blob: fef79ccb2a118da2f6f3144c993c8338c7205351 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Jens Axboeff856ba2006-01-09 16:02:34 +010029#include <linux/interrupt.h>
30#include <linux/cpu.h>
Jens Axboe2056a782006-03-23 20:00:26 +010031#include <linux/blktrace_api.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080032#include <linux/fault-inject.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Jens Axboe8324aa92008-01-29 14:51:59 +010034#include "blk.h"
35
Jens Axboe165125e2007-07-24 09:28:11 +020036static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38/*
39 * For the allocated request tables
40 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010041static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43/*
44 * For queue allocation
45 */
Jens Axboe6728cb02008-01-31 13:03:55 +010046struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 * Controlling structure to kblockd
50 */
Jens Axboeff856ba2006-01-09 16:02:34 +010051static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Jens Axboeff856ba2006-01-09 16:02:34 +010053static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
54
Jens Axboe26b82562008-01-29 13:54:41 +010055static void drive_stat_acct(struct request *rq, int new_io)
56{
Jens Axboe28f13702008-05-07 10:15:46 +020057 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010058 int rw = rq_data_dir(rq);
59
60 if (!blk_fs_request(rq) || !rq->rq_disk)
61 return;
62
Jens Axboe28f13702008-05-07 10:15:46 +020063 part = get_part(rq->rq_disk, rq->sector);
64 if (!new_io)
65 __all_stat_inc(rq->rq_disk, part, merges[rw], rq->sector);
66 else {
Jens Axboe26b82562008-01-29 13:54:41 +010067 disk_round_stats(rq->rq_disk);
68 rq->rq_disk->in_flight++;
Jerome Marchand6f2576a2008-02-08 11:04:35 +010069 if (part) {
70 part_round_stats(part);
71 part->in_flight++;
72 }
Jens Axboe26b82562008-01-29 13:54:41 +010073 }
74}
75
Jens Axboe8324aa92008-01-29 14:51:59 +010076void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070077{
78 int nr;
79
80 nr = q->nr_requests - (q->nr_requests / 8) + 1;
81 if (nr > q->nr_requests)
82 nr = q->nr_requests;
83 q->nr_congestion_on = nr;
84
85 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
86 if (nr < 1)
87 nr = 1;
88 q->nr_congestion_off = nr;
89}
90
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/**
92 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
93 * @bdev: device
94 *
95 * Locates the passed device's request queue and returns the address of its
96 * backing_dev_info
97 *
98 * Will return NULL if the request queue cannot be located.
99 */
100struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
101{
102 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200103 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
105 if (q)
106 ret = &q->backing_dev_info;
107 return ret;
108}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109EXPORT_SYMBOL(blk_get_backing_dev_info);
110
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200111void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200113 memset(rq, 0, sizeof(*rq));
114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100116 INIT_LIST_HEAD(&rq->donelist);
Jens Axboe63a71382008-02-08 12:41:03 +0100117 rq->q = q;
118 rq->sector = rq->hard_sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200119 INIT_HLIST_NODE(&rq->hash);
120 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200121 rq->cmd = rq->__cmd;
Jens Axboe63a71382008-02-08 12:41:03 +0100122 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100123 rq->ref_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200125EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
NeilBrown5bb23a62007-09-27 12:46:13 +0200127static void req_bio_endio(struct request *rq, struct bio *bio,
128 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100129{
Jens Axboe165125e2007-07-24 09:28:11 +0200130 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100131
NeilBrown5bb23a62007-09-27 12:46:13 +0200132 if (&q->bar_rq != rq) {
133 if (error)
134 clear_bit(BIO_UPTODATE, &bio->bi_flags);
135 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
136 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100137
NeilBrown5bb23a62007-09-27 12:46:13 +0200138 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100139 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700140 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200141 nbytes = bio->bi_size;
142 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100143
NeilBrown5bb23a62007-09-27 12:46:13 +0200144 bio->bi_size -= nbytes;
145 bio->bi_sector += (nbytes >> 9);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200146
147 if (bio_integrity(bio))
148 bio_integrity_advance(bio, nbytes);
149
NeilBrown5bb23a62007-09-27 12:46:13 +0200150 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200151 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200152 } else {
153
154 /*
155 * Okay, this is the barrier request in progress, just
156 * record the error;
157 */
158 if (error && !q->orderr)
159 q->orderr = error;
160 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100161}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163void blk_dump_rq_flags(struct request *rq, char *msg)
164{
165 int bit;
166
Jens Axboe6728cb02008-01-31 13:03:55 +0100167 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200168 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
169 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
Jens Axboe6728cb02008-01-31 13:03:55 +0100171 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
172 (unsigned long long)rq->sector,
173 rq->nr_sectors,
174 rq->current_nr_sectors);
175 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
176 rq->bio, rq->biotail,
177 rq->buffer, rq->data,
178 rq->data_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
Jens Axboe4aff5e22006-08-10 08:44:47 +0200180 if (blk_pc_request(rq)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100181 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200182 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 printk("%02x ", rq->cmd[bit]);
184 printk("\n");
185 }
186}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187EXPORT_SYMBOL(blk_dump_rq_flags);
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189/*
190 * "plug" the device if there are no outstanding requests: this will
191 * force the transfer to start only after we have put all the requests
192 * on the list.
193 *
194 * This is called with interrupts off and no requests on the queue and
195 * with the queue lock held.
196 */
Jens Axboe165125e2007-07-24 09:28:11 +0200197void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
199 WARN_ON(!irqs_disabled());
200
201 /*
202 * don't plug a stopped queue, it must be paired with blk_start_queue()
203 * which will restart the queueing
204 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200205 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 return;
207
Jens Axboee48ec692008-07-03 13:18:54 +0200208 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +0100210 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
211 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213EXPORT_SYMBOL(blk_plug_device);
214
215/*
216 * remove the queue from the plugged list, if present. called with
217 * queue lock held and interrupts disabled.
218 */
Jens Axboe165125e2007-07-24 09:28:11 +0200219int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220{
221 WARN_ON(!irqs_disabled());
222
Jens Axboee48ec692008-07-03 13:18:54 +0200223 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 return 0;
225
226 del_timer(&q->unplug_timer);
227 return 1;
228}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229EXPORT_SYMBOL(blk_remove_plug);
230
231/*
232 * remove the plug and let it rip..
233 */
Jens Axboe165125e2007-07-24 09:28:11 +0200234void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200236 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 return;
238
239 if (!blk_remove_plug(q))
240 return;
241
Jens Axboe22e2c502005-06-27 10:55:12 +0200242 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
244EXPORT_SYMBOL(__generic_unplug_device);
245
246/**
247 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200248 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 *
250 * Description:
251 * Linux uses plugging to build bigger requests queues before letting
252 * the device have at them. If a queue is plugged, the I/O scheduler
253 * is still adding and merging requests on the queue. Once the queue
254 * gets unplugged, the request_fn defined for the queue is invoked and
255 * transfers started.
256 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200257void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200259 if (blk_queue_plugged(q)) {
260 spin_lock_irq(q->queue_lock);
261 __generic_unplug_device(q);
262 spin_unlock_irq(q->queue_lock);
263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265EXPORT_SYMBOL(generic_unplug_device);
266
267static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
268 struct page *page)
269{
Jens Axboe165125e2007-07-24 09:28:11 +0200270 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500272 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273}
274
Jens Axboe86db1e22008-01-29 14:53:40 +0100275void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
Jens Axboe165125e2007-07-24 09:28:11 +0200277 struct request_queue *q =
278 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
Jens Axboe2056a782006-03-23 20:00:26 +0100280 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
281 q->rq.count[READ] + q->rq.count[WRITE]);
282
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 q->unplug_fn(q);
284}
285
Jens Axboe86db1e22008-01-29 14:53:40 +0100286void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287{
Jens Axboe165125e2007-07-24 09:28:11 +0200288 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
Jens Axboe2056a782006-03-23 20:00:26 +0100290 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
291 q->rq.count[READ] + q->rq.count[WRITE]);
292
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 kblockd_schedule_work(&q->unplug_work);
294}
295
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500296void blk_unplug(struct request_queue *q)
297{
298 /*
299 * devices don't necessarily have an ->unplug_fn defined
300 */
301 if (q->unplug_fn) {
302 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
303 q->rq.count[READ] + q->rq.count[WRITE]);
304
305 q->unplug_fn(q);
306 }
307}
308EXPORT_SYMBOL(blk_unplug);
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310/**
311 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200312 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 *
314 * Description:
315 * blk_start_queue() will clear the stop flag on the queue, and call
316 * the request_fn for the queue if it was in a stopped state when
317 * entered. Also see blk_stop_queue(). Queue lock must be held.
