blob: 8b919940b2abdb37119cd9927b6f45dbec66aa31 [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>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
7 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
8 */
9
10/*
11 * This handles all read/write requests to block devices
12 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/kernel.h>
14#include <linux/module.h>
15#include <linux/backing-dev.h>
16#include <linux/bio.h>
17#include <linux/blkdev.h>
18#include <linux/highmem.h>
19#include <linux/mm.h>
20#include <linux/kernel_stat.h>
21#include <linux/string.h>
22#include <linux/init.h>
23#include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
24#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>
Jens Axboef5659132007-09-21 10:44:19 +020033#include <linux/scatterlist.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35/*
36 * for max sense size
37 */
38#include <scsi/scsi_cmnd.h>
39
David Howells65f27f32006-11-22 14:55:48 +000040static void blk_unplug_work(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041static void blk_unplug_timeout(unsigned long data);
Jerome Marchandb238b3d2007-10-23 15:05:46 +020042static void drive_stat_acct(struct request *rq, int new_io);
Tejun Heo52d9e672006-01-06 09:49:58 +010043static void init_request_from_bio(struct request *req, struct bio *bio);
Jens Axboe165125e2007-07-24 09:28:11 +020044static int __make_request(struct request_queue *q, struct bio *bio);
Jens Axboeb5deef92006-07-19 23:39:40 +020045static struct io_context *current_io_context(gfp_t gfp_flags, int node);
NeilBrown9dfa5282007-08-16 13:27:52 +020046static void blk_recalc_rq_segments(struct request *rq);
NeilBrown66846572007-08-16 13:31:28 +020047static void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
48 struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50/*
51 * For the allocated request tables
52 */
Christoph Lametere18b8902006-12-06 20:33:20 -080053static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55/*
56 * For queue allocation
57 */
Christoph Lametere18b8902006-12-06 20:33:20 -080058static struct kmem_cache *requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60/*
61 * For io context allocations
62 */
Christoph Lametere18b8902006-12-06 20:33:20 -080063static struct kmem_cache *iocontext_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/*
66 * Controlling structure to kblockd
67 */
Jens Axboeff856ba2006-01-09 16:02:34 +010068static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
70unsigned long blk_max_low_pfn, blk_max_pfn;
71
72EXPORT_SYMBOL(blk_max_low_pfn);
73EXPORT_SYMBOL(blk_max_pfn);
74
Jens Axboeff856ba2006-01-09 16:02:34 +010075static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/* Amount of time in which a process may batch requests */
78#define BLK_BATCH_TIME (HZ/50UL)
79
80/* Number of requests a "batching" process may submit */
81#define BLK_BATCH_REQ 32
82
83/*
84 * Return the threshold (number of used requests) at which the queue is
85 * considered to be congested. It include a little hysteresis to keep the
86 * context switch rate down.
87 */
88static inline int queue_congestion_on_threshold(struct request_queue *q)
89{
90 return q->nr_congestion_on;
91}
92
93/*
94 * The threshold at which a queue is considered to be uncongested
95 */
96static inline int queue_congestion_off_threshold(struct request_queue *q)
97{
98 return q->nr_congestion_off;
99}
100
101static void blk_queue_congestion_threshold(struct request_queue *q)
102{
103 int nr;
104
105 nr = q->nr_requests - (q->nr_requests / 8) + 1;
106 if (nr > q->nr_requests)
107 nr = q->nr_requests;
108 q->nr_congestion_on = nr;
109
110 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
111 if (nr < 1)
112 nr = 1;
113 q->nr_congestion_off = nr;
114}
115
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116/**
117 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
118 * @bdev: device
119 *
120 * Locates the passed device's request queue and returns the address of its
121 * backing_dev_info
122 *
123 * Will return NULL if the request queue cannot be located.
124 */
125struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
126{
127 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200128 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
130 if (q)
131 ret = &q->backing_dev_info;
132 return ret;
133}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134EXPORT_SYMBOL(blk_get_backing_dev_info);
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136/**
137 * blk_queue_prep_rq - set a prepare_request function for queue
138 * @q: queue
139 * @pfn: prepare_request function
140 *
141 * It's possible for a queue to register a prepare_request callback which
142 * is invoked before the request is handed to the request_fn. The goal of
143 * the function is to prepare a request for I/O, it can be used to build a
144 * cdb from the request data for instance.
145 *
146 */
Jens Axboe165125e2007-07-24 09:28:11 +0200147void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
149 q->prep_rq_fn = pfn;
150}
151
152EXPORT_SYMBOL(blk_queue_prep_rq);
153
154/**
155 * blk_queue_merge_bvec - set a merge_bvec function for queue
156 * @q: queue
157 * @mbfn: merge_bvec_fn
158 *
159 * Usually queues have static limitations on the max sectors or segments that
160 * we can put in a request. Stacking drivers may have some settings that
161 * are dynamic, and thus we have to query the queue whether it is ok to
162 * add a new bio_vec to a bio at a given offset or not. If the block device
163 * has such limitations, it needs to register a merge_bvec_fn to control
164 * the size of bio's sent to it. Note that a block device *must* allow a
165 * single page to be added to an empty bio. The block device driver may want
166 * to use the bio_split() function to deal with these bio's. By default
167 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
168 * honored.
169 */
Jens Axboe165125e2007-07-24 09:28:11 +0200170void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
172 q->merge_bvec_fn = mbfn;
173}
174
175EXPORT_SYMBOL(blk_queue_merge_bvec);
176
Jens Axboe165125e2007-07-24 09:28:11 +0200177void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
Jens Axboeff856ba2006-01-09 16:02:34 +0100178{
179 q->softirq_done_fn = fn;
180}
181
182EXPORT_SYMBOL(blk_queue_softirq_done);
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184/**
185 * blk_queue_make_request - define an alternate make_request function for a device
186 * @q: the request queue for the device to be affected
187 * @mfn: the alternate make_request function
188 *
189 * Description:
190 * The normal way for &struct bios to be passed to a device
191 * driver is for them to be collected into requests on a request
192 * queue, and then to allow the device driver to select requests
193 * off that queue when it is ready. This works well for many block
194 * devices. However some block devices (typically virtual devices
195 * such as md or lvm) do not benefit from the processing on the
196 * request queue, and are served best by having the requests passed
197 * directly to them. This can be achieved by providing a function
198 * to blk_queue_make_request().
199 *
200 * Caveat:
201 * The driver that does this *must* be able to deal appropriately
202 * with buffers in "highmemory". This can be accomplished by either calling
203 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
204 * blk_queue_bounce() to create a buffer in normal memory.
205 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200206void blk_queue_make_request(struct request_queue * q, make_request_fn * mfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
208 /*
209 * set defaults
210 */
211 q->nr_requests = BLKDEV_MAX_RQ;
Stuart McLaren309c0a12005-09-06 15:17:47 -0700212 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
213 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 q->make_request_fn = mfn;
215 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
216 q->backing_dev_info.state = 0;
217 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Mike Christiedefd94b2005-12-05 02:37:06 -0600218 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 blk_queue_hardsect_size(q, 512);
220 blk_queue_dma_alignment(q, 511);
221 blk_queue_congestion_threshold(q);
222 q->nr_batching = BLK_BATCH_REQ;
223
224 q->unplug_thresh = 4; /* hmm */
225 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
226 if (q->unplug_delay == 0)
227 q->unplug_delay = 1;
228
David Howells65f27f32006-11-22 14:55:48 +0000229 INIT_WORK(&q->unplug_work, blk_unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
231 q->unplug_timer.function = blk_unplug_timeout;
232 q->unplug_timer.data = (unsigned long)q;
233
234 /*
235 * by default assume old behaviour and bounce for any highmem page
236 */
237 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
240EXPORT_SYMBOL(blk_queue_make_request);
241
Jens Axboe165125e2007-07-24 09:28:11 +0200242static void rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243{
244 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100245 INIT_LIST_HEAD(&rq->donelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
247 rq->errors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 rq->bio = rq->biotail = NULL;
Jens Axboe2e662b62006-07-13 11:55:04 +0200249 INIT_HLIST_NODE(&rq->hash);
250 RB_CLEAR_NODE(&rq->rb_node);
Jens Axboe22e2c502005-06-27 10:55:12 +0200251 rq->ioprio = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 rq->buffer = NULL;
253 rq->ref_count = 1;
254 rq->q = q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 rq->special = NULL;
256 rq->data_len = 0;
257 rq->data = NULL;
Mike Christie df46b9a2005-06-20 14:04:44 +0200258 rq->nr_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 rq->sense = NULL;
260 rq->end_io = NULL;
261 rq->end_io_data = NULL;
Jens Axboeff856ba2006-01-09 16:02:34 +0100262 rq->completion_data = NULL;
FUJITA Tomonoriabae1fd2007-07-16 08:52:14 +0200263 rq->next_rq = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
266/**
267 * blk_queue_ordered - does this queue support ordered writes
Tejun Heo797e7db2006-01-06 09:51:03 +0100268 * @q: the request queue
269 * @ordered: one of QUEUE_ORDERED_*
Jens Axboefddfdea2006-01-31 15:24:34 +0100270 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 *
272 * Description:
273 * For journalled file systems, doing ordered writes on a commit
274 * block instead of explicitly doing wait_on_buffer (which is bad
275 * for performance) can be a big win. Block drivers supporting this
276 * feature should call this function and indicate so.
277 *
278 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200279int blk_queue_ordered(struct request_queue *q, unsigned ordered,
Tejun Heo797e7db2006-01-06 09:51:03 +0100280 prepare_flush_fn *prepare_flush_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
Tejun Heo797e7db2006-01-06 09:51:03 +0100282 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
283 prepare_flush_fn == NULL) {
284 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
285 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100287
288 if (ordered != QUEUE_ORDERED_NONE &&
289 ordered != QUEUE_ORDERED_DRAIN &&
290 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
291 ordered != QUEUE_ORDERED_DRAIN_FUA &&
292 ordered != QUEUE_ORDERED_TAG &&
293 ordered != QUEUE_ORDERED_TAG_FLUSH &&
294 ordered != QUEUE_ORDERED_TAG_FUA) {
295 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
296 return -EINVAL;
297 }
298
Tetsuo Takata60481b12006-01-24 10:34:36 +0100299 q->ordered = ordered;
Tejun Heo797e7db2006-01-06 09:51:03 +0100300 q->next_ordered = ordered;
301 q->prepare_flush_fn = prepare_flush_fn;
302
303 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305
306EXPORT_SYMBOL(blk_queue_ordered);
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308/*
309 * Cache flushing for ordered writes handling
310 */
Jens Axboe165125e2007-07-24 09:28:11 +0200311inline unsigned blk_ordered_cur_seq(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Tejun Heo797e7db2006-01-06 09:51:03 +0100313 if (!q->ordseq)
314 return 0;
315 return 1 << ffz(q->ordseq);
316}
317
318unsigned blk_ordered_req_seq(struct request *rq)
319{
Jens Axboe165125e2007-07-24 09:28:11 +0200320 struct request_queue *q = rq->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
Tejun Heo797e7db2006-01-06 09:51:03 +0100322 BUG_ON(q->ordseq == 0);
Tejun Heo8922e162005-10-20 16:23:44 +0200323
Tejun Heo797e7db2006-01-06 09:51:03 +0100324 if (rq == &q->pre_flush_rq)
325 return QUEUE_ORDSEQ_PREFLUSH;
326 if (rq == &q->bar_rq)
327 return QUEUE_ORDSEQ_BAR;
328 if (rq == &q->post_flush_rq)
329 return QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
Tejun Heobc90ba02007-06-15 13:24:28 +0200331 /*
332 * !fs requests don't need to follow barrier ordering. Always
333 * put them at the front. This fixes the following deadlock.
334 *
335 * http://thread.gmane.org/gmane.linux.kernel/537473
336 */
337 if (!blk_fs_request(rq))
338 return QUEUE_ORDSEQ_DRAIN;
339
Jens Axboe4aff5e22006-08-10 08:44:47 +0200340 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
341 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
Tejun Heo797e7db2006-01-06 09:51:03 +0100342 return QUEUE_ORDSEQ_DRAIN;
343 else
344 return QUEUE_ORDSEQ_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346
Jens Axboe165125e2007-07-24 09:28:11 +0200347void blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
Tejun Heo797e7db2006-01-06 09:51:03 +0100349 struct request *rq;
350 int uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Tejun Heo797e7db2006-01-06 09:51:03 +0100352 if (error && !q->orderr)
353 q->orderr = error;
Tejun Heo8922e162005-10-20 16:23:44 +0200354
Tejun Heo797e7db2006-01-06 09:51:03 +0100355 BUG_ON(q->ordseq & seq);
356 q->ordseq |= seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
Tejun Heo797e7db2006-01-06 09:51:03 +0100358 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
359 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
361 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100362 * Okay, sequence complete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 */
Jens Axboe4fa253f2007-07-18 13:13:10 +0200364 uptodate = 1;
365 if (q->orderr)
366 uptodate = q->orderr;
Tejun Heo797e7db2006-01-06 09:51:03 +0100367
368 q->ordseq = 0;
Jens Axboe4fa253f2007-07-18 13:13:10 +0200369 rq = q->orig_bar_rq;
Tejun Heo797e7db2006-01-06 09:51:03 +0100370
371 end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
372 end_that_request_last(rq, uptodate);
373}
374
375static void pre_flush_end_io(struct request *rq, int error)
376{
377 elv_completed_request(rq->q, rq);
378 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
379}
380
381static void bar_end_io(struct request *rq, int error)
382{
383 elv_completed_request(rq->q, rq);
384 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
385}
386
387static void post_flush_end_io(struct request *rq, int error)
388{
389 elv_completed_request(rq->q, rq);
390 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
391}
392
Jens Axboe165125e2007-07-24 09:28:11 +0200393static void queue_flush(struct request_queue *q, unsigned which)
Tejun Heo797e7db2006-01-06 09:51:03 +0100394{
395 struct request *rq;
396 rq_end_io_fn *end_io;
397
398 if (which == QUEUE_ORDERED_PREFLUSH) {
399 rq = &q->pre_flush_rq;
400 end_io = pre_flush_end_io;
401 } else {
402 rq = &q->post_flush_rq;
403 end_io = post_flush_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 }
405
Jens Axboe4aff5e22006-08-10 08:44:47 +0200406 rq->cmd_flags = REQ_HARDBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100407 rq_init(q, rq);
Tejun Heo797e7db2006-01-06 09:51:03 +0100408 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200409 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100410 rq->rq_disk = q->bar_rq.rq_disk;
Tejun Heo797e7db2006-01-06 09:51:03 +0100411 rq->end_io = end_io;
412 q->prepare_flush_fn(q, rq);
413
Tejun Heo30e96562006-02-08 01:01:31 -0800414 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100415}
416
Jens Axboe165125e2007-07-24 09:28:11 +0200417static inline struct request *start_ordered(struct request_queue *q,
Tejun Heo797e7db2006-01-06 09:51:03 +0100418 struct request *rq)
419{
Tejun Heo797e7db2006-01-06 09:51:03 +0100420 q->orderr = 0;
421 q->ordered = q->next_ordered;
422 q->ordseq |= QUEUE_ORDSEQ_STARTED;
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100425 * Prep proxy barrier request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100427 blkdev_dequeue_request(rq);
428 q->orig_bar_rq = rq;
429 rq = &q->bar_rq;
Jens Axboe4aff5e22006-08-10 08:44:47 +0200430 rq->cmd_flags = 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100431 rq_init(q, rq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200432 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
433 rq->cmd_flags |= REQ_RW;
Jens Axboe4fa253f2007-07-18 13:13:10 +0200434 if (q->ordered & QUEUE_ORDERED_FUA)
435 rq->cmd_flags |= REQ_FUA;
Tejun Heo797e7db2006-01-06 09:51:03 +0100436 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200437 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100438 init_request_from_bio(rq, q->orig_bar_rq->bio);
439 rq->end_io = bar_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
Tejun Heo797e7db2006-01-06 09:51:03 +0100441 /*
442 * Queue ordered sequence. As we stack them at the head, we
443 * need to queue in reverse order. Note that we rely on that
444 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
Jens Axboebf2de6f2007-09-27 13:01:25 +0200445 * request gets inbetween ordered sequence. If this request is
446 * an empty barrier, we don't need to do a postflush ever since
447 * there will be no data written between the pre and post flush.
448 * Hence a single flush will suffice.
Tejun Heo797e7db2006-01-06 09:51:03 +0100449 */
Jens Axboebf2de6f2007-09-27 13:01:25 +0200450 if ((q->ordered & QUEUE_ORDERED_POSTFLUSH) && !blk_empty_barrier(rq))
Tejun Heo797e7db2006-01-06 09:51:03 +0100451 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
452 else
453 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
Tejun Heo30e96562006-02-08 01:01:31 -0800455 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100456
457 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
458 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
459 rq = &q->pre_flush_rq;
460 } else
461 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
462
463 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
464 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
465 else
466 rq = NULL;
467
468 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469}
470
Jens Axboe165125e2007-07-24 09:28:11 +0200471int blk_do_ordered(struct request_queue *q, struct request **rqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472{
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800473 struct request *rq = *rqp;
Jens Axboebf2de6f2007-09-27 13:01:25 +0200474 const int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Tejun Heo797e7db2006-01-06 09:51:03 +0100476 if (!q->ordseq) {
477 if (!is_barrier)
478 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
Tejun Heo797e7db2006-01-06 09:51:03 +0100480 if (q->next_ordered != QUEUE_ORDERED_NONE) {
481 *rqp = start_ordered(q, rq);
482 return 1;
483 } else {
484 /*
485 * This can happen when the queue switches to
486 * ORDERED_NONE while this request is on it.
