]> bbs.cooldavid.org Git - net-next-2.6.git/blame - block/ll_rw_blk.c
Add suspend-related notifications for CPU hotplug
[net-next-2.6.git] / block / ll_rw_blk.c
CommitLineData
1da177e4 1/*
1da177e4
LT
2 * 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 */
1da177e4
LT
13#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>
faccbd4b 28#include <linux/task_io_accounting_ops.h>
ff856bad
JA
29#include <linux/interrupt.h>
30#include <linux/cpu.h>
2056a782 31#include <linux/blktrace_api.h>
c17bb495 32#include <linux/fault-inject.h>
1da177e4
LT
33
34/*
35 * for max sense size
36 */
37#include <scsi/scsi_cmnd.h>
38
65f27f38 39static void blk_unplug_work(struct work_struct *work);
1da177e4 40static void blk_unplug_timeout(unsigned long data);
93d17d3d 41static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io);
52d9e675
TH
42static void init_request_from_bio(struct request *req, struct bio *bio);
43static int __make_request(request_queue_t *q, struct bio *bio);
b5deef90 44static struct io_context *current_io_context(gfp_t gfp_flags, int node);
1da177e4
LT
45
46/*
47 * For the allocated request tables
48 */
e18b890b 49static struct kmem_cache *request_cachep;
1da177e4
LT
50
51/*
52 * For queue allocation
53 */
e18b890b 54static struct kmem_cache *requestq_cachep;
1da177e4
LT
55
56/*
57 * For io context allocations
58 */
e18b890b 59static struct kmem_cache *iocontext_cachep;
1da177e4 60
1da177e4
LT
61/*
62 * Controlling structure to kblockd
63 */
ff856bad 64static struct workqueue_struct *kblockd_workqueue;
1da177e4
LT
65
66unsigned long blk_max_low_pfn, blk_max_pfn;
67
68EXPORT_SYMBOL(blk_max_low_pfn);
69EXPORT_SYMBOL(blk_max_pfn);
70
ff856bad
JA
71static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
72
1da177e4
LT
73/* Amount of time in which a process may batch requests */
74#define BLK_BATCH_TIME (HZ/50UL)
75
76/* Number of requests a "batching" process may submit */
77#define BLK_BATCH_REQ 32
78
79/*
80 * Return the threshold (number of used requests) at which the queue is
81 * considered to be congested. It include a little hysteresis to keep the
82 * context switch rate down.
83 */
84static inline int queue_congestion_on_threshold(struct request_queue *q)
85{
86 return q->nr_congestion_on;
87}
88
89/*
90 * The threshold at which a queue is considered to be uncongested
91 */
92static inline int queue_congestion_off_threshold(struct request_queue *q)
93{
94 return q->nr_congestion_off;
95}
96
97static void blk_queue_congestion_threshold(struct request_queue *q)
98{
99 int nr;
100
101 nr = q->nr_requests - (q->nr_requests / 8) + 1;
102 if (nr > q->nr_requests)
103 nr = q->nr_requests;
104 q->nr_congestion_on = nr;
105
106 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
107 if (nr < 1)
108 nr = 1;
109 q->nr_congestion_off = nr;
110}
111
1da177e4
LT
112/**
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
114 * @bdev: device
115 *
116 * Locates the passed device's request queue and returns the address of its
117 * backing_dev_info
118 *
119 * Will return NULL if the request queue cannot be located.
120 */
121struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
122{
123 struct backing_dev_info *ret = NULL;
124 request_queue_t *q = bdev_get_queue(bdev);
125
126 if (q)
127 ret = &q->backing_dev_info;
128 return ret;
129}
1da177e4
LT
130EXPORT_SYMBOL(blk_get_backing_dev_info);
131
1da177e4
LT
132/**
133 * blk_queue_prep_rq - set a prepare_request function for queue
134 * @q: queue
135 * @pfn: prepare_request function
136 *
137 * It's possible for a queue to register a prepare_request callback which
138 * is invoked before the request is handed to the request_fn. The goal of
139 * the function is to prepare a request for I/O, it can be used to build a
140 * cdb from the request data for instance.
141 *
142 */
143void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
144{
145 q->prep_rq_fn = pfn;
146}
147
148EXPORT_SYMBOL(blk_queue_prep_rq);
149
150/**
151 * blk_queue_merge_bvec - set a merge_bvec function for queue
152 * @q: queue
153 * @mbfn: merge_bvec_fn
154 *
155 * Usually queues have static limitations on the max sectors or segments that
156 * we can put in a request. Stacking drivers may have some settings that
157 * are dynamic, and thus we have to query the queue whether it is ok to
158 * add a new bio_vec to a bio at a given offset or not. If the block device
159 * has such limitations, it needs to register a merge_bvec_fn to control
160 * the size of bio's sent to it. Note that a block device *must* allow a
161 * single page to be added to an empty bio. The block device driver may want
162 * to use the bio_split() function to deal with these bio's. By default
163 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
164 * honored.
165 */
166void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
167{
168 q->merge_bvec_fn = mbfn;
169}
170
171EXPORT_SYMBOL(blk_queue_merge_bvec);
172
ff856bad
JA
173void blk_queue_softirq_done(request_queue_t *q, softirq_done_fn *fn)
174{
175 q->softirq_done_fn = fn;
176}
177
178EXPORT_SYMBOL(blk_queue_softirq_done);
179
1da177e4
LT
180/**
181 * blk_queue_make_request - define an alternate make_request function for a device
182 * @q: the request queue for the device to be affected
183 * @mfn: the alternate make_request function
184 *
185 * Description:
186 * The normal way for &struct bios to be passed to a device
187 * driver is for them to be collected into requests on a request
188 * queue, and then to allow the device driver to select requests
189 * off that queue when it is ready. This works well for many block
190 * devices. However some block devices (typically virtual devices
191 * such as md or lvm) do not benefit from the processing on the
192 * request queue, and are served best by having the requests passed
193 * directly to them. This can be achieved by providing a function
194 * to blk_queue_make_request().
195 *
196 * Caveat:
197 * The driver that does this *must* be able to deal appropriately
198 * with buffers in "highmemory". This can be accomplished by either calling
199 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
200 * blk_queue_bounce() to create a buffer in normal memory.
201 **/
202void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
203{
204 /*
205 * set defaults
206 */
207 q->nr_requests = BLKDEV_MAX_RQ;
309c0a1d
SM
208 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
209 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1da177e4
LT
210 q->make_request_fn = mfn;
211 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
212 q->backing_dev_info.state = 0;
213 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
defd94b7 214 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
1da177e4
LT
215 blk_queue_hardsect_size(q, 512);
216 blk_queue_dma_alignment(q, 511);
217 blk_queue_congestion_threshold(q);
218 q->nr_batching = BLK_BATCH_REQ;
219
220 q->unplug_thresh = 4; /* hmm */
221 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
222 if (q->unplug_delay == 0)
223 q->unplug_delay = 1;
224
65f27f38 225 INIT_WORK(&q->unplug_work, blk_unplug_work);
1da177e4
LT
226
227 q->unplug_timer.function = blk_unplug_timeout;
228 q->unplug_timer.data = (unsigned long)q;
229
230 /*
231 * by default assume old behaviour and bounce for any highmem page
232 */
233 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
1da177e4
LT
234}
235
236EXPORT_SYMBOL(blk_queue_make_request);
237
1ea25ecb 238static void rq_init(request_queue_t *q, struct request *rq)
1da177e4
LT
239{
240 INIT_LIST_HEAD(&rq->queuelist);
ff856bad 241 INIT_LIST_HEAD(&rq->donelist);
1da177e4
LT
242
243 rq->errors = 0;
1da177e4 244 rq->bio = rq->biotail = NULL;
2e662b65
JA
245 INIT_HLIST_NODE(&rq->hash);
246 RB_CLEAR_NODE(&rq->rb_node);
22e2c507 247 rq->ioprio = 0;
1da177e4
LT
248 rq->buffer = NULL;
249 rq->ref_count = 1;
250 rq->q = q;
1da177e4
LT
251 rq->special = NULL;
252 rq->data_len = 0;
253 rq->data = NULL;
df46b9a4 254 rq->nr_phys_segments = 0;
1da177e4
LT
255 rq->sense = NULL;
256 rq->end_io = NULL;
257 rq->end_io_data = NULL;
ff856bad 258 rq->completion_data = NULL;
1da177e4
LT
259}
260
261/**
262 * blk_queue_ordered - does this queue support ordered writes
797e7dbb
TH
263 * @q: the request queue
264 * @ordered: one of QUEUE_ORDERED_*
fddfdeaf 265 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
1da177e4
LT
266 *
267 * Description:
268 * For journalled file systems, doing ordered writes on a commit
269 * block instead of explicitly doing wait_on_buffer (which is bad
270 * for performance) can be a big win. Block drivers supporting this
271 * feature should call this function and indicate so.
272 *
273 **/
797e7dbb
TH
274int blk_queue_ordered(request_queue_t *q, unsigned ordered,
275 prepare_flush_fn *prepare_flush_fn)
276{
277 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
278 prepare_flush_fn == NULL) {
279 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
280 return -EINVAL;
281 }
282
283 if (ordered != QUEUE_ORDERED_NONE &&
284 ordered != QUEUE_ORDERED_DRAIN &&
285 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
286 ordered != QUEUE_ORDERED_DRAIN_FUA &&
287 ordered != QUEUE_ORDERED_TAG &&
288 ordered != QUEUE_ORDERED_TAG_FLUSH &&
289 ordered != QUEUE_ORDERED_TAG_FUA) {
290 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
291 return -EINVAL;
1da177e4 292 }
797e7dbb 293
60481b12 294 q->ordered = ordered;
797e7dbb
TH
295 q->next_ordered = ordered;
296 q->prepare_flush_fn = prepare_flush_fn;
297
298 return 0;
1da177e4
LT
299}
300
301EXPORT_SYMBOL(blk_queue_ordered);
302
303/**
304 * blk_queue_issue_flush_fn - set function for issuing a flush
305 * @q: the request queue
306 * @iff: the function to be called issuing the flush
307 *
308 * Description:
309 * If a driver supports issuing a flush command, the support is notified
310 * to the block layer by defining it through this call.
311 *
312 **/
313void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff)
314{
315 q->issue_flush_fn = iff;
316}
317
318EXPORT_SYMBOL(blk_queue_issue_flush_fn);
319
320/*
321 * Cache flushing for ordered writes handling
322 */
797e7dbb 323inline unsigned blk_ordered_cur_seq(request_queue_t *q)
1da177e4 324{
797e7dbb
TH
325 if (!q->ordseq)
326 return 0;
327 return 1 << ffz(q->ordseq);
1da177e4
LT
328}
329
797e7dbb 330unsigned blk_ordered_req_seq(struct request *rq)
1da177e4 331{
1da177e4
LT
332 request_queue_t *q = rq->q;
333
797e7dbb 334 BUG_ON(q->ordseq == 0);
8922e16c 335
797e7dbb
TH
336 if (rq == &q->pre_flush_rq)
337 return QUEUE_ORDSEQ_PREFLUSH;
338 if (rq == &q->bar_rq)
339 return QUEUE_ORDSEQ_BAR;
340 if (rq == &q->post_flush_rq)
341 return QUEUE_ORDSEQ_POSTFLUSH;
1da177e4 342
4aff5e23
JA
343 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
344 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
797e7dbb
TH
345 return QUEUE_ORDSEQ_DRAIN;
346 else
347 return QUEUE_ORDSEQ_DONE;
1da177e4
LT
348}
349
797e7dbb 350void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error)
1da177e4 351{
797e7dbb
TH
352 struct request *rq;
353 int uptodate;
1da177e4 354
797e7dbb
TH
355 if (error && !q->orderr)
356 q->orderr = error;
1da177e4 357
797e7dbb
TH
358 BUG_ON(q->ordseq & seq);
359 q->ordseq |= seq;
1da177e4 360
797e7dbb
TH
361 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
362 return;
1da177e4
LT
363
364 /*
797e7dbb 365 * Okay, sequence complete.
1da177e4 366 */
797e7dbb
TH
367 rq = q->orig_bar_rq;
368 uptodate = q->orderr ? q->orderr : 1;
1da177e4 369
797e7dbb 370 q->ordseq = 0;
1da177e4 371
797e7dbb
TH
372 end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
373 end_that_request_last(rq, uptodate);
1da177e4
LT
374}
375
797e7dbb 376static void pre_flush_end_io(struct request *rq, int error)
1da177e4 377{
797e7dbb
TH
378 elv_completed_request(rq->q, rq);
379 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
380}
1da177e4 381
797e7dbb
TH
382static void bar_end_io(struct request *rq, int error)
383{
384 elv_completed_request(rq->q, rq);
385 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
386}
1da177e4 387
797e7dbb
TH
388static void post_flush_end_io(struct request *rq, int error)
389{
390 elv_completed_request(rq->q, rq);
391 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
392}
1da177e4 393
797e7dbb
TH
394static void queue_flush(request_queue_t *q, unsigned which)
395{
396 struct request *rq;
397 rq_end_io_fn *end_io;
1da177e4 398
797e7dbb
TH
399 if (which == QUEUE_ORDERED_PREFLUSH) {
400 rq = &q->pre_flush_rq;
401 end_io = pre_flush_end_io;
402 } else {
403 rq = &q->post_flush_rq;
404 end_io = post_flush_end_io;
1da177e4 405 }
797e7dbb 406
4aff5e23 407 rq->cmd_flags = REQ_HARDBARRIER;
797e7dbb 408 rq_init(q, rq);
797e7dbb 409 rq->elevator_private = NULL;
c00895ab 410 rq->elevator_private2 = NULL;
797e7dbb 411 rq->rq_disk = q->bar_rq.rq_disk;
797e7dbb
TH
412 rq->end_io = end_io;
413 q->prepare_flush_fn(q, rq);
414
30e9656c 415 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
1da177e4
LT
416}
417
797e7dbb
TH
418static inline struct request *start_ordered(request_queue_t *q,
419 struct request *rq)
1da177e4 420{
797e7dbb
TH
421 q->bi_size = 0;
422 q->orderr = 0;
423 q->ordered = q->next_ordered;
424 q->ordseq |= QUEUE_ORDSEQ_STARTED;
425
426 /*
427 * Prep proxy barrier request.