318 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200319void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200321 WARN_ON(!irqs_disabled());
322
Nick Piggin75ad23b2008-04-29 14:48:33 +0200323 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
325 /*
326 * one level of recursion is ok and is much faster than kicking
327 * the unplug handling
328 */
Jens Axboee48ec692008-07-03 13:18:54 +0200329 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200331 queue_flag_clear(QUEUE_FLAG_REENTER, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 } else {
333 blk_plug_device(q);
334 kblockd_schedule_work(&q->unplug_work);
335 }
336}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337EXPORT_SYMBOL(blk_start_queue);
338
339/**
340 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200341 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 *
343 * Description:
344 * The Linux block layer assumes that a block driver will consume all
345 * entries on the request queue when the request_fn strategy is called.
346 * Often this will not happen, because of hardware limitations (queue
347 * depth settings). If a device driver gets a 'queue full' response,
348 * or if it simply chooses not to queue more I/O at one point, it can
349 * call this function to prevent the request_fn from being called until
350 * the driver has signalled it's ready to go again. This happens by calling
351 * blk_start_queue() to restart queue operations. Queue lock must be held.
352 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200353void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
355 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200356 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358EXPORT_SYMBOL(blk_stop_queue);
359
360/**
361 * blk_sync_queue - cancel any pending callbacks on a queue
362 * @q: the queue
363 *
364 * Description:
365 * The block layer may perform asynchronous callback activity
366 * on a queue, such as calling the unplug function after a timeout.
367 * A block device may call blk_sync_queue to ensure that any
368 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200369 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 * that its ->make_request_fn will not re-add plugging prior to calling
371 * this function.
372 *
373 */
374void blk_sync_queue(struct request_queue *q)
375{
376 del_timer_sync(&q->unplug_timer);
Oleg Nesterovabbeb882007-10-23 15:08:19 +0200377 kblockd_flush_work(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378}
379EXPORT_SYMBOL(blk_sync_queue);
380
381/**
382 * blk_run_queue - run a single device queue
383 * @q: The queue to run
384 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200385void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200388
389 /*
390 * Only recurse once to avoid overrunning the stack, let the unplug
391 * handling reinvoke the handler shortly if we already got there.
392 */
393 if (!elv_queue_empty(q)) {
Jens Axboee48ec692008-07-03 13:18:54 +0200394 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
Jens Axboedac07ec2006-05-11 08:20:16 +0200395 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200396 queue_flag_clear(QUEUE_FLAG_REENTER, q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200397 } else {
398 blk_plug_device(q);
399 kblockd_schedule_work(&q->unplug_work);
400 }
401 }
Nick Piggin75ad23b2008-04-29 14:48:33 +0200402}
403EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200404
Nick Piggin75ad23b2008-04-29 14:48:33 +0200405/**
406 * blk_run_queue - run a single device queue
407 * @q: The queue to run
408 */
409void blk_run_queue(struct request_queue *q)
410{
411 unsigned long flags;
412
413 spin_lock_irqsave(q->queue_lock, flags);
414 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 spin_unlock_irqrestore(q->queue_lock, flags);
416}
417EXPORT_SYMBOL(blk_run_queue);
418
Jens Axboe165125e2007-07-24 09:28:11 +0200419void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500420{
421 kobject_put(&q->kobj);
422}
Al Viro483f4af2006-03-18 18:34:37 -0500423
Jens Axboe6728cb02008-01-31 13:03:55 +0100424void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500425{
426 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200427 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500428 mutex_unlock(&q->sysfs_lock);
429
430 if (q->elevator)
431 elevator_exit(q->elevator);
432
433 blk_put_queue(q);
434}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435EXPORT_SYMBOL(blk_cleanup_queue);
436
Jens Axboe165125e2007-07-24 09:28:11 +0200437static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438{
439 struct request_list *rl = &q->rq;
440
441 rl->count[READ] = rl->count[WRITE] = 0;
442 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200443 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 init_waitqueue_head(&rl->wait[READ]);
445 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Christoph Lameter19460892005-06-23 00:08:19 -0700447 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
448 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
450 if (!rl->rq_pool)
451 return -ENOMEM;
452
453 return 0;
454}
455
Jens Axboe165125e2007-07-24 09:28:11 +0200456struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
Christoph Lameter19460892005-06-23 00:08:19 -0700458 return blk_alloc_queue_node(gfp_mask, -1);
459}
460EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Jens Axboe165125e2007-07-24 09:28:11 +0200462struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700463{
Jens Axboe165125e2007-07-24 09:28:11 +0200464 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700465 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700466
Jens Axboe8324aa92008-01-29 14:51:59 +0100467 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700468 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 if (!q)
470 return NULL;
471
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700472 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
473 q->backing_dev_info.unplug_io_data = q;
474 err = bdi_init(&q->backing_dev_info);
475 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100476 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700477 return NULL;
478 }
479
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500481
Jens Axboe8324aa92008-01-29 14:51:59 +0100482 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
Al Viro483f4af2006-03-18 18:34:37 -0500484 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700485 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500486
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 return q;
488}
Christoph Lameter19460892005-06-23 00:08:19 -0700489EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
491/**
492 * blk_init_queue - prepare a request queue for use with a block device
493 * @rfn: The function to be called to process requests that have been
494 * placed on the queue.
495 * @lock: Request queue spin lock
496 *
497 * Description:
498 * If a block device wishes to use the standard request handling procedures,
499 * which sorts requests and coalesces adjacent requests, then it must
500 * call blk_init_queue(). The function @rfn will be called when there
501 * are requests on the queue that need to be processed. If the device
502 * supports plugging, then @rfn may not be called immediately when requests
503 * are available on the queue, but may be called at some time later instead.
504 * Plugged queues are generally unplugged when a buffer belonging to one
505 * of the requests on the queue is needed, or due to memory pressure.
506 *
507 * @rfn is not required, or even expected, to remove all requests off the
508 * queue, but only as many as it can handle at a time. If it does leave
509 * requests on the queue, it is responsible for arranging that the requests
510 * get dealt with eventually.
511 *
512 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200513 * request queue; this lock will be taken also from interrupt context, so irq
514 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 *
516 * Function returns a pointer to the initialized request queue, or NULL if
517 * it didn't succeed.
518 *
519 * Note:
520 * blk_init_queue() must be paired with a blk_cleanup_queue() call
521 * when the block device is deactivated (such as at module unload).