487 */
488 blkdev_dequeue_request(rq);
489 end_that_request_first(rq, -EOPNOTSUPP,
490 rq->hard_nr_sectors);
491 end_that_request_last(rq, -EOPNOTSUPP);
492 *rqp = NULL;
493 return 0;
494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800497 /*
498 * Ordered sequence in progress
499 */
500
501 /* Special requests are not subject to ordering rules. */
502 if (!blk_fs_request(rq) &&
503 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
504 return 1;
505
Tejun Heo797e7db2006-01-06 09:51:03 +0100506 if (q->ordered & QUEUE_ORDERED_TAG) {
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800507 /* Ordered by tag. Blocking the next barrier is enough. */
Tejun Heo797e7db2006-01-06 09:51:03 +0100508 if (is_barrier && rq != &q->bar_rq)
509 *rqp = NULL;
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800510 } else {
511 /* Ordered by draining. Wait for turn. */
512 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
513 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
514 *rqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 }
516
517 return 1;
518}
519
NeilBrown5bb23a62007-09-27 12:46:13 +0200520static void req_bio_endio(struct request *rq, struct bio *bio,
521 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100522{
Jens Axboe165125e2007-07-24 09:28:11 +0200523 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100524
NeilBrown5bb23a62007-09-27 12:46:13 +0200525 if (&q->bar_rq != rq) {
526 if (error)
527 clear_bit(BIO_UPTODATE, &bio->bi_flags);
528 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
529 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100530
NeilBrown5bb23a62007-09-27 12:46:13 +0200531 if (unlikely(nbytes > bio->bi_size)) {
532 printk("%s: want %u bytes done, only %u left\n",
533 __FUNCTION__, nbytes, bio->bi_size);
534 nbytes = bio->bi_size;
535 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100536
NeilBrown5bb23a62007-09-27 12:46:13 +0200537 bio->bi_size -= nbytes;
538 bio->bi_sector += (nbytes >> 9);
539 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200540 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200541 } else {
542
543 /*
544 * Okay, this is the barrier request in progress, just
545 * record the error;
546 */
547 if (error && !q->orderr)
548 q->orderr = error;
549 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100550}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552/**
553 * blk_queue_bounce_limit - set bounce buffer limit for queue
554 * @q: the request queue for the device
555 * @dma_addr: bus address limit
556 *
557 * Description:
558 * Different hardware can have different requirements as to what pages
559 * it can do I/O directly to. A low level driver can call
560 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Andi Kleen5ee1af92006-03-08 17:57:26 -0800561 * buffers for doing I/O to pages residing above @page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200563void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564{
565 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800566 int dma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Andi Kleen5ee1af92006-03-08 17:57:26 -0800568 q->bounce_gfp = GFP_NOIO;
569#if BITS_PER_LONG == 64
570 /* Assume anything <= 4GB can be handled by IOMMU.
571 Actually some IOMMUs can handle everything, but I don't
572 know of a way to test this here. */
Andi Kleen82697302006-06-21 14:48:09 +0200573 if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Andi Kleen5ee1af92006-03-08 17:57:26 -0800574 dma = 1;
575 q->bounce_pfn = max_low_pfn;
576#else
577 if (bounce_pfn < blk_max_low_pfn)
578 dma = 1;
579 q->bounce_pfn = bounce_pfn;
580#endif
581 if (dma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 init_emergency_isa_pool();
583 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800584 q->bounce_pfn = bounce_pfn;
585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586}
587
588EXPORT_SYMBOL(blk_queue_bounce_limit);
589
590/**
591 * blk_queue_max_sectors - set max sectors for a request for this queue
592 * @q: the request queue for the device
593 * @max_sectors: max sectors in the usual 512b unit
594 *
595 * Description:
596 * Enables a low level driver to set an upper limit on the size of
597 * received requests.
598 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200599void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
601 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
602 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
603 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
604 }
605
Mike Christiedefd94b2005-12-05 02:37:06 -0600606 if (BLK_DEF_MAX_SECTORS > max_sectors)
607 q->max_hw_sectors = q->max_sectors = max_sectors;
608 else {
609 q->max_sectors = BLK_DEF_MAX_SECTORS;
610 q->max_hw_sectors = max_sectors;
611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
614EXPORT_SYMBOL(blk_queue_max_sectors);
615
616/**
617 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
618 * @q: the request queue for the device
619 * @max_segments: max number of segments
620 *
621 * Description:
622 * Enables a low level driver to set an upper limit on the number of
623 * physical data segments in a request. This would be the largest sized
624 * scatter list the driver could handle.
625 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200626void blk_queue_max_phys_segments(struct request_queue *q,
627 unsigned short max_segments)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
629 if (!max_segments) {
630 max_segments = 1;
631 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
632 }
633
634 q->max_phys_segments = max_segments;
635}
636
637EXPORT_SYMBOL(blk_queue_max_phys_segments);
638
639/**
640 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
641 * @q: the request queue for the device
642 * @max_segments: max number of segments
643 *
644 * Description:
645 * Enables a low level driver to set an upper limit on the number of
646 * hw data segments in a request. This would be the largest number of
647 * address/length pairs the host adapter can actually give as once
648 * to the device.
649 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200650void blk_queue_max_hw_segments(struct request_queue *q,
651 unsigned short max_segments)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
653 if (!max_segments) {
654 max_segments = 1;
655 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
656 }
657
658 q->max_hw_segments = max_segments;
659}
660
661EXPORT_SYMBOL(blk_queue_max_hw_segments);
662
663/**
664 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
665 * @q: the request queue for the device
666 * @max_size: max size of segment in bytes
667 *
668 * Description:
669 * Enables a low level driver to set an upper limit on the size of a
670 * coalesced segment
671 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200672void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
674 if (max_size < PAGE_CACHE_SIZE) {
675 max_size = PAGE_CACHE_SIZE;
676 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
677 }
678
679 q->max_segment_size = max_size;
680}
681
682EXPORT_SYMBOL(blk_queue_max_segment_size);
683
684/**
685 * blk_queue_hardsect_size - set hardware sector size for the queue
686 * @q: the request queue for the device
687 * @size: the hardware sector size, in bytes
688 *
689 * Description:
690 * This should typically be set to the lowest possible sector size
691 * that the hardware can operate on (possible without reverting to
692 * even internal read-modify-write operations). Usually the default
693 * of 512 covers most hardware.
694 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200695void blk_queue_hardsect_size(struct request_queue *q, unsigned short size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
697 q->hardsect_size = size;
698}
699
700EXPORT_SYMBOL(blk_queue_hardsect_size);
701
702/*
703 * Returns the minimum that is _not_ zero, unless both are zero.
704 */
705#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
706
707/**
708 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
709 * @t: the stacking driver (top)
710 * @b: the underlying device (bottom)
711 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200712void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
714 /* zero is "infinity" */
Mike Christiedefd94b2005-12-05 02:37:06 -0600715 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
716 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
718 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
719 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
720 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
721 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
NeilBrown89e5c8b2006-03-27 01:18:02 -0800722 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
723 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724}
725
726EXPORT_SYMBOL(blk_queue_stack_limits);
727
728/**
729 * blk_queue_segment_boundary - set boundary rules for segment merging
730 * @q: the request queue for the device
731 * @mask: the memory boundary mask
732 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200733void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734{
735 if (mask < PAGE_CACHE_SIZE - 1) {
736 mask = PAGE_CACHE_SIZE - 1;
737 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
738 }
739
740 q->seg_boundary_mask = mask;
741}
742
743EXPORT_SYMBOL(blk_queue_segment_boundary);
744
745/**
746 * blk_queue_dma_alignment - set dma length and memory alignment
747 * @q: the request queue for the device
748 * @mask: alignment mask
749 *
750 * description:
751 * set required memory and length aligment for direct dma transactions.
752 * this is used when buiding direct io requests for the queue.
753 *
754 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200755void blk_queue_dma_alignment(struct request_queue *q, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756{
757 q->dma_alignment = mask;
758}
759
760EXPORT_SYMBOL(blk_queue_dma_alignment);
761
762/**
763 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 * @q: The request queue for the device
765 * @tag: The tag of the request
766 *
767 * Notes:
768 * Should be used when a device returns a tag and you want to match
769 * it with a request.
770 *
771 * no locks need be held.
772 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200773struct request *blk_queue_find_tag(struct request_queue *q, int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774{
David C Somayajuluf583f492006-10-04 08:27:25 +0200775 return blk_map_queue_find_tag(q->queue_tags, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
777
778EXPORT_SYMBOL(blk_queue_find_tag);
779
780/**
James Bottomley492dfb42006-08-30 15:48:45 -0400781 * __blk_free_tags - release a given set of tag maintenance info
782 * @bqt: the tag map to free
783 *
784 * Tries to free the specified @bqt@. Returns true if it was
785 * actually freed and false if there are still references using it
786 */
787static int __blk_free_tags(struct blk_queue_tag *bqt)
788{
789 int retval;
790
791 retval = atomic_dec_and_test(&bqt->refcnt);
792 if (retval) {
793 BUG_ON(bqt->busy);
James Bottomley492dfb42006-08-30 15:48:45 -0400794
795 kfree(bqt->tag_index);
796 bqt->tag_index = NULL;
797
798 kfree(bqt->tag_map);
799 bqt->tag_map = NULL;
800
801 kfree(bqt);
802
803 }
804
805 return retval;
806}
807
808/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 * __blk_queue_free_tags - release tag maintenance info
810 * @q: the request queue for the device
811 *
812 * Notes:
813 * blk_cleanup_queue() will take care of calling this function, if tagging
814 * has been used. So there's no need to call this directly.
815 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200816static void __blk_queue_free_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817{
818 struct blk_queue_tag *bqt = q->queue_tags;
819
820 if (!bqt)
821 return;
822
James Bottomley492dfb42006-08-30 15:48:45 -0400823 __blk_free_tags(bqt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
825 q->queue_tags = NULL;
826 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
827}
828
James Bottomley492dfb42006-08-30 15:48:45 -0400829
830/**
831 * blk_free_tags - release a given set of tag maintenance info
832 * @bqt: the tag map to free
833 *
834 * For externally managed @bqt@ frees the map. Callers of this
835 * function must guarantee to have released all the queues that
836 * might have been using this tag map.
837 */
838void blk_free_tags(struct blk_queue_tag *bqt)
839{
840 if (unlikely(!__blk_free_tags(bqt)))
841 BUG();
842}
843EXPORT_SYMBOL(blk_free_tags);
844
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845/**
846 * blk_queue_free_tags - release tag maintenance info
847 * @q: the request queue for the device
848 *
849 * Notes:
850 * This is used to disabled tagged queuing to a device, yet leave
851 * queue in function.
852 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200853void blk_queue_free_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854{
855 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
856}
857
858EXPORT_SYMBOL(blk_queue_free_tags);
859
860static int
Jens Axboe165125e2007-07-24 09:28:11 +0200861init_tag_map(struct request_queue *q, struct blk_queue_tag *tags, int depth)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 struct request **tag_index;
864 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700865 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
James Bottomley492dfb42006-08-30 15:48:45 -0400867 if (q && depth > q->nr_requests * 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 depth = q->nr_requests * 2;
869 printk(KERN_ERR "%s: adjusted depth to %d\n",
870 __FUNCTION__, depth);
871 }
872
Jens Axboef68110f2006-03-08 13:31:44 +0100873 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 if (!tag_index)
875 goto fail;
876
Tejun Heof7d37d02005-06-23 00:08:50 -0700877 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Jens Axboef68110f2006-03-08 13:31:44 +0100878 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 if (!tag_map)
880 goto fail;
881
Tejun Heoba025082005-08-05 13:28:11 -0700882 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 tags->tag_index = tag_index;
885 tags->tag_map = tag_map;
886
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 return 0;
888fail:
889 kfree(tag_index);
890 return -ENOMEM;
891}
892
James Bottomley492dfb42006-08-30 15:48:45 -0400893static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
894 int depth)
895{
896 struct blk_queue_tag *tags;
897
898 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
899 if (!tags)
900 goto fail;
901
902 if (init_tag_map(q, tags, depth))
903 goto fail;
904
James Bottomley492dfb42006-08-30 15:48:45 -0400905 tags->busy = 0;
906 atomic_set(&tags->refcnt, 1);
907 return tags;
908fail:
909 kfree(tags);
910 return NULL;
911}
912
913/**
914 * blk_init_tags - initialize the tag info for an external tag map
915 * @depth: the maximum queue depth supported
916 * @tags: the tag to use
917 **/
918struct blk_queue_tag *blk_init_tags(int depth)
919{
920 return __blk_queue_init_tags(NULL, depth);
921}
922EXPORT_SYMBOL(blk_init_tags);
923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924/**
925 * blk_queue_init_tags - initialize the queue tag info
926 * @q: the request queue for the device
927 * @depth: the maximum queue depth supported
928 * @tags: the tag to use
929 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200930int blk_queue_init_tags(struct request_queue *q, int depth,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 struct blk_queue_tag *tags)
932{
933 int rc;
934
935 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
936
937 if (!tags && !q->queue_tags) {
James Bottomley492dfb42006-08-30 15:48:45 -0400938 tags = __blk_queue_init_tags(q, depth);
939
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 if (!tags)
941 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 } else if (q->queue_tags) {
943 if ((rc = blk_queue_resize_tags(q, depth)))
944 return rc;
945 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
946 return 0;
947 } else
948 atomic_inc(&tags->refcnt);
949
950 /*
951 * assign it, all done
952 */
953 q->queue_tags = tags;
954 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
Jens Axboe6eca9002007-10-25 10:14:47 +0200955 INIT_LIST_HEAD(&q->tag_busy_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 return 0;
957fail:
958 kfree(tags);
959 return -ENOMEM;
960}
961
962EXPORT_SYMBOL(blk_queue_init_tags);
963
964/**
965 * blk_queue_resize_tags - change the queueing depth
966 * @q: the request queue for the device
967 * @new_depth: the new max command queueing depth
968 *
969 * Notes:
970 * Must be called with the queue lock held.
971 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200972int blk_queue_resize_tags(struct request_queue *q, int new_depth)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973{
974 struct blk_queue_tag *bqt = q->queue_tags;
975 struct request **tag_index;
976 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700977 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
979 if (!bqt)
980 return -ENXIO;
981
982 /*
Tejun Heoba025082005-08-05 13:28:11 -0700983 * if we already have large enough real_max_depth. just
984 * adjust max_depth. *NOTE* as requests with tag value
985 * between new_depth and real_max_depth can be in-flight, tag
986 * map can not be shrunk blindly here.
987 */
988 if (new_depth <= bqt->real_max_depth) {
989 bqt->max_depth = new_depth;
990 return 0;
991 }
992
993 /*
James Bottomley492dfb42006-08-30 15:48:45 -0400994 * Currently cannot replace a shared tag map with a new
995 * one, so error out if this is the case
996 */
997 if (atomic_read(&bqt->refcnt) != 1)
998 return -EBUSY;
999
1000 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 * save the old state info, so we can copy it back
1002 */
1003 tag_index = bqt->tag_index;
1004 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -07001005 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
1007 if (init_tag_map(q, bqt, new_depth))
1008 return -ENOMEM;
1009
1010 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -07001011 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -07001012 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
1014 kfree(tag_index);
1015 kfree(tag_map);
1016 return 0;
1017}
1018
1019EXPORT_SYMBOL(blk_queue_resize_tags);
1020
1021/**
1022 * blk_queue_end_tag - end tag operations for a request
1023 * @q: the request queue for the device
1024 * @rq: the request that has completed
1025 *
1026 * Description:
1027 * Typically called when end_that_request_first() returns 0, meaning
1028 * all transfers have been done for a request. It's important to call
1029 * this function before end_that_request_last(), as that will put the
1030 * request back on the free list thus corrupting the internal tag list.
1031 *
1032 * Notes:
1033 * queue lock must be held.
1034 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001035void blk_queue_end_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036{
1037 struct blk_queue_tag *bqt = q->queue_tags;
1038 int tag = rq->tag;
1039
1040 BUG_ON(tag == -1);
1041
Tejun Heoba025082005-08-05 13:28:11 -07001042 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -07001043 /*
1044 * This can happen after tag depth has been reduced.
1045 * FIXME: how about a warning or info message here?