428 */
429 blkdev_dequeue_request(rq);
430 q->orig_bar_rq = rq;
431 rq = &q->bar_rq;
4aff5e23 432 rq->cmd_flags = 0;
797e7dbb 433 rq_init(q, rq);
4aff5e23
JA
434 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
435 rq->cmd_flags |= REQ_RW;
436 rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
797e7dbb 437 rq->elevator_private = NULL;
c00895ab 438 rq->elevator_private2 = NULL;
797e7dbb
TH
439 init_request_from_bio(rq, q->orig_bar_rq->bio);
440 rq->end_io = bar_end_io;
441
442 /*
443 * Queue ordered sequence. As we stack them at the head, we
444 * need to queue in reverse order. Note that we rely on that
445 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
446 * request gets inbetween ordered sequence.
447 */
448 if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
449 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
450 else
451 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
452
30e9656c 453 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
797e7dbb
TH
454
455 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
456 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
457 rq = &q->pre_flush_rq;
458 } else
459 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
1da177e4 460
797e7dbb
TH
461 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
462 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
463 else
464 rq = NULL;
465
466 return rq;
1da177e4
LT
467}
468
797e7dbb 469int blk_do_ordered(request_queue_t *q, struct request **rqp)
1da177e4 470{
9a7a67af 471 struct request *rq = *rqp;
797e7dbb 472 int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
1da177e4 473
797e7dbb
TH
474 if (!q->ordseq) {
475 if (!is_barrier)
476 return 1;
1da177e4 477
797e7dbb
TH
478 if (q->next_ordered != QUEUE_ORDERED_NONE) {
479 *rqp = start_ordered(q, rq);
480 return 1;
481 } else {
482 /*
483 * This can happen when the queue switches to
484 * ORDERED_NONE while this request is on it.
485 */
486 blkdev_dequeue_request(rq);
487 end_that_request_first(rq, -EOPNOTSUPP,
488 rq->hard_nr_sectors);
489 end_that_request_last(rq, -EOPNOTSUPP);
490 *rqp = NULL;
491 return 0;
492 }
493 }
1da177e4 494
9a7a67af
JA
495 /*
496 * Ordered sequence in progress
497 */
498
499 /* Special requests are not subject to ordering rules. */
500 if (!blk_fs_request(rq) &&
501 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
502 return 1;
503
797e7dbb 504 if (q->ordered & QUEUE_ORDERED_TAG) {
9a7a67af 505 /* Ordered by tag. Blocking the next barrier is enough. */
797e7dbb
TH
506 if (is_barrier && rq != &q->bar_rq)
507 *rqp = NULL;
9a7a67af
JA
508 } else {
509 /* Ordered by draining. Wait for turn. */
510 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
511 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
512 *rqp = NULL;
1da177e4
LT
513 }
514
515 return 1;
516}
517
797e7dbb 518static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error)
1da177e4 519{
797e7dbb
TH
520 request_queue_t *q = bio->bi_private;
521 struct bio_vec *bvec;
522 int i;
523
524 /*
525 * This is dry run, restore bio_sector and size. We'll finish
526 * this request again with the original bi_end_io after an
527 * error occurs or post flush is complete.
528 */
529 q->bi_size += bytes;
530
531 if (bio->bi_size)
532 return 1;
533
534 /* Rewind bvec's */
535 bio->bi_idx = 0;
536 bio_for_each_segment(bvec, bio, i) {
537 bvec->bv_len += bvec->bv_offset;
538 bvec->bv_offset = 0;
539 }
540
541 /* Reset bio */
542 set_bit(BIO_UPTODATE, &bio->bi_flags);
543 bio->bi_size = q->bi_size;
544 bio->bi_sector -= (q->bi_size >> 9);
545 q->bi_size = 0;
546
547 return 0;
1da177e4 548}
1da177e4 549
1ea25ecb
JA
550static int ordered_bio_endio(struct request *rq, struct bio *bio,
551 unsigned int nbytes, int error)
1da177e4 552{
797e7dbb
TH
553 request_queue_t *q = rq->q;
554 bio_end_io_t *endio;
555 void *private;
556
557 if (&q->bar_rq != rq)
558 return 0;
559
560 /*
561 * Okay, this is the barrier request in progress, dry finish it.
562 */
563 if (error && !q->orderr)
564 q->orderr = error;
565
566 endio = bio->bi_end_io;
567 private = bio->bi_private;
568 bio->bi_end_io = flush_dry_bio_endio;
569 bio->bi_private = q;
570
571 bio_endio(bio, nbytes, error);
572
573 bio->bi_end_io = endio;
574 bio->bi_private = private;
575
576 return 1;
1da177e4 577}
1da177e4
LT
578
579/**
580 * blk_queue_bounce_limit - set bounce buffer limit for queue
581 * @q: the request queue for the device
582 * @dma_addr: bus address limit
583 *
584 * Description:
585 * Different hardware can have different requirements as to what pages
586 * it can do I/O directly to. A low level driver can call
587 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
5ee1af9f 588 * buffers for doing I/O to pages residing above @page.
1da177e4
LT
589 **/
590void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
591{
592 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
5ee1af9f
AK
593 int dma = 0;
594
595 q->bounce_gfp = GFP_NOIO;
596#if BITS_PER_LONG == 64
597 /* Assume anything <= 4GB can be handled by IOMMU.
598 Actually some IOMMUs can handle everything, but I don't
599 know of a way to test this here. */
8269730b 600 if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
5ee1af9f
AK
601 dma = 1;
602 q->bounce_pfn = max_low_pfn;
603#else
604 if (bounce_pfn < blk_max_low_pfn)
605 dma = 1;
606 q->bounce_pfn = bounce_pfn;
607#endif
608 if (dma) {
1da177e4
LT
609 init_emergency_isa_pool();
610 q->bounce_gfp = GFP_NOIO | GFP_DMA;
5ee1af9f
AK
611 q->bounce_pfn = bounce_pfn;
612 }
1da177e4
LT
613}
614
615EXPORT_SYMBOL(blk_queue_bounce_limit);
616
617/**
618 * blk_queue_max_sectors - set max sectors for a request for this queue
619 * @q: the request queue for the device
620 * @max_sectors: max sectors in the usual 512b unit
621 *
622 * Description:
623 * Enables a low level driver to set an upper limit on the size of
624 * received requests.
625 **/
2cb2e147 626void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors)
1da177e4
LT
627{
628 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
629 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
630 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
631 }
632
defd94b7
MC
633 if (BLK_DEF_MAX_SECTORS > max_sectors)
634 q->max_hw_sectors = q->max_sectors = max_sectors;
635 else {
636 q->max_sectors = BLK_DEF_MAX_SECTORS;
637 q->max_hw_sectors = max_sectors;
638 }
1da177e4
LT
639}
640
641EXPORT_SYMBOL(blk_queue_max_sectors);
642
643/**
644 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
645 * @q: the request queue for the device
646 * @max_segments: max number of segments
647 *
648 * Description:
649 * Enables a low level driver to set an upper limit on the number of
650 * physical data segments in a request. This would be the largest sized
651 * scatter list the driver could handle.
652 **/
653void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
654{
655 if (!max_segments) {
656 max_segments = 1;
657 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
658 }
659
660 q->max_phys_segments = max_segments;
661}
662
663EXPORT_SYMBOL(blk_queue_max_phys_segments);
664
665/**
666 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
667 * @q: the request queue for the device
668 * @max_segments: max number of segments
669 *
670 * Description:
671 * Enables a low level driver to set an upper limit on the number of
672 * hw data segments in a request. This would be the largest number of
673 * address/length pairs the host adapter can actually give as once
674 * to the device.
675 **/
676void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
677{
678 if (!max_segments) {
679 max_segments = 1;
680 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
681 }
682
683 q->max_hw_segments = max_segments;
684}
685
686EXPORT_SYMBOL(blk_queue_max_hw_segments);
687
688/**
689 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
690 * @q: the request queue for the device
691 * @max_size: max size of segment in bytes
692 *
693 * Description:
694 * Enables a low level driver to set an upper limit on the size of a
695 * coalesced segment
696 **/
697void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
698{
699 if (max_size < PAGE_CACHE_SIZE) {
700 max_size = PAGE_CACHE_SIZE;
701 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
702 }
703
704 q->max_segment_size = max_size;
705}
706
707EXPORT_SYMBOL(blk_queue_max_segment_size);
708
709/**
710 * blk_queue_hardsect_size - set hardware sector size for the queue
711 * @q: the request queue for the device
712 * @size: the hardware sector size, in bytes
713 *
714 * Description:
715 * This should typically be set to the lowest possible sector size
716 * that the hardware can operate on (possible without reverting to
717 * even internal read-modify-write operations). Usually the default
718 * of 512 covers most hardware.
719 **/
720void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
721{
722 q->hardsect_size = size;
723}
724
725EXPORT_SYMBOL(blk_queue_hardsect_size);
726
727/*
728 * Returns the minimum that is _not_ zero, unless both are zero.
729 */
730#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
731
732/**
733 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
734 * @t: the stacking driver (top)
735 * @b: the underlying device (bottom)
736 **/
737void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
738{
739 /* zero is "infinity" */
defd94b7
MC
740 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
741 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
1da177e4
LT
742
743 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
744 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
745 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
746 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
89e5c8b5
N
747 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
748 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
1da177e4
LT
749}
750
751EXPORT_SYMBOL(blk_queue_stack_limits);
752
753/**
754 * blk_queue_segment_boundary - set boundary rules for segment merging
755 * @q: the request queue for the device
756 * @mask: the memory boundary mask
757 **/
758void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
759{
760 if (mask < PAGE_CACHE_SIZE - 1) {
761 mask = PAGE_CACHE_SIZE - 1;
762 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
763 }
764
765 q->seg_boundary_mask = mask;
766}
767
768EXPORT_SYMBOL(blk_queue_segment_boundary);
769
770/**
771 * blk_queue_dma_alignment - set dma length and memory alignment
772 * @q: the request queue for the device
773 * @mask: alignment mask
774 *
775 * description:
776 * set required memory and length aligment for direct dma transactions.
777 * this is used when buiding direct io requests for the queue.
778 *
779 **/
780void blk_queue_dma_alignment(request_queue_t *q, int mask)
781{
782 q->dma_alignment = mask;
783}
784
785EXPORT_SYMBOL(blk_queue_dma_alignment);
786
787/**
788 * blk_queue_find_tag - find a request by its tag and queue
1da177e4
LT
789 * @q: The request queue for the device
790 * @tag: The tag of the request
791 *
792 * Notes:
793 * Should be used when a device returns a tag and you want to match
794 * it with a request.
795 *
796 * no locks need be held.
797 **/
798struct request *blk_queue_find_tag(request_queue_t *q, int tag)
799{
f583f492 800 return blk_map_queue_find_tag(q->queue_tags, tag);
1da177e4
LT
801}
802
803EXPORT_SYMBOL(blk_queue_find_tag);
804
805/**
492dfb48
JB
806 * __blk_free_tags - release a given set of tag maintenance info
807 * @bqt: the tag map to free
1da177e4 808 *
492dfb48
JB
809 * Tries to free the specified @bqt@. Returns true if it was
810 * actually freed and false if there are still references using it
811 */
812static int __blk_free_tags(struct blk_queue_tag *bqt)
1da177e4 813{
492dfb48 814 int retval;
1da177e4 815
492dfb48
JB
816 retval = atomic_dec_and_test(&bqt->refcnt);
817 if (retval) {
1da177e4
LT
818 BUG_ON(bqt->busy);
819 BUG_ON(!list_empty(&bqt->busy_list));
820
821 kfree(bqt->tag_index);
822 bqt->tag_index = NULL;
823
824 kfree(bqt->tag_map);
825 bqt->tag_map = NULL;
826
827 kfree(bqt);
492dfb48 828
1da177e4
LT
829 }
830
492dfb48
JB
831 return retval;
832}
833
834/**
835 * __blk_queue_free_tags - release tag maintenance info
836 * @q: the request queue for the device
837 *
838 * Notes:
839 * blk_cleanup_queue() will take care of calling this function, if tagging
840 * has been used. So there's no need to call this directly.
841 **/
842static void __blk_queue_free_tags(request_queue_t *q)
843{
844 struct blk_queue_tag *bqt = q->queue_tags;
845
846 if (!bqt)
847 return;
848
849 __blk_free_tags(bqt);
850
1da177e4
LT
851 q->queue_tags = NULL;
852 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
853}
854
492dfb48
JB
855
856/**
857 * blk_free_tags - release a given set of tag maintenance info
858 * @bqt: the tag map to free
859 *
860 * For externally managed @bqt@ frees the map. Callers of this
861 * function must guarantee to have released all the queues that
862 * might have been using this tag map.
863 */
864void blk_free_tags(struct blk_queue_tag *bqt)
865{
866 if (unlikely(!__blk_free_tags(bqt)))
867 BUG();
868}
869EXPORT_SYMBOL(blk_free_tags);
870
1da177e4
LT
871/**
872 * blk_queue_free_tags - release tag maintenance info
873 * @q: the request queue for the device
874 *
875 * Notes:
876 * This is used to disabled tagged queuing to a device, yet leave
877 * queue in function.