522 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700523
Jens Axboe165125e2007-07-24 09:28:11 +0200524struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
Christoph Lameter19460892005-06-23 00:08:19 -0700526 return blk_init_queue_node(rfn, lock, -1);
527}
528EXPORT_SYMBOL(blk_init_queue);
529
Jens Axboe165125e2007-07-24 09:28:11 +0200530struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700531blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
532{
Jens Axboe165125e2007-07-24 09:28:11 +0200533 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535 if (!q)
536 return NULL;
537
Christoph Lameter19460892005-06-23 00:08:19 -0700538 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -0500539 if (blk_init_free_list(q)) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100540 kmem_cache_free(blk_requestq_cachep, q);
Al Viro8669aaf2006-03-18 13:50:00 -0500541 return NULL;
542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
152587d2005-04-12 16:22:06 -0500544 /*
545 * if caller didn't supply a lock, they get per-queue locking with
546 * our embedded lock
547 */
Neil Browne7e72bf2008-05-14 16:05:54 -0700548 if (!lock)
152587d2005-04-12 16:22:06 -0500549 lock = &q->__queue_lock;
152587d2005-04-12 16:22:06 -0500550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 q->prep_rq_fn = NULL;
553 q->unplug_fn = generic_unplug_device;
554 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
555 q->queue_lock = lock;
556
557 blk_queue_segment_boundary(q, 0xffffffff);
558
559 blk_queue_make_request(q, __make_request);
560 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
561
562 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
563 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
564
Alan Stern44ec9542007-02-20 11:01:57 -0500565 q->sg_reserved_size = INT_MAX;
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 /*
568 * all done
569 */
570 if (!elevator_init(q, NULL)) {
571 blk_queue_congestion_threshold(q);
572 return q;
573 }
574
Al Viro8669aaf2006-03-18 13:50:00 -0500575 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 return NULL;
577}
Christoph Lameter19460892005-06-23 00:08:19 -0700578EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
Jens Axboe165125e2007-07-24 09:28:11 +0200580int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700582 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500583 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 return 0;
585 }
586
587 return 1;
588}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Jens Axboe165125e2007-07-24 09:28:11 +0200590static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200592 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200593 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 mempool_free(rq, q->rq.rq_pool);
595}
596
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200597static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +0200598blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599{
600 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
601
602 if (!rq)
603 return NULL;
604
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200605 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +0200608 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 * see bio.h and blkdev.h
610 */
Jens Axboe49171e52006-08-10 08:59:11 +0200611 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Tejun Heocb98fc82005-10-28 08:29:39 +0200613 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200614 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200615 mempool_free(rq, q->rq.rq_pool);
616 return NULL;
617 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200618 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Tejun Heocb98fc82005-10-28 08:29:39 +0200621 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622}
623
624/*
625 * ioc_batching returns true if the ioc is a valid batching request and
626 * should be given priority access to a request.
627 */
Jens Axboe165125e2007-07-24 09:28:11 +0200628static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
630 if (!ioc)
631 return 0;
632
633 /*
634 * Make sure the process is able to allocate at least 1 request
635 * even if the batch times out, otherwise we could theoretically
636 * lose wakeups.
637 */
638 return ioc->nr_batch_requests == q->nr_batching ||
639 (ioc->nr_batch_requests > 0
640 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
641}
642
643/*
644 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
645 * will cause the process to be a "batcher" on all queues in the system. This
646 * is the behaviour we want though - once it gets a wakeup it should be given
647 * a nice run.
648 */
Jens Axboe165125e2007-07-24 09:28:11 +0200649static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
651 if (!ioc || ioc_batching(q, ioc))
652 return;
653
654 ioc->nr_batch_requests = q->nr_batching;
655 ioc->last_waited = jiffies;
656}
657
Jens Axboe165125e2007-07-24 09:28:11 +0200658static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
660 struct request_list *rl = &q->rq;
661
662 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -0700663 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
665 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 if (waitqueue_active(&rl->wait[rw]))
667 wake_up(&rl->wait[rw]);
668
669 blk_clear_queue_full(q, rw);
670 }
671}
672
673/*
674 * A request has just been released. Account for it, update the full and
675 * congestion status, wake up any waiters. Called under q->queue_lock.
676 */
Jens Axboe165125e2007-07-24 09:28:11 +0200677static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
679 struct request_list *rl = &q->rq;
680
681 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200682 if (priv)
683 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
685 __freed_request(q, rw);
686
687 if (unlikely(rl->starved[rw ^ 1]))
688 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689}
690
691#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
692/*
Nick Piggind6344532005-06-28 20:45:14 -0700693 * Get a free request, queue_lock must be held.
694 * Returns NULL on failure, with queue_lock held.
695 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 */
Jens Axboe165125e2007-07-24 09:28:11 +0200697static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100698 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
700 struct request *rq = NULL;
701 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100702 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +0100703 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100704 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
Jens Axboe7749a8d2006-12-13 13:02:26 +0100706 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100707 if (may_queue == ELV_MQUEUE_NO)
708 goto rq_starved;
709
710 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
711 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200712 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100713 /*
714 * The queue will fill after this allocation, so set
715 * it as full, and mark this process as "batching".
716 * This process will be allowed to complete a batch of
717 * requests, others will be blocked.
718 */
719 if (!blk_queue_full(q, rw)) {
720 ioc_set_batching(q, ioc);
721 blk_set_queue_full(q, rw);
722 } else {
723 if (may_queue != ELV_MQUEUE_MUST
724 && !ioc_batching(q, ioc)) {
725 /*
726 * The queue is full and the allocating
727 * process is not a "batcher", and not
728 * exempted by the IO scheduler
729 */
730 goto out;
731 }
732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 }
Thomas Maier79e2de42006-10-19 23:28:15 -0700734 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 }
736
Jens Axboe082cf692005-06-28 16:35:11 +0200737 /*
738 * Only allow batching queuers to allocate up to 50% over the defined
739 * limit of requests, otherwise we could have thousands of requests
740 * allocated with any setting of ->nr_requests
741 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100742 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200743 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100744
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 rl->count[rw]++;
746 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200747
Jens Axboe64521d12005-10-28 08:30:39 +0200748 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200749 if (priv)
750 rl->elvpriv++;
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 spin_unlock_irq(q->queue_lock);
753
Jens Axboe7749a8d2006-12-13 13:02:26 +0100754 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100755 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 /*
757 * Allocation failed presumably due to memory. Undo anything
758 * we might have messed up.
759 *
760 * Allocating task should really be put onto the front of the
761 * wait queue, but this is pretty rare.
762 */
763 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +0200764 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766 /*
767 * in the very unlikely event that allocation failed and no
768 * requests for this direction was pending, mark us starved
769 * so that freeing of a request in the other direction will
770 * notice us. another possible fix would be to split the
771 * rq mempool into READ and WRITE
772 */
773rq_starved:
774 if (unlikely(rl->count[rw] == 0))
775 rl->starved[rw] = 1;
776
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 goto out;
778 }
779
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100780 /*
781 * ioc may be NULL here, and ioc_batching will be false. That's
782 * OK, if the queue is under the request limit then requests need
783 * not count toward the nr_batch_requests limit. There will always
784 * be some limit enforced by BLK_BATCH_TIME.
785 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 if (ioc_batching(q, ioc))
787 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100788
Jens Axboe2056a782006-03-23 20:00:26 +0100789 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 return rq;
792}
793
794/*
795 * No available requests for this queue, unplug the device and wait for some
796 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700797 *
798 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 */
Jens Axboe165125e2007-07-24 09:28:11 +0200800static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200801 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Jens Axboe7749a8d2006-12-13 13:02:26 +0100803 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 struct request *rq;
805
Jens Axboe7749a8d2006-12-13 13:02:26 +0100806 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700807 while (!rq) {
808 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200809 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 struct request_list *rl = &q->rq;
811
812 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
813 TASK_UNINTERRUPTIBLE);
814
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200815 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200817 __generic_unplug_device(q);
818 spin_unlock_irq(q->queue_lock);
819 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200821 /*
822 * After sleeping, we become a "batching" process and
823 * will be able to allocate at least one request, and
824 * up to a big batch of them for a small period time.
825 * See ioc_batching, ioc_set_batching
826 */
827 ioc = current_io_context(GFP_NOIO, q->node);
828 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100829
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200830 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 finish_wait(&rl->wait[rw], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200832
833 rq = get_request(q, rw_flags, bio, GFP_NOIO);
834 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
836 return rq;
837}
838
Jens Axboe165125e2007-07-24 09:28:11 +0200839struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840{
841 struct request *rq;
842
843 BUG_ON(rw != READ && rw != WRITE);
844
Nick Piggind6344532005-06-28 20:45:14 -0700845 spin_lock_irq(q->queue_lock);
846 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200847 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700848 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200849 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700850 if (!rq)
851 spin_unlock_irq(q->queue_lock);
852 }
853 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855 return rq;
856}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857EXPORT_SYMBOL(blk_get_request);
858
859/**
Jens Axboedc72ef42006-07-20 14:54:05 +0200860 * blk_start_queueing - initiate dispatch of requests to device
861 * @q: request queue to kick into gear
862 *
863 * This is basically a helper to remove the need to know whether a queue
864 * is plugged or not if someone just wants to initiate dispatch of requests
865 * for this queue.