1046 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 return;
1048
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001050 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 rq->tag = -1;
1052
1053 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001054 printk(KERN_ERR "%s: tag %d is missing\n",
1055 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
1057 bqt->tag_index[tag] = NULL;
Jens Axboef3da54b2007-09-13 14:26:53 +02001058
Nick Pigginadb4ddb2007-10-24 10:54:38 +02001059 if (unlikely(!test_bit(tag, bqt->tag_map))) {
Jens Axboef3da54b2007-09-13 14:26:53 +02001060 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1061 __FUNCTION__, tag);
1062 return;
1063 }
Nick Pigginadb4ddb2007-10-24 10:54:38 +02001064 /*
1065 * The tag_map bit acts as a lock for tag_index[bit], so we need
1066 * unlock memory barrier semantics.
1067 */
1068 clear_bit_unlock(tag, bqt->tag_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 bqt->busy--;
1070}
1071
1072EXPORT_SYMBOL(blk_queue_end_tag);
1073
1074/**
1075 * blk_queue_start_tag - find a free tag and assign it
1076 * @q: the request queue for the device
1077 * @rq: the block request that needs tagging
1078 *
1079 * Description:
1080 * This can either be used as a stand-alone helper, or possibly be
1081 * assigned as the queue &prep_rq_fn (in which case &struct request
1082 * automagically gets a tag assigned). Note that this function
1083 * assumes that any type of request can be queued! if this is not
1084 * true for your device, you must check the request type before
1085 * calling this function. The request will also be removed from
1086 * the request queue, so it's the drivers responsibility to readd
1087 * it if it should need to be restarted for some reason.
1088 *
1089 * Notes:
1090 * queue lock must be held.
1091 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001092int blk_queue_start_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093{
1094 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001095 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Jens Axboe4aff5e22006-08-10 08:44:47 +02001097 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001099 "%s: request %p for device [%s] already tagged %d",
1100 __FUNCTION__, rq,
1101 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 BUG();
1103 }
1104
Jens Axboe059af492006-09-21 20:37:22 +02001105 /*
1106 * Protect against shared tag maps, as we may not have exclusive
1107 * access to the tag map.
1108 */
1109 do {
1110 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1111 if (tag >= bqt->max_depth)
1112 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
Nick Pigginadb4ddb2007-10-24 10:54:38 +02001114 } while (test_and_set_bit_lock(tag, bqt->tag_map));
Nick Piggindd941252007-09-14 08:41:12 +02001115 /*
Nick Pigginadb4ddb2007-10-24 10:54:38 +02001116 * We need lock ordering semantics given by test_and_set_bit_lock.
1117 * See blk_queue_end_tag for details.
Nick Piggindd941252007-09-14 08:41:12 +02001118 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
Jens Axboe4aff5e22006-08-10 08:44:47 +02001120 rq->cmd_flags |= REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 rq->tag = tag;
1122 bqt->tag_index[tag] = rq;
1123 blkdev_dequeue_request(rq);
Jens Axboe6eca9002007-10-25 10:14:47 +02001124 list_add(&rq->queuelist, &q->tag_busy_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 bqt->busy++;
1126 return 0;
1127}
1128
1129EXPORT_SYMBOL(blk_queue_start_tag);
1130
1131/**
1132 * blk_queue_invalidate_tags - invalidate all pending tags
1133 * @q: the request queue for the device
1134 *
1135 * Description:
1136 * Hardware conditions may dictate a need to stop all pending requests.
1137 * In this case, we will safely clear the block side of the tag queue and
1138 * readd all requests to the request queue in the right order.
1139 *
1140 * Notes:
1141 * queue lock must be held.
1142 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001143void blk_queue_invalidate_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 struct list_head *tmp, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Jens Axboed85532e2007-11-09 12:52:45 +01001147 list_for_each_safe(tmp, n, &q->tag_busy_list)
1148 blk_requeue_request(q, list_entry_rq(tmp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149}
1150
1151EXPORT_SYMBOL(blk_queue_invalidate_tags);
1152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153void blk_dump_rq_flags(struct request *rq, char *msg)
1154{
1155 int bit;
1156
Jens Axboe4aff5e22006-08-10 08:44:47 +02001157 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1158 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1159 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
1161 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1162 rq->nr_sectors,
1163 rq->current_nr_sectors);
1164 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1165
Jens Axboe4aff5e22006-08-10 08:44:47 +02001166 if (blk_pc_request(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 printk("cdb: ");
1168 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1169 printk("%02x ", rq->cmd[bit]);
1170 printk("\n");
1171 }
1172}
1173
1174EXPORT_SYMBOL(blk_dump_rq_flags);
1175
Jens Axboe165125e2007-07-24 09:28:11 +02001176void blk_recount_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177{
NeilBrown9dfa5282007-08-16 13:27:52 +02001178 struct request rq;
1179 struct bio *nxt = bio->bi_next;
1180 rq.q = q;
1181 rq.bio = rq.biotail = bio;
1182 bio->bi_next = NULL;
1183 blk_recalc_rq_segments(&rq);
1184 bio->bi_next = nxt;
1185 bio->bi_phys_segments = rq.nr_phys_segments;
1186 bio->bi_hw_segments = rq.nr_hw_segments;
1187 bio->bi_flags |= (1 << BIO_SEG_VALID);
1188}
1189EXPORT_SYMBOL(blk_recount_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
NeilBrown9dfa5282007-08-16 13:27:52 +02001191static void blk_recalc_rq_segments(struct request *rq)
1192{
1193 int nr_phys_segs;
1194 int nr_hw_segs;
1195 unsigned int phys_size;
1196 unsigned int hw_size;
1197 struct bio_vec *bv, *bvprv = NULL;
1198 int seg_size;
1199 int hw_seg_size;
1200 int cluster;
NeilBrown5705f702007-09-25 12:35:59 +02001201 struct req_iterator iter;
NeilBrown9dfa5282007-08-16 13:27:52 +02001202 int high, highprv = 1;
1203 struct request_queue *q = rq->q;
1204
1205 if (!rq->bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 return;
1207
1208 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
NeilBrown9dfa5282007-08-16 13:27:52 +02001209 hw_seg_size = seg_size = 0;
1210 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
NeilBrown5705f702007-09-25 12:35:59 +02001211 rq_for_each_segment(bv, rq, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 /*
1213 * the trick here is making sure that a high page is never
1214 * considered part of another segment, since that might
1215 * change with the bounce page.
1216 */
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02001217 high = page_to_pfn(bv->bv_page) > q->bounce_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 if (high || highprv)
1219 goto new_hw_segment;
1220 if (cluster) {
1221 if (seg_size + bv->bv_len > q->max_segment_size)
1222 goto new_segment;
1223 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1224 goto new_segment;
1225 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1226 goto new_segment;
1227 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1228 goto new_hw_segment;
1229
1230 seg_size += bv->bv_len;
1231 hw_seg_size += bv->bv_len;
1232 bvprv = bv;
1233 continue;
1234 }
1235new_segment:
1236 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
NeilBrown9dfa5282007-08-16 13:27:52 +02001237 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 hw_seg_size += bv->bv_len;
NeilBrown9dfa5282007-08-16 13:27:52 +02001239 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240new_hw_segment:
NeilBrown9dfa5282007-08-16 13:27:52 +02001241 if (nr_hw_segs == 1 &&
1242 hw_seg_size > rq->bio->bi_hw_front_size)
1243 rq->bio->bi_hw_front_size = hw_seg_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1245 nr_hw_segs++;
1246 }
1247
1248 nr_phys_segs++;
1249 bvprv = bv;
1250 seg_size = bv->bv_len;
1251 highprv = high;
1252 }
NeilBrown9dfa5282007-08-16 13:27:52 +02001253
1254 if (nr_hw_segs == 1 &&
1255 hw_seg_size > rq->bio->bi_hw_front_size)
1256 rq->bio->bi_hw_front_size = hw_seg_size;
1257 if (hw_seg_size > rq->biotail->bi_hw_back_size)
1258 rq->biotail->bi_hw_back_size = hw_seg_size;
1259 rq->nr_phys_segments = nr_phys_segs;
1260 rq->nr_hw_segments = nr_hw_segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
Jens Axboe165125e2007-07-24 09:28:11 +02001263static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 struct bio *nxt)
1265{
1266 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1267 return 0;
1268
1269 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1270 return 0;
1271 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1272 return 0;
1273
1274 /*
1275 * bio and nxt are contigous in memory, check if the queue allows
1276 * these two to be merged into one
1277 */
1278 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1279 return 1;
1280
1281 return 0;
1282}
1283
Jens Axboe165125e2007-07-24 09:28:11 +02001284static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 struct bio *nxt)
1286{
1287 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1288 blk_recount_segments(q, bio);
1289 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1290 blk_recount_segments(q, nxt);
1291 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
Jens Axboe32eef962007-06-19 09:09:27 +02001292 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 return 0;
Jens Axboe32eef962007-06-19 09:09:27 +02001294 if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 return 0;
1296
1297 return 1;
1298}
1299
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300/*
1301 * map a request to scatterlist, return number of sg entries setup. Caller
1302 * must make sure sg can hold rq->nr_phys_segments entries
1303 */
Jens Axboe165125e2007-07-24 09:28:11 +02001304int blk_rq_map_sg(struct request_queue *q, struct request *rq,
Jens Axboef5659132007-09-21 10:44:19 +02001305 struct scatterlist *sglist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306{
1307 struct bio_vec *bvec, *bvprv;
NeilBrown5705f702007-09-25 12:35:59 +02001308 struct req_iterator iter;
Jens Axboeba951842007-10-17 19:34:11 +02001309 struct scatterlist *sg;
NeilBrown5705f702007-09-25 12:35:59 +02001310 int nsegs, cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312 nsegs = 0;
1313 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1314
1315 /*
1316 * for each bio in rq
1317 */
1318 bvprv = NULL;
Jens Axboeba951842007-10-17 19:34:11 +02001319 sg = NULL;
NeilBrown5705f702007-09-25 12:35:59 +02001320 rq_for_each_segment(bvec, rq, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +02001321 int nbytes = bvec->bv_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
Jens Axboe6c92e692007-08-16 13:43:12 +02001323 if (bvprv && cluster) {
Jens Axboef5659132007-09-21 10:44:19 +02001324 if (sg->length + nbytes > q->max_segment_size)
Jens Axboe6c92e692007-08-16 13:43:12 +02001325 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
Jens Axboe6c92e692007-08-16 13:43:12 +02001327 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1328 goto new_segment;
1329 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1330 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
Jens Axboef5659132007-09-21 10:44:19 +02001332 sg->length += nbytes;
Jens Axboe6c92e692007-08-16 13:43:12 +02001333 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334new_segment:
Jens Axboeba951842007-10-17 19:34:11 +02001335 if (!sg)
1336 sg = sglist;
Jens Axboe7aeacf92007-10-23 09:49:25 +02001337 else {
1338 /*
1339 * If the driver previously mapped a shorter
1340 * list, we could see a termination bit
1341 * prematurely unless it fully inits the sg
1342 * table on each mapping. We KNOW that there
1343 * must be more entries here or the driver
1344 * would be buggy, so force clear the
1345 * termination bit to avoid doing a full
1346 * sg_init_table() in drivers for each command.
1347 */
1348 sg->page_link &= ~0x02;
Jens Axboeba951842007-10-17 19:34:11 +02001349 sg = sg_next(sg);
Jens Axboe7aeacf92007-10-23 09:49:25 +02001350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Jens Axboe642f1492007-10-24 11:20:47 +02001352 sg_set_page(sg, bvec->bv_page, nbytes, bvec->bv_offset);
Jens Axboe6c92e692007-08-16 13:43:12 +02001353 nsegs++;
1354 }
1355 bvprv = bvec;
NeilBrown5705f702007-09-25 12:35:59 +02001356 } /* segments in rq */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
Jens Axboe9b617642007-10-22 19:39:33 +02001358 if (sg)
Jens Axboec46f2332007-10-31 12:06:37 +01001359 sg_mark_end(sg);
Jens Axboe9b617642007-10-22 19:39:33 +02001360
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 return nsegs;
1362}
1363
1364EXPORT_SYMBOL(blk_rq_map_sg);
1365
1366/*
1367 * the standard queue merge functions, can be overridden with device
1368 * specific ones if so desired
1369 */
1370
Jens Axboe165125e2007-07-24 09:28:11 +02001371static inline int ll_new_mergeable(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 struct request *req,
1373 struct bio *bio)
1374{
1375 int nr_phys_segs = bio_phys_segments(q, bio);
1376
1377 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001378 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 if (req == q->last_merge)
1380 q->last_merge = NULL;
1381 return 0;
1382 }
1383
1384 /*
1385 * A hw segment is just getting larger, bump just the phys
1386 * counter.
1387 */
1388 req->nr_phys_segments += nr_phys_segs;
1389 return 1;
1390}
1391
Jens Axboe165125e2007-07-24 09:28:11 +02001392static inline int ll_new_hw_segment(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 struct request *req,
1394 struct bio *bio)
1395{
1396 int nr_hw_segs = bio_hw_segments(q, bio);
1397 int nr_phys_segs = bio_phys_segments(q, bio);
1398
1399 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1400 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001401 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 if (req == q->last_merge)
1403 q->last_merge = NULL;
1404 return 0;
1405 }
1406
1407 /*
1408 * This will form the start of a new hw segment. Bump both
1409 * counters.
1410 */
1411 req->nr_hw_segments += nr_hw_segs;
1412 req->nr_phys_segments += nr_phys_segs;
1413 return 1;
1414}
1415
NeilBrown3001ca72007-08-16 13:31:27 +02001416static int ll_back_merge_fn(struct request_queue *q, struct request *req,
1417 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418{
Mike Christiedefd94b2005-12-05 02:37:06 -06001419 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 int len;
1421
Mike Christiedefd94b2005-12-05 02:37:06 -06001422 if (unlikely(blk_pc_request(req)))
1423 max_sectors = q->max_hw_sectors;
1424 else
1425 max_sectors = q->max_sectors;
1426
1427 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001428 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 if (req == q->last_merge)
1430 q->last_merge = NULL;
1431 return 0;
1432 }
1433 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1434 blk_recount_segments(q, req->biotail);
1435 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1436 blk_recount_segments(q, bio);
1437 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1438 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1439 !BIOVEC_VIRT_OVERSIZE(len)) {
1440 int mergeable = ll_new_mergeable(q, req, bio);
1441
1442 if (mergeable) {
1443 if (req->nr_hw_segments == 1)
1444 req->bio->bi_hw_front_size = len;
1445 if (bio->bi_hw_segments == 1)
1446 bio->bi_hw_back_size = len;
1447 }
1448 return mergeable;
1449 }
1450
1451 return ll_new_hw_segment(q, req, bio);
1452}
1453
Jens Axboe165125e2007-07-24 09:28:11 +02001454static int ll_front_merge_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 struct bio *bio)
1456{
Mike Christiedefd94b2005-12-05 02:37:06 -06001457 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 int len;
1459
Mike Christiedefd94b2005-12-05 02:37:06 -06001460 if (unlikely(blk_pc_request(req)))
1461 max_sectors = q->max_hw_sectors;
1462 else
1463 max_sectors = q->max_sectors;
1464
1465
1466 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001467 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 if (req == q->last_merge)
1469 q->last_merge = NULL;
1470 return 0;
1471 }
1472 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1473 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1474 blk_recount_segments(q, bio);
1475 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1476 blk_recount_segments(q, req->bio);
1477 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1478 !BIOVEC_VIRT_OVERSIZE(len)) {
1479 int mergeable = ll_new_mergeable(q, req, bio);
1480
1481 if (mergeable) {
1482 if (bio->bi_hw_segments == 1)
1483 bio->bi_hw_front_size = len;
1484 if (req->nr_hw_segments == 1)
1485 req->biotail->bi_hw_back_size = len;
1486 }
1487 return mergeable;
1488 }
1489
1490 return ll_new_hw_segment(q, req, bio);
1491}
1492
Jens Axboe165125e2007-07-24 09:28:11 +02001493static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 struct request *next)
1495{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001496 int total_phys_segments;
1497 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498
1499 /*
1500 * First check if the either of the requests are re-queued
1501 * requests. Can't merge them if they are.
1502 */
1503 if (req->special || next->special)
1504 return 0;
1505
1506 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001507 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 */
1509 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1510 return 0;
1511
1512 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1513 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1514 total_phys_segments--;
1515
1516 if (total_phys_segments > q->max_phys_segments)
1517 return 0;
1518
1519 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1520 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1521 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1522 /*
1523 * propagate the combined length to the end of the requests
1524 */
1525 if (req->nr_hw_segments == 1)
1526 req->bio->bi_hw_front_size = len;
1527 if (next->nr_hw_segments == 1)
1528 next->biotail->bi_hw_back_size = len;
1529 total_hw_segments--;
1530 }
1531
1532 if (total_hw_segments > q->max_hw_segments)
1533 return 0;
1534
1535 /* Merge is OK... */
1536 req->nr_phys_segments = total_phys_segments;
1537 req->nr_hw_segments = total_hw_segments;
1538 return 1;
1539}
1540
1541/*
1542 * "plug" the device if there are no outstanding requests: this will
1543 * force the transfer to start only after we have put all the requests
1544 * on the list.
1545 *
1546 * This is called with interrupts off and no requests on the queue and
1547 * with the queue lock held.
