blob: 2a438a93f7233d8bc2033956f23e3f436f0e1f44 [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>
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{
57 int rw = rq_data_dir(rq);
58
59 if (!blk_fs_request(rq) || !rq->rq_disk)
60 return;
61
62 if (!new_io) {
Jerome Marchand6f2576a2008-02-08 11:04:35 +010063 __all_stat_inc(rq->rq_disk, merges[rw], rq->sector);
Jens Axboe26b82562008-01-29 13:54:41 +010064 } else {
Jerome Marchand6f2576a2008-02-08 11:04:35 +010065 struct hd_struct *part = get_part(rq->rq_disk, rq->sector);
Jens Axboe26b82562008-01-29 13:54:41 +010066 disk_round_stats(rq->rq_disk);
67 rq->rq_disk->in_flight++;
Jerome Marchand6f2576a2008-02-08 11:04:35 +010068 if (part) {
69 part_round_stats(part);
70 part->in_flight++;
71 }
Jens Axboe26b82562008-01-29 13:54:41 +010072 }
73}
74
Jens Axboe8324aa92008-01-29 14:51:59 +010075void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
77 int nr;
78
79 nr = q->nr_requests - (q->nr_requests / 8) + 1;
80 if (nr > q->nr_requests)
81 nr = q->nr_requests;
82 q->nr_congestion_on = nr;
83
84 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
85 if (nr < 1)
86 nr = 1;
87 q->nr_congestion_off = nr;
88}
89
Linus Torvalds1da177e2005-04-16 15:20:36 -070090/**
91 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
92 * @bdev: device
93 *
94 * Locates the passed device's request queue and returns the address of its
95 * backing_dev_info
96 *
97 * Will return NULL if the request queue cannot be located.
98 */
99struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
100{
101 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200102 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104 if (q)
105 ret = &q->backing_dev_info;
106 return ret;
107}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108EXPORT_SYMBOL(blk_get_backing_dev_info);
109
Jens Axboe63a71382008-02-08 12:41:03 +0100110/*
111 * We can't just memset() the structure, since the allocation path
112 * already stored some information in the request.
113 */
Jens Axboe86db1e22008-01-29 14:53:40 +0100114void rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115{
116 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100117 INIT_LIST_HEAD(&rq->donelist);
Jens Axboe63a71382008-02-08 12:41:03 +0100118 rq->q = q;
119 rq->sector = rq->hard_sector = (sector_t) -1;
120 rq->nr_sectors = rq->hard_nr_sectors = 0;
121 rq->current_nr_sectors = rq->hard_cur_sectors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 rq->bio = rq->biotail = NULL;
Jens Axboe2e662b62006-07-13 11:55:04 +0200123 INIT_HLIST_NODE(&rq->hash);
124 RB_CLEAR_NODE(&rq->rb_node);
Jens Axboe63a71382008-02-08 12:41:03 +0100125 rq->rq_disk = NULL;
Mike Christie df46b9a2005-06-20 14:04:44 +0200126 rq->nr_phys_segments = 0;
Jens Axboe63a71382008-02-08 12:41:03 +0100127 rq->nr_hw_segments = 0;
128 rq->ioprio = 0;
129 rq->special = NULL;
130 rq->buffer = NULL;
131 rq->tag = -1;
132 rq->errors = 0;
133 rq->ref_count = 1;
134 rq->cmd_len = 0;
135 memset(rq->cmd, 0, sizeof(rq->cmd));
136 rq->data_len = 0;
FUJITA Tomonori7a85f882008-03-04 11:17:11 +0100137 rq->extra_len = 0;
Jens Axboe63a71382008-02-08 12:41:03 +0100138 rq->sense_len = 0;
139 rq->data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 rq->sense = NULL;
141 rq->end_io = NULL;
142 rq->end_io_data = NULL;
FUJITA Tomonoriabae1fd2007-07-16 08:52:14 +0200143 rq->next_rq = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144}
145
NeilBrown5bb23a62007-09-27 12:46:13 +0200146static void req_bio_endio(struct request *rq, struct bio *bio,
147 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100148{
Jens Axboe165125e2007-07-24 09:28:11 +0200149 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100150
NeilBrown5bb23a62007-09-27 12:46:13 +0200151 if (&q->bar_rq != rq) {
152 if (error)
153 clear_bit(BIO_UPTODATE, &bio->bi_flags);
154 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
155 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100156
NeilBrown5bb23a62007-09-27 12:46:13 +0200157 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100158 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
NeilBrown5bb23a62007-09-27 12:46:13 +0200159 __FUNCTION__, nbytes, bio->bi_size);
160 nbytes = bio->bi_size;
161 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100162
NeilBrown5bb23a62007-09-27 12:46:13 +0200163 bio->bi_size -= nbytes;
164 bio->bi_sector += (nbytes >> 9);
165 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200166 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200167 } else {
168
169 /*
170 * Okay, this is the barrier request in progress, just
171 * record the error;
172 */
173 if (error && !q->orderr)
174 q->orderr = error;
175 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100176}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178void blk_dump_rq_flags(struct request *rq, char *msg)
179{
180 int bit;
181
Jens Axboe6728cb02008-01-31 13:03:55 +0100182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
184 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Jens Axboe6728cb02008-01-31 13:03:55 +0100186 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
187 (unsigned long long)rq->sector,
188 rq->nr_sectors,
189 rq->current_nr_sectors);
190 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
191 rq->bio, rq->biotail,
192 rq->buffer, rq->data,
193 rq->data_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Jens Axboe4aff5e22006-08-10 08:44:47 +0200195 if (blk_pc_request(rq)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100196 printk(KERN_INFO " cdb: ");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 for (bit = 0; bit < sizeof(rq->cmd); bit++)
198 printk("%02x ", rq->cmd[bit]);
199 printk("\n");
200 }
201}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202EXPORT_SYMBOL(blk_dump_rq_flags);
203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204/*
205 * "plug" the device if there are no outstanding requests: this will
206 * force the transfer to start only after we have put all the requests
207 * on the list.
208 *
209 * This is called with interrupts off and no requests on the queue and
210 * with the queue lock held.
211 */
Jens Axboe165125e2007-07-24 09:28:11 +0200212void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
214 WARN_ON(!irqs_disabled());
215
216 /*
217 * don't plug a stopped queue, it must be paired with blk_start_queue()
218 * which will restart the queueing
219 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200220 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 return;
222
Jens Axboe2056a782006-03-23 20:00:26 +0100223 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +0100225 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228EXPORT_SYMBOL(blk_plug_device);
229
230/*
231 * remove the queue from the plugged list, if present. called with
232 * queue lock held and interrupts disabled.
233 */
Jens Axboe165125e2007-07-24 09:28:11 +0200234int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
236 WARN_ON(!irqs_disabled());
237
238 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
239 return 0;
240
241 del_timer(&q->unplug_timer);
242 return 1;
243}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244EXPORT_SYMBOL(blk_remove_plug);
245
246/*
247 * remove the plug and let it rip..
248 */
Jens Axboe165125e2007-07-24 09:28:11 +0200249void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200251 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 return;
253
254 if (!blk_remove_plug(q))
255 return;
256
Jens Axboe22e2c502005-06-27 10:55:12 +0200257 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
259EXPORT_SYMBOL(__generic_unplug_device);
260
261/**
262 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200263 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 *
265 * Description:
266 * Linux uses plugging to build bigger requests queues before letting
267 * the device have at them. If a queue is plugged, the I/O scheduler
268 * is still adding and merging requests on the queue. Once the queue
269 * gets unplugged, the request_fn defined for the queue is invoked and
270 * transfers started.