878 **/
879void blk_queue_free_tags(request_queue_t *q)
880{
881 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
882}
883
884EXPORT_SYMBOL(blk_queue_free_tags);
885
886static int
887init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
888{
1da177e4
LT
889 struct request **tag_index;
890 unsigned long *tag_map;
fa72b903 891 int nr_ulongs;
1da177e4 892
492dfb48 893 if (q && depth > q->nr_requests * 2) {
1da177e4
LT
894 depth = q->nr_requests * 2;
895 printk(KERN_ERR "%s: adjusted depth to %d\n",
896 __FUNCTION__, depth);
897 }
898
f68110fc 899 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
1da177e4
LT
900 if (!tag_index)
901 goto fail;
902
f7d37d02 903 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
f68110fc 904 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
1da177e4
LT
905 if (!tag_map)
906 goto fail;
907
ba025082 908 tags->real_max_depth = depth;
1da177e4 909 tags->max_depth = depth;
1da177e4
LT
910 tags->tag_index = tag_index;
911 tags->tag_map = tag_map;
912
1da177e4
LT
913 return 0;
914fail:
915 kfree(tag_index);
916 return -ENOMEM;
917}
918
492dfb48
JB
919static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
920 int depth)
921{
922 struct blk_queue_tag *tags;
923
924 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
925 if (!tags)
926 goto fail;
927
928 if (init_tag_map(q, tags, depth))
929 goto fail;
930
931 INIT_LIST_HEAD(&tags->busy_list);
932 tags->busy = 0;
933 atomic_set(&tags->refcnt, 1);
934 return tags;
935fail:
936 kfree(tags);
937 return NULL;
938}
939
940/**
941 * blk_init_tags - initialize the tag info for an external tag map
942 * @depth: the maximum queue depth supported
943 * @tags: the tag to use
944 **/
945struct blk_queue_tag *blk_init_tags(int depth)
946{
947 return __blk_queue_init_tags(NULL, depth);
948}
949EXPORT_SYMBOL(blk_init_tags);
950
1da177e4
LT
951/**
952 * blk_queue_init_tags - initialize the queue tag info
953 * @q: the request queue for the device
954 * @depth: the maximum queue depth supported
955 * @tags: the tag to use
956 **/
957int blk_queue_init_tags(request_queue_t *q, int depth,
958 struct blk_queue_tag *tags)
959{
960 int rc;
961
962 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
963
964 if (!tags && !q->queue_tags) {
492dfb48 965 tags = __blk_queue_init_tags(q, depth);
1da177e4 966
492dfb48 967 if (!tags)
1da177e4 968 goto fail;
1da177e4
LT
969 } else if (q->queue_tags) {
970 if ((rc = blk_queue_resize_tags(q, depth)))
971 return rc;
972 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
973 return 0;
974 } else
975 atomic_inc(&tags->refcnt);
976
977 /*
978 * assign it, all done
979 */
980 q->queue_tags = tags;
981 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
982 return 0;
983fail:
984 kfree(tags);
985 return -ENOMEM;
986}
987
988EXPORT_SYMBOL(blk_queue_init_tags);
989
990/**
991 * blk_queue_resize_tags - change the queueing depth
992 * @q: the request queue for the device
993 * @new_depth: the new max command queueing depth
994 *
995 * Notes:
996 * Must be called with the queue lock held.
997 **/
998int blk_queue_resize_tags(request_queue_t *q, int new_depth)
999{
1000 struct blk_queue_tag *bqt = q->queue_tags;
1001 struct request **tag_index;
1002 unsigned long *tag_map;
fa72b903 1003 int max_depth, nr_ulongs;
1da177e4
LT
1004
1005 if (!bqt)
1006 return -ENXIO;
1007
ba025082
TH
1008 /*
1009 * if we already have large enough real_max_depth. just
1010 * adjust max_depth. *NOTE* as requests with tag value
1011 * between new_depth and real_max_depth can be in-flight, tag
1012 * map can not be shrunk blindly here.
1013 */
1014 if (new_depth <= bqt->real_max_depth) {
1015 bqt->max_depth = new_depth;
1016 return 0;
1017 }
1018
492dfb48
JB
1019 /*
1020 * Currently cannot replace a shared tag map with a new
1021 * one, so error out if this is the case
1022 */
1023 if (atomic_read(&bqt->refcnt) != 1)
1024 return -EBUSY;
1025
1da177e4
LT
1026 /*
1027 * save the old state info, so we can copy it back
1028 */
1029 tag_index = bqt->tag_index;
1030 tag_map = bqt->tag_map;
ba025082 1031 max_depth = bqt->real_max_depth;
1da177e4
LT
1032
1033 if (init_tag_map(q, bqt, new_depth))
1034 return -ENOMEM;
1035
1036 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
f7d37d02 1037 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
fa72b903 1038 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
1da177e4
LT
1039
1040 kfree(tag_index);
1041 kfree(tag_map);
1042 return 0;
1043}
1044
1045EXPORT_SYMBOL(blk_queue_resize_tags);
1046
1047/**
1048 * blk_queue_end_tag - end tag operations for a request
1049 * @q: the request queue for the device
1050 * @rq: the request that has completed
1051 *
1052 * Description:
1053 * Typically called when end_that_request_first() returns 0, meaning
1054 * all transfers have been done for a request. It's important to call
1055 * this function before end_that_request_last(), as that will put the
1056 * request back on the free list thus corrupting the internal tag list.
1057 *
1058 * Notes:
1059 * queue lock must be held.
1060 **/
1061void blk_queue_end_tag(request_queue_t *q, struct request *rq)
1062{
1063 struct blk_queue_tag *bqt = q->queue_tags;
1064 int tag = rq->tag;
1065
1066 BUG_ON(tag == -1);
1067
ba025082 1068 if (unlikely(tag >= bqt->real_max_depth))
040c928c
TH
1069 /*
1070 * This can happen after tag depth has been reduced.
1071 * FIXME: how about a warning or info message here?
1072 */
1da177e4
LT
1073 return;
1074
1075 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
040c928c
TH
1076 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1077 __FUNCTION__, tag);
1da177e4
LT
1078 return;
1079 }
1080
1081 list_del_init(&rq->queuelist);
4aff5e23 1082 rq->cmd_flags &= ~REQ_QUEUED;
1da177e4
LT
1083 rq->tag = -1;
1084
1085 if (unlikely(bqt->tag_index[tag] == NULL))
040c928c
TH
1086 printk(KERN_ERR "%s: tag %d is missing\n",
1087 __FUNCTION__, tag);
1da177e4
LT
1088
1089 bqt->tag_index[tag] = NULL;
1090 bqt->busy--;
1091}
1092
1093EXPORT_SYMBOL(blk_queue_end_tag);
1094
1095/**
1096 * blk_queue_start_tag - find a free tag and assign it
1097 * @q: the request queue for the device
1098 * @rq: the block request that needs tagging
1099 *
1100 * Description:
1101 * This can either be used as a stand-alone helper, or possibly be
1102 * assigned as the queue &prep_rq_fn (in which case &struct request
1103 * automagically gets a tag assigned). Note that this function
1104 * assumes that any type of request can be queued! if this is not
1105 * true for your device, you must check the request type before
1106 * calling this function. The request will also be removed from
1107 * the request queue, so it's the drivers responsibility to readd
1108 * it if it should need to be restarted for some reason.
1109 *
1110 * Notes:
1111 * queue lock must be held.
1112 **/
1113int blk_queue_start_tag(request_queue_t *q, struct request *rq)
1114{
1115 struct blk_queue_tag *bqt = q->queue_tags;
2bf0fdad 1116 int tag;
1da177e4 1117
4aff5e23 1118 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
1da177e4 1119 printk(KERN_ERR
040c928c
TH
1120 "%s: request %p for device [%s] already tagged %d",
1121 __FUNCTION__, rq,
1122 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
1da177e4
LT
1123 BUG();
1124 }
1125
059af497
JA
1126 /*
1127 * Protect against shared tag maps, as we may not have exclusive
1128 * access to the tag map.
1129 */
1130 do {
1131 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1132 if (tag >= bqt->max_depth)
1133 return 1;
1da177e4 1134
059af497 1135 } while (test_and_set_bit(tag, bqt->tag_map));
1da177e4 1136
4aff5e23 1137 rq->cmd_flags |= REQ_QUEUED;
1da177e4
LT
1138 rq->tag = tag;
1139 bqt->tag_index[tag] = rq;
1140 blkdev_dequeue_request(rq);
1141 list_add(&rq->queuelist, &bqt->busy_list);
1142 bqt->busy++;
1143 return 0;
1144}
1145
1146EXPORT_SYMBOL(blk_queue_start_tag);
1147
1148/**
1149 * blk_queue_invalidate_tags - invalidate all pending tags
1150 * @q: the request queue for the device
1151 *
1152 * Description:
1153 * Hardware conditions may dictate a need to stop all pending requests.
1154 * In this case, we will safely clear the block side of the tag queue and
1155 * readd all requests to the request queue in the right order.
1156 *
1157 * Notes:
1158 * queue lock must be held.
1159 **/
1160void blk_queue_invalidate_tags(request_queue_t *q)
1161{
1162 struct blk_queue_tag *bqt = q->queue_tags;
1163 struct list_head *tmp, *n;
1164 struct request *rq;
1165
1166 list_for_each_safe(tmp, n, &bqt->busy_list) {
1167 rq = list_entry_rq(tmp);
1168
1169 if (rq->tag == -1) {
040c928c
TH
1170 printk(KERN_ERR
1171 "%s: bad tag found on list\n", __FUNCTION__);
1da177e4 1172 list_del_init(&rq->queuelist);
4aff5e23 1173 rq->cmd_flags &= ~REQ_QUEUED;
1da177e4
LT
1174 } else
1175 blk_queue_end_tag(q, rq);
1176
4aff5e23 1177 rq->cmd_flags &= ~REQ_STARTED;
1da177e4
LT
1178 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1179 }
1180}
1181
1182EXPORT_SYMBOL(blk_queue_invalidate_tags);
1183
1da177e4
LT
1184void blk_dump_rq_flags(struct request *rq, char *msg)
1185{
1186 int bit;
1187
4aff5e23
JA
1188 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1189 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1190 rq->cmd_flags);
1da177e4
LT
1191
1192 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1193 rq->nr_sectors,
1194 rq->current_nr_sectors);
1195 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1196
4aff5e23 1197 if (blk_pc_request(rq)) {
1da177e4
LT
1198 printk("cdb: ");
1199 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1200 printk("%02x ", rq->cmd[bit]);
1201 printk("\n");
1202 }
1203}
1204
1205EXPORT_SYMBOL(blk_dump_rq_flags);
1206
1207void blk_recount_segments(request_queue_t *q, struct bio *bio)
1208{
1209 struct bio_vec *bv, *bvprv = NULL;
1210 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1211 int high, highprv = 1;
1212
1213 if (unlikely(!bio->bi_io_vec))
1214 return;
1215
1216 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1217 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1218 bio_for_each_segment(bv, bio, i) {
1219 /*
1220 * the trick here is making sure that a high page is never
1221 * considered part of another segment, since that might
1222 * change with the bounce page.
1223 */
f772b3d9 1224 high = page_to_pfn(bv->bv_page) > q->bounce_pfn;
1da177e4
LT
1225 if (high || highprv)
1226 goto new_hw_segment;
1227 if (cluster) {
1228 if (seg_size + bv->bv_len > q->max_segment_size)
1229 goto new_segment;
1230 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1231 goto new_segment;
1232 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1233 goto new_segment;
1234 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1235 goto new_hw_segment;
1236
1237 seg_size += bv->bv_len;
1238 hw_seg_size += bv->bv_len;
1239 bvprv = bv;
1240 continue;
1241 }
1242new_segment:
1243 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1244 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1245 hw_seg_size += bv->bv_len;
1246 } else {
1247new_hw_segment:
1248 if (hw_seg_size > bio->bi_hw_front_size)
1249 bio->bi_hw_front_size = hw_seg_size;
1250 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1251 nr_hw_segs++;
1252 }
1253
1254 nr_phys_segs++;
1255 bvprv = bv;
1256 seg_size = bv->bv_len;
1257 highprv = high;
1258 }
1259 if (hw_seg_size > bio->bi_hw_back_size)
1260 bio->bi_hw_back_size = hw_seg_size;
1261 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1262 bio->bi_hw_front_size = hw_seg_size;
1263 bio->bi_phys_segments = nr_phys_segs;
1264 bio->bi_hw_segments = nr_hw_segs;
1265 bio->bi_flags |= (1 << BIO_SEG_VALID);
1266}
387bb173 1267EXPORT_SYMBOL(blk_recount_segments);
1da177e4 1268
93d17d3d 1269static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
1da177e4
LT
1270 struct bio *nxt)
1271{
1272 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1273 return 0;
1274
1275 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1276 return 0;
1277 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1278 return 0;
1279
1280 /*
1281 * bio and nxt are contigous in memory, check if the queue allows
1282 * these two to be merged into one
1283 */
1284 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1285 return 1;
1286
1287 return 0;
1288}
1289
93d17d3d 1290static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
1da177e4
LT
1291 struct bio *nxt)
1292{
1293 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1294 blk_recount_segments(q, bio);
1295 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1296 blk_recount_segments(q, nxt);
1297 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1298 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1299 return 0;
1300 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1301 return 0;
1302
1303 return 1;
1304}
1305
1da177e4
LT
1306/*
1307 * map a request to scatterlist, return number of sg entries setup. Caller
1308 * must make sure sg can hold rq->nr_phys_segments entries
1309 */
1310int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1311{
1312 struct bio_vec *bvec, *bvprv;
1313 struct bio *bio;
1314 int nsegs, i, cluster;
1315
1316 nsegs = 0;
1317 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1318
1319 /*
1320 * for each bio in rq
1321 */
1322 bvprv = NULL;
1323 rq_for_each_bio(bio, rq) {
1324 /*
1325 * for each segment in bio
1326 */
1327 bio_for_each_segment(bvec, bio, i) {
1328 int nbytes = bvec->bv_len;
1329
1330 if (bvprv && cluster) {
1331 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1332 goto new_segment;
1333
1334 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1335 goto new_segment;
1336 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1337 goto new_segment;
1338
1339 sg[nsegs - 1].length += nbytes;
1340 } else {
1341new_segment:
1342 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1343 sg[nsegs].page = bvec->bv_page;
1344 sg[nsegs].length = nbytes;
1345 sg[nsegs].offset = bvec->bv_offset;
1346
1347 nsegs++;
1348 }
1349 bvprv = bvec;
1350 } /* segments in bio */
1351 } /* bios in rq */
1352
1353 return nsegs;
1354}
1355
1356EXPORT_SYMBOL(blk_rq_map_sg);
1357
1358/*
1359 * the standard queue merge functions, can be overridden with device
1360 * specific ones if so desired
1361 */
1362
1363static inline int ll_new_mergeable(request_queue_t *q,
1364 struct request *req,
1365 struct bio *bio)
1366{
1367 int nr_phys_segs = bio_phys_segments(q, bio);
1368
1369 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
4aff5e23 1370 req->cmd_flags |= REQ_NOMERGE;
1da177e4
LT
1371 if (req == q->last_merge)
1372 q->last_merge = NULL;
1373 return 0;
1374 }
1375
1376 /*
1377 * A hw segment is just getting larger, bump just the phys
1378 * counter.