1548 */
Jens Axboe165125e2007-07-24 09:28:11 +02001549void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
1551 WARN_ON(!irqs_disabled());
1552
1553 /*
1554 * don't plug a stopped queue, it must be paired with blk_start_queue()
1555 * which will restart the queueing
1556 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001557 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 return;
1559
Jens Axboe2056a782006-03-23 20:00:26 +01001560 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +01001562 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1563 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564}
1565
1566EXPORT_SYMBOL(blk_plug_device);
1567
1568/*
1569 * remove the queue from the plugged list, if present. called with
1570 * queue lock held and interrupts disabled.
1571 */
Jens Axboe165125e2007-07-24 09:28:11 +02001572int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573{
1574 WARN_ON(!irqs_disabled());
1575
1576 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1577 return 0;
1578
1579 del_timer(&q->unplug_timer);
1580 return 1;
1581}
1582
1583EXPORT_SYMBOL(blk_remove_plug);
1584
1585/*
1586 * remove the plug and let it rip..
1587 */
Jens Axboe165125e2007-07-24 09:28:11 +02001588void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001590 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 return;
1592
1593 if (!blk_remove_plug(q))
1594 return;
1595
Jens Axboe22e2c502005-06-27 10:55:12 +02001596 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597}
1598EXPORT_SYMBOL(__generic_unplug_device);
1599
1600/**
1601 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +02001602 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 *
1604 * Description:
1605 * Linux uses plugging to build bigger requests queues before letting
1606 * the device have at them. If a queue is plugged, the I/O scheduler
1607 * is still adding and merging requests on the queue. Once the queue
1608 * gets unplugged, the request_fn defined for the queue is invoked and
1609 * transfers started.
1610 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001611void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612{
1613 spin_lock_irq(q->queue_lock);
1614 __generic_unplug_device(q);
1615 spin_unlock_irq(q->queue_lock);
1616}
1617EXPORT_SYMBOL(generic_unplug_device);
1618
1619static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1620 struct page *page)
1621{
Jens Axboe165125e2007-07-24 09:28:11 +02001622 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -05001624 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
1626
David Howells65f27f32006-11-22 14:55:48 +00001627static void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628{
Jens Axboe165125e2007-07-24 09:28:11 +02001629 struct request_queue *q =
1630 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
Jens Axboe2056a782006-03-23 20:00:26 +01001632 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1633 q->rq.count[READ] + q->rq.count[WRITE]);
1634
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 q->unplug_fn(q);
1636}
1637
1638static void blk_unplug_timeout(unsigned long data)
1639{
Jens Axboe165125e2007-07-24 09:28:11 +02001640 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
Jens Axboe2056a782006-03-23 20:00:26 +01001642 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1643 q->rq.count[READ] + q->rq.count[WRITE]);
1644
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 kblockd_schedule_work(&q->unplug_work);
1646}
1647
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -05001648void blk_unplug(struct request_queue *q)
1649{
1650 /*
1651 * devices don't necessarily have an ->unplug_fn defined
1652 */
1653 if (q->unplug_fn) {
1654 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1655 q->rq.count[READ] + q->rq.count[WRITE]);
1656
1657 q->unplug_fn(q);
1658 }
1659}
1660EXPORT_SYMBOL(blk_unplug);
1661
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662/**
1663 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +02001664 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 *
1666 * Description:
1667 * blk_start_queue() will clear the stop flag on the queue, and call
1668 * the request_fn for the queue if it was in a stopped state when
1669 * entered. Also see blk_stop_queue(). Queue lock must be held.
1670 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001671void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001673 WARN_ON(!irqs_disabled());
1674
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1676
1677 /*
1678 * one level of recursion is ok and is much faster than kicking
1679 * the unplug handling
1680 */
1681 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1682 q->request_fn(q);
1683 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1684 } else {
1685 blk_plug_device(q);
1686 kblockd_schedule_work(&q->unplug_work);
1687 }
1688}
1689
1690EXPORT_SYMBOL(blk_start_queue);
1691
1692/**
1693 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +02001694 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 *
1696 * Description:
1697 * The Linux block layer assumes that a block driver will consume all
1698 * entries on the request queue when the request_fn strategy is called.
1699 * Often this will not happen, because of hardware limitations (queue
1700 * depth settings). If a device driver gets a 'queue full' response,
1701 * or if it simply chooses not to queue more I/O at one point, it can
1702 * call this function to prevent the request_fn from being called until
1703 * the driver has signalled it's ready to go again. This happens by calling
1704 * blk_start_queue() to restart queue operations. Queue lock must be held.
1705 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001706void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707{
1708 blk_remove_plug(q);
1709 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1710}
1711EXPORT_SYMBOL(blk_stop_queue);
1712
1713/**
1714 * blk_sync_queue - cancel any pending callbacks on a queue
1715 * @q: the queue
1716 *
1717 * Description:
1718 * The block layer may perform asynchronous callback activity
1719 * on a queue, such as calling the unplug function after a timeout.
1720 * A block device may call blk_sync_queue to ensure that any
1721 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +02001722 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 * that its ->make_request_fn will not re-add plugging prior to calling
1724 * this function.
1725 *
1726 */
1727void blk_sync_queue(struct request_queue *q)
1728{
1729 del_timer_sync(&q->unplug_timer);
Oleg Nesterovabbeb882007-10-23 15:08:19 +02001730 kblockd_flush_work(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731}
1732EXPORT_SYMBOL(blk_sync_queue);
1733
1734/**
1735 * blk_run_queue - run a single device queue
1736 * @q: The queue to run
1737 */
1738void blk_run_queue(struct request_queue *q)
1739{
1740 unsigned long flags;
1741
1742 spin_lock_irqsave(q->queue_lock, flags);
1743 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +02001744
1745 /*
1746 * Only recurse once to avoid overrunning the stack, let the unplug
1747 * handling reinvoke the handler shortly if we already got there.
1748 */
1749 if (!elv_queue_empty(q)) {
1750 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1751 q->request_fn(q);
1752 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1753 } else {
1754 blk_plug_device(q);
1755 kblockd_schedule_work(&q->unplug_work);
1756 }
1757 }
1758
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 spin_unlock_irqrestore(q->queue_lock, flags);
1760}
1761EXPORT_SYMBOL(blk_run_queue);
1762
1763/**
Jens Axboe165125e2007-07-24 09:28:11 +02001764 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
Martin Waitza5802902006-04-02 13:59:55 +02001765 * @kobj: the kobj belonging of the request queue to be released
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 *
1767 * Description:
1768 * blk_cleanup_queue is the pair to blk_init_queue() or
1769 * blk_queue_make_request(). It should be called when a request queue is
1770 * being released; typically when a block device is being de-registered.
1771 * Currently, its primary task it to free all the &struct request
1772 * structures that were allocated to the queue and the queue itself.
1773 *
1774 * Caveat:
1775 * Hopefully the low level driver will have finished any
1776 * outstanding requests first...
1777 **/
Al Viro483f4af2006-03-18 18:34:37 -05001778static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779{
Jens Axboe165125e2007-07-24 09:28:11 +02001780 struct request_queue *q =
1781 container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 struct request_list *rl = &q->rq;
1783
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 blk_sync_queue(q);
1785
1786 if (rl->rq_pool)
1787 mempool_destroy(rl->rq_pool);
1788
1789 if (q->queue_tags)
1790 __blk_queue_free_tags(q);
1791
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -07001792 blk_trace_shutdown(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001793
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001794 bdi_destroy(&q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 kmem_cache_free(requestq_cachep, q);
1796}
1797
Jens Axboe165125e2007-07-24 09:28:11 +02001798void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -05001799{
1800 kobject_put(&q->kobj);
1801}
1802EXPORT_SYMBOL(blk_put_queue);
1803
Jens Axboe165125e2007-07-24 09:28:11 +02001804void blk_cleanup_queue(struct request_queue * q)
Al Viro483f4af2006-03-18 18:34:37 -05001805{
1806 mutex_lock(&q->sysfs_lock);
1807 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1808 mutex_unlock(&q->sysfs_lock);
1809
1810 if (q->elevator)
1811 elevator_exit(q->elevator);
1812
1813 blk_put_queue(q);
1814}
1815
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816EXPORT_SYMBOL(blk_cleanup_queue);
1817
Jens Axboe165125e2007-07-24 09:28:11 +02001818static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819{
1820 struct request_list *rl = &q->rq;
1821
1822 rl->count[READ] = rl->count[WRITE] = 0;
1823 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001824 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 init_waitqueue_head(&rl->wait[READ]);
1826 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
Christoph Lameter19460892005-06-23 00:08:19 -07001828 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1829 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830
1831 if (!rl->rq_pool)
1832 return -ENOMEM;
1833
1834 return 0;
1835}
1836
Jens Axboe165125e2007-07-24 09:28:11 +02001837struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838{
Christoph Lameter19460892005-06-23 00:08:19 -07001839 return blk_alloc_queue_node(gfp_mask, -1);
1840}
1841EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
Al Viro483f4af2006-03-18 18:34:37 -05001843static struct kobj_type queue_ktype;
1844
Jens Axboe165125e2007-07-24 09:28:11 +02001845struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001846{
Jens Axboe165125e2007-07-24 09:28:11 +02001847 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001848 int err;
Christoph Lameter19460892005-06-23 00:08:19 -07001849
Christoph Lameter94f60302007-07-17 04:03:29 -07001850 q = kmem_cache_alloc_node(requestq_cachep,
1851 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 if (!q)
1853 return NULL;
1854
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001855 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1856 q->backing_dev_info.unplug_io_data = q;
1857 err = bdi_init(&q->backing_dev_info);
1858 if (err) {
1859 kmem_cache_free(requestq_cachep, q);
1860 return NULL;
1861 }
1862
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001864
Greg Kroah-Hartman19c38de2007-09-12 15:06:57 -07001865 kobject_set_name(&q->kobj, "%s", "queue");
Al Viro483f4af2006-03-18 18:34:37 -05001866 q->kobj.ktype = &queue_ktype;
1867 kobject_init(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868
Al Viro483f4af2006-03-18 18:34:37 -05001869 mutex_init(&q->sysfs_lock);
1870
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 return q;
1872}
Christoph Lameter19460892005-06-23 00:08:19 -07001873EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874
1875/**
1876 * blk_init_queue - prepare a request queue for use with a block device
1877 * @rfn: The function to be called to process requests that have been
1878 * placed on the queue.
1879 * @lock: Request queue spin lock
1880 *
1881 * Description:
1882 * If a block device wishes to use the standard request handling procedures,
1883 * which sorts requests and coalesces adjacent requests, then it must
1884 * call blk_init_queue(). The function @rfn will be called when there
1885 * are requests on the queue that need to be processed. If the device
1886 * supports plugging, then @rfn may not be called immediately when requests
1887 * are available on the queue, but may be called at some time later instead.
1888 * Plugged queues are generally unplugged when a buffer belonging to one
1889 * of the requests on the queue is needed, or due to memory pressure.
1890 *
1891 * @rfn is not required, or even expected, to remove all requests off the
1892 * queue, but only as many as it can handle at a time. If it does leave
1893 * requests on the queue, it is responsible for arranging that the requests
1894 * get dealt with eventually.
1895 *
1896 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001897 * request queue; this lock will be taken also from interrupt context, so irq
1898 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 *
1900 * Function returns a pointer to the initialized request queue, or NULL if
1901 * it didn't succeed.
1902 *
1903 * Note:
1904 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1905 * when the block device is deactivated (such as at module unload).
1906 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001907
Jens Axboe165125e2007-07-24 09:28:11 +02001908struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909{
Christoph Lameter19460892005-06-23 00:08:19 -07001910 return blk_init_queue_node(rfn, lock, -1);
1911}
1912EXPORT_SYMBOL(blk_init_queue);
1913
Jens Axboe165125e2007-07-24 09:28:11 +02001914struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -07001915blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1916{
Jens Axboe165125e2007-07-24 09:28:11 +02001917 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
1919 if (!q)
1920 return NULL;
1921
Christoph Lameter19460892005-06-23 00:08:19 -07001922 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001923 if (blk_init_free_list(q)) {
1924 kmem_cache_free(requestq_cachep, q);
1925 return NULL;
1926 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
152587d2005-04-12 16:22:06 -05001928 /*
1929 * if caller didn't supply a lock, they get per-queue locking with
1930 * our embedded lock
1931 */
1932 if (!lock) {
1933 spin_lock_init(&q->__queue_lock);
1934 lock = &q->__queue_lock;
1935 }
1936
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 q->prep_rq_fn = NULL;
1939 q->unplug_fn = generic_unplug_device;
1940 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1941 q->queue_lock = lock;
1942
1943 blk_queue_segment_boundary(q, 0xffffffff);
1944
1945 blk_queue_make_request(q, __make_request);
1946 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1947
1948 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1949 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1950
Alan Stern44ec9542007-02-20 11:01:57 -05001951 q->sg_reserved_size = INT_MAX;
1952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 /*
1954 * all done
1955 */
1956 if (!elevator_init(q, NULL)) {
1957 blk_queue_congestion_threshold(q);
1958 return q;
1959 }
1960
Al Viro8669aaf2006-03-18 13:50:00 -05001961 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 return NULL;
1963}
Christoph Lameter19460892005-06-23 00:08:19 -07001964EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
Jens Axboe165125e2007-07-24 09:28:11 +02001966int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001968 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001969 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 return 0;
1971 }
1972
1973 return 1;
1974}
1975
1976EXPORT_SYMBOL(blk_get_queue);
1977
Jens Axboe165125e2007-07-24 09:28:11 +02001978static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001980 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +02001981 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 mempool_free(rq, q->rq.rq_pool);
1983}
1984
Jens Axboe1ea25ecb2006-07-18 22:24:11 +02001985static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +02001986blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987{
1988 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1989
1990 if (!rq)
1991 return NULL;
1992
1993 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +02001994 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 * see bio.h and blkdev.h
1996 */
Jens Axboe49171e52006-08-10 08:59:11 +02001997 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
Tejun Heocb98fc82005-10-28 08:29:39 +02001999 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +02002000 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +02002001 mempool_free(rq, q->rq.rq_pool);
2002 return NULL;
2003 }
Jens Axboe4aff5e22006-08-10 08:44:47 +02002004 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +02002005 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
Tejun Heocb98fc82005-10-28 08:29:39 +02002007 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008}
2009
2010/*
2011 * ioc_batching returns true if the ioc is a valid batching request and
2012 * should be given priority access to a request.
2013 */
Jens Axboe165125e2007-07-24 09:28:11 +02002014static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015{
2016 if (!ioc)
2017 return 0;
2018
2019 /*
2020 * Make sure the process is able to allocate at least 1 request
2021 * even if the batch times out, otherwise we could theoretically
2022 * lose wakeups.
2023 */
2024 return ioc->nr_batch_requests == q->nr_batching ||
2025 (ioc->nr_batch_requests > 0
2026 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2027}
2028
2029/*
2030 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2031 * will cause the process to be a "batcher" on all queues in the system. This
2032 * is the behaviour we want though - once it gets a wakeup it should be given
2033 * a nice run.
2034 */
Jens Axboe165125e2007-07-24 09:28:11 +02002035static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036{
2037 if (!ioc || ioc_batching(q, ioc))
2038 return;
2039
2040 ioc->nr_batch_requests = q->nr_batching;
2041 ioc->last_waited = jiffies;
2042}
2043
Jens Axboe165125e2007-07-24 09:28:11 +02002044static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045{
2046 struct request_list *rl = &q->rq;
2047
2048 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07002049 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050
2051 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 if (waitqueue_active(&rl->wait[rw]))
2053 wake_up(&rl->wait[rw]);
2054
2055 blk_clear_queue_full(q, rw);
2056 }
2057}
2058
2059/*
2060 * A request has just been released. Account for it, update the full and
2061 * congestion status, wake up any waiters. Called under q->queue_lock.
2062 */
Jens Axboe165125e2007-07-24 09:28:11 +02002063static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064{
2065 struct request_list *rl = &q->rq;
2066
2067 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002068 if (priv)
2069 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
2071 __freed_request(q, rw);
2072
2073 if (unlikely(rl->starved[rw ^ 1]))
2074 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075}
2076
2077#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2078/*
Nick Piggind6344532005-06-28 20:45:14 -07002079 * Get a free request, queue_lock must be held.
2080 * Returns NULL on failure, with queue_lock held.
2081 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 */
Jens Axboe165125e2007-07-24 09:28:11 +02002083static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +01002084 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085{
2086 struct request *rq = NULL;
2087 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002088 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +01002089 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002090 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
Jens Axboe7749a8d2006-12-13 13:02:26 +01002092 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002093 if (may_queue == ELV_MQUEUE_NO)
2094 goto rq_starved;
2095
2096 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2097 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +02002098 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002099 /*
2100 * The queue will fill after this allocation, so set
2101 * it as full, and mark this process as "batching".
2102 * This process will be allowed to complete a batch of
2103 * requests, others will be blocked.