271 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200272void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
274 spin_lock_irq(q->queue_lock);
275 __generic_unplug_device(q);
276 spin_unlock_irq(q->queue_lock);
277}
278EXPORT_SYMBOL(generic_unplug_device);
279
280static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
281 struct page *page)
282{
Jens Axboe165125e2007-07-24 09:28:11 +0200283 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500285 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286}
287
Jens Axboe86db1e22008-01-29 14:53:40 +0100288void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289{
Jens Axboe165125e2007-07-24 09:28:11 +0200290 struct request_queue *q =
291 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
Jens Axboe2056a782006-03-23 20:00:26 +0100293 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
294 q->rq.count[READ] + q->rq.count[WRITE]);
295
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 q->unplug_fn(q);
297}
298
Jens Axboe86db1e22008-01-29 14:53:40 +0100299void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300{
Jens Axboe165125e2007-07-24 09:28:11 +0200301 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Jens Axboe2056a782006-03-23 20:00:26 +0100303 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
304 q->rq.count[READ] + q->rq.count[WRITE]);
305
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 kblockd_schedule_work(&q->unplug_work);
307}
308
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500309void blk_unplug(struct request_queue *q)
310{
311 /*
312 * devices don't necessarily have an ->unplug_fn defined
313 */
314 if (q->unplug_fn) {
315 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
316 q->rq.count[READ] + q->rq.count[WRITE]);
317
318 q->unplug_fn(q);
319 }
320}
321EXPORT_SYMBOL(blk_unplug);
322
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323/**
324 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200325 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 *
327 * Description:
328 * blk_start_queue() will clear the stop flag on the queue, and call
329 * the request_fn for the queue if it was in a stopped state when
330 * entered. Also see blk_stop_queue(). Queue lock must be held.
331 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200332void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200334 WARN_ON(!irqs_disabled());
335
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
337
338 /*
339 * one level of recursion is ok and is much faster than kicking
340 * the unplug handling
341 */
342 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
343 q->request_fn(q);
344 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
345 } else {
346 blk_plug_device(q);
347 kblockd_schedule_work(&q->unplug_work);
348 }
349}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350EXPORT_SYMBOL(blk_start_queue);
351
352/**
353 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200354 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 *
356 * Description:
357 * The Linux block layer assumes that a block driver will consume all
358 * entries on the request queue when the request_fn strategy is called.
359 * Often this will not happen, because of hardware limitations (queue
360 * depth settings). If a device driver gets a 'queue full' response,
361 * or if it simply chooses not to queue more I/O at one point, it can
362 * call this function to prevent the request_fn from being called until
363 * the driver has signalled it's ready to go again. This happens by calling
364 * blk_start_queue() to restart queue operations. Queue lock must be held.
365 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200366void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367{
368 blk_remove_plug(q);
369 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
370}
371EXPORT_SYMBOL(blk_stop_queue);
372
373/**
374 * blk_sync_queue - cancel any pending callbacks on a queue
375 * @q: the queue
376 *
377 * Description:
378 * The block layer may perform asynchronous callback activity
379 * on a queue, such as calling the unplug function after a timeout.
380 * A block device may call blk_sync_queue to ensure that any
381 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200382 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 * that its ->make_request_fn will not re-add plugging prior to calling
384 * this function.
385 *
386 */
387void blk_sync_queue(struct request_queue *q)
388{
389 del_timer_sync(&q->unplug_timer);
Oleg Nesterovabbeb882007-10-23 15:08:19 +0200390 kblockd_flush_work(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392EXPORT_SYMBOL(blk_sync_queue);
393
394/**
395 * blk_run_queue - run a single device queue
396 * @q: The queue to run
397 */
398void blk_run_queue(struct request_queue *q)
399{
400 unsigned long flags;
401
402 spin_lock_irqsave(q->queue_lock, flags);
403 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200404
405 /*
406 * Only recurse once to avoid overrunning the stack, let the unplug
407 * handling reinvoke the handler shortly if we already got there.
408 */
409 if (!elv_queue_empty(q)) {
410 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
411 q->request_fn(q);
412 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
413 } else {
414 blk_plug_device(q);
415 kblockd_schedule_work(&q->unplug_work);
416 }
417 }
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 spin_unlock_irqrestore(q->queue_lock, flags);
420}
421EXPORT_SYMBOL(blk_run_queue);
422
Jens Axboe165125e2007-07-24 09:28:11 +0200423void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500424{
425 kobject_put(&q->kobj);
426}
Al Viro483f4af2006-03-18 18:34:37 -0500427
Jens Axboe6728cb02008-01-31 13:03:55 +0100428void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500429{
430 mutex_lock(&q->sysfs_lock);
431 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
432 mutex_unlock(&q->sysfs_lock);
433
434 if (q->elevator)
435 elevator_exit(q->elevator);
436
437 blk_put_queue(q);
438}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439EXPORT_SYMBOL(blk_cleanup_queue);
440
Jens Axboe165125e2007-07-24 09:28:11 +0200441static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442{
443 struct request_list *rl = &q->rq;
444
445 rl->count[READ] = rl->count[WRITE] = 0;
446 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200447 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 init_waitqueue_head(&rl->wait[READ]);
449 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Christoph Lameter19460892005-06-23 00:08:19 -0700451 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
452 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454 if (!rl->rq_pool)
455 return -ENOMEM;
456
457 return 0;
458}
459
Jens Axboe165125e2007-07-24 09:28:11 +0200460struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
Christoph Lameter19460892005-06-23 00:08:19 -0700462 return blk_alloc_queue_node(gfp_mask, -1);
463}
464EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Jens Axboe165125e2007-07-24 09:28:11 +0200466struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700467{
Jens Axboe165125e2007-07-24 09:28:11 +0200468 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700469 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700470
Jens Axboe8324aa92008-01-29 14:51:59 +0100471 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700472 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 if (!q)
474 return NULL;
475
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700476 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
477 q->backing_dev_info.unplug_io_data = q;
478 err = bdi_init(&q->backing_dev_info);
479 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100480 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700481 return NULL;
482 }
483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500485
Jens Axboe8324aa92008-01-29 14:51:59 +0100486 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
Al Viro483f4af2006-03-18 18:34:37 -0500488 mutex_init(&q->sysfs_lock);
489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 return q;
491}
Christoph Lameter19460892005-06-23 00:08:19 -0700492EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494/**
495 * blk_init_queue - prepare a request queue for use with a block device
496 * @rfn: The function to be called to process requests that have been
497 * placed on the queue.
498 * @lock: Request queue spin lock
499 *
500 * Description:
501 * If a block device wishes to use the standard request handling procedures,
502 * which sorts requests and coalesces adjacent requests, then it must
503 * call blk_init_queue(). The function @rfn will be called when there
504 * are requests on the queue that need to be processed. If the device
505 * supports plugging, then @rfn may not be called immediately when requests
506 * are available on the queue, but may be called at some time later instead.
507 * Plugged queues are generally unplugged when a buffer belonging to one
508 * of the requests on the queue is needed, or due to memory pressure.
509 *
510 * @rfn is not required, or even expected, to remove all requests off the
511 * queue, but only as many as it can handle at a time. If it does leave
512 * requests on the queue, it is responsible for arranging that the requests
513 * get dealt with eventually.
514 *
515 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200516 * request queue; this lock will be taken also from interrupt context, so irq
517 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 *
519 * Function returns a pointer to the initialized request queue, or NULL if
520 * it didn't succeed.
521 *
522 * Note:
523 * blk_init_queue() must be paired with a blk_cleanup_queue() call
524 * when the block device is deactivated (such as at module unload).