1379 */
1380 req->nr_phys_segments += nr_phys_segs;
1381 return 1;
1382}
1383
1384static inline int ll_new_hw_segment(request_queue_t *q,
1385 struct request *req,
1386 struct bio *bio)
1387{
1388 int nr_hw_segs = bio_hw_segments(q, bio);
1389 int nr_phys_segs = bio_phys_segments(q, bio);
1390
1391 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1392 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
4aff5e23 1393 req->cmd_flags |= REQ_NOMERGE;
1da177e4
LT
1394 if (req == q->last_merge)
1395 q->last_merge = NULL;
1396 return 0;
1397 }
1398
1399 /*
1400 * This will form the start of a new hw segment. Bump both
1401 * counters.
1402 */
1403 req->nr_hw_segments += nr_hw_segs;
1404 req->nr_phys_segments += nr_phys_segs;
1405 return 1;
1406}
1407
1aa4f24f 1408int ll_back_merge_fn(request_queue_t *q, struct request *req, struct bio *bio)
1da177e4 1409{
defd94b7 1410 unsigned short max_sectors;
1da177e4
LT
1411 int len;
1412
defd94b7
MC
1413 if (unlikely(blk_pc_request(req)))
1414 max_sectors = q->max_hw_sectors;
1415 else
1416 max_sectors = q->max_sectors;
1417
1418 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
4aff5e23 1419 req->cmd_flags |= REQ_NOMERGE;
1da177e4
LT
1420 if (req == q->last_merge)
1421 q->last_merge = NULL;
1422 return 0;
1423 }
1424 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1425 blk_recount_segments(q, req->biotail);
1426 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1427 blk_recount_segments(q, bio);
1428 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1429 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1430 !BIOVEC_VIRT_OVERSIZE(len)) {
1431 int mergeable = ll_new_mergeable(q, req, bio);
1432
1433 if (mergeable) {
1434 if (req->nr_hw_segments == 1)
1435 req->bio->bi_hw_front_size = len;
1436 if (bio->bi_hw_segments == 1)
1437 bio->bi_hw_back_size = len;
1438 }
1439 return mergeable;
1440 }
1441
1442 return ll_new_hw_segment(q, req, bio);
1443}
1aa4f24f 1444EXPORT_SYMBOL(ll_back_merge_fn);
1da177e4
LT
1445
1446static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1447 struct bio *bio)
1448{
defd94b7 1449 unsigned short max_sectors;
1da177e4
LT
1450 int len;
1451
defd94b7
MC
1452 if (unlikely(blk_pc_request(req)))
1453 max_sectors = q->max_hw_sectors;
1454 else
1455 max_sectors = q->max_sectors;
1456
1457
1458 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
4aff5e23 1459 req->cmd_flags |= REQ_NOMERGE;
1da177e4
LT
1460 if (req == q->last_merge)
1461 q->last_merge = NULL;
1462 return 0;
1463 }
1464 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1465 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1466 blk_recount_segments(q, bio);
1467 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1468 blk_recount_segments(q, req->bio);
1469 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1470 !BIOVEC_VIRT_OVERSIZE(len)) {
1471 int mergeable = ll_new_mergeable(q, req, bio);
1472
1473 if (mergeable) {
1474 if (bio->bi_hw_segments == 1)
1475 bio->bi_hw_front_size = len;
1476 if (req->nr_hw_segments == 1)
1477 req->biotail->bi_hw_back_size = len;
1478 }
1479 return mergeable;
1480 }
1481
1482 return ll_new_hw_segment(q, req, bio);
1483}
1484
1485static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1486 struct request *next)
1487{
dfa1a553
ND
1488 int total_phys_segments;
1489 int total_hw_segments;
1da177e4
LT
1490
1491 /*
1492 * First check if the either of the requests are re-queued
1493 * requests. Can't merge them if they are.
1494 */
1495 if (req->special || next->special)
1496 return 0;
1497
1498 /*
dfa1a553 1499 * Will it become too large?
1da177e4
LT
1500 */
1501 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1502 return 0;
1503
1504 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1505 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1506 total_phys_segments--;
1507
1508 if (total_phys_segments > q->max_phys_segments)
1509 return 0;
1510
1511 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1512 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1513 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1514 /*
1515 * propagate the combined length to the end of the requests
1516 */
1517 if (req->nr_hw_segments == 1)
1518 req->bio->bi_hw_front_size = len;
1519 if (next->nr_hw_segments == 1)
1520 next->biotail->bi_hw_back_size = len;
1521 total_hw_segments--;
1522 }
1523
1524 if (total_hw_segments > q->max_hw_segments)
1525 return 0;
1526
1527 /* Merge is OK... */
1528 req->nr_phys_segments = total_phys_segments;
1529 req->nr_hw_segments = total_hw_segments;
1530 return 1;
1531}
1532
1533/*
1534 * "plug" the device if there are no outstanding requests: this will
1535 * force the transfer to start only after we have put all the requests
1536 * on the list.
1537 *
1538 * This is called with interrupts off and no requests on the queue and
1539 * with the queue lock held.
1540 */
1541void blk_plug_device(request_queue_t *q)
1542{
1543 WARN_ON(!irqs_disabled());
1544
1545 /*
1546 * don't plug a stopped queue, it must be paired with blk_start_queue()
1547 * which will restart the queueing
1548 */
7daac490 1549 if (blk_queue_stopped(q))
1da177e4
LT
1550 return;
1551
2056a782 1552 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
1da177e4 1553 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
2056a782
JA
1554 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1555 }
1da177e4
LT
1556}
1557
1558EXPORT_SYMBOL(blk_plug_device);
1559
1560/*
1561 * remove the queue from the plugged list, if present. called with
1562 * queue lock held and interrupts disabled.
1563 */
1564int blk_remove_plug(request_queue_t *q)
1565{
1566 WARN_ON(!irqs_disabled());
1567
1568 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1569 return 0;
1570
1571 del_timer(&q->unplug_timer);
1572 return 1;
1573}
1574
1575EXPORT_SYMBOL(blk_remove_plug);
1576
1577/*
1578 * remove the plug and let it rip..
1579 */
1580void __generic_unplug_device(request_queue_t *q)
1581{
7daac490 1582 if (unlikely(blk_queue_stopped(q)))
1da177e4
LT
1583 return;
1584
1585 if (!blk_remove_plug(q))
1586 return;
1587
22e2c507 1588 q->request_fn(q);
1da177e4
LT
1589}
1590EXPORT_SYMBOL(__generic_unplug_device);
1591
1592/**
1593 * generic_unplug_device - fire a request queue
1594 * @q: The &request_queue_t in question
1595 *
1596 * Description:
1597 * Linux uses plugging to build bigger requests queues before letting
1598 * the device have at them. If a queue is plugged, the I/O scheduler
1599 * is still adding and merging requests on the queue. Once the queue
1600 * gets unplugged, the request_fn defined for the queue is invoked and
1601 * transfers started.
1602 **/
1603void generic_unplug_device(request_queue_t *q)
1604{
1605 spin_lock_irq(q->queue_lock);
1606 __generic_unplug_device(q);
1607 spin_unlock_irq(q->queue_lock);
1608}
1609EXPORT_SYMBOL(generic_unplug_device);
1610
1611static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1612 struct page *page)
1613{
1614 request_queue_t *q = bdi->unplug_io_data;
1615
1616 /*
1617 * devices don't necessarily have an ->unplug_fn defined
1618 */
2056a782
JA
1619 if (q->unplug_fn) {
1620 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1621 q->rq.count[READ] + q->rq.count[WRITE]);
1622
1da177e4 1623 q->unplug_fn(q);
2056a782 1624 }
1da177e4
LT
1625}
1626
65f27f38 1627static void blk_unplug_work(struct work_struct *work)
1da177e4 1628{
65f27f38 1629 request_queue_t *q = container_of(work, request_queue_t, unplug_work);
1da177e4 1630
2056a782
JA
1631 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1632 q->rq.count[READ] + q->rq.count[WRITE]);
1633
1da177e4
LT
1634 q->unplug_fn(q);
1635}
1636
1637static void blk_unplug_timeout(unsigned long data)
1638{
1639 request_queue_t *q = (request_queue_t *)data;
1640
2056a782
JA
1641 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1642 q->rq.count[READ] + q->rq.count[WRITE]);
1643
1da177e4
LT
1644 kblockd_schedule_work(&q->unplug_work);
1645}
1646
1647/**
1648 * blk_start_queue - restart a previously stopped queue
1649 * @q: The &request_queue_t in question
1650 *
1651 * Description:
1652 * blk_start_queue() will clear the stop flag on the queue, and call
1653 * the request_fn for the queue if it was in a stopped state when
1654 * entered. Also see blk_stop_queue(). Queue lock must be held.
1655 **/
1656void blk_start_queue(request_queue_t *q)
1657{
a038e253
PBG
1658 WARN_ON(!irqs_disabled());
1659
1da177e4
LT
1660 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1661
1662 /*
1663 * one level of recursion is ok and is much faster than kicking
1664 * the unplug handling
1665 */
1666 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1667 q->request_fn(q);
1668 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1669 } else {
1670 blk_plug_device(q);
1671 kblockd_schedule_work(&q->unplug_work);
1672 }
1673}
1674
1675EXPORT_SYMBOL(blk_start_queue);
1676
1677/**
1678 * blk_stop_queue - stop a queue
1679 * @q: The &request_queue_t in question
1680 *
1681 * Description:
1682 * The Linux block layer assumes that a block driver will consume all
1683 * entries on the request queue when the request_fn strategy is called.
1684 * Often this will not happen, because of hardware limitations (queue
1685 * depth settings). If a device driver gets a 'queue full' response,
1686 * or if it simply chooses not to queue more I/O at one point, it can
1687 * call this function to prevent the request_fn from being called until
1688 * the driver has signalled it's ready to go again. This happens by calling
1689 * blk_start_queue() to restart queue operations. Queue lock must be held.
1690 **/
1691void blk_stop_queue(request_queue_t *q)
1692{
1693 blk_remove_plug(q);
1694 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1695}
1696EXPORT_SYMBOL(blk_stop_queue);
1697
1698/**
1699 * blk_sync_queue - cancel any pending callbacks on a queue
1700 * @q: the queue
1701 *
1702 * Description:
1703 * The block layer may perform asynchronous callback activity
1704 * on a queue, such as calling the unplug function after a timeout.
1705 * A block device may call blk_sync_queue to ensure that any
1706 * such activity is cancelled, thus allowing it to release resources
1707 * the the callbacks might use. The caller must already have made sure
1708 * that its ->make_request_fn will not re-add plugging prior to calling
1709 * this function.
1710 *
1711 */
1712void blk_sync_queue(struct request_queue *q)
1713{
1714 del_timer_sync(&q->unplug_timer);
1da177e4
LT
1715}
1716EXPORT_SYMBOL(blk_sync_queue);
1717
1718/**
1719 * blk_run_queue - run a single device queue
1720 * @q: The queue to run
1721 */
1722void blk_run_queue(struct request_queue *q)
1723{
1724 unsigned long flags;
1725
1726 spin_lock_irqsave(q->queue_lock, flags);
1727 blk_remove_plug(q);
dac07ec1
JA
1728
1729 /*
1730 * Only recurse once to avoid overrunning the stack, let the unplug
1731 * handling reinvoke the handler shortly if we already got there.
1732 */
1733 if (!elv_queue_empty(q)) {
1734 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1735 q->request_fn(q);
1736 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1737 } else {
1738 blk_plug_device(q);
1739 kblockd_schedule_work(&q->unplug_work);
1740 }
1741 }
1742
1da177e4
LT
1743 spin_unlock_irqrestore(q->queue_lock, flags);
1744}
1745EXPORT_SYMBOL(blk_run_queue);
1746
1747/**
1748 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
a580290c 1749 * @kobj: the kobj belonging of the request queue to be released
1da177e4
LT
1750 *
1751 * Description:
1752 * blk_cleanup_queue is the pair to blk_init_queue() or
1753 * blk_queue_make_request(). It should be called when a request queue is
1754 * being released; typically when a block device is being de-registered.