2104 */
2105 if (!blk_queue_full(q, rw)) {
2106 ioc_set_batching(q, ioc);
2107 blk_set_queue_full(q, rw);
2108 } else {
2109 if (may_queue != ELV_MQUEUE_MUST
2110 && !ioc_batching(q, ioc)) {
2111 /*
2112 * The queue is full and the allocating
2113 * process is not a "batcher", and not
2114 * exempted by the IO scheduler
2115 */
2116 goto out;
2117 }
2118 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 }
Thomas Maier79e2de42006-10-19 23:28:15 -07002120 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 }
2122
Jens Axboe082cf692005-06-28 16:35:11 +02002123 /*
2124 * Only allow batching queuers to allocate up to 50% over the defined
2125 * limit of requests, otherwise we could have thousands of requests
2126 * allocated with any setting of ->nr_requests
2127 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002128 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002129 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002130
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 rl->count[rw]++;
2132 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002133
Jens Axboe64521d12005-10-28 08:30:39 +02002134 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002135 if (priv)
2136 rl->elvpriv++;
2137
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 spin_unlock_irq(q->queue_lock);
2139
Jens Axboe7749a8d2006-12-13 13:02:26 +01002140 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002141 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 /*
2143 * Allocation failed presumably due to memory. Undo anything
2144 * we might have messed up.
2145 *
2146 * Allocating task should really be put onto the front of the
2147 * wait queue, but this is pretty rare.
2148 */
2149 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002150 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151
2152 /*
2153 * in the very unlikely event that allocation failed and no
2154 * requests for this direction was pending, mark us starved
2155 * so that freeing of a request in the other direction will
2156 * notice us. another possible fix would be to split the
2157 * rq mempool into READ and WRITE
2158 */
2159rq_starved:
2160 if (unlikely(rl->count[rw] == 0))
2161 rl->starved[rw] = 1;
2162
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 goto out;
2164 }
2165
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002166 /*
2167 * ioc may be NULL here, and ioc_batching will be false. That's
2168 * OK, if the queue is under the request limit then requests need
2169 * not count toward the nr_batch_requests limit. There will always
2170 * be some limit enforced by BLK_BATCH_TIME.
2171 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 if (ioc_batching(q, ioc))
2173 ioc->nr_batch_requests--;
2174
2175 rq_init(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01002176
2177 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 return rq;
2180}
2181
2182/*
2183 * No available requests for this queue, unplug the device and wait for some
2184 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002185 *
2186 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 */
Jens Axboe165125e2007-07-24 09:28:11 +02002188static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +02002189 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190{
Jens Axboe7749a8d2006-12-13 13:02:26 +01002191 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 struct request *rq;
2193
Jens Axboe7749a8d2006-12-13 13:02:26 +01002194 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -07002195 while (!rq) {
2196 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 struct request_list *rl = &q->rq;
2198
2199 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2200 TASK_UNINTERRUPTIBLE);
2201
Jens Axboe7749a8d2006-12-13 13:02:26 +01002202 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
2204 if (!rq) {
2205 struct io_context *ioc;
2206
Jens Axboe2056a782006-03-23 20:00:26 +01002207 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2208
Nick Piggind6344532005-06-28 20:45:14 -07002209 __generic_unplug_device(q);
2210 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 io_schedule();
2212
2213 /*
2214 * After sleeping, we become a "batching" process and
2215 * will be able to allocate at least one request, and
2216 * up to a big batch of them for a small period time.
2217 * See ioc_batching, ioc_set_batching
2218 */
Jens Axboeb5deef92006-07-19 23:39:40 +02002219 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002221
2222 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 }
2224 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226
2227 return rq;
2228}
2229
Jens Axboe165125e2007-07-24 09:28:11 +02002230struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231{
2232 struct request *rq;
2233
2234 BUG_ON(rw != READ && rw != WRITE);
2235
Nick Piggind6344532005-06-28 20:45:14 -07002236 spin_lock_irq(q->queue_lock);
2237 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002238 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002239 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002240 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002241 if (!rq)
2242 spin_unlock_irq(q->queue_lock);
2243 }
2244 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
2246 return rq;
2247}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248EXPORT_SYMBOL(blk_get_request);
2249
2250/**
Jens Axboedc72ef42006-07-20 14:54:05 +02002251 * blk_start_queueing - initiate dispatch of requests to device
2252 * @q: request queue to kick into gear
2253 *
2254 * This is basically a helper to remove the need to know whether a queue
2255 * is plugged or not if someone just wants to initiate dispatch of requests
2256 * for this queue.
2257 *
2258 * The queue lock must be held with interrupts disabled.
2259 */
Jens Axboe165125e2007-07-24 09:28:11 +02002260void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +02002261{
2262 if (!blk_queue_plugged(q))
2263 q->request_fn(q);
2264 else
2265 __generic_unplug_device(q);
2266}
2267EXPORT_SYMBOL(blk_start_queueing);
2268
2269/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 * blk_requeue_request - put a request back on queue
2271 * @q: request queue where request should be inserted
2272 * @rq: request to be inserted
2273 *
2274 * Description:
2275 * Drivers often keep queueing requests until the hardware cannot accept
2276 * more, when that condition happens we need to put the request back
2277 * on the queue. Must be called with queue lock held.
2278 */
Jens Axboe165125e2007-07-24 09:28:11 +02002279void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280{
Jens Axboe2056a782006-03-23 20:00:26 +01002281 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2282
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 if (blk_rq_tagged(rq))
2284 blk_queue_end_tag(q, rq);
2285
2286 elv_requeue_request(q, rq);
2287}
2288
2289EXPORT_SYMBOL(blk_requeue_request);
2290
2291/**
2292 * blk_insert_request - insert a special request in to a request queue
2293 * @q: request queue where request should be inserted
2294 * @rq: request to be inserted
2295 * @at_head: insert request at head or tail of queue
2296 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 *
2298 * Description:
2299 * Many block devices need to execute commands asynchronously, so they don't
2300 * block the whole kernel from preemption during request execution. This is
2301 * accomplished normally by inserting aritficial requests tagged as
2302 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2303 * scheduled for actual execution by the request queue.
2304 *
2305 * We have the option of inserting the head or the tail of the queue.
2306 * Typically we use the tail for new ioctls and so forth. We use the head
2307 * of the queue for things like a QUEUE_FULL message from a device, or a
2308 * host that is unable to accept a particular command.
2309 */
Jens Axboe165125e2007-07-24 09:28:11 +02002310void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002311 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312{
Tejun Heo 867d1192005-04-24 02:06:05 -05002313 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 unsigned long flags;
2315
2316 /*
2317 * tell I/O scheduler that this isn't a regular read/write (ie it
2318 * must not attempt merges on this) and that it acts as a soft
2319 * barrier
2320 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002321 rq->cmd_type = REQ_TYPE_SPECIAL;
2322 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323
2324 rq->special = data;
2325
2326 spin_lock_irqsave(q->queue_lock, flags);
2327
2328 /*
2329 * If command is tagged, release the tag
2330 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002331 if (blk_rq_tagged(rq))
2332 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333
Jerome Marchandb238b3d2007-10-23 15:05:46 +02002334 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -05002335 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +02002336 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 spin_unlock_irqrestore(q->queue_lock, flags);
2338}
2339
2340EXPORT_SYMBOL(blk_insert_request);
2341
Mike Christie0e75f902006-12-01 10:40:55 +01002342static int __blk_rq_unmap_user(struct bio *bio)
2343{
2344 int ret = 0;
2345
2346 if (bio) {
2347 if (bio_flagged(bio, BIO_USER_MAPPED))
2348 bio_unmap_user(bio);
2349 else
2350 ret = bio_uncopy_user(bio);
2351 }
2352
2353 return ret;
2354}
2355
NeilBrown3001ca72007-08-16 13:31:27 +02002356int blk_rq_append_bio(struct request_queue *q, struct request *rq,
2357 struct bio *bio)
2358{
2359 if (!rq->bio)
2360 blk_rq_bio_prep(q, rq, bio);
2361 else if (!ll_back_merge_fn(q, rq, bio))
2362 return -EINVAL;
2363 else {
2364 rq->biotail->bi_next = bio;
2365 rq->biotail = bio;
2366
2367 rq->data_len += bio->bi_size;
2368 }
2369 return 0;
2370}
2371EXPORT_SYMBOL(blk_rq_append_bio);
2372
Jens Axboe165125e2007-07-24 09:28:11 +02002373static int __blk_rq_map_user(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002374 void __user *ubuf, unsigned int len)
2375{
2376 unsigned long uaddr;
2377 struct bio *bio, *orig_bio;
2378 int reading, ret;
2379
2380 reading = rq_data_dir(rq) == READ;
2381
2382 /*
2383 * if alignment requirement is satisfied, map in user pages for
2384 * direct dma. else, set up kernel bounce buffers
2385 */
2386 uaddr = (unsigned long) ubuf;
2387 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2388 bio = bio_map_user(q, NULL, uaddr, len, reading);
2389 else
2390 bio = bio_copy_user(q, uaddr, len, reading);
2391
Jens Axboe29852592006-12-19 08:27:31 +01002392 if (IS_ERR(bio))
Mike Christie0e75f902006-12-01 10:40:55 +01002393 return PTR_ERR(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002394
2395 orig_bio = bio;
2396 blk_queue_bounce(q, &bio);
Jens Axboe29852592006-12-19 08:27:31 +01002397
Mike Christie0e75f902006-12-01 10:40:55 +01002398 /*
2399 * We link the bounce buffer in and could have to traverse it
2400 * later so we have to get a ref to prevent it from being freed
2401 */
2402 bio_get(bio);
2403
NeilBrown3001ca72007-08-16 13:31:27 +02002404 ret = blk_rq_append_bio(q, rq, bio);
2405 if (!ret)
2406 return bio->bi_size;
Mike Christie0e75f902006-12-01 10:40:55 +01002407
Mike Christie0e75f902006-12-01 10:40:55 +01002408 /* if it was boucned we must call the end io function */
NeilBrown6712ecf2007-09-27 12:47:43 +02002409 bio_endio(bio, 0);
Mike Christie0e75f902006-12-01 10:40:55 +01002410 __blk_rq_unmap_user(orig_bio);
2411 bio_put(bio);
2412 return ret;
2413}
2414
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415/**
2416 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2417 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002418 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 * @ubuf: the user buffer
2420 * @len: length of user data
2421 *
2422 * Description:
2423 * Data will be mapped directly for zero copy io, if possible. Otherwise
2424 * a kernel bounce buffer is used.
2425 *
2426 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2427 * still in process context.
2428 *
2429 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2430 * before being submitted to the device, as pages mapped may be out of
2431 * reach. It's the callers responsibility to make sure this happens. The
2432 * original bio must be passed back in to blk_rq_unmap_user() for proper
2433 * unmapping.
2434 */
Jens Axboe165125e2007-07-24 09:28:11 +02002435int blk_rq_map_user(struct request_queue *q, struct request *rq,
2436 void __user *ubuf, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437{
Mike Christie0e75f902006-12-01 10:40:55 +01002438 unsigned long bytes_read = 0;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002439 struct bio *bio = NULL;
Mike Christie0e75f902006-12-01 10:40:55 +01002440 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Mike Christiedefd94b2005-12-05 02:37:06 -06002442 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002443 return -EINVAL;
2444 if (!len || !ubuf)
2445 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446
Mike Christie0e75f902006-12-01 10:40:55 +01002447 while (bytes_read != len) {
2448 unsigned long map_len, end, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
Mike Christie0e75f902006-12-01 10:40:55 +01002450 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
2451 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
2452 >> PAGE_SHIFT;
2453 start = (unsigned long)ubuf >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454
Mike Christie0e75f902006-12-01 10:40:55 +01002455 /*
2456 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2457 * pages. If this happens we just lower the requested
2458 * mapping len by a page so that we can fit
2459 */
2460 if (end - start > BIO_MAX_PAGES)
2461 map_len -= PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462
Mike Christie0e75f902006-12-01 10:40:55 +01002463 ret = __blk_rq_map_user(q, rq, ubuf, map_len);
2464 if (ret < 0)
2465 goto unmap_rq;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002466 if (!bio)
2467 bio = rq->bio;
Mike Christie0e75f902006-12-01 10:40:55 +01002468 bytes_read += ret;
2469 ubuf += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 }
2471
Mike Christie0e75f902006-12-01 10:40:55 +01002472 rq->buffer = rq->data = NULL;
2473 return 0;
2474unmap_rq:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002475 blk_rq_unmap_user(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002476 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477}
2478
2479EXPORT_SYMBOL(blk_rq_map_user);
2480
2481/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002482 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2483 * @q: request queue where request should be inserted
2484 * @rq: request to map data to
2485 * @iov: pointer to the iovec
2486 * @iov_count: number of elements in the iovec
Randy Dunlapaf9997e2006-12-22 01:06:52 -08002487 * @len: I/O byte count
James Bottomley f1970ba2005-06-20 14:06:52 +02002488 *
2489 * Description:
2490 * Data will be mapped directly for zero copy io, if possible. Otherwise
2491 * a kernel bounce buffer is used.
2492 *
2493 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2494 * still in process context.
2495 *
2496 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2497 * before being submitted to the device, as pages mapped may be out of
2498 * reach. It's the callers responsibility to make sure this happens. The
2499 * original bio must be passed back in to blk_rq_unmap_user() for proper
2500 * unmapping.
2501 */
Jens Axboe165125e2007-07-24 09:28:11 +02002502int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002503 struct sg_iovec *iov, int iov_count, unsigned int len)
James Bottomley f1970ba2005-06-20 14:06:52 +02002504{
2505 struct bio *bio;
2506
2507 if (!iov || iov_count <= 0)
2508 return -EINVAL;
2509
2510 /* we don't allow misaligned data like bio_map_user() does. If the
2511 * user is using sg, they're expected to know the alignment constraints
2512 * and respect them accordingly */
2513 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2514 if (IS_ERR(bio))
2515 return PTR_ERR(bio);
2516
Mike Christie0e75f902006-12-01 10:40:55 +01002517 if (bio->bi_size != len) {
NeilBrown6712ecf2007-09-27 12:47:43 +02002518 bio_endio(bio, 0);
Mike Christie0e75f902006-12-01 10:40:55 +01002519 bio_unmap_user(bio);
2520 return -EINVAL;
2521 }
2522
2523 bio_get(bio);
James Bottomley f1970ba2005-06-20 14:06:52 +02002524 blk_rq_bio_prep(q, rq, bio);
2525 rq->buffer = rq->data = NULL;
James Bottomley f1970ba2005-06-20 14:06:52 +02002526 return 0;
2527}
2528
2529EXPORT_SYMBOL(blk_rq_map_user_iov);
2530
2531/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 * blk_rq_unmap_user - unmap a request with user data
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002533 * @bio: start of bio list
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 *
2535 * Description:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002536 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2537 * supply the original rq->bio from the blk_rq_map_user() return, since
2538 * the io completion may have changed rq->bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 */
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002540int blk_rq_unmap_user(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541{
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002542 struct bio *mapped_bio;
Jens Axboe48785bb2006-12-19 11:07:59 +01002543 int ret = 0, ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002545 while (bio) {
2546 mapped_bio = bio;
2547 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
Mike Christie0e75f902006-12-01 10:40:55 +01002548 mapped_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549
Jens Axboe48785bb2006-12-19 11:07:59 +01002550 ret2 = __blk_rq_unmap_user(mapped_bio);
2551 if (ret2 && !ret)
2552 ret = ret2;
2553
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002554 mapped_bio = bio;
2555 bio = bio->bi_next;
2556 bio_put(mapped_bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002557 }
Jens Axboe48785bb2006-12-19 11:07:59 +01002558
2559 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560}
2561
2562EXPORT_SYMBOL(blk_rq_unmap_user);
2563
2564/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002565 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2566 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002567 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002568 * @kbuf: the kernel buffer
2569 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002570 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002571 */
Jens Axboe165125e2007-07-24 09:28:11 +02002572int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002573 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002574{
Mike Christie df46b9a2005-06-20 14:04:44 +02002575 struct bio *bio;
2576
Mike Christiedefd94b2005-12-05 02:37:06 -06002577 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002578 return -EINVAL;
2579 if (!len || !kbuf)
2580 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002581
2582 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002583 if (IS_ERR(bio))
2584 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002585
Jens Axboedd1cab92005-06-20 14:06:01 +02002586 if (rq_data_dir(rq) == WRITE)
2587 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002588
Jens Axboedd1cab92005-06-20 14:06:01 +02002589 blk_rq_bio_prep(q, rq, bio);
Mike Christie821de3a2007-05-08 19:12:23 +02002590 blk_queue_bounce(q, &rq->bio);
Jens Axboedd1cab92005-06-20 14:06:01 +02002591 rq->buffer = rq->data = NULL;
Jens Axboedd1cab92005-06-20 14:06:01 +02002592 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002593}
2594
2595EXPORT_SYMBOL(blk_rq_map_kern);
2596
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002597/**
2598 * blk_execute_rq_nowait - insert a request into queue for execution
2599 * @q: queue to insert the request in
2600 * @bd_disk: matching gendisk
2601 * @rq: request to insert
2602 * @at_head: insert request at head or tail of queue
2603 * @done: I/O completion handler
2604 *
2605 * Description:
2606 * Insert a fully prepared request at the back of the io scheduler queue
2607 * for execution. Don't wait for completion.