525 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700526
Jens Axboe165125e2007-07-24 09:28:11 +0200527struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528{
Christoph Lameter19460892005-06-23 00:08:19 -0700529 return blk_init_queue_node(rfn, lock, -1);
530}
531EXPORT_SYMBOL(blk_init_queue);
532
Jens Axboe165125e2007-07-24 09:28:11 +0200533struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700534blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
535{
Jens Axboe165125e2007-07-24 09:28:11 +0200536 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
538 if (!q)
539 return NULL;
540
Christoph Lameter19460892005-06-23 00:08:19 -0700541 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -0500542 if (blk_init_free_list(q)) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100543 kmem_cache_free(blk_requestq_cachep, q);
Al Viro8669aaf2006-03-18 13:50:00 -0500544 return NULL;
545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
152587d2005-04-12 16:22:06 -0500547 /*
548 * if caller didn't supply a lock, they get per-queue locking with
549 * our embedded lock
550 */
551 if (!lock) {
552 spin_lock_init(&q->__queue_lock);
553 lock = &q->__queue_lock;
554 }
555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 q->prep_rq_fn = NULL;
558 q->unplug_fn = generic_unplug_device;
559 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
560 q->queue_lock = lock;
561
562 blk_queue_segment_boundary(q, 0xffffffff);
563
564 blk_queue_make_request(q, __make_request);
565 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
566
567 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
568 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
569
Alan Stern44ec9542007-02-20 11:01:57 -0500570 q->sg_reserved_size = INT_MAX;
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 /*
573 * all done
574 */
575 if (!elevator_init(q, NULL)) {
576 blk_queue_congestion_threshold(q);
577 return q;
578 }
579
Al Viro8669aaf2006-03-18 13:50:00 -0500580 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 return NULL;
582}
Christoph Lameter19460892005-06-23 00:08:19 -0700583EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Jens Axboe165125e2007-07-24 09:28:11 +0200585int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700587 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500588 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 return 0;
590 }
591
592 return 1;
593}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Jens Axboe165125e2007-07-24 09:28:11 +0200595static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200597 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200598 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 mempool_free(rq, q->rq.rq_pool);
600}
601
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200602static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +0200603blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604{
605 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
606
607 if (!rq)
608 return NULL;
609
610 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +0200611 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 * see bio.h and blkdev.h
613 */
Jens Axboe49171e52006-08-10 08:59:11 +0200614 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
Tejun Heocb98fc82005-10-28 08:29:39 +0200616 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200617 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200618 mempool_free(rq, q->rq.rq_pool);
619 return NULL;
620 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200621 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200622 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
Tejun Heocb98fc82005-10-28 08:29:39 +0200624 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
626
627/*
628 * ioc_batching returns true if the ioc is a valid batching request and
629 * should be given priority access to a request.
630 */
Jens Axboe165125e2007-07-24 09:28:11 +0200631static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
633 if (!ioc)
634 return 0;
635
636 /*
637 * Make sure the process is able to allocate at least 1 request
638 * even if the batch times out, otherwise we could theoretically
639 * lose wakeups.
640 */
641 return ioc->nr_batch_requests == q->nr_batching ||
642 (ioc->nr_batch_requests > 0
643 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
644}
645
646/*
647 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
648 * will cause the process to be a "batcher" on all queues in the system. This
649 * is the behaviour we want though - once it gets a wakeup it should be given
650 * a nice run.
651 */
Jens Axboe165125e2007-07-24 09:28:11 +0200652static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653{
654 if (!ioc || ioc_batching(q, ioc))
655 return;
656
657 ioc->nr_batch_requests = q->nr_batching;
658 ioc->last_waited = jiffies;
659}
660
Jens Axboe165125e2007-07-24 09:28:11 +0200661static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
663 struct request_list *rl = &q->rq;
664
665 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -0700666 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
668 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 if (waitqueue_active(&rl->wait[rw]))
670 wake_up(&rl->wait[rw]);
671
672 blk_clear_queue_full(q, rw);
673 }
674}
675
676/*
677 * A request has just been released. Account for it, update the full and
678 * congestion status, wake up any waiters. Called under q->queue_lock.
679 */
Jens Axboe165125e2007-07-24 09:28:11 +0200680static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
682 struct request_list *rl = &q->rq;
683
684 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200685 if (priv)
686 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
688 __freed_request(q, rw);
689
690 if (unlikely(rl->starved[rw ^ 1]))
691 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692}
693
694#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
695/*
Nick Piggind6344532005-06-28 20:45:14 -0700696 * Get a free request, queue_lock must be held.
697 * Returns NULL on failure, with queue_lock held.
698 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 */
Jens Axboe165125e2007-07-24 09:28:11 +0200700static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100701 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
703 struct request *rq = NULL;
704 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100705 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +0100706 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100707 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
Jens Axboe7749a8d2006-12-13 13:02:26 +0100709 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100710 if (may_queue == ELV_MQUEUE_NO)
711 goto rq_starved;
712
713 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
714 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200715 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100716 /*
717 * The queue will fill after this allocation, so set
718 * it as full, and mark this process as "batching".
719 * This process will be allowed to complete a batch of
720 * requests, others will be blocked.
721 */
722 if (!blk_queue_full(q, rw)) {
723 ioc_set_batching(q, ioc);
724 blk_set_queue_full(q, rw);
725 } else {
726 if (may_queue != ELV_MQUEUE_MUST
727 && !ioc_batching(q, ioc)) {
728 /*
729 * The queue is full and the allocating
730 * process is not a "batcher", and not
731 * exempted by the IO scheduler
732 */
733 goto out;
734 }
735 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 }
Thomas Maier79e2de42006-10-19 23:28:15 -0700737 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 }
739
Jens Axboe082cf692005-06-28 16:35:11 +0200740 /*
741 * Only allow batching queuers to allocate up to 50% over the defined
742 * limit of requests, otherwise we could have thousands of requests
743 * allocated with any setting of ->nr_requests
744 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100745 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200746 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100747
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 rl->count[rw]++;
749 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200750
Jens Axboe64521d12005-10-28 08:30:39 +0200751 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200752 if (priv)
753 rl->elvpriv++;
754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 spin_unlock_irq(q->queue_lock);
756
Jens Axboe7749a8d2006-12-13 13:02:26 +0100757 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100758 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 /*
760 * Allocation failed presumably due to memory. Undo anything
761 * we might have messed up.
762 *
763 * Allocating task should really be put onto the front of the
764 * wait queue, but this is pretty rare.
765 */
766 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +0200767 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
769 /*
770 * in the very unlikely event that allocation failed and no
771 * requests for this direction was pending, mark us starved
772 * so that freeing of a request in the other direction will
773 * notice us. another possible fix would be to split the
774 * rq mempool into READ and WRITE
775 */
776rq_starved:
777 if (unlikely(rl->count[rw] == 0))
778 rl->starved[rw] = 1;
779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 goto out;
781 }
782
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100783 /*
784 * ioc may be NULL here, and ioc_batching will be false. That's
785 * OK, if the queue is under the request limit then requests need
786 * not count toward the nr_batch_requests limit. There will always
787 * be some limit enforced by BLK_BATCH_TIME.
788 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 if (ioc_batching(q, ioc))
790 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100791
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 rq_init(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100793
794 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 return rq;
797}
798
799/*
800 * No available requests for this queue, unplug the device and wait for some
801 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700802 *
803 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 */
Jens Axboe165125e2007-07-24 09:28:11 +0200805static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200806 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
Jens Axboe7749a8d2006-12-13 13:02:26 +0100808 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 struct request *rq;
810
Jens Axboe7749a8d2006-12-13 13:02:26 +0100811 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700812 while (!rq) {
813 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 struct request_list *rl = &q->rq;
815
816 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
817 TASK_UNINTERRUPTIBLE);
818
Jens Axboe7749a8d2006-12-13 13:02:26 +0100819 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821 if (!rq) {
822 struct io_context *ioc;
823
Jens Axboe2056a782006-03-23 20:00:26 +0100824 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
825
Nick Piggind6344532005-06-28 20:45:14 -0700826 __generic_unplug_device(q);
827 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 io_schedule();
829
830 /*
831 * After sleeping, we become a "batching" process and
832 * will be able to allocate at least one request, and
833 * up to a big batch of them for a small period time.