1755 * Currently, its primary task it to free all the &struct request
1756 * structures that were allocated to the queue and the queue itself.
1757 *
1758 * Caveat:
1759 * Hopefully the low level driver will have finished any
1760 * outstanding requests first...
1761 **/
483f4afc 1762static void blk_release_queue(struct kobject *kobj)
1da177e4 1763{
483f4afc 1764 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
1da177e4
LT
1765 struct request_list *rl = &q->rq;
1766
1da177e4
LT
1767 blk_sync_queue(q);
1768
1769 if (rl->rq_pool)
1770 mempool_destroy(rl->rq_pool);
1771
1772 if (q->queue_tags)
1773 __blk_queue_free_tags(q);
1774
6c5c9341 1775 blk_trace_shutdown(q);
2056a782 1776
1da177e4
LT
1777 kmem_cache_free(requestq_cachep, q);
1778}
1779
483f4afc
AV
1780void blk_put_queue(request_queue_t *q)
1781{
1782 kobject_put(&q->kobj);
1783}
1784EXPORT_SYMBOL(blk_put_queue);
1785
1786void blk_cleanup_queue(request_queue_t * q)
1787{
1788 mutex_lock(&q->sysfs_lock);
1789 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1790 mutex_unlock(&q->sysfs_lock);
1791
1792 if (q->elevator)
1793 elevator_exit(q->elevator);
1794
1795 blk_put_queue(q);
1796}
1797
1da177e4
LT
1798EXPORT_SYMBOL(blk_cleanup_queue);
1799
1800static int blk_init_free_list(request_queue_t *q)
1801{
1802 struct request_list *rl = &q->rq;
1803
1804 rl->count[READ] = rl->count[WRITE] = 0;
1805 rl->starved[READ] = rl->starved[WRITE] = 0;
cb98fc8b 1806 rl->elvpriv = 0;
1da177e4
LT
1807 init_waitqueue_head(&rl->wait[READ]);
1808 init_waitqueue_head(&rl->wait[WRITE]);
1da177e4 1809
1946089a
CL
1810 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1811 mempool_free_slab, request_cachep, q->node);
1da177e4
LT
1812
1813 if (!rl->rq_pool)
1814 return -ENOMEM;
1815
1816 return 0;
1817}
1818
8267e268 1819request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
1da177e4 1820{
1946089a
CL
1821 return blk_alloc_queue_node(gfp_mask, -1);
1822}
1823EXPORT_SYMBOL(blk_alloc_queue);
1da177e4 1824
483f4afc
AV
1825static struct kobj_type queue_ktype;
1826
8267e268 1827request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
1946089a
CL
1828{
1829 request_queue_t *q;
1830
1831 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
1da177e4
LT
1832 if (!q)
1833 return NULL;
1834
1835 memset(q, 0, sizeof(*q));
1836 init_timer(&q->unplug_timer);
483f4afc
AV
1837
1838 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
1839 q->kobj.ktype = &queue_ktype;
1840 kobject_init(&q->kobj);
1da177e4
LT
1841
1842 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1843 q->backing_dev_info.unplug_io_data = q;
1844
483f4afc
AV
1845 mutex_init(&q->sysfs_lock);
1846
1da177e4
LT
1847 return q;
1848}
1946089a 1849EXPORT_SYMBOL(blk_alloc_queue_node);
1da177e4
LT
1850
1851/**
1852 * blk_init_queue - prepare a request queue for use with a block device
1853 * @rfn: The function to be called to process requests that have been
1854 * placed on the queue.
1855 * @lock: Request queue spin lock
1856 *
1857 * Description:
1858 * If a block device wishes to use the standard request handling procedures,
1859 * which sorts requests and coalesces adjacent requests, then it must
1860 * call blk_init_queue(). The function @rfn will be called when there
1861 * are requests on the queue that need to be processed. If the device
1862 * supports plugging, then @rfn may not be called immediately when requests
1863 * are available on the queue, but may be called at some time later instead.
1864 * Plugged queues are generally unplugged when a buffer belonging to one
1865 * of the requests on the queue is needed, or due to memory pressure.
1866 *
1867 * @rfn is not required, or even expected, to remove all requests off the
1868 * queue, but only as many as it can handle at a time. If it does leave
1869 * requests on the queue, it is responsible for arranging that the requests
1870 * get dealt with eventually.
1871 *
1872 * The queue spin lock must be held while manipulating the requests on the
a038e253
PBG
1873 * request queue; this lock will be taken also from interrupt context, so irq
1874 * disabling is needed for it.
1da177e4
LT
1875 *
1876 * Function returns a pointer to the initialized request queue, or NULL if
1877 * it didn't succeed.
1878 *
1879 * Note:
1880 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1881 * when the block device is deactivated (such as at module unload).
1882 **/
1946089a 1883
1da177e4
LT
1884request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1885{
1946089a
CL
1886 return blk_init_queue_node(rfn, lock, -1);
1887}
1888EXPORT_SYMBOL(blk_init_queue);
1889
1890request_queue_t *
1891blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1892{
1893 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
1da177e4
LT
1894
1895 if (!q)
1896 return NULL;
1897
1946089a 1898 q->node = node_id;
8669aafd
AV
1899 if (blk_init_free_list(q)) {
1900 kmem_cache_free(requestq_cachep, q);
1901 return NULL;
1902 }
1da177e4 1903
152587de
JA
1904 /*
1905 * if caller didn't supply a lock, they get per-queue locking with
1906 * our embedded lock
1907 */
1908 if (!lock) {
1909 spin_lock_init(&q->__queue_lock);
1910 lock = &q->__queue_lock;
1911 }
1912
1da177e4 1913 q->request_fn = rfn;
1da177e4
LT
1914 q->prep_rq_fn = NULL;
1915 q->unplug_fn = generic_unplug_device;
1916 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1917 q->queue_lock = lock;
1918
1919 blk_queue_segment_boundary(q, 0xffffffff);
1920
1921 blk_queue_make_request(q, __make_request);
1922 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1923
1924 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1925 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1926
44ec9542
AS
1927 q->sg_reserved_size = INT_MAX;
1928
1da177e4
LT
1929 /*
1930 * all done
1931 */
1932 if (!elevator_init(q, NULL)) {
1933 blk_queue_congestion_threshold(q);
1934 return q;
1935 }
1936
8669aafd 1937 blk_put_queue(q);
1da177e4
LT
1938 return NULL;
1939}
1946089a 1940EXPORT_SYMBOL(blk_init_queue_node);
1da177e4
LT
1941
1942int blk_get_queue(request_queue_t *q)
1943{
fde6ad22 1944 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
483f4afc 1945 kobject_get(&q->kobj);
1da177e4
LT
1946 return 0;
1947 }
1948
1949 return 1;
1950}
1951
1952EXPORT_SYMBOL(blk_get_queue);
1953
1954static inline void blk_free_request(request_queue_t *q, struct request *rq)
1955{
4aff5e23 1956 if (rq->cmd_flags & REQ_ELVPRIV)
cb98fc8b 1957 elv_put_request(q, rq);
1da177e4
LT
1958 mempool_free(rq, q->rq.rq_pool);
1959}
1960
1ea25ecb 1961static struct request *
cb78b285 1962blk_alloc_request(request_queue_t *q, int rw, int priv, gfp_t gfp_mask)
1da177e4
LT
1963{
1964 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1965
1966 if (!rq)
1967 return NULL;
1968
1969 /*
4aff5e23 1970 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
1da177e4
LT
1971 * see bio.h and blkdev.h
1972 */
49171e5c 1973 rq->cmd_flags = rw | REQ_ALLOCED;
1da177e4 1974
cb98fc8b 1975 if (priv) {
cb78b285 1976 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
cb98fc8b
TH
1977 mempool_free(rq, q->rq.rq_pool);
1978 return NULL;
1979 }
4aff5e23 1980 rq->cmd_flags |= REQ_ELVPRIV;
cb98fc8b 1981 }
1da177e4 1982
cb98fc8b 1983 return rq;
1da177e4
LT
1984}
1985
1986/*
1987 * ioc_batching returns true if the ioc is a valid batching request and
1988 * should be given priority access to a request.
1989 */
1990static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
1991{
1992 if (!ioc)
1993 return 0;
1994
1995 /*
1996 * Make sure the process is able to allocate at least 1 request
1997 * even if the batch times out, otherwise we could theoretically
1998 * lose wakeups.
1999 */
2000 return ioc->nr_batch_requests == q->nr_batching ||
2001 (ioc->nr_batch_requests > 0
2002 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2003}
2004
2005/*
2006 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2007 * will cause the process to be a "batcher" on all queues in the system. This
2008 * is the behaviour we want though - once it gets a wakeup it should be given
2009 * a nice run.
2010 */
93d17d3d 2011static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
1da177e4
LT
2012{
2013 if (!ioc || ioc_batching(q, ioc))
2014 return;
2015
2016 ioc->nr_batch_requests = q->nr_batching;
2017 ioc->last_waited = jiffies;
2018}
2019
2020static void __freed_request(request_queue_t *q, int rw)
2021{
2022 struct request_list *rl = &q->rq;
2023
2024 if (rl->count[rw] < queue_congestion_off_threshold(q))
79e2de4b 2025 blk_clear_queue_congested(q, rw);
1da177e4
LT
2026
2027 if (rl->count[rw] + 1 <= q->nr_requests) {
1da177e4
LT
2028 if (waitqueue_active(&rl->wait[rw]))
2029 wake_up(&rl->wait[rw]);
2030
2031 blk_clear_queue_full(q, rw);
2032 }
2033}
2034
2035/*
2036 * A request has just been released. Account for it, update the full and
2037 * congestion status, wake up any waiters. Called under q->queue_lock.
2038 */
cb98fc8b 2039static void freed_request(request_queue_t *q, int rw, int priv)
1da177e4
LT
2040{
2041 struct request_list *rl = &q->rq;
2042
2043 rl->count[rw]--;
cb98fc8b
TH
2044 if (priv)
2045 rl->elvpriv--;
1da177e4
LT
2046
2047 __freed_request(q, rw);
2048
2049 if (unlikely(rl->starved[rw ^ 1]))
2050 __freed_request(q, rw ^ 1);
1da177e4
LT
2051}
2052
2053#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2054/*
d6344532
NP
2055 * Get a free request, queue_lock must be held.
2056 * Returns NULL on failure, with queue_lock held.
2057 * Returns !NULL on success, with queue_lock *not held*.
1da177e4 2058 */
7749a8d4
JA
2059static struct request *get_request(request_queue_t *q, int rw_flags,
2060 struct bio *bio, gfp_t gfp_mask)
1da177e4
LT
2061{
2062 struct request *rq = NULL;
2063 struct request_list *rl = &q->rq;
88ee5ef1 2064 struct io_context *ioc = NULL;
7749a8d4 2065 const int rw = rw_flags & 0x01;
88ee5ef1
JA
2066 int may_queue, priv;
2067
7749a8d4 2068 may_queue = elv_may_queue(q, rw_flags);
88ee5ef1
JA
2069 if (may_queue == ELV_MQUEUE_NO)
2070 goto rq_starved;
2071
2072 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2073 if (rl->count[rw]+1 >= q->nr_requests) {
b5deef90 2074 ioc = current_io_context(GFP_ATOMIC, q->node);
88ee5ef1
JA
2075 /*
2076 * The queue will fill after this allocation, so set
2077 * it as full, and mark this process as "batching".
2078 * This process will be allowed to complete a batch of
2079 * requests, others will be blocked.
2080 */
2081 if (!blk_queue_full(q, rw)) {
2082 ioc_set_batching(q, ioc);
2083 blk_set_queue_full(q, rw);
2084 } else {
2085 if (may_queue != ELV_MQUEUE_MUST
2086 && !ioc_batching(q, ioc)) {
2087 /*
2088 * The queue is full and the allocating
2089 * process is not a "batcher", and not
2090 * exempted by the IO scheduler
2091 */
2092 goto out;
2093 }
2094 }
1da177e4 2095 }
79e2de4b 2096 blk_set_queue_congested(q, rw);
1da177e4
LT
2097 }
2098
082cf69e
JA
2099 /*
2100 * Only allow batching queuers to allocate up to 50% over the defined
2101 * limit of requests, otherwise we could have thousands of requests
2102 * allocated with any setting of ->nr_requests
2103 */
fd782a4a 2104 if (rl->count[rw] >= (3 * q->nr_requests / 2))
082cf69e 2105 goto out;
fd782a4a 2106
1da177e4
LT
2107 rl->count[rw]++;
2108 rl->starved[rw] = 0;
cb98fc8b 2109
64521d1a 2110 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
cb98fc8b
TH
2111 if (priv)
2112 rl->elvpriv++;
2113
1da177e4
LT
2114 spin_unlock_irq(q->queue_lock);
2115
7749a8d4 2116 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
88ee5ef1 2117 if (unlikely(!rq)) {
1da177e4
LT
2118 /*
2119 * Allocation failed presumably due to memory. Undo anything
2120 * we might have messed up.
2121 *
2122 * Allocating task should really be put onto the front of the
2123 * wait queue, but this is pretty rare.