866 *
867 * The queue lock must be held with interrupts disabled.
868 */
Jens Axboe165125e2007-07-24 09:28:11 +0200869void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +0200870{
871 if (!blk_queue_plugged(q))
872 q->request_fn(q);
873 else
874 __generic_unplug_device(q);
875}
876EXPORT_SYMBOL(blk_start_queueing);
877
878/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 * blk_requeue_request - put a request back on queue
880 * @q: request queue where request should be inserted
881 * @rq: request to be inserted
882 *
883 * Description:
884 * Drivers often keep queueing requests until the hardware cannot accept
885 * more, when that condition happens we need to put the request back
886 * on the queue. Must be called with queue lock held.
887 */
Jens Axboe165125e2007-07-24 09:28:11 +0200888void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889{
Jens Axboe2056a782006-03-23 20:00:26 +0100890 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
891
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 if (blk_rq_tagged(rq))
893 blk_queue_end_tag(q, rq);
894
895 elv_requeue_request(q, rq);
896}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897EXPORT_SYMBOL(blk_requeue_request);
898
899/**
900 * blk_insert_request - insert a special request in to a request queue
901 * @q: request queue where request should be inserted
902 * @rq: request to be inserted
903 * @at_head: insert request at head or tail of queue
904 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 *
906 * Description:
907 * Many block devices need to execute commands asynchronously, so they don't
908 * block the whole kernel from preemption during request execution. This is
909 * accomplished normally by inserting aritficial requests tagged as
910 * REQ_SPECIAL in to the corresponding request queue, and letting them be
911 * scheduled for actual execution by the request queue.
912 *
913 * We have the option of inserting the head or the tail of the queue.
914 * Typically we use the tail for new ioctls and so forth. We use the head
915 * of the queue for things like a QUEUE_FULL message from a device, or a
916 * host that is unable to accept a particular command.
917 */
Jens Axboe165125e2007-07-24 09:28:11 +0200918void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500919 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920{
Tejun Heo 867d1192005-04-24 02:06:05 -0500921 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 unsigned long flags;
923
924 /*
925 * tell I/O scheduler that this isn't a regular read/write (ie it
926 * must not attempt merges on this) and that it acts as a soft
927 * barrier
928 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200929 rq->cmd_type = REQ_TYPE_SPECIAL;
930 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
932 rq->special = data;
933
934 spin_lock_irqsave(q->queue_lock, flags);
935
936 /*
937 * If command is tagged, release the tag
938 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500939 if (blk_rq_tagged(rq))
940 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200942 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -0500943 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +0200944 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 spin_unlock_irqrestore(q->queue_lock, flags);
946}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947EXPORT_SYMBOL(blk_insert_request);
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949/*
950 * add-request adds a request to the linked list.
951 * queue lock is held and interrupts disabled, as we muck with the
952 * request queue list.
953 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100954static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955{
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200956 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 /*
959 * elevator indicated where it wants this request to be
960 * inserted at elevator_merge time
961 */
962 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
963}
Jens Axboe6728cb02008-01-31 13:03:55 +0100964
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965/*
966 * disk_round_stats() - Round off the performance stats on a struct
967 * disk_stats.
968 *
969 * The average IO queue length and utilisation statistics are maintained
970 * by observing the current state of the queue length and the amount of
971 * time it has been in this state for.
972 *
973 * Normally, that accounting is done on IO completion, but that can result
974 * in more than a second's worth of IO being accounted for within any one
975 * second, leading to >100% utilisation. To deal with that, we call this
976 * function to do a round-off before returning the results when reading
977 * /proc/diskstats. This accounts immediately for all queue usage up to
978 * the current jiffies and restarts the counters again.
979 */
980void disk_round_stats(struct gendisk *disk)
981{
982 unsigned long now = jiffies;
983
Chen, Kenneth Wb2982642005-10-13 21:49:29 +0200984 if (now == disk->stamp)
985 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
Chen, Kenneth W20e5c812005-10-13 21:48:42 +0200987 if (disk->in_flight) {
988 __disk_stat_add(disk, time_in_queue,
989 disk->in_flight * (now - disk->stamp));
990 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
991 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993}
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800994EXPORT_SYMBOL_GPL(disk_round_stats);
995
Jerome Marchand6f2576a2008-02-08 11:04:35 +0100996void part_round_stats(struct hd_struct *part)
997{
998 unsigned long now = jiffies;
999
1000 if (now == part->stamp)
1001 return;
1002
1003 if (part->in_flight) {
1004 __part_stat_add(part, time_in_queue,
1005 part->in_flight * (now - part->stamp));
1006 __part_stat_add(part, io_ticks, (now - part->stamp));
1007 }
1008 part->stamp = now;
1009}
1010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011/*
1012 * queue lock must be held
1013 */
Jens Axboe165125e2007-07-24 09:28:11 +02001014void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 if (unlikely(!q))
1017 return;
1018 if (unlikely(--req->ref_count))
1019 return;
1020
Tejun Heo8922e162005-10-20 16:23:44 +02001021 elv_completed_request(q, req);
1022
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 /*
1024 * Request may not have originated from ll_rw_blk. if not,
1025 * it didn't come out of our reserved rq pools
1026 */
Jens Axboe49171e52006-08-10 08:59:11 +02001027 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001029 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001032 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
1034 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02001035 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
1037}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001038EXPORT_SYMBOL_GPL(__blk_put_request);
1039
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040void blk_put_request(struct request *req)
1041{
Tejun Heo8922e162005-10-20 16:23:44 +02001042 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001043 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001045 spin_lock_irqsave(q->queue_lock, flags);
1046 __blk_put_request(q, req);
1047 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049EXPORT_SYMBOL(blk_put_request);
1050
Jens Axboe86db1e22008-01-29 14:53:40 +01001051void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001052{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001053 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001054
1055 /*
1056 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1057 */
1058 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001059 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01001060
1061 /*
1062 * REQ_BARRIER implies no merging, but lets make it explicit
1063 */
1064 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001065 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01001066
Jens Axboeb31dc662006-06-13 08:26:10 +02001067 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001068 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02001069 if (bio_rw_meta(bio))
1070 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02001071
Tejun Heo52d9e672006-01-06 09:49:58 +01001072 req->errors = 0;
1073 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001074 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001075 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02001076 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001077}
1078
Jens Axboe165125e2007-07-24 09:28:11 +02001079static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
Nick Piggin450991b2005-06-28 20:45:13 -07001081 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02001082 int el_ret, nr_sectors, barrier, err;
1083 const unsigned short prio = bio_prio(bio);
1084 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01001085 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
1089 /*
1090 * low level driver can indicate that it wants pages above a
1091 * certain limit bounced to low memory (ie for highmem, or even
1092 * ISA dma in theory)
1093 */
1094 blk_queue_bounce(q, &bio);
1095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01001097 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 err = -EOPNOTSUPP;
1099 goto end_io;
1100 }
1101
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 spin_lock_irq(q->queue_lock);
1103
Nick Piggin450991b2005-06-28 20:45:13 -07001104 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 goto get_rq;
1106
1107 el_ret = elv_merge(q, &req, bio);
1108 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001109 case ELEVATOR_BACK_MERGE:
1110 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
Jens Axboe6728cb02008-01-31 13:03:55 +01001112 if (!ll_back_merge_fn(q, req, bio))
1113 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Jens Axboe6728cb02008-01-31 13:03:55 +01001115 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001116
Jens Axboe6728cb02008-01-31 13:03:55 +01001117 req->biotail->bi_next = bio;
1118 req->biotail = bio;
1119 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1120 req->ioprio = ioprio_best(req->ioprio, prio);
1121 drive_stat_acct(req, 0);
1122 if (!attempt_back_merge(q, req))
1123 elv_merged_request(q, req, el_ret);
1124 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
Jens Axboe6728cb02008-01-31 13:03:55 +01001126 case ELEVATOR_FRONT_MERGE:
1127 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Jens Axboe6728cb02008-01-31 13:03:55 +01001129 if (!ll_front_merge_fn(q, req, bio))
1130 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
Jens Axboe6728cb02008-01-31 13:03:55 +01001132 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001133
Jens Axboe6728cb02008-01-31 13:03:55 +01001134 bio->bi_next = req->bio;
1135 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
Jens Axboe6728cb02008-01-31 13:03:55 +01001137 /*
1138 * may not be valid. if the low level driver said
1139 * it didn't need a bounce buffer then it better
1140 * not touch req->buffer either...