834 * See ioc_batching, ioc_set_batching
835 */
Jens Axboeb5deef92006-07-19 23:39:40 +0200836 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -0700838
839 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 }
841 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -0700842 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
844 return rq;
845}
846
Jens Axboe165125e2007-07-24 09:28:11 +0200847struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848{
849 struct request *rq;
850
851 BUG_ON(rw != READ && rw != WRITE);
852
Nick Piggind6344532005-06-28 20:45:14 -0700853 spin_lock_irq(q->queue_lock);
854 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200855 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700856 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200857 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700858 if (!rq)
859 spin_unlock_irq(q->queue_lock);
860 }
861 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863 return rq;
864}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865EXPORT_SYMBOL(blk_get_request);
866
867/**
Jens Axboedc72ef42006-07-20 14:54:05 +0200868 * blk_start_queueing - initiate dispatch of requests to device
869 * @q: request queue to kick into gear
870 *
871 * This is basically a helper to remove the need to know whether a queue
872 * is plugged or not if someone just wants to initiate dispatch of requests
873 * for this queue.
874 *
875 * The queue lock must be held with interrupts disabled.
876 */
Jens Axboe165125e2007-07-24 09:28:11 +0200877void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +0200878{
879 if (!blk_queue_plugged(q))
880 q->request_fn(q);
881 else
882 __generic_unplug_device(q);
883}
884EXPORT_SYMBOL(blk_start_queueing);
885
886/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 * blk_requeue_request - put a request back on queue
888 * @q: request queue where request should be inserted
889 * @rq: request to be inserted
890 *
891 * Description:
892 * Drivers often keep queueing requests until the hardware cannot accept
893 * more, when that condition happens we need to put the request back
894 * on the queue. Must be called with queue lock held.
895 */
Jens Axboe165125e2007-07-24 09:28:11 +0200896void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897{
Jens Axboe2056a782006-03-23 20:00:26 +0100898 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
899
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 if (blk_rq_tagged(rq))
901 blk_queue_end_tag(q, rq);
902
903 elv_requeue_request(q, rq);
904}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905EXPORT_SYMBOL(blk_requeue_request);
906
907/**
908 * blk_insert_request - insert a special request in to a request queue
909 * @q: request queue where request should be inserted
910 * @rq: request to be inserted
911 * @at_head: insert request at head or tail of queue
912 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 *
914 * Description:
915 * Many block devices need to execute commands asynchronously, so they don't
916 * block the whole kernel from preemption during request execution. This is
917 * accomplished normally by inserting aritficial requests tagged as
918 * REQ_SPECIAL in to the corresponding request queue, and letting them be
919 * scheduled for actual execution by the request queue.
920 *
921 * We have the option of inserting the head or the tail of the queue.
922 * Typically we use the tail for new ioctls and so forth. We use the head
923 * of the queue for things like a QUEUE_FULL message from a device, or a
924 * host that is unable to accept a particular command.
925 */
Jens Axboe165125e2007-07-24 09:28:11 +0200926void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500927 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
Tejun Heo 867d1192005-04-24 02:06:05 -0500929 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 unsigned long flags;
931
932 /*
933 * tell I/O scheduler that this isn't a regular read/write (ie it
934 * must not attempt merges on this) and that it acts as a soft
935 * barrier
936 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200937 rq->cmd_type = REQ_TYPE_SPECIAL;
938 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
940 rq->special = data;
941
942 spin_lock_irqsave(q->queue_lock, flags);
943
944 /*
945 * If command is tagged, release the tag
946 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500947 if (blk_rq_tagged(rq))
948 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200950 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -0500951 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +0200952 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 spin_unlock_irqrestore(q->queue_lock, flags);
954}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955EXPORT_SYMBOL(blk_insert_request);
956
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957/*
958 * add-request adds a request to the linked list.
959 * queue lock is held and interrupts disabled, as we muck with the
960 * request queue list.
961 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100962static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963{
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200964 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 /*
967 * elevator indicated where it wants this request to be
968 * inserted at elevator_merge time
969 */
970 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
971}
Jens Axboe6728cb02008-01-31 13:03:55 +0100972
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973/*
974 * disk_round_stats() - Round off the performance stats on a struct
975 * disk_stats.
976 *
977 * The average IO queue length and utilisation statistics are maintained
978 * by observing the current state of the queue length and the amount of
979 * time it has been in this state for.
980 *
981 * Normally, that accounting is done on IO completion, but that can result
982 * in more than a second's worth of IO being accounted for within any one
983 * second, leading to >100% utilisation. To deal with that, we call this
984 * function to do a round-off before returning the results when reading
985 * /proc/diskstats. This accounts immediately for all queue usage up to
986 * the current jiffies and restarts the counters again.
987 */
988void disk_round_stats(struct gendisk *disk)
989{
990 unsigned long now = jiffies;
991
Chen, Kenneth Wb2982642005-10-13 21:49:29 +0200992 if (now == disk->stamp)
993 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Chen, Kenneth W20e5c812005-10-13 21:48:42 +0200995 if (disk->in_flight) {
996 __disk_stat_add(disk, time_in_queue,
997 disk->in_flight * (now - disk->stamp));
998 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001}
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08001002EXPORT_SYMBOL_GPL(disk_round_stats);
1003
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001004void part_round_stats(struct hd_struct *part)
1005{
1006 unsigned long now = jiffies;
1007
1008 if (now == part->stamp)
1009 return;
1010
1011 if (part->in_flight) {
1012 __part_stat_add(part, time_in_queue,
1013 part->in_flight * (now - part->stamp));
1014 __part_stat_add(part, io_ticks, (now - part->stamp));
1015 }
1016 part->stamp = now;
1017}
1018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019/*
1020 * queue lock must be held
1021 */
Jens Axboe165125e2007-07-24 09:28:11 +02001022void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 if (unlikely(!q))
1025 return;
1026 if (unlikely(--req->ref_count))
1027 return;
1028
Tejun Heo8922e162005-10-20 16:23:44 +02001029 elv_completed_request(q, req);
1030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 /*
1032 * Request may not have originated from ll_rw_blk. if not,
1033 * it didn't come out of our reserved rq pools
1034 */
Jens Axboe49171e52006-08-10 08:59:11 +02001035 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001037 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001040 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02001043 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 }
1045}
Mike Christie6e39b692005-11-11 05:30:24 -06001046EXPORT_SYMBOL_GPL(__blk_put_request);
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048void blk_put_request(struct request *req)
1049{
Tejun Heo8922e162005-10-20 16:23:44 +02001050 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001051 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
Tejun Heo8922e162005-10-20 16:23:44 +02001053 /*
1054 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
1055 * following if (q) test.
1056 */
1057 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 spin_lock_irqsave(q->queue_lock, flags);
1059 __blk_put_request(q, req);
1060 spin_unlock_irqrestore(q->queue_lock, flags);
1061 }
1062}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063EXPORT_SYMBOL(blk_put_request);
1064
Jens Axboe86db1e22008-01-29 14:53:40 +01001065void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001066{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001067 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001068
1069 /*
1070 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1071 */
1072 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001073 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01001074
1075 /*
1076 * REQ_BARRIER implies no merging, but lets make it explicit
1077 */
1078 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001079 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01001080
Jens Axboeb31dc662006-06-13 08:26:10 +02001081 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001082 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02001083 if (bio_rw_meta(bio))
1084 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02001085
Tejun Heo52d9e672006-01-06 09:49:58 +01001086 req->errors = 0;
1087 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001088 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001089 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02001090 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001091}
1092
Jens Axboe165125e2007-07-24 09:28:11 +02001093static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094{
Nick Piggin450991b2005-06-28 20:45:13 -07001095 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02001096 int el_ret, nr_sectors, barrier, err;
1097 const unsigned short prio = bio_prio(bio);
1098 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01001099 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
1103 /*
1104 * low level driver can indicate that it wants pages above a
1105 * certain limit bounced to low memory (ie for highmem, or even
1106 * ISA dma in theory)
1107 */
1108 blk_queue_bounce(q, &bio);
1109
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01001111 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 err = -EOPNOTSUPP;
1113 goto end_io;
1114 }
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 spin_lock_irq(q->queue_lock);
1117
Nick Piggin450991b2005-06-28 20:45:13 -07001118 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 goto get_rq;
1120
1121 el_ret = elv_merge(q, &req, bio);
1122 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001123 case ELEVATOR_BACK_MERGE:
1124 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
Jens Axboe6728cb02008-01-31 13:03:55 +01001126 if (!ll_back_merge_fn(q, req, bio))
1127 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Jens Axboe6728cb02008-01-31 13:03:55 +01001129 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001130
Jens Axboe6728cb02008-01-31 13:03:55 +01001131 req->biotail->bi_next = bio;
1132 req->biotail = bio;
1133 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1134 req->ioprio = ioprio_best(req->ioprio, prio);
1135 drive_stat_acct(req, 0);
1136 if (!attempt_back_merge(q, req))
1137 elv_merged_request(q, req, el_ret);
1138 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
Jens Axboe6728cb02008-01-31 13:03:55 +01001140 case ELEVATOR_FRONT_MERGE:
1141 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Jens Axboe6728cb02008-01-31 13:03:55 +01001143 if (!ll_front_merge_fn(q, req, bio))
1144 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Jens Axboe6728cb02008-01-31 13:03:55 +01001146 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
Jens Axboe2056a782006-03-23 20:00:26 +01001147
Jens Axboe6728cb02008-01-31 13:03:55 +01001148 bio->bi_next = req->bio;
1149 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
Jens Axboe6728cb02008-01-31 13:03:55 +01001151 /*
1152 * may not be valid. if the low level driver said
1153 * it didn't need a bounce buffer then it better
1154 * not touch req->buffer either...