2124 */
2125 spin_lock_irq(q->queue_lock);
cb98fc8b 2126 freed_request(q, rw, priv);
1da177e4
LT
2127
2128 /*
2129 * in the very unlikely event that allocation failed and no
2130 * requests for this direction was pending, mark us starved
2131 * so that freeing of a request in the other direction will
2132 * notice us. another possible fix would be to split the
2133 * rq mempool into READ and WRITE
2134 */
2135rq_starved:
2136 if (unlikely(rl->count[rw] == 0))
2137 rl->starved[rw] = 1;
2138
1da177e4
LT
2139 goto out;
2140 }
2141
88ee5ef1
JA
2142 /*
2143 * ioc may be NULL here, and ioc_batching will be false. That's
2144 * OK, if the queue is under the request limit then requests need
2145 * not count toward the nr_batch_requests limit. There will always
2146 * be some limit enforced by BLK_BATCH_TIME.
2147 */
1da177e4
LT
2148 if (ioc_batching(q, ioc))
2149 ioc->nr_batch_requests--;
2150
2151 rq_init(q, rq);
2056a782
JA
2152
2153 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
1da177e4 2154out:
1da177e4
LT
2155 return rq;
2156}
2157
2158/*
2159 * No available requests for this queue, unplug the device and wait for some
2160 * requests to become available.
d6344532
NP
2161 *
2162 * Called with q->queue_lock held, and returns with it unlocked.
1da177e4 2163 */
7749a8d4 2164static struct request *get_request_wait(request_queue_t *q, int rw_flags,
22e2c507 2165 struct bio *bio)
1da177e4 2166{
7749a8d4 2167 const int rw = rw_flags & 0x01;
1da177e4
LT
2168 struct request *rq;
2169
7749a8d4 2170 rq = get_request(q, rw_flags, bio, GFP_NOIO);
450991bc
NP
2171 while (!rq) {
2172 DEFINE_WAIT(wait);
1da177e4
LT
2173 struct request_list *rl = &q->rq;
2174
2175 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2176 TASK_UNINTERRUPTIBLE);
2177
7749a8d4 2178 rq = get_request(q, rw_flags, bio, GFP_NOIO);
1da177e4
LT
2179
2180 if (!rq) {
2181 struct io_context *ioc;
2182
2056a782
JA
2183 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2184
d6344532
NP
2185 __generic_unplug_device(q);
2186 spin_unlock_irq(q->queue_lock);
1da177e4
LT
2187 io_schedule();
2188
2189 /*
2190 * After sleeping, we become a "batching" process and
2191 * will be able to allocate at least one request, and
2192 * up to a big batch of them for a small period time.
2193 * See ioc_batching, ioc_set_batching
2194 */
b5deef90 2195 ioc = current_io_context(GFP_NOIO, q->node);
1da177e4 2196 ioc_set_batching(q, ioc);
d6344532
NP
2197
2198 spin_lock_irq(q->queue_lock);
1da177e4
LT
2199 }
2200 finish_wait(&rl->wait[rw], &wait);
450991bc 2201 }
1da177e4
LT
2202
2203 return rq;
2204}
2205
8267e268 2206struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
1da177e4
LT
2207{
2208 struct request *rq;
2209
2210 BUG_ON(rw != READ && rw != WRITE);
2211
d6344532
NP
2212 spin_lock_irq(q->queue_lock);
2213 if (gfp_mask & __GFP_WAIT) {
22e2c507 2214 rq = get_request_wait(q, rw, NULL);
d6344532 2215 } else {
22e2c507 2216 rq = get_request(q, rw, NULL, gfp_mask);
d6344532
NP
2217 if (!rq)
2218 spin_unlock_irq(q->queue_lock);
2219 }
2220 /* q->queue_lock is unlocked at this point */
1da177e4
LT
2221
2222 return rq;
2223}
1da177e4
LT
2224EXPORT_SYMBOL(blk_get_request);
2225
dc72ef4a
JA
2226/**
2227 * blk_start_queueing - initiate dispatch of requests to device
2228 * @q: request queue to kick into gear
2229 *
2230 * This is basically a helper to remove the need to know whether a queue
2231 * is plugged or not if someone just wants to initiate dispatch of requests
2232 * for this queue.
2233 *
2234 * The queue lock must be held with interrupts disabled.
2235 */
2236void blk_start_queueing(request_queue_t *q)
2237{
2238 if (!blk_queue_plugged(q))
2239 q->request_fn(q);
2240 else
2241 __generic_unplug_device(q);
2242}
2243EXPORT_SYMBOL(blk_start_queueing);
2244
1da177e4
LT
2245/**
2246 * blk_requeue_request - put a request back on queue
2247 * @q: request queue where request should be inserted
2248 * @rq: request to be inserted
2249 *
2250 * Description:
2251 * Drivers often keep queueing requests until the hardware cannot accept
2252 * more, when that condition happens we need to put the request back
2253 * on the queue. Must be called with queue lock held.
2254 */
2255void blk_requeue_request(request_queue_t *q, struct request *rq)
2256{
2056a782
JA
2257 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2258
1da177e4
LT
2259 if (blk_rq_tagged(rq))
2260 blk_queue_end_tag(q, rq);
2261
2262 elv_requeue_request(q, rq);
2263}
2264
2265EXPORT_SYMBOL(blk_requeue_request);
2266
2267/**
2268 * blk_insert_request - insert a special request in to a request queue
2269 * @q: request queue where request should be inserted
2270 * @rq: request to be inserted
2271 * @at_head: insert request at head or tail of queue
2272 * @data: private data
1da177e4
LT
2273 *
2274 * Description:
2275 * Many block devices need to execute commands asynchronously, so they don't
2276 * block the whole kernel from preemption during request execution. This is
2277 * accomplished normally by inserting aritficial requests tagged as
2278 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2279 * scheduled for actual execution by the request queue.
2280 *
2281 * We have the option of inserting the head or the tail of the queue.
2282 * Typically we use the tail for new ioctls and so forth. We use the head
2283 * of the queue for things like a QUEUE_FULL message from a device, or a
2284 * host that is unable to accept a particular command.
2285 */
2286void blk_insert_request(request_queue_t *q, struct request *rq,
867d1191 2287 int at_head, void *data)
1da177e4 2288{
867d1191 2289 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
1da177e4
LT
2290 unsigned long flags;
2291
2292 /*
2293 * tell I/O scheduler that this isn't a regular read/write (ie it
2294 * must not attempt merges on this) and that it acts as a soft
2295 * barrier
2296 */
4aff5e23
JA
2297 rq->cmd_type = REQ_TYPE_SPECIAL;
2298 rq->cmd_flags |= REQ_SOFTBARRIER;
1da177e4
LT
2299
2300 rq->special = data;
2301
2302 spin_lock_irqsave(q->queue_lock, flags);
2303
2304 /*
2305 * If command is tagged, release the tag
2306 */
867d1191
TH
2307 if (blk_rq_tagged(rq))
2308 blk_queue_end_tag(q, rq);
1da177e4 2309
867d1191
TH
2310 drive_stat_acct(rq, rq->nr_sectors, 1);
2311 __elv_add_request(q, rq, where, 0);
dc72ef4a 2312 blk_start_queueing(q);
1da177e4
LT
2313 spin_unlock_irqrestore(q->queue_lock, flags);
2314}
2315
2316EXPORT_SYMBOL(blk_insert_request);
2317
0e75f906
MC
2318static int __blk_rq_unmap_user(struct bio *bio)
2319{
2320 int ret = 0;
2321
2322 if (bio) {
2323 if (bio_flagged(bio, BIO_USER_MAPPED))
2324 bio_unmap_user(bio);
2325 else
2326 ret = bio_uncopy_user(bio);
2327 }
2328
2329 return ret;
2330}
2331
2332static int __blk_rq_map_user(request_queue_t *q, struct request *rq,
2333 void __user *ubuf, unsigned int len)
2334{
2335 unsigned long uaddr;
2336 struct bio *bio, *orig_bio;
2337 int reading, ret;
2338
2339 reading = rq_data_dir(rq) == READ;
2340
2341 /*
2342 * if alignment requirement is satisfied, map in user pages for
2343 * direct dma. else, set up kernel bounce buffers
2344 */
2345 uaddr = (unsigned long) ubuf;
2346 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2347 bio = bio_map_user(q, NULL, uaddr, len, reading);
2348 else
2349 bio = bio_copy_user(q, uaddr, len, reading);
2350
2985259b 2351 if (IS_ERR(bio))
0e75f906 2352 return PTR_ERR(bio);
0e75f906
MC
2353
2354 orig_bio = bio;
2355 blk_queue_bounce(q, &bio);
2985259b 2356
0e75f906
MC
2357 /*
2358 * We link the bounce buffer in and could have to traverse it
2359 * later so we have to get a ref to prevent it from being freed
2360 */
2361 bio_get(bio);
2362
0e75f906
MC
2363 if (!rq->bio)
2364 blk_rq_bio_prep(q, rq, bio);
1aa4f24f 2365 else if (!ll_back_merge_fn(q, rq, bio)) {
0e75f906 2366 ret = -EINVAL;
0e75f906
MC
2367 goto unmap_bio;
2368 } else {
2369 rq->biotail->bi_next = bio;
2370 rq->biotail = bio;
2371
0e75f906
MC
2372 rq->data_len += bio->bi_size;
2373 }
0e75f906
MC
2374
2375 return bio->bi_size;
2376
2377unmap_bio:
2378 /* if it was boucned we must call the end io function */
2379 bio_endio(bio, bio->bi_size, 0);
2380 __blk_rq_unmap_user(orig_bio);
2381 bio_put(bio);
2382 return ret;
2383}
2384
1da177e4
LT
2385/**
2386 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2387 * @q: request queue where request should be inserted
73747aed 2388 * @rq: request structure to fill
1da177e4
LT
2389 * @ubuf: the user buffer
2390 * @len: length of user data
2391 *
2392 * Description:
2393 * Data will be mapped directly for zero copy io, if possible. Otherwise
2394 * a kernel bounce buffer is used.
2395 *
2396 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2397 * still in process context.
2398 *
2399 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2400 * before being submitted to the device, as pages mapped may be out of
2401 * reach. It's the callers responsibility to make sure this happens. The
2402 * original bio must be passed back in to blk_rq_unmap_user() for proper
2403 * unmapping.
2404 */
dd1cab95 2405int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
0e75f906 2406 unsigned long len)
1da177e4 2407{
0e75f906 2408 unsigned long bytes_read = 0;
8e5cfc45 2409 struct bio *bio = NULL;
0e75f906 2410 int ret;
1da177e4 2411
defd94b7 2412 if (len > (q->max_hw_sectors << 9))
dd1cab95
JA
2413 return -EINVAL;
2414 if (!len || !ubuf)
2415 return -EINVAL;
1da177e4 2416
0e75f906
MC
2417 while (bytes_read != len) {
2418 unsigned long map_len, end, start;
1da177e4 2419
0e75f906
MC
2420 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
2421 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
2422 >> PAGE_SHIFT;
2423 start = (unsigned long)ubuf >> PAGE_SHIFT;
1da177e4 2424
0e75f906
MC
2425 /*
2426 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2427 * pages. If this happens we just lower the requested
2428 * mapping len by a page so that we can fit
2429 */
2430 if (end - start > BIO_MAX_PAGES)
2431 map_len -= PAGE_SIZE;
1da177e4 2432
0e75f906
MC
2433 ret = __blk_rq_map_user(q, rq, ubuf, map_len);
2434 if (ret < 0)
2435 goto unmap_rq;
8e5cfc45
JA
2436 if (!bio)
2437 bio = rq->bio;
0e75f906
MC
2438 bytes_read += ret;
2439 ubuf += ret;
1da177e4
LT
2440 }
2441
0e75f906
MC
2442 rq->buffer = rq->data = NULL;
2443 return 0;
2444unmap_rq:
8e5cfc45 2445 blk_rq_unmap_user(bio);
0e75f906 2446 return ret;
1da177e4
LT
2447}
2448
2449EXPORT_SYMBOL(blk_rq_map_user);
2450
f1970baf
JB
2451/**
2452 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2453 * @q: request queue where request should be inserted
2454 * @rq: request to map data to
2455 * @iov: pointer to the iovec
2456 * @iov_count: number of elements in the iovec
af9997e4 2457 * @len: I/O byte count
f1970baf
JB
2458 *
2459 * Description:
2460 * Data will be mapped directly for zero copy io, if possible. Otherwise
2461 * a kernel bounce buffer is used.
2462 *
2463 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2464 * still in process context.
2465 *
2466 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2467 * before being submitted to the device, as pages mapped may be out of
2468 * reach. It's the callers responsibility to make sure this happens. The
2469 * original bio must be passed back in to blk_rq_unmap_user() for proper
2470 * unmapping.
2471 */
2472int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
0e75f906 2473 struct sg_iovec *iov, int iov_count, unsigned int len)
f1970baf
JB
2474{
2475 struct bio *bio;
2476
2477 if (!iov || iov_count <= 0)
2478 return -EINVAL;
2479
2480 /* we don't allow misaligned data like bio_map_user() does. If the
2481 * user is using sg, they're expected to know the alignment constraints
2482 * and respect them accordingly */
2483 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2484 if (IS_ERR(bio))
2485 return PTR_ERR(bio);
2486
0e75f906
MC
2487 if (bio->bi_size != len) {
2488 bio_endio(bio, bio->bi_size, 0);
2489 bio_unmap_user(bio);
2490 return -EINVAL;
2491 }
2492
2493 bio_get(bio);
f1970baf
JB
2494 blk_rq_bio_prep(q, rq, bio);
2495 rq->buffer = rq->data = NULL;
f1970baf
JB
2496 return 0;
2497}
2498
2499EXPORT_SYMBOL(blk_rq_map_user_iov);
2500
1da177e4
LT
2501/**
2502 * blk_rq_unmap_user - unmap a request with user data
8e5cfc45 2503 * @bio: start of bio list
1da177e4
LT
2504 *
2505 * Description:
8e5cfc45
JA
2506 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2507 * supply the original rq->bio from the blk_rq_map_user() return, since
2508 * the io completion may have changed rq->bio.