2608 */
Jens Axboe165125e2007-07-24 09:28:11 +02002609void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley f1970ba2005-06-20 14:06:52 +02002610 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002611 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002612{
2613 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2614
2615 rq->rq_disk = bd_disk;
Jens Axboe4aff5e22006-08-10 08:44:47 +02002616 rq->cmd_flags |= REQ_NOMERGE;
James Bottomley f1970ba2005-06-20 14:06:52 +02002617 rq->end_io = done;
Andrew Morton4c5d0bb2006-03-22 08:08:01 +01002618 WARN_ON(irqs_disabled());
2619 spin_lock_irq(q->queue_lock);
2620 __elv_add_request(q, rq, where, 1);
2621 __generic_unplug_device(q);
2622 spin_unlock_irq(q->queue_lock);
James Bottomley f1970ba2005-06-20 14:06:52 +02002623}
Mike Christie6e39b69e2005-11-11 05:30:24 -06002624EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2625
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626/**
2627 * blk_execute_rq - insert a request into queue for execution
2628 * @q: queue to insert the request in
2629 * @bd_disk: matching gendisk
2630 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002631 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 *
2633 * Description:
2634 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002635 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 */
Jens Axboe165125e2007-07-24 09:28:11 +02002637int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002638 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639{
Ingo Molnar60be6b92006-07-03 00:25:26 -07002640 DECLARE_COMPLETION_ONSTACK(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 char sense[SCSI_SENSE_BUFFERSIZE];
2642 int err = 0;
2643
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 /*
2645 * we need an extra reference to the request, so we can look at
2646 * it after io completion
2647 */
2648 rq->ref_count++;
2649
2650 if (!rq->sense) {
2651 memset(sense, 0, sizeof(sense));
2652 rq->sense = sense;
2653 rq->sense_len = 0;
2654 }
2655
Jens Axboec00895a2006-09-30 20:29:12 +02002656 rq->end_io_data = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002657 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 wait_for_completion(&wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659
2660 if (rq->errors)
2661 err = -EIO;
2662
2663 return err;
2664}
2665
2666EXPORT_SYMBOL(blk_execute_rq);
2667
Jens Axboefd5d8062007-10-16 11:05:02 +02002668static void bio_end_empty_barrier(struct bio *bio, int err)
2669{
2670 if (err)
2671 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2672
2673 complete(bio->bi_private);
2674}
2675
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676/**
2677 * blkdev_issue_flush - queue a flush
2678 * @bdev: blockdev to issue flush for
2679 * @error_sector: error sector
2680 *
2681 * Description:
2682 * Issue a flush for the block device in question. Caller can supply
2683 * room for storing the error offset in case of a flush error, if they
2684 * wish to. Caller must run wait_for_completion() on its own.
2685 */
2686int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2687{
Jens Axboefd5d8062007-10-16 11:05:02 +02002688 DECLARE_COMPLETION_ONSTACK(wait);
Jens Axboe165125e2007-07-24 09:28:11 +02002689 struct request_queue *q;
Jens Axboefd5d8062007-10-16 11:05:02 +02002690 struct bio *bio;
2691 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692
2693 if (bdev->bd_disk == NULL)
2694 return -ENXIO;
2695
2696 q = bdev_get_queue(bdev);
2697 if (!q)
2698 return -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
Jens Axboefd5d8062007-10-16 11:05:02 +02002700 bio = bio_alloc(GFP_KERNEL, 0);
2701 if (!bio)
2702 return -ENOMEM;
2703
2704 bio->bi_end_io = bio_end_empty_barrier;
2705 bio->bi_private = &wait;
2706 bio->bi_bdev = bdev;
2707 submit_bio(1 << BIO_RW_BARRIER, bio);
2708
2709 wait_for_completion(&wait);
2710
2711 /*
2712 * The driver must store the error location in ->bi_sector, if
2713 * it supports it. For non-stacked drivers, this should be copied
2714 * from rq->sector.
2715 */
2716 if (error_sector)
2717 *error_sector = bio->bi_sector;
2718
2719 ret = 0;
2720 if (!bio_flagged(bio, BIO_UPTODATE))
2721 ret = -EIO;
2722
2723 bio_put(bio);
2724 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725}
2726
2727EXPORT_SYMBOL(blkdev_issue_flush);
2728
Jerome Marchandb238b3d2007-10-23 15:05:46 +02002729static void drive_stat_acct(struct request *rq, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730{
2731 int rw = rq_data_dir(rq);
2732
2733 if (!blk_fs_request(rq) || !rq->rq_disk)
2734 return;
2735
Jens Axboed72d9042005-11-01 08:35:42 +01002736 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002737 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002738 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 disk_round_stats(rq->rq_disk);
2740 rq->rq_disk->in_flight++;
2741 }
2742}
2743
2744/*
2745 * add-request adds a request to the linked list.
2746 * queue lock is held and interrupts disabled, as we muck with the
2747 * request queue list.
2748 */
Jens Axboe165125e2007-07-24 09:28:11 +02002749static inline void add_request(struct request_queue * q, struct request * req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750{
Jerome Marchandb238b3d2007-10-23 15:05:46 +02002751 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 /*
2754 * elevator indicated where it wants this request to be
2755 * inserted at elevator_merge time
2756 */
2757 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2758}
2759
2760/*
2761 * disk_round_stats() - Round off the performance stats on a struct
2762 * disk_stats.
2763 *
2764 * The average IO queue length and utilisation statistics are maintained
2765 * by observing the current state of the queue length and the amount of
2766 * time it has been in this state for.
2767 *
2768 * Normally, that accounting is done on IO completion, but that can result
2769 * in more than a second's worth of IO being accounted for within any one
2770 * second, leading to >100% utilisation. To deal with that, we call this
2771 * function to do a round-off before returning the results when reading
2772 * /proc/diskstats. This accounts immediately for all queue usage up to
2773 * the current jiffies and restarts the counters again.
2774 */
2775void disk_round_stats(struct gendisk *disk)
2776{
2777 unsigned long now = jiffies;
2778
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002779 if (now == disk->stamp)
2780 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002782 if (disk->in_flight) {
2783 __disk_stat_add(disk, time_in_queue,
2784 disk->in_flight * (now - disk->stamp));
2785 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788}
2789
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002790EXPORT_SYMBOL_GPL(disk_round_stats);
2791
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792/*
2793 * queue lock must be held
2794 */
Jens Axboe165125e2007-07-24 09:28:11 +02002795void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 if (unlikely(!q))
2798 return;
2799 if (unlikely(--req->ref_count))
2800 return;
2801
Tejun Heo8922e162005-10-20 16:23:44 +02002802 elv_completed_request(q, req);
2803
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 /*
2805 * Request may not have originated from ll_rw_blk. if not,
2806 * it didn't come out of our reserved rq pools
2807 */
Jens Axboe49171e52006-08-10 08:59:11 +02002808 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02002810 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02002813 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814
2815 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002816 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 }
2818}
2819
Mike Christie6e39b69e2005-11-11 05:30:24 -06002820EXPORT_SYMBOL_GPL(__blk_put_request);
2821
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822void blk_put_request(struct request *req)
2823{
Tejun Heo8922e162005-10-20 16:23:44 +02002824 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02002825 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826
Tejun Heo8922e162005-10-20 16:23:44 +02002827 /*
2828 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2829 * following if (q) test.
2830 */
2831 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 spin_lock_irqsave(q->queue_lock, flags);
2833 __blk_put_request(q, req);
2834 spin_unlock_irqrestore(q->queue_lock, flags);
2835 }
2836}
2837
2838EXPORT_SYMBOL(blk_put_request);
2839
2840/**
2841 * blk_end_sync_rq - executes a completion event on a request
2842 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002843 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002845void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846{
Jens Axboec00895a2006-09-30 20:29:12 +02002847 struct completion *waiting = rq->end_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848
Jens Axboec00895a2006-09-30 20:29:12 +02002849 rq->end_io_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 __blk_put_request(rq->q, rq);
2851
2852 /*
2853 * complete last, if this is a stack request the process (and thus
2854 * the rq pointer) could be invalid right after this complete()
2855 */
2856 complete(waiting);
2857}
2858EXPORT_SYMBOL(blk_end_sync_rq);
2859
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860/*
2861 * Has to be called with the request spinlock acquired
2862 */
Jens Axboe165125e2007-07-24 09:28:11 +02002863static int attempt_merge(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 struct request *next)
2865{
2866 if (!rq_mergeable(req) || !rq_mergeable(next))
2867 return 0;
2868
2869 /*
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02002870 * not contiguous
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 */
2872 if (req->sector + req->nr_sectors != next->sector)
2873 return 0;
2874
2875 if (rq_data_dir(req) != rq_data_dir(next)
2876 || req->rq_disk != next->rq_disk
Jens Axboec00895a2006-09-30 20:29:12 +02002877 || next->special)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 return 0;
2879
2880 /*
2881 * If we are allowed to merge, then append bio list
2882 * from next to rq and release next. merge_requests_fn
2883 * will have updated segment counts, update sector
2884 * counts here.
2885 */
Jens Axboe1aa4f242006-12-19 08:33:11 +01002886 if (!ll_merge_requests_fn(q, req, next))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 return 0;
2888
2889 /*
2890 * At this point we have either done a back merge
2891 * or front merge. We need the smaller start_time of
2892 * the merged requests to be the current request
2893 * for accounting purposes.
2894 */
2895 if (time_after(req->start_time, next->start_time))
2896 req->start_time = next->start_time;
2897
2898 req->biotail->bi_next = next->bio;
2899 req->biotail = next->biotail;
2900
2901 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2902
2903 elv_merge_requests(q, req, next);
2904
2905 if (req->rq_disk) {
2906 disk_round_stats(req->rq_disk);
2907 req->rq_disk->in_flight--;
2908 }
2909
Jens Axboe22e2c502005-06-27 10:55:12 +02002910 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2911
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 __blk_put_request(q, next);
2913 return 1;
2914}
2915
Jens Axboe165125e2007-07-24 09:28:11 +02002916static inline int attempt_back_merge(struct request_queue *q,
2917 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918{
2919 struct request *next = elv_latter_request(q, rq);
2920
2921 if (next)
2922 return attempt_merge(q, rq, next);
2923
2924 return 0;
2925}
2926
Jens Axboe165125e2007-07-24 09:28:11 +02002927static inline int attempt_front_merge(struct request_queue *q,
2928 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929{
2930 struct request *prev = elv_former_request(q, rq);
2931
2932 if (prev)
2933 return attempt_merge(q, prev, rq);
2934
2935 return 0;
2936}
2937
Tejun Heo52d9e672006-01-06 09:49:58 +01002938static void init_request_from_bio(struct request *req, struct bio *bio)
2939{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002940 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01002941
2942 /*
2943 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2944 */
2945 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002946 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01002947
2948 /*
2949 * REQ_BARRIER implies no merging, but lets make it explicit
2950 */
2951 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002952 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01002953
Jens Axboeb31dc662006-06-13 08:26:10 +02002954 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002955 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02002956 if (bio_rw_meta(bio))
2957 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02002958
Tejun Heo52d9e672006-01-06 09:49:58 +01002959 req->errors = 0;
2960 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01002961 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01002962 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02002963 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01002964}
2965
Jens Axboe165125e2007-07-24 09:28:11 +02002966static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967{
Nick Piggin450991b2005-06-28 20:45:13 -07002968 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02002969 int el_ret, nr_sectors, barrier, err;
2970 const unsigned short prio = bio_prio(bio);
2971 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01002972 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
2976 /*
2977 * low level driver can indicate that it wants pages above a
2978 * certain limit bounced to low memory (ie for highmem, or even
2979 * ISA dma in theory)
2980 */
2981 blk_queue_bounce(q, &bio);
2982
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002984 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 err = -EOPNOTSUPP;
2986 goto end_io;
2987 }
2988
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 spin_lock_irq(q->queue_lock);
2990
Nick Piggin450991b2005-06-28 20:45:13 -07002991 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 goto get_rq;
2993
2994 el_ret = elv_merge(q, &req, bio);
2995 switch (el_ret) {
2996 case ELEVATOR_BACK_MERGE:
2997 BUG_ON(!rq_mergeable(req));
2998
Jens Axboe1aa4f242006-12-19 08:33:11 +01002999 if (!ll_back_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 break;
3001
Jens Axboe2056a782006-03-23 20:00:26 +01003002 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
3003
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 req->biotail->bi_next = bio;
3005 req->biotail = bio;
3006 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02003007 req->ioprio = ioprio_best(req->ioprio, prio);
Jerome Marchandb238b3d2007-10-23 15:05:46 +02003008 drive_stat_acct(req, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009 if (!attempt_back_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02003010 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 goto out;
3012
3013 case ELEVATOR_FRONT_MERGE:
3014 BUG_ON(!rq_mergeable(req));
3015
Jens Axboe1aa4f242006-12-19 08:33:11 +01003016 if (!ll_front_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 break;
3018
Jens Axboe2056a782006-03-23 20:00:26 +01003019 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
3020
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 bio->bi_next = req->bio;
3022 req->bio = bio;
3023
3024 /*
3025 * may not be valid. if the low level driver said
3026 * it didn't need a bounce buffer then it better
3027 * not touch req->buffer either...
3028 */
3029 req->buffer = bio_data(bio);
Jens Axboe51da90f2006-07-18 04:14:45 +02003030 req->current_nr_sectors = bio_cur_sectors(bio);
3031 req->hard_cur_sectors = req->current_nr_sectors;
3032 req->sector = req->hard_sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02003034 req->ioprio = ioprio_best(req->ioprio, prio);
Jerome Marchandb238b3d2007-10-23 15:05:46 +02003035 drive_stat_acct(req, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 if (!attempt_front_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02003037 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 goto out;
3039
Nick Piggin450991b2005-06-28 20:45:13 -07003040 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 default:
Nick Piggin450991b2005-06-28 20:45:13 -07003042 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 }
3044
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07003046 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01003047 * This sync check and mask will be re-done in init_request_from_bio(),
3048 * but we need to set it earlier to expose the sync flag to the
3049 * rq allocator and io schedulers.
3050 */
3051 rw_flags = bio_data_dir(bio);
3052 if (sync)
3053 rw_flags |= REQ_RW_SYNC;
3054
3055 /*
Nick Piggin450991b2005-06-28 20:45:13 -07003056 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07003057 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07003058 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01003059 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07003060
Nick Piggin450991b2005-06-28 20:45:13 -07003061 /*
3062 * After dropping the lock and possibly sleeping here, our request
3063 * may now be mergeable after it had proven unmergeable (above).
3064 * We don't worry about that case for efficiency. It won't happen
3065 * often, and the elevators are able to handle it.
3066 */
Tejun Heo52d9e672006-01-06 09:49:58 +01003067 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068
Nick Piggin450991b2005-06-28 20:45:13 -07003069 spin_lock_irq(q->queue_lock);
3070 if (elv_queue_empty(q))
3071 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 add_request(q, req);
3073out:
Jens Axboe4a534f92005-04-16 15:25:40 -07003074 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 __generic_unplug_device(q);
3076
3077 spin_unlock_irq(q->queue_lock);
3078 return 0;
3079
3080end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02003081 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 return 0;
3083}
3084
3085/*
3086 * If bio->bi_dev is a partition, remap the location
3087 */
3088static inline void blk_partition_remap(struct bio *bio)
3089{
3090 struct block_device *bdev = bio->bi_bdev;
3091
Jens Axboebf2de6f2007-09-27 13:01:25 +02003092 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01003094 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095
Jens Axboea3623572005-11-01 09:26:16 +01003096 p->sectors[rw] += bio_sectors(bio);
3097 p->ios[rw]++;
3098
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 bio->bi_sector += p->start_sect;
3100 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02003101
3102 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
3103 bdev->bd_dev, bio->bi_sector,
3104 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 }
3106}
3107
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108static void handle_bad_sector(struct bio *bio)
3109{
3110 char b[BDEVNAME_SIZE];
3111
3112 printk(KERN_INFO "attempt to access beyond end of device\n");
3113 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3114 bdevname(bio->bi_bdev, b),
3115 bio->bi_rw,
3116 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3117 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3118
3119 set_bit(BIO_EOF, &bio->bi_flags);
3120}
3121
Akinobu Mitac17bb492006-12-08 02:39:46 -08003122#ifdef CONFIG_FAIL_MAKE_REQUEST
3123
3124static DECLARE_FAULT_ATTR(fail_make_request);
3125
3126static int __init setup_fail_make_request(char *str)
3127{
3128 return setup_fault_attr(&fail_make_request, str);
3129}
3130__setup("fail_make_request=", setup_fail_make_request);
3131
3132static int should_fail_request(struct bio *bio)
3133{
3134 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
3135 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
3136 return should_fail(&fail_make_request, bio->bi_size);
3137
3138 return 0;
3139}
3140
3141static int __init fail_make_request_debugfs(void)
3142{
3143 return init_fault_attr_dentries(&fail_make_request,
3144 "fail_make_request");
3145}
3146
3147late_initcall(fail_make_request_debugfs);
3148
3149#else /* CONFIG_FAIL_MAKE_REQUEST */
3150
3151static inline int should_fail_request(struct bio *bio)
3152{
3153 return 0;
3154}
3155
3156#endif /* CONFIG_FAIL_MAKE_REQUEST */
3157
Jens Axboec07e2b42007-07-18 13:27:58 +02003158/*
3159 * Check whether this bio extends beyond the end of the device.