1155 */
1156 req->buffer = bio_data(bio);
1157 req->current_nr_sectors = bio_cur_sectors(bio);
1158 req->hard_cur_sectors = req->current_nr_sectors;
1159 req->sector = req->hard_sector = bio->bi_sector;
1160 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1161 req->ioprio = ioprio_best(req->ioprio, prio);
1162 drive_stat_acct(req, 0);
1163 if (!attempt_front_merge(q, req))
1164 elv_merged_request(q, req, el_ret);
1165 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
Jens Axboe6728cb02008-01-31 13:03:55 +01001167 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1168 default:
1169 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 }
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001173 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001174 * This sync check and mask will be re-done in init_request_from_bio(),
1175 * but we need to set it earlier to expose the sync flag to the
1176 * rq allocator and io schedulers.
1177 */
1178 rw_flags = bio_data_dir(bio);
1179 if (sync)
1180 rw_flags |= REQ_RW_SYNC;
1181
1182 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001183 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001184 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001185 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001186 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001187
Nick Piggin450991b2005-06-28 20:45:13 -07001188 /*
1189 * After dropping the lock and possibly sleeping here, our request
1190 * may now be mergeable after it had proven unmergeable (above).
1191 * We don't worry about that case for efficiency. It won't happen
1192 * often, and the elevators are able to handle it.
1193 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001194 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
Nick Piggin450991b2005-06-28 20:45:13 -07001196 spin_lock_irq(q->queue_lock);
1197 if (elv_queue_empty(q))
1198 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 add_request(q, req);
1200out:
Jens Axboe4a534f92005-04-16 15:25:40 -07001201 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 __generic_unplug_device(q);
1203
1204 spin_unlock_irq(q->queue_lock);
1205 return 0;
1206
1207end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02001208 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 return 0;
1210}
1211
1212/*
1213 * If bio->bi_dev is a partition, remap the location
1214 */
1215static inline void blk_partition_remap(struct bio *bio)
1216{
1217 struct block_device *bdev = bio->bi_bdev;
1218
Jens Axboebf2de6f2007-09-27 13:01:25 +02001219 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001221
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 bio->bi_sector += p->start_sect;
1223 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001224
1225 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
1226 bdev->bd_dev, bio->bi_sector,
1227 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 }
1229}
1230
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231static void handle_bad_sector(struct bio *bio)
1232{
1233 char b[BDEVNAME_SIZE];
1234
1235 printk(KERN_INFO "attempt to access beyond end of device\n");
1236 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1237 bdevname(bio->bi_bdev, b),
1238 bio->bi_rw,
1239 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1240 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1241
1242 set_bit(BIO_EOF, &bio->bi_flags);
1243}
1244
Akinobu Mitac17bb492006-12-08 02:39:46 -08001245#ifdef CONFIG_FAIL_MAKE_REQUEST
1246
1247static DECLARE_FAULT_ATTR(fail_make_request);
1248
1249static int __init setup_fail_make_request(char *str)
1250{
1251 return setup_fault_attr(&fail_make_request, str);
1252}
1253__setup("fail_make_request=", setup_fail_make_request);
1254
1255static int should_fail_request(struct bio *bio)
1256{
1257 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
1258 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
1259 return should_fail(&fail_make_request, bio->bi_size);
1260
1261 return 0;
1262}
1263
1264static int __init fail_make_request_debugfs(void)
1265{
1266 return init_fault_attr_dentries(&fail_make_request,
1267 "fail_make_request");
1268}
1269
1270late_initcall(fail_make_request_debugfs);
1271
1272#else /* CONFIG_FAIL_MAKE_REQUEST */
1273
1274static inline int should_fail_request(struct bio *bio)
1275{
1276 return 0;
1277}
1278
1279#endif /* CONFIG_FAIL_MAKE_REQUEST */
1280
Jens Axboec07e2b42007-07-18 13:27:58 +02001281/*
1282 * Check whether this bio extends beyond the end of the device.
1283 */
1284static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1285{
1286 sector_t maxsector;
1287
1288 if (!nr_sectors)
1289 return 0;
1290
1291 /* Test device or partition size, when known. */
1292 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1293 if (maxsector) {
1294 sector_t sector = bio->bi_sector;
1295
1296 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1297 /*
1298 * This may well happen - the kernel calls bread()
1299 * without checking the size of the device, e.g., when
1300 * mounting a device.
1301 */
1302 handle_bad_sector(bio);
1303 return 1;
1304 }
1305 }
1306
1307 return 0;
1308}
1309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310/**
1311 * generic_make_request: hand a buffer to its device driver for I/O
1312 * @bio: The bio describing the location in memory and on the device.
1313 *
1314 * generic_make_request() is used to make I/O requests of block
1315 * devices. It is passed a &struct bio, which describes the I/O that needs
1316 * to be done.
1317 *
1318 * generic_make_request() does not return any status. The
1319 * success/failure status of the request, along with notification of
1320 * completion, is delivered asynchronously through the bio->bi_end_io
1321 * function described (one day) else where.
1322 *
1323 * The caller of generic_make_request must make sure that bi_io_vec
1324 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1325 * set to describe the device address, and the
1326 * bi_end_io and optionally bi_private are set to describe how
1327 * completion notification should be signaled.
1328 *
1329 * generic_make_request and the drivers it calls may use bi_next if this
1330 * bio happens to be merged with someone else, and may change bi_dev and
1331 * bi_sector for remaps as it sees fit. So the values of these fields
1332 * should NOT be depended on after the call to generic_make_request.
1333 */
Neil Brownd89d8792007-05-01 09:53:42 +02001334static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335{
Jens Axboe165125e2007-07-24 09:28:11 +02001336 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001337 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001339 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001340 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
1342 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
Jens Axboec07e2b42007-07-18 13:27:58 +02001344 if (bio_check_eod(bio, nr_sectors))
1345 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347 /*
1348 * Resolve the mapping until finished. (drivers are
1349 * still free to implement/resolve their own stacking
1350 * by explicitly returning 0)
1351 *
1352 * NOTE: we don't repeat the blk_size check for each new device.
1353 * Stacking drivers are expected to know what they are doing.