1da177e4 2509 */
8e5cfc45 2510int blk_rq_unmap_user(struct bio *bio)
1da177e4 2511{
8e5cfc45 2512 struct bio *mapped_bio;
48785bb9 2513 int ret = 0, ret2;
1da177e4 2514
8e5cfc45
JA
2515 while (bio) {
2516 mapped_bio = bio;
2517 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
0e75f906 2518 mapped_bio = bio->bi_private;
1da177e4 2519
48785bb9
JA
2520 ret2 = __blk_rq_unmap_user(mapped_bio);
2521 if (ret2 && !ret)
2522 ret = ret2;
2523
8e5cfc45
JA
2524 mapped_bio = bio;
2525 bio = bio->bi_next;
2526 bio_put(mapped_bio);
0e75f906 2527 }
48785bb9
JA
2528
2529 return ret;
1da177e4
LT
2530}
2531
2532EXPORT_SYMBOL(blk_rq_unmap_user);
2533
df46b9a4
MC
2534/**
2535 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2536 * @q: request queue where request should be inserted
73747aed 2537 * @rq: request to fill
df46b9a4
MC
2538 * @kbuf: the kernel buffer
2539 * @len: length of user data
73747aed 2540 * @gfp_mask: memory allocation flags
df46b9a4 2541 */
dd1cab95 2542int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
8267e268 2543 unsigned int len, gfp_t gfp_mask)
df46b9a4 2544{
df46b9a4
MC
2545 struct bio *bio;
2546
defd94b7 2547 if (len > (q->max_hw_sectors << 9))
dd1cab95
JA
2548 return -EINVAL;
2549 if (!len || !kbuf)
2550 return -EINVAL;
df46b9a4
MC
2551
2552 bio = bio_map_kern(q, kbuf, len, gfp_mask);
dd1cab95
JA
2553 if (IS_ERR(bio))
2554 return PTR_ERR(bio);
df46b9a4 2555
dd1cab95
JA
2556 if (rq_data_dir(rq) == WRITE)
2557 bio->bi_rw |= (1 << BIO_RW);
df46b9a4 2558
dd1cab95 2559 blk_rq_bio_prep(q, rq, bio);
821de3a2 2560 blk_queue_bounce(q, &rq->bio);
dd1cab95 2561 rq->buffer = rq->data = NULL;
dd1cab95 2562 return 0;
df46b9a4
MC
2563}
2564
2565EXPORT_SYMBOL(blk_rq_map_kern);
2566
73747aed
CH
2567/**
2568 * blk_execute_rq_nowait - insert a request into queue for execution
2569 * @q: queue to insert the request in
2570 * @bd_disk: matching gendisk
2571 * @rq: request to insert
2572 * @at_head: insert request at head or tail of queue
2573 * @done: I/O completion handler
2574 *
2575 * Description:
2576 * Insert a fully prepared request at the back of the io scheduler queue
2577 * for execution. Don't wait for completion.
2578 */
f1970baf
JB
2579void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
2580 struct request *rq, int at_head,
8ffdc655 2581 rq_end_io_fn *done)
f1970baf
JB
2582{
2583 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2584
2585 rq->rq_disk = bd_disk;
4aff5e23 2586 rq->cmd_flags |= REQ_NOMERGE;
f1970baf 2587 rq->end_io = done;
4c5d0bbd
AM
2588 WARN_ON(irqs_disabled());
2589 spin_lock_irq(q->queue_lock);
2590 __elv_add_request(q, rq, where, 1);
2591 __generic_unplug_device(q);
2592 spin_unlock_irq(q->queue_lock);
f1970baf 2593}
6e39b69e
MC
2594EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2595
1da177e4
LT
2596/**
2597 * blk_execute_rq - insert a request into queue for execution
2598 * @q: queue to insert the request in
2599 * @bd_disk: matching gendisk
2600 * @rq: request to insert
994ca9a1 2601 * @at_head: insert request at head or tail of queue
1da177e4
LT
2602 *
2603 * Description:
2604 * Insert a fully prepared request at the back of the io scheduler queue
73747aed 2605 * for execution and wait for completion.
1da177e4
LT
2606 */
2607int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
994ca9a1 2608 struct request *rq, int at_head)
1da177e4 2609{
60be6b9a 2610 DECLARE_COMPLETION_ONSTACK(wait);
1da177e4
LT
2611 char sense[SCSI_SENSE_BUFFERSIZE];
2612 int err = 0;
2613
1da177e4
LT
2614 /*
2615 * we need an extra reference to the request, so we can look at
2616 * it after io completion
2617 */
2618 rq->ref_count++;
2619
2620 if (!rq->sense) {
2621 memset(sense, 0, sizeof(sense));
2622 rq->sense = sense;
2623 rq->sense_len = 0;
2624 }
2625
c00895ab 2626 rq->end_io_data = &wait;
994ca9a1 2627 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
1da177e4 2628 wait_for_completion(&wait);
1da177e4
LT
2629
2630 if (rq->errors)
2631 err = -EIO;
2632
2633 return err;
2634}
2635
2636EXPORT_SYMBOL(blk_execute_rq);
2637
2638/**
2639 * blkdev_issue_flush - queue a flush
2640 * @bdev: blockdev to issue flush for
2641 * @error_sector: error sector
2642 *
2643 * Description:
2644 * Issue a flush for the block device in question. Caller can supply
2645 * room for storing the error offset in case of a flush error, if they
2646 * wish to. Caller must run wait_for_completion() on its own.
2647 */
2648int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2649{
2650 request_queue_t *q;
2651
2652 if (bdev->bd_disk == NULL)
2653 return -ENXIO;
2654
2655 q = bdev_get_queue(bdev);
2656 if (!q)
2657 return -ENXIO;
2658 if (!q->issue_flush_fn)
2659 return -EOPNOTSUPP;
2660
2661 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2662}
2663
2664EXPORT_SYMBOL(blkdev_issue_flush);
2665
93d17d3d 2666static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
1da177e4
LT
2667{
2668 int rw = rq_data_dir(rq);
2669
2670 if (!blk_fs_request(rq) || !rq->rq_disk)
2671 return;
2672
d72d904a 2673 if (!new_io) {
a362357b 2674 __disk_stat_inc(rq->rq_disk, merges[rw]);
d72d904a 2675 } else {
1da177e4
LT
2676 disk_round_stats(rq->rq_disk);
2677 rq->rq_disk->in_flight++;
2678 }
2679}
2680
2681/*
2682 * add-request adds a request to the linked list.
2683 * queue lock is held and interrupts disabled, as we muck with the
2684 * request queue list.
2685 */
2686static inline void add_request(request_queue_t * q, struct request * req)
2687{
2688 drive_stat_acct(req, req->nr_sectors, 1);
2689
1da177e4
LT
2690 /*
2691 * elevator indicated where it wants this request to be
2692 * inserted at elevator_merge time
2693 */
2694 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2695}
2696
2697/*
2698 * disk_round_stats() - Round off the performance stats on a struct
2699 * disk_stats.
2700 *
2701 * The average IO queue length and utilisation statistics are maintained
2702 * by observing the current state of the queue length and the amount of
2703 * time it has been in this state for.
2704 *
2705 * Normally, that accounting is done on IO completion, but that can result
2706 * in more than a second's worth of IO being accounted for within any one
2707 * second, leading to >100% utilisation. To deal with that, we call this
2708 * function to do a round-off before returning the results when reading
2709 * /proc/diskstats. This accounts immediately for all queue usage up to
2710 * the current jiffies and restarts the counters again.
2711 */
2712void disk_round_stats(struct gendisk *disk)
2713{
2714 unsigned long now = jiffies;
2715
b2982649
KC
2716 if (now == disk->stamp)
2717 return;
1da177e4 2718
20e5c81f
KC
2719 if (disk->in_flight) {
2720 __disk_stat_add(disk, time_in_queue,
2721 disk->in_flight * (now - disk->stamp));
2722 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2723 }
1da177e4 2724 disk->stamp = now;
1da177e4
LT
2725}
2726
3eaf840e
JNN
2727EXPORT_SYMBOL_GPL(disk_round_stats);
2728
1da177e4
LT
2729/*
2730 * queue lock must be held
2731 */
6e39b69e 2732void __blk_put_request(request_queue_t *q, struct request *req)
1da177e4 2733{
1da177e4
LT
2734 if (unlikely(!q))
2735 return;
2736 if (unlikely(--req->ref_count))
2737 return;
2738
8922e16c
TH
2739 elv_completed_request(q, req);
2740
1da177e4
LT
2741 /*
2742 * Request may not have originated from ll_rw_blk. if not,
2743 * it didn't come out of our reserved rq pools
2744 */
49171e5c 2745 if (req->cmd_flags & REQ_ALLOCED) {
1da177e4 2746 int rw = rq_data_dir(req);
4aff5e23 2747 int priv = req->cmd_flags & REQ_ELVPRIV;
1da177e4 2748
1da177e4 2749 BUG_ON(!list_empty(&req->queuelist));
9817064b 2750 BUG_ON(!hlist_unhashed(&req->hash));
1da177e4
LT
2751
2752 blk_free_request(q, req);
cb98fc8b 2753 freed_request(q, rw, priv);
1da177e4
LT
2754 }
2755}
2756
6e39b69e
MC
2757EXPORT_SYMBOL_GPL(__blk_put_request);
2758
1da177e4
LT
2759void blk_put_request(struct request *req)
2760{
8922e16c
TH
2761 unsigned long flags;
2762 request_queue_t *q = req->q;
2763
1da177e4 2764 /*
8922e16c
TH
2765 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2766 * following if (q) test.
1da177e4 2767 */
8922e16c 2768 if (q) {
1da177e4
LT
2769 spin_lock_irqsave(q->queue_lock, flags);
2770 __blk_put_request(q, req);
2771 spin_unlock_irqrestore(q->queue_lock, flags);
2772 }
2773}
2774
2775EXPORT_SYMBOL(blk_put_request);
2776
2777/**
2778 * blk_end_sync_rq - executes a completion event on a request
2779 * @rq: request to complete
fddfdeaf 2780 * @error: end io status of the request
1da177e4 2781 */
8ffdc655 2782void blk_end_sync_rq(struct request *rq, int error)
1da177e4 2783{
c00895ab 2784 struct completion *waiting = rq->end_io_data;
1da177e4 2785
c00895ab 2786 rq->end_io_data = NULL;
1da177e4
LT
2787 __blk_put_request(rq->q, rq);
2788
2789 /*
2790 * complete last, if this is a stack request the process (and thus
2791 * the rq pointer) could be invalid right after this complete()
2792 */
2793 complete(waiting);
2794}
2795EXPORT_SYMBOL(blk_end_sync_rq);
2796
1da177e4
LT
2797/*
2798 * Has to be called with the request spinlock acquired
2799 */
2800static int attempt_merge(request_queue_t *q, struct request *req,
2801 struct request *next)
2802{
2803 if (!rq_mergeable(req) || !rq_mergeable(next))
2804 return 0;
2805
2806 /*
d6e05edc 2807 * not contiguous
1da177e4
LT
2808 */
2809 if (req->sector + req->nr_sectors != next->sector)
2810 return 0;
2811
2812 if (rq_data_dir(req) != rq_data_dir(next)
2813 || req->rq_disk != next->rq_disk
c00895ab 2814 || next->special)
1da177e4
LT
2815 return 0;
2816
2817 /*
2818 * If we are allowed to merge, then append bio list
2819 * from next to rq and release next. merge_requests_fn
2820 * will have updated segment counts, update sector
2821 * counts here.
2822 */
1aa4f24f 2823 if (!ll_merge_requests_fn(q, req, next))
1da177e4
LT
2824 return 0;
2825
2826 /*
2827 * At this point we have either done a back merge
2828 * or front merge. We need the smaller start_time of
2829 * the merged requests to be the current request
2830 * for accounting purposes.