3160 */
3161static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
3162{
3163 sector_t maxsector;
3164
3165 if (!nr_sectors)
3166 return 0;
3167
3168 /* Test device or partition size, when known. */
3169 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3170 if (maxsector) {
3171 sector_t sector = bio->bi_sector;
3172
3173 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3174 /*
3175 * This may well happen - the kernel calls bread()
3176 * without checking the size of the device, e.g., when
3177 * mounting a device.
3178 */
3179 handle_bad_sector(bio);
3180 return 1;
3181 }
3182 }
3183
3184 return 0;
3185}
3186
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187/**
3188 * generic_make_request: hand a buffer to its device driver for I/O
3189 * @bio: The bio describing the location in memory and on the device.
3190 *
3191 * generic_make_request() is used to make I/O requests of block
3192 * devices. It is passed a &struct bio, which describes the I/O that needs
3193 * to be done.
3194 *
3195 * generic_make_request() does not return any status. The
3196 * success/failure status of the request, along with notification of
3197 * completion, is delivered asynchronously through the bio->bi_end_io
3198 * function described (one day) else where.
3199 *
3200 * The caller of generic_make_request must make sure that bi_io_vec
3201 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3202 * set to describe the device address, and the
3203 * bi_end_io and optionally bi_private are set to describe how
3204 * completion notification should be signaled.
3205 *
3206 * generic_make_request and the drivers it calls may use bi_next if this
3207 * bio happens to be merged with someone else, and may change bi_dev and
3208 * bi_sector for remaps as it sees fit. So the values of these fields
3209 * should NOT be depended on after the call to generic_make_request.
3210 */
Neil Brownd89d8792007-05-01 09:53:42 +02003211static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212{
Jens Axboe165125e2007-07-24 09:28:11 +02003213 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08003214 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01003216 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01003217 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218
3219 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220
Jens Axboec07e2b42007-07-18 13:27:58 +02003221 if (bio_check_eod(bio, nr_sectors))
3222 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223
3224 /*
3225 * Resolve the mapping until finished. (drivers are
3226 * still free to implement/resolve their own stacking
3227 * by explicitly returning 0)
3228 *
3229 * NOTE: we don't repeat the blk_size check for each new device.
3230 * Stacking drivers are expected to know what they are doing.
3231 */
NeilBrown5ddfe962006-10-30 22:07:21 -08003232 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01003233 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234 do {
3235 char b[BDEVNAME_SIZE];
3236
3237 q = bdev_get_queue(bio->bi_bdev);
3238 if (!q) {
3239 printk(KERN_ERR
3240 "generic_make_request: Trying to access "
3241 "nonexistent block-device %s (%Lu)\n",
3242 bdevname(bio->bi_bdev, b),
3243 (long long) bio->bi_sector);
3244end_io:
Jens Axboe51fd77b2007-11-02 08:49:08 +01003245 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246 break;
3247 }
3248
Jens Axboe4fa253f2007-07-18 13:13:10 +02003249 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250 printk("bio too big device %s (%u > %u)\n",
3251 bdevname(bio->bi_bdev, b),
3252 bio_sectors(bio),
3253 q->max_hw_sectors);
3254 goto end_io;
3255 }
3256
Nick Pigginfde6ad22005-06-23 00:08:53 -07003257 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 goto end_io;
3259
Akinobu Mitac17bb492006-12-08 02:39:46 -08003260 if (should_fail_request(bio))
3261 goto end_io;
3262
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 /*
3264 * If this device has partitions, remap block n
3265 * of partition p to block n+start(p) of the disk.
3266 */
3267 blk_partition_remap(bio);
3268
NeilBrown5ddfe962006-10-30 22:07:21 -08003269 if (old_sector != -1)
Jens Axboe4fa253f2007-07-18 13:13:10 +02003270 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08003271 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01003272
3273 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3274
NeilBrown5ddfe962006-10-30 22:07:21 -08003275 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01003276 old_dev = bio->bi_bdev->bd_dev;
3277
Jens Axboec07e2b42007-07-18 13:27:58 +02003278 if (bio_check_eod(bio, nr_sectors))
3279 goto end_io;
Jens Axboe51fd77b2007-11-02 08:49:08 +01003280 if (bio_empty_barrier(bio) && !q->prepare_flush_fn) {
3281 err = -EOPNOTSUPP;
3282 goto end_io;
3283 }
NeilBrown5ddfe962006-10-30 22:07:21 -08003284
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 ret = q->make_request_fn(q, bio);
3286 } while (ret);
3287}
3288
Neil Brownd89d8792007-05-01 09:53:42 +02003289/*
3290 * We only want one ->make_request_fn to be active at a time,
3291 * else stack usage with stacked devices could be a problem.
3292 * So use current->bio_{list,tail} to keep a list of requests
3293 * submited by a make_request_fn function.
3294 * current->bio_tail is also used as a flag to say if
3295 * generic_make_request is currently active in this task or not.
3296 * If it is NULL, then no make_request is active. If it is non-NULL,
3297 * then a make_request is active, and new requests should be added
3298 * at the tail
3299 */
3300void generic_make_request(struct bio *bio)
3301{
3302 if (current->bio_tail) {
3303 /* make_request is active */
3304 *(current->bio_tail) = bio;
3305 bio->bi_next = NULL;
3306 current->bio_tail = &bio->bi_next;
3307 return;
3308 }
3309 /* following loop may be a bit non-obvious, and so deserves some
3310 * explanation.
3311 * Before entering the loop, bio->bi_next is NULL (as all callers
3312 * ensure that) so we have a list with a single bio.
3313 * We pretend that we have just taken it off a longer list, so
3314 * we assign bio_list to the next (which is NULL) and bio_tail
3315 * to &bio_list, thus initialising the bio_list of new bios to be
3316 * added. __generic_make_request may indeed add some more bios
3317 * through a recursive call to generic_make_request. If it
3318 * did, we find a non-NULL value in bio_list and re-enter the loop
3319 * from the top. In this case we really did just take the bio
3320 * of the top of the list (no pretending) and so fixup bio_list and
3321 * bio_tail or bi_next, and call into __generic_make_request again.
3322 *
3323 * The loop was structured like this to make only one call to
3324 * __generic_make_request (which is important as it is large and
3325 * inlined) and to keep the structure simple.
3326 */
3327 BUG_ON(bio->bi_next);
3328 do {
3329 current->bio_list = bio->bi_next;
3330 if (bio->bi_next == NULL)
3331 current->bio_tail = &current->bio_list;
3332 else
3333 bio->bi_next = NULL;
3334 __generic_make_request(bio);
3335 bio = current->bio_list;
3336 } while (bio);
3337 current->bio_tail = NULL; /* deactivate */
3338}
3339
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340EXPORT_SYMBOL(generic_make_request);
3341
3342/**
3343 * submit_bio: submit a bio to the block device layer for I/O
3344 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3345 * @bio: The &struct bio which describes the I/O
3346 *
3347 * submit_bio() is very similar in purpose to generic_make_request(), and
3348 * uses that function to do most of the work. Both are fairly rough
3349 * interfaces, @bio must be presetup and ready for I/O.
3350 *
3351 */
3352void submit_bio(int rw, struct bio *bio)
3353{
3354 int count = bio_sectors(bio);
3355
Jens Axboe22e2c502005-06-27 10:55:12 +02003356 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357
Jens Axboebf2de6f2007-09-27 13:01:25 +02003358 /*
3359 * If it's a regular read/write or a barrier with data attached,
3360 * go through the normal accounting stuff before submission.
3361 */
3362 if (!bio_empty_barrier(bio)) {
3363
3364 BIO_BUG_ON(!bio->bi_size);
3365 BIO_BUG_ON(!bio->bi_io_vec);
3366
3367 if (rw & WRITE) {
3368 count_vm_events(PGPGOUT, count);
3369 } else {
3370 task_io_account_read(bio->bi_size);
3371 count_vm_events(PGPGIN, count);
3372 }
3373
3374 if (unlikely(block_dump)) {
3375 char b[BDEVNAME_SIZE];
3376 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07003377 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02003378 (rw & WRITE) ? "WRITE" : "READ",
3379 (unsigned long long)bio->bi_sector,
3380 bdevname(bio->bi_bdev,b));
3381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382 }
3383
3384 generic_make_request(bio);
3385}
3386
3387EXPORT_SYMBOL(submit_bio);
3388
Adrian Bunk93d17d32005-06-25 14:59:10 -07003389static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390{
3391 if (blk_fs_request(rq)) {
3392 rq->hard_sector += nsect;
3393 rq->hard_nr_sectors -= nsect;
3394
3395 /*
3396 * Move the I/O submission pointers ahead if required.
3397 */
3398 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3399 (rq->sector <= rq->hard_sector)) {
3400 rq->sector = rq->hard_sector;
3401 rq->nr_sectors = rq->hard_nr_sectors;
3402 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3403 rq->current_nr_sectors = rq->hard_cur_sectors;
3404 rq->buffer = bio_data(rq->bio);
3405 }
3406
3407 /*
3408 * if total number of sectors is less than the first segment
3409 * size, something has gone terribly wrong
3410 */
3411 if (rq->nr_sectors < rq->current_nr_sectors) {
3412 printk("blk: request botched\n");
3413 rq->nr_sectors = rq->current_nr_sectors;
3414 }
3415 }
3416}
3417
3418static int __end_that_request_first(struct request *req, int uptodate,
3419 int nr_bytes)
3420{
3421 int total_bytes, bio_nbytes, error, next_idx = 0;
3422 struct bio *bio;
3423
Jens Axboe2056a782006-03-23 20:00:26 +01003424 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3425
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 /*
3427 * extend uptodate bool to allow < 0 value to be direct io error
3428 */
3429 error = 0;
3430 if (end_io_error(uptodate))
3431 error = !uptodate ? -EIO : uptodate;
3432
3433 /*
3434 * for a REQ_BLOCK_PC request, we want to carry any eventual
3435 * sense key with us all the way through
3436 */
3437 if (!blk_pc_request(req))
3438 req->errors = 0;
3439
3440 if (!uptodate) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02003441 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442 printk("end_request: I/O error, dev %s, sector %llu\n",
3443 req->rq_disk ? req->rq_disk->disk_name : "?",
3444 (unsigned long long)req->sector);
3445 }
3446
Jens Axboed72d9042005-11-01 08:35:42 +01003447 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003448 const int rw = rq_data_dir(req);
3449
Jens Axboe53e86062006-01-17 11:09:27 +01003450 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003451 }
3452
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 total_bytes = bio_nbytes = 0;
3454 while ((bio = req->bio) != NULL) {
3455 int nbytes;
3456
Jens Axboebf2de6f2007-09-27 13:01:25 +02003457 /*
3458 * For an empty barrier request, the low level driver must
3459 * store a potential error location in ->sector. We pass
3460 * that back up in ->bi_sector.
3461 */
3462 if (blk_empty_barrier(req))
3463 bio->bi_sector = req->sector;
3464
Linus Torvalds1da177e2005-04-16 15:20:36 -07003465 if (nr_bytes >= bio->bi_size) {
3466 req->bio = bio->bi_next;
3467 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02003468 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469 next_idx = 0;
3470 bio_nbytes = 0;
3471 } else {
3472 int idx = bio->bi_idx + next_idx;
3473
3474 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3475 blk_dump_rq_flags(req, "__end_that");
3476 printk("%s: bio idx %d >= vcnt %d\n",
3477 __FUNCTION__,
3478 bio->bi_idx, bio->bi_vcnt);
3479 break;
3480 }
3481
3482 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3483 BIO_BUG_ON(nbytes > bio->bi_size);
3484
3485 /*
3486 * not a complete bvec done
3487 */
3488 if (unlikely(nbytes > nr_bytes)) {
3489 bio_nbytes += nr_bytes;
3490 total_bytes += nr_bytes;
3491 break;
3492 }
3493
3494 /*
3495 * advance to the next vector
3496 */
3497 next_idx++;
3498 bio_nbytes += nbytes;
3499 }
3500
3501 total_bytes += nbytes;
3502 nr_bytes -= nbytes;
3503
3504 if ((bio = req->bio)) {
3505 /*
3506 * end more in this run, or just return 'not-done'
3507 */
3508 if (unlikely(nr_bytes <= 0))
3509 break;
3510 }
3511 }
3512
3513 /*
3514 * completely done
3515 */
3516 if (!req->bio)
3517 return 0;
3518
3519 /*
3520 * if the request wasn't completed, update state
3521 */
3522 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02003523 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524 bio->bi_idx += next_idx;
3525 bio_iovec(bio)->bv_offset += nr_bytes;
3526 bio_iovec(bio)->bv_len -= nr_bytes;
3527 }
3528
3529 blk_recalc_rq_sectors(req, total_bytes >> 9);
3530 blk_recalc_rq_segments(req);
3531 return 1;
3532}
3533
3534/**
3535 * end_that_request_first - end I/O on a request
3536 * @req: the request being processed
3537 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3538 * @nr_sectors: number of sectors to end I/O on
3539 *
3540 * Description:
3541 * Ends I/O on a number of sectors attached to @req, and sets it up
3542 * for the next range of segments (if any) in the cluster.
3543 *
3544 * Return:
3545 * 0 - we are done with this request, call end_that_request_last()
3546 * 1 - still buffers pending for this request
3547 **/
3548int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3549{
3550 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3551}
3552
3553EXPORT_SYMBOL(end_that_request_first);
3554
3555/**
3556 * end_that_request_chunk - end I/O on a request
3557 * @req: the request being processed
3558 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3559 * @nr_bytes: number of bytes to complete
3560 *
3561 * Description:
3562 * Ends I/O on a number of bytes attached to @req, and sets it up
3563 * for the next range of segments (if any). Like end_that_request_first(),
3564 * but deals with bytes instead of sectors.
3565 *
3566 * Return:
3567 * 0 - we are done with this request, call end_that_request_last()
3568 * 1 - still buffers pending for this request
3569 **/
3570int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3571{
3572 return __end_that_request_first(req, uptodate, nr_bytes);
3573}
3574
3575EXPORT_SYMBOL(end_that_request_chunk);
3576
3577/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003578 * splice the completion data to a local structure and hand off to
3579 * process_completion_queue() to complete the requests
3580 */
3581static void blk_done_softirq(struct softirq_action *h)
3582{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003583 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01003584
3585 local_irq_disable();
3586 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003587 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01003588 local_irq_enable();
3589
3590 while (!list_empty(&local_list)) {
3591 struct request *rq = list_entry(local_list.next, struct request, donelist);
3592
3593 list_del_init(&rq->donelist);
3594 rq->q->softirq_done_fn(rq);
3595 }
3596}
3597
Satyam Sharmadb47d472007-08-23 09:29:40 +02003598static int __cpuinit blk_cpu_notify(struct notifier_block *self, unsigned long action,
Jens Axboeff856ba2006-01-09 16:02:34 +01003599 void *hcpu)
3600{
3601 /*
3602 * If a CPU goes away, splice its entries to the current CPU
3603 * and trigger a run of the softirq
3604 */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003605 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Jens Axboeff856ba2006-01-09 16:02:34 +01003606 int cpu = (unsigned long) hcpu;
3607
3608 local_irq_disable();
3609 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3610 &__get_cpu_var(blk_cpu_done));
3611 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3612 local_irq_enable();
3613 }
3614
3615 return NOTIFY_OK;
3616}
3617
3618
Satyam Sharmadb47d472007-08-23 09:29:40 +02003619static struct notifier_block blk_cpu_notifier __cpuinitdata = {
Jens Axboeff856ba2006-01-09 16:02:34 +01003620 .notifier_call = blk_cpu_notify,
3621};
3622
Jens Axboeff856ba2006-01-09 16:02:34 +01003623/**
3624 * blk_complete_request - end I/O on a request
3625 * @req: the request being processed
3626 *
3627 * Description:
3628 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02003629 * unless the driver actually implements this in its completion callback
Jens Axboe4fa253f2007-07-18 13:13:10 +02003630 * through requeueing. The actual completion happens out-of-order,
Jens Axboeff856ba2006-01-09 16:02:34 +01003631 * through a softirq handler. The user must have registered a completion
3632 * callback through blk_queue_softirq_done().
3633 **/
3634
3635void blk_complete_request(struct request *req)
3636{
3637 struct list_head *cpu_list;
3638 unsigned long flags;
3639
3640 BUG_ON(!req->q->softirq_done_fn);
3641
3642 local_irq_save(flags);
3643
3644 cpu_list = &__get_cpu_var(blk_cpu_done);
3645 list_add_tail(&req->donelist, cpu_list);
3646 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3647
3648 local_irq_restore(flags);
3649}
3650
3651EXPORT_SYMBOL(blk_complete_request);
3652
3653/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654 * queue lock must be held
3655 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003656void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657{
3658 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003659 int error;
3660
3661 /*
3662 * extend uptodate bool to allow < 0 value to be direct io error
3663 */
3664 error = 0;
3665 if (end_io_error(uptodate))
3666 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667
3668 if (unlikely(laptop_mode) && blk_fs_request(req))
3669 laptop_io_completion();
3670
Jens Axboefd0ff8a2006-05-23 11:23:49 +02003671 /*
3672 * Account IO completion. bar_rq isn't accounted as a normal
3673 * IO on queueing nor completion. Accounting the containing
3674 * request is enough.