1354 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001355 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001356 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 do {
1358 char b[BDEVNAME_SIZE];
1359
1360 q = bdev_get_queue(bio->bi_bdev);
1361 if (!q) {
1362 printk(KERN_ERR
1363 "generic_make_request: Trying to access "
1364 "nonexistent block-device %s (%Lu)\n",
1365 bdevname(bio->bi_bdev, b),
1366 (long long) bio->bi_sector);
1367end_io:
Jens Axboe51fd77b2007-11-02 08:49:08 +01001368 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 break;
1370 }
1371
Jens Axboe4fa253f2007-07-18 13:13:10 +02001372 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001373 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 bdevname(bio->bi_bdev, b),
1375 bio_sectors(bio),
1376 q->max_hw_sectors);
1377 goto end_io;
1378 }
1379
Nick Pigginfde6ad22005-06-23 00:08:53 -07001380 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 goto end_io;
1382
Akinobu Mitac17bb492006-12-08 02:39:46 -08001383 if (should_fail_request(bio))
1384 goto end_io;
1385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 /*
1387 * If this device has partitions, remap block n
1388 * of partition p to block n+start(p) of the disk.
1389 */
1390 blk_partition_remap(bio);
1391
NeilBrown5ddfe962006-10-30 22:07:21 -08001392 if (old_sector != -1)
Jens Axboe4fa253f2007-07-18 13:13:10 +02001393 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08001394 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001395
1396 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
1397
NeilBrown5ddfe962006-10-30 22:07:21 -08001398 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001399 old_dev = bio->bi_bdev->bd_dev;
1400
Jens Axboec07e2b42007-07-18 13:27:58 +02001401 if (bio_check_eod(bio, nr_sectors))
1402 goto end_io;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001403 if (bio_empty_barrier(bio) && !q->prepare_flush_fn) {
1404 err = -EOPNOTSUPP;
1405 goto end_io;
1406 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 ret = q->make_request_fn(q, bio);
1409 } while (ret);
1410}
1411
Neil Brownd89d8792007-05-01 09:53:42 +02001412/*
1413 * We only want one ->make_request_fn to be active at a time,
1414 * else stack usage with stacked devices could be a problem.
1415 * So use current->bio_{list,tail} to keep a list of requests
1416 * submited by a make_request_fn function.
1417 * current->bio_tail is also used as a flag to say if
1418 * generic_make_request is currently active in this task or not.
1419 * If it is NULL, then no make_request is active. If it is non-NULL,
1420 * then a make_request is active, and new requests should be added
1421 * at the tail
1422 */
1423void generic_make_request(struct bio *bio)
1424{
1425 if (current->bio_tail) {
1426 /* make_request is active */
1427 *(current->bio_tail) = bio;
1428 bio->bi_next = NULL;
1429 current->bio_tail = &bio->bi_next;
1430 return;
1431 }
1432 /* following loop may be a bit non-obvious, and so deserves some
1433 * explanation.
1434 * Before entering the loop, bio->bi_next is NULL (as all callers
1435 * ensure that) so we have a list with a single bio.
1436 * We pretend that we have just taken it off a longer list, so
1437 * we assign bio_list to the next (which is NULL) and bio_tail
1438 * to &bio_list, thus initialising the bio_list of new bios to be
1439 * added. __generic_make_request may indeed add some more bios
1440 * through a recursive call to generic_make_request. If it
1441 * did, we find a non-NULL value in bio_list and re-enter the loop
1442 * from the top. In this case we really did just take the bio
1443 * of the top of the list (no pretending) and so fixup bio_list and
1444 * bio_tail or bi_next, and call into __generic_make_request again.
1445 *
1446 * The loop was structured like this to make only one call to
1447 * __generic_make_request (which is important as it is large and
1448 * inlined) and to keep the structure simple.
1449 */
1450 BUG_ON(bio->bi_next);
1451 do {
1452 current->bio_list = bio->bi_next;
1453 if (bio->bi_next == NULL)
1454 current->bio_tail = &current->bio_list;
1455 else
1456 bio->bi_next = NULL;
1457 __generic_make_request(bio);
1458 bio = current->bio_list;
1459 } while (bio);
1460 current->bio_tail = NULL; /* deactivate */
1461}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462EXPORT_SYMBOL(generic_make_request);
1463
1464/**
1465 * submit_bio: submit a bio to the block device layer for I/O
1466 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1467 * @bio: The &struct bio which describes the I/O
1468 *
1469 * submit_bio() is very similar in purpose to generic_make_request(), and
1470 * uses that function to do most of the work. Both are fairly rough
1471 * interfaces, @bio must be presetup and ready for I/O.
1472 *
1473 */
1474void submit_bio(int rw, struct bio *bio)
1475{
1476 int count = bio_sectors(bio);
1477
Jens Axboe22e2c502005-06-27 10:55:12 +02001478 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
Jens Axboebf2de6f2007-09-27 13:01:25 +02001480 /*
1481 * If it's a regular read/write or a barrier with data attached,
1482 * go through the normal accounting stuff before submission.
1483 */
1484 if (!bio_empty_barrier(bio)) {
1485
1486 BIO_BUG_ON(!bio->bi_size);
1487 BIO_BUG_ON(!bio->bi_io_vec);
1488
1489 if (rw & WRITE) {
1490 count_vm_events(PGPGOUT, count);
1491 } else {
1492 task_io_account_read(bio->bi_size);
1493 count_vm_events(PGPGIN, count);
1494 }
1495
1496 if (unlikely(block_dump)) {
1497 char b[BDEVNAME_SIZE];
1498 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001499 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001500 (rw & WRITE) ? "WRITE" : "READ",
1501 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001502 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 }
1505
1506 generic_make_request(bio);
1507}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508EXPORT_SYMBOL(submit_bio);
1509
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001510/**
1511 * __end_that_request_first - end I/O on a request
1512 * @req: the request being processed
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001513 * @error: 0 for success, < 0 for error
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001514 * @nr_bytes: number of bytes to complete
1515 *
1516 * Description:
1517 * Ends I/O on a number of bytes attached to @req, and sets it up
1518 * for the next range of segments (if any) in the cluster.
1519 *
1520 * Return:
1521 * 0 - we are done with this request, call end_that_request_last()
1522 * 1 - still buffers pending for this request
1523 **/
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001524static int __end_that_request_first(struct request *req, int error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 int nr_bytes)
1526{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001527 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 struct bio *bio;
1529
Jens Axboe2056a782006-03-23 20:00:26 +01001530 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
1531
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 * for a REQ_BLOCK_PC request, we want to carry any eventual
1534 * sense key with us all the way through
1535 */
1536 if (!blk_pc_request(req))
1537 req->errors = 0;
1538
Jens Axboe6728cb02008-01-31 13:03:55 +01001539 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1540 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 req->rq_disk ? req->rq_disk->disk_name : "?",
1542 (unsigned long long)req->sector);
1543 }
1544
Jens Axboed72d9042005-11-01 08:35:42 +01001545 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01001546 const int rw = rq_data_dir(req);
1547
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001548 all_stat_add(req->rq_disk, sectors[rw],
1549 nr_bytes >> 9, req->sector);
Jens Axboed72d9042005-11-01 08:35:42 +01001550 }
1551
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 total_bytes = bio_nbytes = 0;
1553 while ((bio = req->bio) != NULL) {
1554 int nbytes;
1555
Jens Axboebf2de6f2007-09-27 13:01:25 +02001556 /*
1557 * For an empty barrier request, the low level driver must
1558 * store a potential error location in ->sector. We pass
1559 * that back up in ->bi_sector.