1141 */
1142 req->buffer = bio_data(bio);
1143 req->current_nr_sectors = bio_cur_sectors(bio);
1144 req->hard_cur_sectors = req->current_nr_sectors;
1145 req->sector = req->hard_sector = bio->bi_sector;
1146 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1147 req->ioprio = ioprio_best(req->ioprio, prio);
1148 drive_stat_acct(req, 0);
1149 if (!attempt_front_merge(q, req))
1150 elv_merged_request(q, req, el_ret);
1151 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
Jens Axboe6728cb02008-01-31 13:03:55 +01001153 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1154 default:
1155 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 }
1157
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001159 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001160 * This sync check and mask will be re-done in init_request_from_bio(),
1161 * but we need to set it earlier to expose the sync flag to the
1162 * rq allocator and io schedulers.
1163 */
1164 rw_flags = bio_data_dir(bio);
1165 if (sync)
1166 rw_flags |= REQ_RW_SYNC;
1167
1168 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001169 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001170 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001171 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001172 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001173
Nick Piggin450991b2005-06-28 20:45:13 -07001174 /*
1175 * After dropping the lock and possibly sleeping here, our request
1176 * may now be mergeable after it had proven unmergeable (above).
1177 * We don't worry about that case for efficiency. It won't happen
1178 * often, and the elevators are able to handle it.
1179 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001180 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
Nick Piggin450991b2005-06-28 20:45:13 -07001182 spin_lock_irq(q->queue_lock);
1183 if (elv_queue_empty(q))
1184 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 add_request(q, req);
1186out:
Jens Axboe4a534f92005-04-16 15:25:40 -07001187 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 __generic_unplug_device(q);
1189
1190 spin_unlock_irq(q->queue_lock);
1191 return 0;
1192
1193end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02001194 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 return 0;
1196}
1197
1198/*
1199 * If bio->bi_dev is a partition, remap the location
1200 */
1201static inline void blk_partition_remap(struct bio *bio)
1202{
1203 struct block_device *bdev = bio->bi_bdev;
1204
Jens Axboebf2de6f2007-09-27 13:01:25 +02001205 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001207
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 bio->bi_sector += p->start_sect;
1209 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001210
1211 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
1212 bdev->bd_dev, bio->bi_sector,
1213 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
1215}
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217static void handle_bad_sector(struct bio *bio)
1218{
1219 char b[BDEVNAME_SIZE];
1220
1221 printk(KERN_INFO "attempt to access beyond end of device\n");
1222 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1223 bdevname(bio->bi_bdev, b),
1224 bio->bi_rw,
1225 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1226 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1227
1228 set_bit(BIO_EOF, &bio->bi_flags);
1229}
1230
Akinobu Mitac17bb492006-12-08 02:39:46 -08001231#ifdef CONFIG_FAIL_MAKE_REQUEST
1232
1233static DECLARE_FAULT_ATTR(fail_make_request);
1234
1235static int __init setup_fail_make_request(char *str)
1236{
1237 return setup_fault_attr(&fail_make_request, str);
1238}
1239__setup("fail_make_request=", setup_fail_make_request);
1240
1241static int should_fail_request(struct bio *bio)
1242{
1243 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
1244 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
1245 return should_fail(&fail_make_request, bio->bi_size);
1246
1247 return 0;
1248}
1249
1250static int __init fail_make_request_debugfs(void)
1251{
1252 return init_fault_attr_dentries(&fail_make_request,
1253 "fail_make_request");
1254}
1255
1256late_initcall(fail_make_request_debugfs);
1257
1258#else /* CONFIG_FAIL_MAKE_REQUEST */
1259
1260static inline int should_fail_request(struct bio *bio)
1261{
1262 return 0;
1263}
1264
1265#endif /* CONFIG_FAIL_MAKE_REQUEST */
1266
Jens Axboec07e2b42007-07-18 13:27:58 +02001267/*
1268 * Check whether this bio extends beyond the end of the device.
1269 */
1270static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1271{
1272 sector_t maxsector;
1273
1274 if (!nr_sectors)
1275 return 0;
1276
1277 /* Test device or partition size, when known. */
1278 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1279 if (maxsector) {
1280 sector_t sector = bio->bi_sector;
1281
1282 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1283 /*
1284 * This may well happen - the kernel calls bread()
1285 * without checking the size of the device, e.g., when
1286 * mounting a device.
1287 */
1288 handle_bad_sector(bio);
1289 return 1;
1290 }
1291 }
1292
1293 return 0;
1294}
1295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296/**
1297 * generic_make_request: hand a buffer to its device driver for I/O
1298 * @bio: The bio describing the location in memory and on the device.
1299 *
1300 * generic_make_request() is used to make I/O requests of block
1301 * devices. It is passed a &struct bio, which describes the I/O that needs
1302 * to be done.
1303 *
1304 * generic_make_request() does not return any status. The
1305 * success/failure status of the request, along with notification of
1306 * completion, is delivered asynchronously through the bio->bi_end_io
1307 * function described (one day) else where.
1308 *
1309 * The caller of generic_make_request must make sure that bi_io_vec
1310 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1311 * set to describe the device address, and the
1312 * bi_end_io and optionally bi_private are set to describe how
1313 * completion notification should be signaled.
1314 *
1315 * generic_make_request and the drivers it calls may use bi_next if this
1316 * bio happens to be merged with someone else, and may change bi_dev and
1317 * bi_sector for remaps as it sees fit. So the values of these fields
1318 * should NOT be depended on after the call to generic_make_request.
1319 */
Neil Brownd89d8792007-05-01 09:53:42 +02001320static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321{
Jens Axboe165125e2007-07-24 09:28:11 +02001322 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001323 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001325 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001326 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
1328 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Jens Axboec07e2b42007-07-18 13:27:58 +02001330 if (bio_check_eod(bio, nr_sectors))
1331 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 /*
1334 * Resolve the mapping until finished. (drivers are
1335 * still free to implement/resolve their own stacking
1336 * by explicitly returning 0)
1337 *
1338 * NOTE: we don't repeat the blk_size check for each new device.
1339 * Stacking drivers are expected to know what they are doing.
1340 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001341 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001342 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 do {
1344 char b[BDEVNAME_SIZE];
1345
1346 q = bdev_get_queue(bio->bi_bdev);
1347 if (!q) {
1348 printk(KERN_ERR
1349 "generic_make_request: Trying to access "
1350 "nonexistent block-device %s (%Lu)\n",
1351 bdevname(bio->bi_bdev, b),
1352 (long long) bio->bi_sector);
1353end_io:
Jens Axboe51fd77b2007-11-02 08:49:08 +01001354 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 break;
1356 }
1357
Jens Axboe4fa253f2007-07-18 13:13:10 +02001358 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001359 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 bdevname(bio->bi_bdev, b),
1361 bio_sectors(bio),
1362 q->max_hw_sectors);
1363 goto end_io;
1364 }
1365
Nick Pigginfde6ad22005-06-23 00:08:53 -07001366 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 goto end_io;
1368
Akinobu Mitac17bb492006-12-08 02:39:46 -08001369 if (should_fail_request(bio))
1370 goto end_io;
1371
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 /*
1373 * If this device has partitions, remap block n
1374 * of partition p to block n+start(p) of the disk.