3675 */
3676 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003677 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003678 const int rw = rq_data_dir(req);
3679
3680 __disk_stat_inc(disk, ios[rw]);
3681 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682 disk_round_stats(disk);
3683 disk->in_flight--;
3684 }
3685 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003686 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 else
3688 __blk_put_request(req->q, req);
3689}
3690
3691EXPORT_SYMBOL(end_that_request_last);
3692
Jens Axboea0cd1282007-09-21 10:41:07 +02003693static inline void __end_request(struct request *rq, int uptodate,
3694 unsigned int nr_bytes, int dequeue)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695{
Jens Axboea0cd1282007-09-21 10:41:07 +02003696 if (!end_that_request_chunk(rq, uptodate, nr_bytes)) {
3697 if (dequeue)
3698 blkdev_dequeue_request(rq);
3699 add_disk_randomness(rq->rq_disk);
3700 end_that_request_last(rq, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701 }
3702}
3703
Jens Axboea0cd1282007-09-21 10:41:07 +02003704static unsigned int rq_byte_size(struct request *rq)
3705{
3706 if (blk_fs_request(rq))
3707 return rq->hard_nr_sectors << 9;
3708
3709 return rq->data_len;
3710}
3711
3712/**
3713 * end_queued_request - end all I/O on a queued request
3714 * @rq: the request being processed
3715 * @uptodate: error value or 0/1 uptodate flag
3716 *
3717 * Description:
3718 * Ends all I/O on a request, and removes it from the block layer queues.
3719 * Not suitable for normal IO completion, unless the driver still has
3720 * the request attached to the block layer.
3721 *
3722 **/
3723void end_queued_request(struct request *rq, int uptodate)
3724{
3725 __end_request(rq, uptodate, rq_byte_size(rq), 1);
3726}
3727EXPORT_SYMBOL(end_queued_request);
3728
3729/**
3730 * end_dequeued_request - end all I/O on a dequeued request
3731 * @rq: the request being processed
3732 * @uptodate: error value or 0/1 uptodate flag
3733 *
3734 * Description:
3735 * Ends all I/O on a request. The request must already have been
3736 * dequeued using blkdev_dequeue_request(), as is normally the case
3737 * for most drivers.
3738 *
3739 **/
3740void end_dequeued_request(struct request *rq, int uptodate)
3741{
3742 __end_request(rq, uptodate, rq_byte_size(rq), 0);
3743}
3744EXPORT_SYMBOL(end_dequeued_request);
3745
3746
3747/**
3748 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07003749 * @req: the request being processed
Jens Axboea0cd1282007-09-21 10:41:07 +02003750 * @uptodate: error value or 0/1 uptodate flag
3751 *
3752 * Description:
3753 * Ends I/O on the current segment of a request. If that is the only
3754 * remaining segment, the request is also completed and freed.
3755 *
3756 * This is a remnant of how older block drivers handled IO completions.
3757 * Modern drivers typically end IO on the full request in one go, unless
3758 * they have a residual value to account for. For that case this function
3759 * isn't really useful, unless the residual just happens to be the
3760 * full current segment. In other words, don't use this function in new
3761 * code. Either use end_request_completely(), or the
3762 * end_that_request_chunk() (along with end_that_request_last()) for
3763 * partial completions.
3764 *
3765 **/
3766void end_request(struct request *req, int uptodate)
3767{
3768 __end_request(req, uptodate, req->hard_cur_sectors << 9, 1);
3769}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770EXPORT_SYMBOL(end_request);
3771
NeilBrown66846572007-08-16 13:31:28 +02003772static void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3773 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003774{
Jens Axboe4aff5e22006-08-10 08:44:47 +02003775 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3776 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777
3778 rq->nr_phys_segments = bio_phys_segments(q, bio);
3779 rq->nr_hw_segments = bio_hw_segments(q, bio);
3780 rq->current_nr_sectors = bio_cur_sectors(bio);
3781 rq->hard_cur_sectors = rq->current_nr_sectors;
3782 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3783 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01003784 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785
3786 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787
NeilBrown66846572007-08-16 13:31:28 +02003788 if (bio->bi_bdev)
3789 rq->rq_disk = bio->bi_bdev->bd_disk;
3790}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791
3792int kblockd_schedule_work(struct work_struct *work)
3793{
3794 return queue_work(kblockd_workqueue, work);
3795}
3796
3797EXPORT_SYMBOL(kblockd_schedule_work);
3798
Andrew Morton19a75d82007-05-09 02:33:56 -07003799void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07003801 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802}
Andrew Morton19a75d82007-05-09 02:33:56 -07003803EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804
3805int __init blk_dev_init(void)
3806{
Jens Axboeff856ba2006-01-09 16:02:34 +01003807 int i;
3808
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809 kblockd_workqueue = create_workqueue("kblockd");
3810 if (!kblockd_workqueue)
3811 panic("Failed to create kblockd\n");
3812
3813 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003814 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815
3816 requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003817 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818
3819 iocontext_cachep = kmem_cache_create("blkdev_ioc",
Paul Mundt20c2df82007-07-20 10:11:58 +09003820 sizeof(struct io_context), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08003822 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003823 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3824
3825 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07003826 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01003827
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02003828 blk_max_low_pfn = max_low_pfn - 1;
3829 blk_max_pfn = max_pfn - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830
3831 return 0;
3832}
3833
3834/*
3835 * IO Context helper functions
3836 */
3837void put_io_context(struct io_context *ioc)
3838{
3839 if (ioc == NULL)
3840 return;
3841
3842 BUG_ON(atomic_read(&ioc->refcount) == 0);
3843
3844 if (atomic_dec_and_test(&ioc->refcount)) {
Jens Axboee2d74ac2006-03-28 08:59:01 +02003845 struct cfq_io_context *cic;
3846
Al Viro334e94d2006-03-18 15:05:53 -05003847 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003848 if (ioc->aic && ioc->aic->dtor)
3849 ioc->aic->dtor(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003850 if (ioc->cic_root.rb_node != NULL) {
Jens Axboe7143dd42006-03-28 09:00:28 +02003851 struct rb_node *n = rb_first(&ioc->cic_root);
3852
3853 cic = rb_entry(n, struct cfq_io_context, rb_node);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003854 cic->dtor(ioc);
3855 }
Al Viro334e94d2006-03-18 15:05:53 -05003856 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857
3858 kmem_cache_free(iocontext_cachep, ioc);
3859 }
3860}
3861EXPORT_SYMBOL(put_io_context);
3862
3863/* Called by the exitting task */
3864void exit_io_context(void)
3865{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866 struct io_context *ioc;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003867 struct cfq_io_context *cic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868
Jens Axboe22e2c502005-06-27 10:55:12 +02003869 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003870 ioc = current->io_context;
3871 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003872 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873
Oleg Nesterov25034d72006-08-29 09:15:14 +02003874 ioc->task = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 if (ioc->aic && ioc->aic->exit)
3876 ioc->aic->exit(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003877 if (ioc->cic_root.rb_node != NULL) {
3878 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3879 cic->exit(ioc);
3880 }
Oleg Nesterov25034d72006-08-29 09:15:14 +02003881
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 put_io_context(ioc);
3883}
3884
3885/*
3886 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003887 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003889 * This returned IO context doesn't have a specifically elevated refcount,
3890 * but since the current task itself holds a reference, the context can be
3891 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003893static struct io_context *current_io_context(gfp_t gfp_flags, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894{
3895 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896 struct io_context *ret;
3897
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003899 if (likely(ret))
3900 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901
Jens Axboeb5deef92006-07-19 23:39:40 +02003902 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903 if (ret) {
3904 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003905 ret->task = current;
Jens Axboefc463792006-08-29 09:05:44 +02003906 ret->ioprio_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907 ret->last_waited = jiffies; /* doesn't matter... */
3908 ret->nr_batch_requests = 0; /* because this is 0 */
3909 ret->aic = NULL;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003910 ret->cic_root.rb_node = NULL;
Jens Axboe4e521c22007-04-24 21:17:33 +02003911 ret->ioc_data = NULL;
Oleg Nesterov9f83e452006-08-21 08:34:15 +02003912 /* make sure set_task_ioprio() sees the settings above */
3913 smp_wmb();
Nick Pigginfb3cc432005-06-28 20:45:15 -07003914 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915 }
3916
3917 return ret;
3918}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003919
3920/*
3921 * If the current task has no IO context then create one and initialise it.
3922 * If it does have a context, take a ref on it.
3923 *
3924 * This is always called in the context of the task which submitted the I/O.
3925 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003926struct io_context *get_io_context(gfp_t gfp_flags, int node)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003927{
3928 struct io_context *ret;
Jens Axboeb5deef92006-07-19 23:39:40 +02003929 ret = current_io_context(gfp_flags, node);
Nick Pigginfb3cc432005-06-28 20:45:15 -07003930 if (likely(ret))
3931 atomic_inc(&ret->refcount);
3932 return ret;
3933}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934EXPORT_SYMBOL(get_io_context);
3935
3936void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3937{
3938 struct io_context *src = *psrc;
3939 struct io_context *dst = *pdst;
3940
3941 if (src) {
3942 BUG_ON(atomic_read(&src->refcount) == 0);
3943 atomic_inc(&src->refcount);
3944 put_io_context(dst);
3945 *pdst = src;
3946 }
3947}
3948EXPORT_SYMBOL(copy_io_context);
3949
3950void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3951{
3952 struct io_context *temp;
3953 temp = *ioc1;
3954 *ioc1 = *ioc2;
3955 *ioc2 = temp;
3956}
3957EXPORT_SYMBOL(swap_io_context);
3958
3959/*
3960 * sysfs parts below
3961 */
3962struct queue_sysfs_entry {
3963 struct attribute attr;
3964 ssize_t (*show)(struct request_queue *, char *);
3965 ssize_t (*store)(struct request_queue *, const char *, size_t);
3966};
3967
3968static ssize_t
3969queue_var_show(unsigned int var, char *page)
3970{
3971 return sprintf(page, "%d\n", var);
3972}
3973
3974static ssize_t
3975queue_var_store(unsigned long *var, const char *page, size_t count)
3976{
3977 char *p = (char *) page;
3978
3979 *var = simple_strtoul(p, &p, 10);
3980 return count;
3981}
3982
3983static ssize_t queue_requests_show(struct request_queue *q, char *page)
3984{
3985 return queue_var_show(q->nr_requests, (page));
3986}
3987
3988static ssize_t
3989queue_requests_store(struct request_queue *q, const char *page, size_t count)
3990{
3991 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05003992 unsigned long nr;
3993 int ret = queue_var_store(&nr, page, count);
3994 if (nr < BLKDEV_MIN_RQ)
3995 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996
Al Viroc981ff92006-03-18 13:51:29 -05003997 spin_lock_irq(q->queue_lock);
3998 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999 blk_queue_congestion_threshold(q);
4000
4001 if (rl->count[READ] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07004002 blk_set_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003 else if (rl->count[READ] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07004004 blk_clear_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005
4006 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07004007 blk_set_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07004009 blk_clear_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004010
4011 if (rl->count[READ] >= q->nr_requests) {
4012 blk_set_queue_full(q, READ);
4013 } else if (rl->count[READ]+1 <= q->nr_requests) {
4014 blk_clear_queue_full(q, READ);
4015 wake_up(&rl->wait[READ]);
4016 }
4017
4018 if (rl->count[WRITE] >= q->nr_requests) {
4019 blk_set_queue_full(q, WRITE);
4020 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
4021 blk_clear_queue_full(q, WRITE);
4022 wake_up(&rl->wait[WRITE]);
4023 }
Al Viroc981ff92006-03-18 13:51:29 -05004024 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025 return ret;
4026}
4027
4028static ssize_t queue_ra_show(struct request_queue *q, char *page)
4029{
4030 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
4031
4032 return queue_var_show(ra_kb, (page));
4033}
4034
4035static ssize_t
4036queue_ra_store(struct request_queue *q, const char *page, size_t count)
4037{
4038 unsigned long ra_kb;
4039 ssize_t ret = queue_var_store(&ra_kb, page, count);
4040
4041 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
4043 spin_unlock_irq(q->queue_lock);
4044
4045 return ret;
4046}
4047
4048static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
4049{
4050 int max_sectors_kb = q->max_sectors >> 1;
4051
4052 return queue_var_show(max_sectors_kb, (page));
4053}
4054
4055static ssize_t
4056queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
4057{
4058 unsigned long max_sectors_kb,
4059 max_hw_sectors_kb = q->max_hw_sectors >> 1,
4060 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
4061 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062
4063 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
4064 return -EINVAL;
4065 /*
4066 * Take the queue lock to update the readahead and max_sectors
4067 * values synchronously:
4068 */
4069 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070 q->max_sectors = max_sectors_kb << 1;
4071 spin_unlock_irq(q->queue_lock);
4072
4073 return ret;
4074}
4075
4076static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
4077{
4078 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
4079
4080 return queue_var_show(max_hw_sectors_kb, (page));
4081}
4082
4083
4084static struct queue_sysfs_entry queue_requests_entry = {
4085 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
4086 .show = queue_requests_show,
4087 .store = queue_requests_store,
4088};
4089
4090static struct queue_sysfs_entry queue_ra_entry = {
4091 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
4092 .show = queue_ra_show,
4093 .store = queue_ra_store,
4094};
4095
4096static struct queue_sysfs_entry queue_max_sectors_entry = {
4097 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
4098 .show = queue_max_sectors_show,
4099 .store = queue_max_sectors_store,
4100};
4101
4102static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
4103 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
4104 .show = queue_max_hw_sectors_show,
4105};
4106
4107static struct queue_sysfs_entry queue_iosched_entry = {
4108 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
4109 .show = elv_iosched_show,
4110 .store = elv_iosched_store,
4111};
4112
4113static struct attribute *default_attrs[] = {
4114 &queue_requests_entry.attr,
4115 &queue_ra_entry.attr,
4116 &queue_max_hw_sectors_entry.attr,
4117 &queue_max_sectors_entry.attr,
4118 &queue_iosched_entry.attr,
4119 NULL,
4120};
4121
4122#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4123
4124static ssize_t
4125queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4126{
4127 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004128 struct request_queue *q =
4129 container_of(kobj, struct request_queue, kobj);
Al Viro483f4af2006-03-18 18:34:37 -05004130 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131
Linus Torvalds1da177e2005-04-16 15:20:36 -07004132 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004133 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004134 mutex_lock(&q->sysfs_lock);
4135 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4136 mutex_unlock(&q->sysfs_lock);
4137 return -ENOENT;
4138 }
4139 res = entry->show(q, page);
4140 mutex_unlock(&q->sysfs_lock);
4141 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004142}
4143
4144static ssize_t
4145queue_attr_store(struct kobject *kobj, struct attribute *attr,
4146 const char *page, size_t length)
4147{
4148 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004149 struct request_queue *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004150
Al Viro483f4af2006-03-18 18:34:37 -05004151 ssize_t res;
4152
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004154 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004155 mutex_lock(&q->sysfs_lock);
4156 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4157 mutex_unlock(&q->sysfs_lock);
4158 return -ENOENT;
4159 }
4160 res = entry->store(q, page, length);
4161 mutex_unlock(&q->sysfs_lock);
4162 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004163}
4164
4165static struct sysfs_ops queue_sysfs_ops = {
4166 .show = queue_attr_show,
4167 .store = queue_attr_store,
4168};
4169
Adrian Bunk93d17d32005-06-25 14:59:10 -07004170static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004171 .sysfs_ops = &queue_sysfs_ops,
4172 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05004173 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004174};
4175
4176int blk_register_queue(struct gendisk *disk)
4177{
4178 int ret;
4179
Jens Axboe165125e2007-07-24 09:28:11 +02004180 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181
4182 if (!q || !q->request_fn)
4183 return -ENXIO;
4184
4185 q->kobj.parent = kobject_get(&disk->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004186
Al Viro483f4af2006-03-18 18:34:37 -05004187 ret = kobject_add(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 if (ret < 0)
4189 return ret;
4190
Al Viro483f4af2006-03-18 18:34:37 -05004191 kobject_uevent(&q->kobj, KOBJ_ADD);
4192
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193 ret = elv_register_queue(q);
4194 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05004195 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4196 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004197 return ret;
4198 }
4199
4200 return 0;
4201}
4202
4203void blk_unregister_queue(struct gendisk *disk)
4204{
Jens Axboe165125e2007-07-24 09:28:11 +02004205 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004206
4207 if (q && q->request_fn) {
4208 elv_unregister_queue(q);
4209
Al Viro483f4af2006-03-18 18:34:37 -05004210 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4211 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004212 kobject_put(&disk->kobj);
4213 }
4214}