1560 */
1561 if (blk_empty_barrier(req))
1562 bio->bi_sector = req->sector;
1563
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 if (nr_bytes >= bio->bi_size) {
1565 req->bio = bio->bi_next;
1566 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02001567 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 next_idx = 0;
1569 bio_nbytes = 0;
1570 } else {
1571 int idx = bio->bi_idx + next_idx;
1572
1573 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1574 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01001575 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1576 __FUNCTION__, bio->bi_idx,
1577 bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 break;
1579 }
1580
1581 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1582 BIO_BUG_ON(nbytes > bio->bi_size);
1583
1584 /*
1585 * not a complete bvec done
1586 */
1587 if (unlikely(nbytes > nr_bytes)) {
1588 bio_nbytes += nr_bytes;
1589 total_bytes += nr_bytes;
1590 break;
1591 }
1592
1593 /*
1594 * advance to the next vector
1595 */
1596 next_idx++;
1597 bio_nbytes += nbytes;
1598 }
1599
1600 total_bytes += nbytes;
1601 nr_bytes -= nbytes;
1602
Jens Axboe6728cb02008-01-31 13:03:55 +01001603 bio = req->bio;
1604 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 /*
1606 * end more in this run, or just return 'not-done'
1607 */
1608 if (unlikely(nr_bytes <= 0))
1609 break;
1610 }
1611 }
1612
1613 /*
1614 * completely done
1615 */
1616 if (!req->bio)
1617 return 0;
1618
1619 /*
1620 * if the request wasn't completed, update state
1621 */
1622 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02001623 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 bio->bi_idx += next_idx;
1625 bio_iovec(bio)->bv_offset += nr_bytes;
1626 bio_iovec(bio)->bv_len -= nr_bytes;
1627 }
1628
1629 blk_recalc_rq_sectors(req, total_bytes >> 9);
1630 blk_recalc_rq_segments(req);
1631 return 1;
1632}
1633
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634/*
Jens Axboeff856ba2006-01-09 16:02:34 +01001635 * splice the completion data to a local structure and hand off to
1636 * process_completion_queue() to complete the requests
1637 */
1638static void blk_done_softirq(struct softirq_action *h)
1639{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07001640 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01001641
1642 local_irq_disable();
1643 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07001644 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01001645 local_irq_enable();
1646
1647 while (!list_empty(&local_list)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001648 struct request *rq;
Jens Axboeff856ba2006-01-09 16:02:34 +01001649
Jens Axboe6728cb02008-01-31 13:03:55 +01001650 rq = list_entry(local_list.next, struct request, donelist);
Jens Axboeff856ba2006-01-09 16:02:34 +01001651 list_del_init(&rq->donelist);
1652 rq->q->softirq_done_fn(rq);
1653 }
1654}
1655
Jens Axboe6728cb02008-01-31 13:03:55 +01001656static int __cpuinit blk_cpu_notify(struct notifier_block *self,
1657 unsigned long action, void *hcpu)
Jens Axboeff856ba2006-01-09 16:02:34 +01001658{
1659 /*
1660 * If a CPU goes away, splice its entries to the current CPU
1661 * and trigger a run of the softirq
1662 */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001663 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Jens Axboeff856ba2006-01-09 16:02:34 +01001664 int cpu = (unsigned long) hcpu;
1665
1666 local_irq_disable();
1667 list_splice_init(&per_cpu(blk_cpu_done, cpu),
1668 &__get_cpu_var(blk_cpu_done));
1669 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1670 local_irq_enable();
1671 }
1672
1673 return NOTIFY_OK;
1674}
1675
1676
Satyam Sharmadb47d472007-08-23 09:29:40 +02001677static struct notifier_block blk_cpu_notifier __cpuinitdata = {
Jens Axboeff856ba2006-01-09 16:02:34 +01001678 .notifier_call = blk_cpu_notify,
1679};
1680
Jens Axboeff856ba2006-01-09 16:02:34 +01001681/**
1682 * blk_complete_request - end I/O on a request
1683 * @req: the request being processed
1684 *
1685 * Description:
1686 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02001687 * unless the driver actually implements this in its completion callback
Jens Axboe4fa253f2007-07-18 13:13:10 +02001688 * through requeueing. The actual completion happens out-of-order,
Jens Axboeff856ba2006-01-09 16:02:34 +01001689 * through a softirq handler. The user must have registered a completion
1690 * callback through blk_queue_softirq_done().
1691 **/
1692
1693void blk_complete_request(struct request *req)
1694{
1695 struct list_head *cpu_list;
1696 unsigned long flags;
1697
1698 BUG_ON(!req->q->softirq_done_fn);
Jens Axboe6728cb02008-01-31 13:03:55 +01001699
Jens Axboeff856ba2006-01-09 16:02:34 +01001700 local_irq_save(flags);
1701
1702 cpu_list = &__get_cpu_var(blk_cpu_done);
1703 list_add_tail(&req->donelist, cpu_list);
1704 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1705
1706 local_irq_restore(flags);
1707}
Jens Axboeff856ba2006-01-09 16:02:34 +01001708EXPORT_SYMBOL(blk_complete_request);
Jens Axboe6728cb02008-01-31 13:03:55 +01001709
Jens Axboeff856ba2006-01-09 16:02:34 +01001710/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 * queue lock must be held
1712 */
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001713static void end_that_request_last(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714{
1715 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01001716
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001717 if (blk_rq_tagged(req))
1718 blk_queue_end_tag(req->q, req);
1719
1720 if (blk_queued_rq(req))
1721 blkdev_dequeue_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722
1723 if (unlikely(laptop_mode) && blk_fs_request(req))
1724 laptop_io_completion();
1725
Jens Axboefd0ff8a2006-05-23 11:23:49 +02001726 /*
1727 * Account IO completion. bar_rq isn't accounted as a normal
1728 * IO on queueing nor completion. Accounting the containing
1729 * request is enough.
1730 */
1731 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01001733 const int rw = rq_data_dir(req);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001734 struct hd_struct *part = get_part(disk, req->sector);
Jens Axboea3623572005-11-01 09:26:16 +01001735
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001736 __all_stat_inc(disk, ios[rw], req->sector);
1737 __all_stat_add(disk, ticks[rw], duration, req->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 disk_round_stats(disk);
1739 disk->in_flight--;
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001740 if (part) {
1741 part_round_stats(part);
1742 part->in_flight--;
1743 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 }
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001745
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01001747 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001748 else {
1749 if (blk_bidi_rq(req))
1750 __blk_put_request(req->next_rq->q, req->next_rq);
1751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 }
1754}
1755
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756static inline void __end_request(struct request *rq, int uptodate,
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001757 unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001759 int error = 0;
1760
1761 if (uptodate <= 0)
1762 error = uptodate ? uptodate : -EIO;
1763
1764 __blk_end_request(rq, error, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765}
1766
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001767/**
1768 * blk_rq_bytes - Returns bytes left to complete in the entire request
Randy Dunlap5d87a052008-02-20 09:01:22 +01001769 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001770 **/
1771unsigned int blk_rq_bytes(struct request *rq)
Jens Axboea0cd1282007-09-21 10:41:07 +02001772{
1773 if (blk_fs_request(rq))
1774 return rq->hard_nr_sectors << 9;
1775
1776 return rq->data_len;
1777}
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001778EXPORT_SYMBOL_GPL(blk_rq_bytes);
1779
1780/**
1781 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
Randy Dunlap5d87a052008-02-20 09:01:22 +01001782 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001783 **/
1784unsigned int blk_rq_cur_bytes(struct request *rq)
1785{
1786 if (blk_fs_request(rq))
1787 return rq->current_nr_sectors << 9;
1788
1789 if (rq->bio)
1790 return rq->bio->bi_size;
1791
1792 return rq->data_len;
1793}
1794EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
Jens Axboea0cd1282007-09-21 10:41:07 +02001795
1796/**
1797 * end_queued_request - end all I/O on a queued request
1798 * @rq: the request being processed
1799 * @uptodate: error value or 0/1 uptodate flag
1800 *
1801 * Description:
1802 * Ends all I/O on a request, and removes it from the block layer queues.
1803 * Not suitable for normal IO completion, unless the driver still has
1804 * the request attached to the block layer.
1805 *
1806 **/
1807void end_queued_request(struct request *rq, int uptodate)
1808{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001809 __end_request(rq, uptodate, blk_rq_bytes(rq));
Jens Axboea0cd1282007-09-21 10:41:07 +02001810}
1811EXPORT_SYMBOL(end_queued_request);
1812
1813/**
1814 * end_dequeued_request - end all I/O on a dequeued request
1815 * @rq: the request being processed
1816 * @uptodate: error value or 0/1 uptodate flag
1817 *
1818 * Description:
1819 * Ends all I/O on a request. The request must already have been
1820 * dequeued using blkdev_dequeue_request(), as is normally the case
1821 * for most drivers.