1375 */
1376 blk_partition_remap(bio);
1377
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001378 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1379 goto end_io;
1380
NeilBrown5ddfe962006-10-30 22:07:21 -08001381 if (old_sector != -1)
Jens Axboe4fa253f2007-07-18 13:13:10 +02001382 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08001383 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001384
1385 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
1386
NeilBrown5ddfe962006-10-30 22:07:21 -08001387 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001388 old_dev = bio->bi_bdev->bd_dev;
1389
Jens Axboec07e2b42007-07-18 13:27:58 +02001390 if (bio_check_eod(bio, nr_sectors))
1391 goto end_io;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001392 if (bio_empty_barrier(bio) && !q->prepare_flush_fn) {
1393 err = -EOPNOTSUPP;
1394 goto end_io;
1395 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 ret = q->make_request_fn(q, bio);
1398 } while (ret);
1399}
1400
Neil Brownd89d8792007-05-01 09:53:42 +02001401/*
1402 * We only want one ->make_request_fn to be active at a time,
1403 * else stack usage with stacked devices could be a problem.
1404 * So use current->bio_{list,tail} to keep a list of requests
1405 * submited by a make_request_fn function.
1406 * current->bio_tail is also used as a flag to say if
1407 * generic_make_request is currently active in this task or not.
1408 * If it is NULL, then no make_request is active. If it is non-NULL,
1409 * then a make_request is active, and new requests should be added
1410 * at the tail
1411 */
1412void generic_make_request(struct bio *bio)
1413{
1414 if (current->bio_tail) {
1415 /* make_request is active */
1416 *(current->bio_tail) = bio;
1417 bio->bi_next = NULL;
1418 current->bio_tail = &bio->bi_next;
1419 return;
1420 }
1421 /* following loop may be a bit non-obvious, and so deserves some
1422 * explanation.
1423 * Before entering the loop, bio->bi_next is NULL (as all callers
1424 * ensure that) so we have a list with a single bio.
1425 * We pretend that we have just taken it off a longer list, so
1426 * we assign bio_list to the next (which is NULL) and bio_tail
1427 * to &bio_list, thus initialising the bio_list of new bios to be
1428 * added. __generic_make_request may indeed add some more bios
1429 * through a recursive call to generic_make_request. If it
1430 * did, we find a non-NULL value in bio_list and re-enter the loop
1431 * from the top. In this case we really did just take the bio
1432 * of the top of the list (no pretending) and so fixup bio_list and
1433 * bio_tail or bi_next, and call into __generic_make_request again.
1434 *
1435 * The loop was structured like this to make only one call to
1436 * __generic_make_request (which is important as it is large and
1437 * inlined) and to keep the structure simple.
1438 */
1439 BUG_ON(bio->bi_next);
1440 do {
1441 current->bio_list = bio->bi_next;
1442 if (bio->bi_next == NULL)
1443 current->bio_tail = &current->bio_list;
1444 else
1445 bio->bi_next = NULL;
1446 __generic_make_request(bio);
1447 bio = current->bio_list;
1448 } while (bio);
1449 current->bio_tail = NULL; /* deactivate */
1450}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451EXPORT_SYMBOL(generic_make_request);
1452
1453/**
1454 * submit_bio: submit a bio to the block device layer for I/O
1455 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1456 * @bio: The &struct bio which describes the I/O
1457 *
1458 * submit_bio() is very similar in purpose to generic_make_request(), and
1459 * uses that function to do most of the work. Both are fairly rough
1460 * interfaces, @bio must be presetup and ready for I/O.
1461 *
1462 */
1463void submit_bio(int rw, struct bio *bio)
1464{
1465 int count = bio_sectors(bio);
1466
Jens Axboe22e2c502005-06-27 10:55:12 +02001467 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
Jens Axboebf2de6f2007-09-27 13:01:25 +02001469 /*
1470 * If it's a regular read/write or a barrier with data attached,
1471 * go through the normal accounting stuff before submission.
1472 */
1473 if (!bio_empty_barrier(bio)) {
1474
1475 BIO_BUG_ON(!bio->bi_size);
1476 BIO_BUG_ON(!bio->bi_io_vec);
1477
1478 if (rw & WRITE) {
1479 count_vm_events(PGPGOUT, count);
1480 } else {
1481 task_io_account_read(bio->bi_size);
1482 count_vm_events(PGPGIN, count);
1483 }
1484
1485 if (unlikely(block_dump)) {
1486 char b[BDEVNAME_SIZE];
1487 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001488 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001489 (rw & WRITE) ? "WRITE" : "READ",
1490 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001491 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001492 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 }
1494
1495 generic_make_request(bio);
1496}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497EXPORT_SYMBOL(submit_bio);
1498
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001499/**
1500 * __end_that_request_first - end I/O on a request
1501 * @req: the request being processed
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001502 * @error: 0 for success, < 0 for error
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001503 * @nr_bytes: number of bytes to complete
1504 *
1505 * Description:
1506 * Ends I/O on a number of bytes attached to @req, and sets it up
1507 * for the next range of segments (if any) in the cluster.
1508 *
1509 * Return:
1510 * 0 - we are done with this request, call end_that_request_last()
1511 * 1 - still buffers pending for this request
1512 **/
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001513static int __end_that_request_first(struct request *req, int error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 int nr_bytes)
1515{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001516 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 struct bio *bio;
1518
Jens Axboe2056a782006-03-23 20:00:26 +01001519 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
1520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 * for a REQ_BLOCK_PC request, we want to carry any eventual
1523 * sense key with us all the way through
1524 */
1525 if (!blk_pc_request(req))
1526 req->errors = 0;
1527
Jens Axboe6728cb02008-01-31 13:03:55 +01001528 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1529 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 req->rq_disk ? req->rq_disk->disk_name : "?",
1531 (unsigned long long)req->sector);
1532 }
1533
Jens Axboed72d9042005-11-01 08:35:42 +01001534 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboe28f13702008-05-07 10:15:46 +02001535 struct hd_struct *part = get_part(req->rq_disk, req->sector);
Jens Axboea3623572005-11-01 09:26:16 +01001536 const int rw = rq_data_dir(req);
1537
Jens Axboe28f13702008-05-07 10:15:46 +02001538 all_stat_add(req->rq_disk, part, sectors[rw],
1539 nr_bytes >> 9, req->sector);
Jens Axboed72d9042005-11-01 08:35:42 +01001540 }
1541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 total_bytes = bio_nbytes = 0;
1543 while ((bio = req->bio) != NULL) {
1544 int nbytes;
1545
Jens Axboebf2de6f2007-09-27 13:01:25 +02001546 /*
1547 * For an empty barrier request, the low level driver must
1548 * store a potential error location in ->sector. We pass
1549 * that back up in ->bi_sector.
