUG_MAGIC;
2729 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002730 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002731 plug->should_sort = 0;
2732
2733 /*
2734 * If this is a nested plug, don't actually assign it. It will be
2735 * flushed on its own.
2736 */
2737 if (!tsk->plug) {
2738 /*
2739 * Store ordering should not be needed here, since a potential
2740 * preempt will imply a full memory barrier
2741 */
2742 tsk->plug = plug;
2743 }
2744}
2745EXPORT_SYMBOL(blk_start_plug);
2746
2747static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2748{
2749 struct request *rqa = container_of(a, struct request, queuelist);
2750 struct request *rqb = container_of(b, struct request, queuelist);
2751
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002752 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002753}
2754
Jens Axboe49cac012011-04-16 13:51:05 +02002755/*
2756 * If 'from_schedule' is true, then postpone the dispatch of requests
2757 * until a safe kblockd context. We due this to avoid accidental big
2758 * additional stack usage in driver dispatch, in places where the originally
2759 * plugger did not intend it.
2760 */
Jens Axboef6603782011-04-15 15:49:07 +02002761static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002762 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002763 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002764{
Jens Axboe49cac012011-04-16 13:51:05 +02002765 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002766
2767 /*
Tejun Heo8ba61432011-12-14 00:33:37 +01002768 * Don't mess with dead queue.
2769 */
2770 if (unlikely(blk_queue_dead(q))) {
2771 spin_unlock(q->queue_lock);
2772 return;
2773 }
2774
2775 /*
Jens Axboe99e22592011-04-18 09:59:55 +02002776 * If we are punting this to kblockd, then we can safely drop
2777 * the queue_lock before waking kblockd (which needs to take
2778 * this lock).
2779 */
2780 if (from_schedule) {
2781 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002782 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002783 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002784 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002785 spin_unlock(q->queue_lock);
2786 }
2787
Jens Axboe94b5eb22011-04-12 10:12:19 +02002788}
2789
NeilBrown048c9372011-04-18 09:52:22 +02002790static void flush_plug_callbacks(struct blk_plug *plug)
2791{
2792 LIST_HEAD(callbacks);
2793
2794 if (list_empty(&plug->cb_list))
2795 return;
2796
2797 list_splice_init(&plug->cb_list, &callbacks);
2798
2799 while (!list_empty(&callbacks)) {
2800 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2801 struct blk_plug_cb,
2802 list);
2803 list_del(&cb->list);
2804 cb->callback(cb);
2805 }
2806}
2807
Jens Axboe49cac012011-04-16 13:51:05 +02002808void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002809{
2810 struct request_queue *q;
2811 unsigned long flags;
2812 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002813 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002814 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002815
2816 BUG_ON(plug->magic != PLUG_MAGIC);
2817
NeilBrown048c9372011-04-18 09:52:22 +02002818 flush_plug_callbacks(plug);
Jens Axboe73c10102011-03-08 13:19:51 +01002819 if (list_empty(&plug->list))
2820 return;
2821
NeilBrown109b8122011-04-11 14:13:10 +02002822 list_splice_init(&plug->list, &list);
2823
2824 if (plug->should_sort) {
2825 list_sort(NULL, &list, plug_rq_cmp);
2826 plug->should_sort = 0;
2827 }
Jens Axboe73c10102011-03-08 13:19:51 +01002828
2829 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002830 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002831
2832 /*
2833 * Save and disable interrupts here, to avoid doing it for every
2834 * queue lock we have to take.
2835 */
Jens Axboe73c10102011-03-08 13:19:51 +01002836 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002837 while (!list_empty(&list)) {
2838 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002839 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002840 BUG_ON(!rq->q);
2841 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002842 /*
2843 * This drops the queue lock
2844 */
2845 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002846 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002847 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002848 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002849 spin_lock(q->queue_lock);
2850 }
Tejun Heo8ba61432011-12-14 00:33:37 +01002851
2852 /*
2853 * Short-circuit if @q is dead
2854 */
2855 if (unlikely(blk_queue_dead(q))) {
2856 __blk_end_request_all(rq, -ENODEV);
2857 continue;
2858 }
2859
Jens Axboe73c10102011-03-08 13:19:51 +01002860 /*
2861 * rq is already accounted, so use raw insert
2862 */
Jens Axboe401a18e2011-03-25 16:57:52 +01002863 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2864 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2865 else
2866 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002867
2868 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01002869 }
2870
Jens Axboe99e22592011-04-18 09:59:55 +02002871 /*
2872 * This drops the queue lock
2873 */
2874 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002875 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002876
Jens Axboe73c10102011-03-08 13:19:51 +01002877 local_irq_restore(flags);
2878}
Jens Axboe73c10102011-03-08 13:19:51 +01002879
2880void blk_finish_plug(struct blk_plug *plug)
2881{
Jens Axboef6603782011-04-15 15:49:07 +02002882 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02002883
2884 if (plug == current->plug)
2885 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01002886}
2887EXPORT_SYMBOL(blk_finish_plug);
2888
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889int __init blk_dev_init(void)
2890{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002891 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2892 sizeof(((struct request *)0)->cmd_flags));
2893
Tejun Heo89b90be2011-01-03 15:01:47 +01002894 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2895 kblockd_workqueue = alloc_workqueue("kblockd",
2896 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 if (!kblockd_workqueue)
2898 panic("Failed to create kblockd\n");
2899
2900 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002901 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902
Jens Axboe8324aa92008-01-29 14:51:59 +01002903 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002904 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 return 0;
2907}