1822 *
1823 **/
1824void end_dequeued_request(struct request *rq, int uptodate)
1825{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001826 __end_request(rq, uptodate, blk_rq_bytes(rq));
Jens Axboea0cd1282007-09-21 10:41:07 +02001827}
1828EXPORT_SYMBOL(end_dequeued_request);
1829
1830
1831/**
1832 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07001833 * @req: the request being processed
Jens Axboea0cd1282007-09-21 10:41:07 +02001834 * @uptodate: error value or 0/1 uptodate flag
1835 *
1836 * Description:
1837 * Ends I/O on the current segment of a request. If that is the only
1838 * remaining segment, the request is also completed and freed.
1839 *
1840 * This is a remnant of how older block drivers handled IO completions.
1841 * Modern drivers typically end IO on the full request in one go, unless
1842 * they have a residual value to account for. For that case this function
1843 * isn't really useful, unless the residual just happens to be the
1844 * full current segment. In other words, don't use this function in new
1845 * code. Either use end_request_completely(), or the
1846 * end_that_request_chunk() (along with end_that_request_last()) for
1847 * partial completions.
1848 *
1849 **/
1850void end_request(struct request *req, int uptodate)
1851{
Kiyoshi Ueda9e6e39f2007-12-11 17:41:54 -05001852 __end_request(req, uptodate, req->hard_cur_sectors << 9);
Jens Axboea0cd1282007-09-21 10:41:07 +02001853}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854EXPORT_SYMBOL(end_request);
1855
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001856/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001857 * blk_end_io - Generic end_io function to complete a request.
1858 * @rq: the request being processed
1859 * @error: 0 for success, < 0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001860 * @nr_bytes: number of bytes to complete @rq
1861 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001862 * @drv_callback: function called between completion of bios in the request
1863 * and completion of the request.
1864 * If the callback returns non 0, this helper returns without
1865 * completion of the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001866 *
1867 * Description:
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001868 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001869 * If @rq has leftover, sets it up for the next range of segments.
1870 *
1871 * Return:
1872 * 0 - we are done with this request
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001873 * 1 - this request is not freed yet, it still has pending buffers.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001874 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001875static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1876 unsigned int bidi_bytes,
1877 int (drv_callback)(struct request *))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001878{
1879 struct request_queue *q = rq->q;
1880 unsigned long flags = 0UL;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001881
1882 if (blk_fs_request(rq) || blk_pc_request(rq)) {
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001883 if (__end_that_request_first(rq, error, nr_bytes))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001884 return 1;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001885
1886 /* Bidi request must be completed as a whole */
1887 if (blk_bidi_rq(rq) &&
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001888 __end_that_request_first(rq->next_rq, error, bidi_bytes))
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001889 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001890 }
1891
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001892 /* Special feature for tricky drivers */
1893 if (drv_callback && drv_callback(rq))
1894 return 1;
1895
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001896 add_disk_randomness(rq->rq_disk);
1897
1898 spin_lock_irqsave(q->queue_lock, flags);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001899 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001900 spin_unlock_irqrestore(q->queue_lock, flags);
1901
1902 return 0;
1903}
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001904
1905/**
1906 * blk_end_request - Helper function for drivers to complete the request.
1907 * @rq: the request being processed
1908 * @error: 0 for success, < 0 for error
1909 * @nr_bytes: number of bytes to complete
1910 *
1911 * Description:
1912 * Ends I/O on a number of bytes attached to @rq.
1913 * If @rq has leftover, sets it up for the next range of segments.
1914 *
1915 * Return:
1916 * 0 - we are done with this request
1917 * 1 - still buffers pending for this request
1918 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001919int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001920{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001921 return blk_end_io(rq, error, nr_bytes, 0, NULL);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001922}
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001923EXPORT_SYMBOL_GPL(blk_end_request);
1924
1925/**
1926 * __blk_end_request - Helper function for drivers to complete the request.
1927 * @rq: the request being processed
1928 * @error: 0 for success, < 0 for error
1929 * @nr_bytes: number of bytes to complete
1930 *
1931 * Description:
1932 * Must be called with queue lock held unlike blk_end_request().
1933 *
1934 * Return:
1935 * 0 - we are done with this request
1936 * 1 - still buffers pending for this request
1937 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001938int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001939{
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001940 if (blk_fs_request(rq) || blk_pc_request(rq)) {
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001941 if (__end_that_request_first(rq, error, nr_bytes))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001942 return 1;
1943 }
1944
1945 add_disk_randomness(rq->rq_disk);
1946
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001947 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001948
1949 return 0;
1950}
1951EXPORT_SYMBOL_GPL(__blk_end_request);
1952
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001953/**
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001954 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1955 * @rq: the bidi request being processed
1956 * @error: 0 for success, < 0 for error
1957 * @nr_bytes: number of bytes to complete @rq
1958 * @bidi_bytes: number of bytes to complete @rq->next_rq
1959 *
1960 * Description:
1961 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1962 *
1963 * Return:
1964 * 0 - we are done with this request
1965 * 1 - still buffers pending for this request
1966 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001967int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
1968 unsigned int bidi_bytes)
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001969{
1970 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
1971}
1972EXPORT_SYMBOL_GPL(blk_end_bidi_request);
1973
1974/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001975 * blk_end_request_callback - Special helper function for tricky drivers
1976 * @rq: the request being processed
1977 * @error: 0 for success, < 0 for error
1978 * @nr_bytes: number of bytes to complete
1979 * @drv_callback: function called between completion of bios in the request
1980 * and completion of the request.
1981 * If the callback returns non 0, this helper returns without
1982 * completion of the request.
1983 *
1984 * Description:
1985 * Ends I/O on a number of bytes attached to @rq.
1986 * If @rq has leftover, sets it up for the next range of segments.
1987 *
1988 * This special helper function is used only for existing tricky drivers.
1989 * (e.g. cdrom_newpc_intr() of ide-cd)
1990 * This interface will be removed when such drivers are rewritten.
1991 * Don't use this interface in other places anymore.
1992 *
1993 * Return:
1994 * 0 - we are done with this request
1995 * 1 - this request is not freed yet.
1996 * this request still has pending buffers or
1997 * the driver doesn't want to finish this request yet.
1998 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001999int blk_end_request_callback(struct request *rq, int error,
2000 unsigned int nr_bytes,
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002001 int (drv_callback)(struct request *))
2002{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002003 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002004}
2005EXPORT_SYMBOL_GPL(blk_end_request_callback);
2006
Jens Axboe86db1e22008-01-29 14:53:40 +01002007void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2008 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002010 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
2011 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
2013 rq->nr_phys_segments = bio_phys_segments(q, bio);
2014 rq->nr_hw_segments = bio_hw_segments(q, bio);
2015 rq->current_nr_sectors = bio_cur_sectors(bio);
2016 rq->hard_cur_sectors = rq->current_nr_sectors;
2017 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
2018 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002019 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020
2021 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
NeilBrown66846572007-08-16 13:31:28 +02002023 if (bio->bi_bdev)
2024 rq->rq_disk = bio->bi_bdev->bd_disk;
2025}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
2027int kblockd_schedule_work(struct work_struct *work)
2028{
2029 return queue_work(kblockd_workqueue, work);
2030}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031EXPORT_SYMBOL(kblockd_schedule_work);
2032
Andrew Morton19a75d82007-05-09 02:33:56 -07002033void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07002035 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036}
Andrew Morton19a75d82007-05-09 02:33:56 -07002037EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
2039int __init blk_dev_init(void)
2040{
Jens Axboeff856ba2006-01-09 16:02:34 +01002041 int i;
2042
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 kblockd_workqueue = create_workqueue("kblockd");
2044 if (!kblockd_workqueue)
2045 panic("Failed to create kblockd\n");
2046
2047 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002048 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Jens Axboe8324aa92008-01-29 14:51:59 +01002050 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002051 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002053 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01002054 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
2055
2056 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002057 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01002058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 return 0;
2060}
2061