1550 */
1551 if (blk_empty_barrier(req))
1552 bio->bi_sector = req->sector;
1553
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 if (nr_bytes >= bio->bi_size) {
1555 req->bio = bio->bi_next;
1556 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02001557 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 next_idx = 0;
1559 bio_nbytes = 0;
1560 } else {
1561 int idx = bio->bi_idx + next_idx;
1562
1563 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1564 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01001565 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -07001566 __func__, bio->bi_idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 break;
1568 }
1569
1570 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1571 BIO_BUG_ON(nbytes > bio->bi_size);
1572
1573 /*
1574 * not a complete bvec done
1575 */
1576 if (unlikely(nbytes > nr_bytes)) {
1577 bio_nbytes += nr_bytes;
1578 total_bytes += nr_bytes;
1579 break;
1580 }
1581
1582 /*
1583 * advance to the next vector
1584 */
1585 next_idx++;
1586 bio_nbytes += nbytes;
1587 }
1588
1589 total_bytes += nbytes;
1590 nr_bytes -= nbytes;
1591
Jens Axboe6728cb02008-01-31 13:03:55 +01001592 bio = req->bio;
1593 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 /*
1595 * end more in this run, or just return 'not-done'
1596 */
1597 if (unlikely(nr_bytes <= 0))
1598 break;
1599 }
1600 }
1601
1602 /*
1603 * completely done
1604 */
1605 if (!req->bio)
1606 return 0;
1607
1608 /*
1609 * if the request wasn't completed, update state
1610 */
1611 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02001612 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 bio->bi_idx += next_idx;
1614 bio_iovec(bio)->bv_offset += nr_bytes;
1615 bio_iovec(bio)->bv_len -= nr_bytes;
1616 }
1617
1618 blk_recalc_rq_sectors(req, total_bytes >> 9);
1619 blk_recalc_rq_segments(req);
1620 return 1;
1621}
1622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623/*
Jens Axboeff856ba2006-01-09 16:02:34 +01001624 * splice the completion data to a local structure and hand off to
1625 * process_completion_queue() to complete the requests
1626 */
1627static void blk_done_softirq(struct softirq_action *h)
1628{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07001629 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01001630
1631 local_irq_disable();
1632 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07001633 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01001634 local_irq_enable();
1635
1636 while (!list_empty(&local_list)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001637 struct request *rq;
Jens Axboeff856ba2006-01-09 16:02:34 +01001638
Jens Axboe6728cb02008-01-31 13:03:55 +01001639 rq = list_entry(local_list.next, struct request, donelist);
Jens Axboeff856ba2006-01-09 16:02:34 +01001640 list_del_init(&rq->donelist);
1641 rq->q->softirq_done_fn(rq);
1642 }
1643}
1644
Jens Axboe6728cb02008-01-31 13:03:55 +01001645static int __cpuinit blk_cpu_notify(struct notifier_block *self,
1646 unsigned long action, void *hcpu)
Jens Axboeff856ba2006-01-09 16:02:34 +01001647{
1648 /*
1649 * If a CPU goes away, splice its entries to the current CPU
1650 * and trigger a run of the softirq
1651 */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001652 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Jens Axboeff856ba2006-01-09 16:02:34 +01001653 int cpu = (unsigned long) hcpu;
1654
1655 local_irq_disable();
1656 list_splice_init(&per_cpu(blk_cpu_done, cpu),
1657 &__get_cpu_var(blk_cpu_done));
1658 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1659 local_irq_enable();
1660 }
1661
1662 return NOTIFY_OK;
1663}
1664
1665
Satyam Sharmadb47d472007-08-23 09:29:40 +02001666static struct notifier_block blk_cpu_notifier __cpuinitdata = {
Jens Axboeff856ba2006-01-09 16:02:34 +01001667 .notifier_call = blk_cpu_notify,
1668};
1669
Jens Axboeff856ba2006-01-09 16:02:34 +01001670/**
1671 * blk_complete_request - end I/O on a request
1672 * @req: the request being processed
1673 *
1674 * Description:
1675 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02001676 * unless the driver actually implements this in its completion callback
Jens Axboe4fa253f2007-07-18 13:13:10 +02001677 * through requeueing. The actual completion happens out-of-order,
Jens Axboeff856ba2006-01-09 16:02:34 +01001678 * through a softirq handler. The user must have registered a completion
1679 * callback through blk_queue_softirq_done().
1680 **/
1681
1682void blk_complete_request(struct request *req)
1683{
1684 struct list_head *cpu_list;
1685 unsigned long flags;
1686
1687 BUG_ON(!req->q->softirq_done_fn);
Jens Axboe6728cb02008-01-31 13:03:55 +01001688
Jens Axboeff856ba2006-01-09 16:02:34 +01001689 local_irq_save(flags);
1690
1691 cpu_list = &__get_cpu_var(blk_cpu_done);
1692 list_add_tail(&req->donelist, cpu_list);
1693 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1694
1695 local_irq_restore(flags);
1696}
Jens Axboeff856ba2006-01-09 16:02:34 +01001697EXPORT_SYMBOL(blk_complete_request);
Jens Axboe6728cb02008-01-31 13:03:55 +01001698
Jens Axboeff856ba2006-01-09 16:02:34 +01001699/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 * queue lock must be held
1701 */
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001702static void end_that_request_last(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703{
1704 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01001705
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001706 if (blk_rq_tagged(req))
1707 blk_queue_end_tag(req->q, req);
1708
1709 if (blk_queued_rq(req))
1710 blkdev_dequeue_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
1712 if (unlikely(laptop_mode) && blk_fs_request(req))
1713 laptop_io_completion();
1714
Jens Axboefd0ff8a2006-05-23 11:23:49 +02001715 /*
1716 * Account IO completion. bar_rq isn't accounted as a normal
1717 * IO on queueing nor completion. Accounting the containing
1718 * request is enough.
1719 */
1720 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01001722 const int rw = rq_data_dir(req);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001723 struct hd_struct *part = get_part(disk, req->sector);
Jens Axboea3623572005-11-01 09:26:16 +01001724
Jens Axboe28f13702008-05-07 10:15:46 +02001725 __all_stat_inc(disk, part, ios[rw], req->sector);
1726 __all_stat_add(disk, part, ticks[rw], duration, req->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 disk_round_stats(disk);
1728 disk->in_flight--;
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001729 if (part) {
1730 part_round_stats(part);
1731 part->in_flight--;
1732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 }
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001734
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01001736 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001737 else {
1738 if (blk_bidi_rq(req))
1739 __blk_put_request(req->next_rq->q, req->next_rq);
1740
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 }
1743}
1744
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745static inline void __end_request(struct request *rq, int uptodate,
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001746 unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001748 int error = 0;
1749
1750 if (uptodate <= 0)
1751 error = uptodate ? uptodate : -EIO;
1752
1753 __blk_end_request(rq, error, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754}
1755
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001756/**
1757 * blk_rq_bytes - Returns bytes left to complete in the entire request
Randy Dunlap5d87a052008-02-20 09:01:22 +01001758 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001759 **/
1760unsigned int blk_rq_bytes(struct request *rq)
Jens Axboea0cd1282007-09-21 10:41:07 +02001761{
1762 if (blk_fs_request(rq))
1763 return rq->hard_nr_sectors << 9;
1764
1765 return rq->data_len;
1766}
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001767EXPORT_SYMBOL_GPL(blk_rq_bytes);
1768
1769/**
1770 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
Randy Dunlap5d87a052008-02-20 09:01:22 +01001771 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001772 **/
1773unsigned int blk_rq_cur_bytes(struct request *rq)
1774{
1775 if (blk_fs_request(rq))
1776 return rq->current_nr_sectors << 9;
1777
1778 if (rq->bio)
1779 return rq->bio->bi_size;
1780
1781 return rq->data_len;
1782}
1783EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
Jens Axboea0cd1282007-09-21 10:41:07 +02001784
1785/**
1786 * end_queued_request - end all I/O on a queued request
1787 * @rq: the request being processed
1788 * @uptodate: error value or 0/1 uptodate flag
1789 *
1790 * Description:
1791 * Ends all I/O on a request, and removes it from the block layer queues.
1792 * Not suitable for normal IO completion, unless the driver still has
1793 * the request attached to the block layer.
1794 *
1795 **/
1796void end_queued_request(struct request *rq, int uptodate)
1797{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001798 __end_request(rq, uptodate, blk_rq_bytes(rq));
Jens Axboea0cd1282007-09-21 10:41:07 +02001799}
1800EXPORT_SYMBOL(end_queued_request);
1801
1802/**
1803 * end_dequeued_request - end all I/O on a dequeued request
1804 * @rq: the request being processed
1805 * @uptodate: error value or 0/1 uptodate flag
1806 *
1807 * Description:
1808 * Ends all I/O on a request. The request must already have been
1809 * dequeued using blkdev_dequeue_request(), as is normally the case
1810 * for most drivers.
1811 *
1812 **/
1813void end_dequeued_request(struct request *rq, int uptodate)
1814{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001815 __end_request(rq, uptodate, blk_rq_bytes(rq));
Jens Axboea0cd1282007-09-21 10:41:07 +02001816}
1817EXPORT_SYMBOL(end_dequeued_request);
1818
1819
1820/**
1821 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07001822 * @req: the request being processed
Jens Axboea0cd1282007-09-21 10:41:07 +02001823 * @uptodate: error value or 0/1 uptodate flag
1824 *
1825 * Description:
1826 * Ends I/O on the current segment of a request. If that is the only
1827 * remaining segment, the request is also completed and freed.
1828 *
1829 * This is a remnant of how older block drivers handled IO completions.
1830 * Modern drivers typically end IO on the full request in one go, unless
1831 * they have a residual value to account for. For that case this function
1832 * isn't really useful, unless the residual just happens to be the
1833 * full current segment. In other words, don't use this function in new
1834 * code. Either use end_request_completely(), or the
1835 * end_that_request_chunk() (along with end_that_request_last()) for
1836 * partial completions.
1837 *
1838 **/
1839void end_request(struct request *req, int uptodate)
1840{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001841 __end_request(req, uptodate, req->hard_cur_sectors << 9);
Jens Axboea0cd1282007-09-21 10:41:07 +02001842}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843EXPORT_SYMBOL(end_request);
1844
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001845/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001846 * blk_end_io - Generic end_io function to complete a request.
1847 * @rq: the request being processed
1848 * @error: 0 for success, < 0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001849 * @nr_bytes: number of bytes to complete @rq
1850 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001851 * @drv_callback: function called between completion of bios in the request
1852 * and completion of the request.
1853 * If the callback returns non 0, this helper returns without
1854 * completion of the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001855 *
1856 * Description:
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001857 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001858 * If @rq has leftover, sets it up for the next range of segments.
1859 *
1860 * Return:
1861 * 0 - we are done with this request
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001862 * 1 - this request is not freed yet, it still has pending buffers.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001863 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001864static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1865 unsigned int bidi_bytes,
1866 int (drv_callback)(struct request *))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001867{
1868 struct request_queue *q = rq->q;
1869 unsigned long flags = 0UL;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001870
1871 if (blk_fs_request(rq) || blk_pc_request(rq)) {
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001872 if (__end_that_request_first(rq, error, nr_bytes))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001873 return 1;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001874
1875 /* Bidi request must be completed as a whole */
1876 if (blk_bidi_rq(rq) &&
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001877 __end_that_request_first(rq->next_rq, error, bidi_bytes))
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001878 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001879 }
1880
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001881 /* Special feature for tricky drivers */
1882 if (drv_callback && drv_callback(rq))
1883 return 1;
1884
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001885 add_disk_randomness(rq->rq_disk);
1886
1887 spin_lock_irqsave(q->queue_lock, flags);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001888 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001889 spin_unlock_irqrestore(q->queue_lock, flags);
1890
1891 return 0;
1892}
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001893
1894/**
1895 * blk_end_request - Helper function for drivers to complete the request.
1896 * @rq: the request being processed
1897 * @error: 0 for success, < 0 for error
1898 * @nr_bytes: number of bytes to complete
1899 *
1900 * Description:
1901 * Ends I/O on a number of bytes attached to @rq.
1902 * If @rq has leftover, sets it up for the next range of segments.
1903 *
1904 * Return:
1905 * 0 - we are done with this request
1906 * 1 - still buffers pending for this request
1907 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001908int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001909{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001910 return blk_end_io(rq, error, nr_bytes, 0, NULL);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001911}
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001912EXPORT_SYMBOL_GPL(blk_end_request);
1913
1914/**
1915 * __blk_end_request - Helper function for drivers to complete the request.
1916 * @rq: the request being processed
1917 * @error: 0 for success, < 0 for error
1918 * @nr_bytes: number of bytes to complete
1919 *
1920 * Description:
1921 * Must be called with queue lock held unlike blk_end_request().
1922 *
1923 * Return:
1924 * 0 - we are done with this request
1925 * 1 - still buffers pending for this request
1926 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001927int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001928{
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001929 if (blk_fs_request(rq) || blk_pc_request(rq)) {
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001930 if (__end_that_request_first(rq, error, nr_bytes))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001931 return 1;
1932 }
1933
1934 add_disk_randomness(rq->rq_disk);
1935
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001936 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001937
1938 return 0;
1939}
1940EXPORT_SYMBOL_GPL(__blk_end_request);
1941
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001942/**
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001943 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1944 * @rq: the bidi request being processed
1945 * @error: 0 for success, < 0 for error
1946 * @nr_bytes: number of bytes to complete @rq
1947 * @bidi_bytes: number of bytes to complete @rq->next_rq
1948 *
1949 * Description:
1950 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1951 *
1952 * Return:
1953 * 0 - we are done with this request
1954 * 1 - still buffers pending for this request
1955 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001956int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
1957 unsigned int bidi_bytes)
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001958{
1959 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
1960}
1961EXPORT_SYMBOL_GPL(blk_end_bidi_request);
1962
1963/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001964 * blk_end_request_callback - Special helper function for tricky drivers
1965 * @rq: the request being processed
1966 * @error: 0 for success, < 0 for error
1967 * @nr_bytes: number of bytes to complete
1968 * @drv_callback: function called between completion of bios in the request
1969 * and completion of the request.
1970 * If the callback returns non 0, this helper returns without
1971 * completion of the request.
1972 *
1973 * Description:
1974 * Ends I/O on a number of bytes attached to @rq.
1975 * If @rq has leftover, sets it up for the next range of segments.
1976 *
1977 * This special helper function is used only for existing tricky drivers.
1978 * (e.g. cdrom_newpc_intr() of ide-cd)
1979 * This interface will be removed when such drivers are rewritten.
1980 * Don't use this interface in other places anymore.
1981 *
1982 * Return:
1983 * 0 - we are done with this request
1984 * 1 - this request is not freed yet.
1985 * this request still has pending buffers or
1986 * the driver doesn't want to finish this request yet.
1987 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001988int blk_end_request_callback(struct request *rq, int error,
1989 unsigned int nr_bytes,
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001990 int (drv_callback)(struct request *))
1991{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001992 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001993}
1994EXPORT_SYMBOL_GPL(blk_end_request_callback);
1995
Jens Axboe86db1e22008-01-29 14:53:40 +01001996void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
1997 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001999 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
2000 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001
2002 rq->nr_phys_segments = bio_phys_segments(q, bio);
2003 rq->nr_hw_segments = bio_hw_segments(q, bio);
2004 rq->current_nr_sectors = bio_cur_sectors(bio);
2005 rq->hard_cur_sectors = rq->current_nr_sectors;
2006 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
2007 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002008 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009
2010 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
NeilBrown66846572007-08-16 13:31:28 +02002012 if (bio->bi_bdev)
2013 rq->rq_disk = bio->bi_bdev->bd_disk;
2014}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
2016int kblockd_schedule_work(struct work_struct *work)
2017{
2018 return queue_work(kblockd_workqueue, work);
2019}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020EXPORT_SYMBOL(kblockd_schedule_work);
2021
Andrew Morton19a75d82007-05-09 02:33:56 -07002022void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07002024 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025}
Andrew Morton19a75d82007-05-09 02:33:56 -07002026EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
2028int __init blk_dev_init(void)
2029{
Jens Axboeff856ba2006-01-09 16:02:34 +01002030 int i;
2031
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 kblockd_workqueue = create_workqueue("kblockd");
2033 if (!kblockd_workqueue)
2034 panic("Failed to create kblockd\n");
2035
2036 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002037 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
Jens Axboe8324aa92008-01-29 14:51:59 +01002039 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002040 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002042 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01002043 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
2044
Carlos R. Mafra962cf362008-05-15 11:15:37 -03002045 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002046 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01002047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return 0;
2049}
2050