2831 */
2832 if (time_after(req->start_time, next->start_time))
2833 req->start_time = next->start_time;
2834
2835 req->biotail->bi_next = next->bio;
2836 req->biotail = next->biotail;
2837
2838 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2839
2840 elv_merge_requests(q, req, next);
2841
2842 if (req->rq_disk) {
2843 disk_round_stats(req->rq_disk);
2844 req->rq_disk->in_flight--;
2845 }
2846
22e2c507
JA
2847 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2848
1da177e4
LT
2849 __blk_put_request(q, next);
2850 return 1;
2851}
2852
2853static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2854{
2855 struct request *next = elv_latter_request(q, rq);
2856
2857 if (next)
2858 return attempt_merge(q, rq, next);
2859
2860 return 0;
2861}
2862
2863static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2864{
2865 struct request *prev = elv_former_request(q, rq);
2866
2867 if (prev)
2868 return attempt_merge(q, prev, rq);
2869
2870 return 0;
2871}
2872
52d9e675
TH
2873static void init_request_from_bio(struct request *req, struct bio *bio)
2874{
4aff5e23 2875 req->cmd_type = REQ_TYPE_FS;
52d9e675
TH
2876
2877 /*
2878 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2879 */
2880 if (bio_rw_ahead(bio) || bio_failfast(bio))
4aff5e23 2881 req->cmd_flags |= REQ_FAILFAST;
52d9e675
TH
2882
2883 /*
2884 * REQ_BARRIER implies no merging, but lets make it explicit
2885 */
2886 if (unlikely(bio_barrier(bio)))
4aff5e23 2887 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
52d9e675 2888
b31dc66a 2889 if (bio_sync(bio))
4aff5e23 2890 req->cmd_flags |= REQ_RW_SYNC;
5404bc7a
JA
2891 if (bio_rw_meta(bio))
2892 req->cmd_flags |= REQ_RW_META;
b31dc66a 2893
52d9e675
TH
2894 req->errors = 0;
2895 req->hard_sector = req->sector = bio->bi_sector;
2896 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
2897 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
2898 req->nr_phys_segments = bio_phys_segments(req->q, bio);
2899 req->nr_hw_segments = bio_hw_segments(req->q, bio);
2900 req->buffer = bio_data(bio); /* see ->buffer comment above */
52d9e675
TH
2901 req->bio = req->biotail = bio;
2902 req->ioprio = bio_prio(bio);
2903 req->rq_disk = bio->bi_bdev->bd_disk;
2904 req->start_time = jiffies;
2905}
2906
1da177e4
LT
2907static int __make_request(request_queue_t *q, struct bio *bio)
2908{
450991bc 2909 struct request *req;
51da90fc
JA
2910 int el_ret, nr_sectors, barrier, err;
2911 const unsigned short prio = bio_prio(bio);
2912 const int sync = bio_sync(bio);
7749a8d4 2913 int rw_flags;
1da177e4 2914
1da177e4 2915 nr_sectors = bio_sectors(bio);
1da177e4
LT
2916
2917 /*
2918 * low level driver can indicate that it wants pages above a
2919 * certain limit bounced to low memory (ie for highmem, or even
2920 * ISA dma in theory)
2921 */
2922 blk_queue_bounce(q, &bio);
2923
1da177e4 2924 barrier = bio_barrier(bio);
797e7dbb 2925 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
1da177e4
LT
2926 err = -EOPNOTSUPP;
2927 goto end_io;
2928 }
2929
1da177e4
LT
2930 spin_lock_irq(q->queue_lock);
2931
450991bc 2932 if (unlikely(barrier) || elv_queue_empty(q))
1da177e4
LT
2933 goto get_rq;
2934
2935 el_ret = elv_merge(q, &req, bio);
2936 switch (el_ret) {
2937 case ELEVATOR_BACK_MERGE:
2938 BUG_ON(!rq_mergeable(req));
2939
1aa4f24f 2940 if (!ll_back_merge_fn(q, req, bio))
1da177e4
LT
2941 break;
2942
2056a782
JA
2943 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
2944
1da177e4
LT
2945 req->biotail->bi_next = bio;
2946 req->biotail = bio;
2947 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
22e2c507 2948 req->ioprio = ioprio_best(req->ioprio, prio);
1da177e4
LT
2949 drive_stat_acct(req, nr_sectors, 0);
2950 if (!attempt_back_merge(q, req))
2e662b65 2951 elv_merged_request(q, req, el_ret);
1da177e4
LT
2952 goto out;
2953
2954 case ELEVATOR_FRONT_MERGE:
2955 BUG_ON(!rq_mergeable(req));
2956
1aa4f24f 2957 if (!ll_front_merge_fn(q, req, bio))
1da177e4
LT
2958 break;
2959
2056a782
JA
2960 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
2961
1da177e4
LT
2962 bio->bi_next = req->bio;
2963 req->bio = bio;
2964
2965 /*
2966 * may not be valid. if the low level driver said
2967 * it didn't need a bounce buffer then it better
2968 * not touch req->buffer either...
2969 */
2970 req->buffer = bio_data(bio);
51da90fc
JA
2971 req->current_nr_sectors = bio_cur_sectors(bio);
2972 req->hard_cur_sectors = req->current_nr_sectors;
2973 req->sector = req->hard_sector = bio->bi_sector;
1da177e4 2974 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
22e2c507 2975 req->ioprio = ioprio_best(req->ioprio, prio);
1da177e4
LT
2976 drive_stat_acct(req, nr_sectors, 0);
2977 if (!attempt_front_merge(q, req))
2e662b65 2978 elv_merged_request(q, req, el_ret);
1da177e4
LT
2979 goto out;
2980
450991bc 2981 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1da177e4 2982 default:
450991bc 2983 ;
1da177e4
LT
2984 }
2985
450991bc 2986get_rq:
7749a8d4
JA
2987 /*
2988 * This sync check and mask will be re-done in init_request_from_bio(),
2989 * but we need to set it earlier to expose the sync flag to the
2990 * rq allocator and io schedulers.
2991 */
2992 rw_flags = bio_data_dir(bio);
2993 if (sync)
2994 rw_flags |= REQ_RW_SYNC;
2995
1da177e4 2996 /*
450991bc 2997 * Grab a free request. This is might sleep but can not fail.
d6344532 2998 * Returns with the queue unlocked.
450991bc 2999 */
7749a8d4 3000 req = get_request_wait(q, rw_flags, bio);
d6344532 3001
450991bc
NP
3002 /*
3003 * After dropping the lock and possibly sleeping here, our request
3004 * may now be mergeable after it had proven unmergeable (above).
3005 * We don't worry about that case for efficiency. It won't happen
3006 * often, and the elevators are able to handle it.
1da177e4 3007 */
52d9e675 3008 init_request_from_bio(req, bio);
1da177e4 3009
450991bc
NP
3010 spin_lock_irq(q->queue_lock);
3011 if (elv_queue_empty(q))
3012 blk_plug_device(q);
1da177e4
LT
3013 add_request(q, req);
3014out:
4a534f93 3015 if (sync)
1da177e4
LT
3016 __generic_unplug_device(q);
3017
3018 spin_unlock_irq(q->queue_lock);
3019 return 0;
3020
3021end_io:
3022 bio_endio(bio, nr_sectors << 9, err);
3023 return 0;
3024}
3025
3026/*
3027 * If bio->bi_dev is a partition, remap the location
3028 */
3029static inline void blk_partition_remap(struct bio *bio)
3030{
3031 struct block_device *bdev = bio->bi_bdev;
3032
3033 if (bdev != bdev->bd_contains) {
3034 struct hd_struct *p = bdev->bd_part;
a362357b
JA
3035 const int rw = bio_data_dir(bio);
3036
3037 p->sectors[rw] += bio_sectors(bio);
3038 p->ios[rw]++;
1da177e4 3039
1da177e4
LT
3040 bio->bi_sector += p->start_sect;
3041 bio->bi_bdev = bdev->bd_contains;
3042 }
3043}
3044
1da177e4
LT
3045static void handle_bad_sector(struct bio *bio)
3046{
3047 char b[BDEVNAME_SIZE];
3048
3049 printk(KERN_INFO "attempt to access beyond end of device\n");
3050 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3051 bdevname(bio->bi_bdev, b),
3052 bio->bi_rw,
3053 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3054 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3055
3056 set_bit(BIO_EOF, &bio->bi_flags);
3057}
3058
c17bb495
AM
3059#ifdef CONFIG_FAIL_MAKE_REQUEST
3060
3061static DECLARE_FAULT_ATTR(fail_make_request);
3062
3063static int __init setup_fail_make_request(char *str)
3064{
3065 return setup_fault_attr(&fail_make_request, str);
3066}
3067__setup("fail_make_request=", setup_fail_make_request);
3068
3069static int should_fail_request(struct bio *bio)
3070{
3071 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
3072 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
3073 return should_fail(&fail_make_request, bio->bi_size);
3074
3075 return 0;
3076}
3077
3078static int __init fail_make_request_debugfs(void)
3079{
3080 return init_fault_attr_dentries(&fail_make_request,
3081 "fail_make_request");
3082}
3083
3084late_initcall(fail_make_request_debugfs);
3085
3086#else /* CONFIG_FAIL_MAKE_REQUEST */
3087
3088static inline int should_fail_request(struct bio *bio)
3089{
3090 return 0;
3091}
3092
3093#endif /* CONFIG_FAIL_MAKE_REQUEST */
3094
1da177e4
LT
3095/**
3096 * generic_make_request: hand a buffer to its device driver for I/O
3097 * @bio: The bio describing the location in memory and on the device.
3098 *
3099 * generic_make_request() is used to make I/O requests of block
3100 * devices. It is passed a &struct bio, which describes the I/O that needs
3101 * to be done.
3102 *
3103 * generic_make_request() does not return any status. The
3104 * success/failure status of the request, along with notification of
3105 * completion, is delivered asynchronously through the bio->bi_end_io
3106 * function described (one day) else where.
3107 *
3108 * The caller of generic_make_request must make sure that bi_io_vec
3109 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3110 * set to describe the device address, and the
3111 * bi_end_io and optionally bi_private are set to describe how
3112 * completion notification should be signaled.
3113 *
3114 * generic_make_request and the drivers it calls may use bi_next if this
3115 * bio happens to be merged with someone else, and may change bi_dev and
3116 * bi_sector for remaps as it sees fit. So the values of these fields
3117 * should NOT be depended on after the call to generic_make_request.
3118 */
3119void generic_make_request(struct bio *bio)
3120{
3121 request_queue_t *q;
3122 sector_t maxsector;
5ddfe969 3123 sector_t old_sector;
1da177e4 3124 int ret, nr_sectors = bio_sectors(bio);
2056a782 3125 dev_t old_dev;
1da177e4
LT
3126
3127 might_sleep();
3128 /* Test device or partition size, when known. */
3129 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3130 if (maxsector) {
3131 sector_t sector = bio->bi_sector;
3132
3133 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3134 /*
3135 * This may well happen - the kernel calls bread()
3136 * without checking the size of the device, e.g., when
3137 * mounting a device.
3138 */
3139 handle_bad_sector(bio);
3140 goto end_io;
3141 }
3142 }
3143
3144 /*
3145 * Resolve the mapping until finished. (drivers are
3146 * still free to implement/resolve their own stacking
3147 * by explicitly returning 0)
3148 *
3149 * NOTE: we don't repeat the blk_size check for each new device.
3150 * Stacking drivers are expected to know what they are doing.
3151 */
5ddfe969 3152 old_sector = -1;
2056a782 3153 old_dev = 0;
1da177e4
LT
3154 do {
3155 char b[BDEVNAME_SIZE];
3156
3157 q = bdev_get_queue(bio->bi_bdev);
3158 if (!q) {
3159 printk(KERN_ERR
3160 "generic_make_request: Trying to access "
3161 "nonexistent block-device %s (%Lu)\n",
3162 bdevname(bio->bi_bdev, b),
3163 (long long) bio->bi_sector);
3164end_io:
3165 bio_endio(bio, bio->bi_size, -EIO);
3166 break;
3167 }
3168
3169 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
3170 printk("bio too big device %s (%u > %u)\n",
3171 bdevname(bio->bi_bdev, b),
3172 bio_sectors(bio),
3173 q->max_hw_sectors);
3174 goto end_io;
3175 }
3176
fde6ad22 3177 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1da177e4
LT
3178 goto end_io;
3179
c17bb495
AM
3180 if (should_fail_request(bio))
3181 goto end_io;
3182
1da177e4
LT
3183 /*
3184 * If this device has partitions, remap block n
3185 * of partition p to block n+start(p) of the disk.
3186 */
3187 blk_partition_remap(bio);
3188
5ddfe969 3189 if (old_sector != -1)
2056a782 3190 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
5ddfe969 3191 old_sector);
2056a782
JA
3192
3193 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3194
5ddfe969 3195 old_sector = bio->bi_sector;
2056a782
JA
3196 old_dev = bio->bi_bdev->bd_dev;
3197
5ddfe969
N
3198 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3199 if (maxsector) {
3200 sector_t sector = bio->bi_sector;
3201
df66b855
AM
3202 if (maxsector < nr_sectors ||
3203 maxsector - nr_sectors < sector) {
5ddfe969 3204 /*
df66b855
AM
3205 * This may well happen - partitions are not
3206 * checked to make sure they are within the size
3207 * of the whole device.
5ddfe969
N
3208 */
3209 handle_bad_sector(bio);
3210 goto end_io;
3211 }
3212 }
3213
1da177e4
LT
3214 ret = q->make_request_fn(q, bio);
3215 } while (ret);
3216}
3217
3218EXPORT_SYMBOL(generic_make_request);
3219
3220/**
3221 * submit_bio: submit a bio to the block device layer for I/O
3222 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3223 * @bio: The &struct bio which describes the I/O
3224 *
3225 * submit_bio() is very similar in purpose to generic_make_request(), and
3226 * uses that function to do most of the work. Both are fairly rough
3227 * interfaces, @bio must be presetup and ready for I/O.
3228 *
3229 */
3230void submit_bio(int rw, struct bio *bio)
3231{
3232 int count = bio_sectors(bio);
3233
3234 BIO_BUG_ON(!bio->bi_size);
3235 BIO_BUG_ON(!bio->bi_io_vec);
22e2c507 3236 bio->bi_rw |= rw;
faccbd4b 3237 if (rw & WRITE) {
f8891e5e 3238 count_vm_events(PGPGOUT, count);
faccbd4b
AM
3239 } else {
3240 task_io_account_read(bio->bi_size);
f8891e5e 3241 count_vm_events(PGPGIN, count);
faccbd4b 3242 }
1da177e4
LT
3243
3244 if (unlikely(block_dump)) {
3245 char b[BDEVNAME_SIZE];
3246 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3247 current->comm, current->pid,
3248 (rw & WRITE) ? "WRITE" : "READ",
3249 (unsigned long long)bio->bi_sector,
3250 bdevname(bio->bi_bdev,b));
3251 }
3252
3253 generic_make_request(bio);
3254}
3255
3256EXPORT_SYMBOL(submit_bio);
3257
93d17d3d 3258static void blk_recalc_rq_segments(struct request *rq)
1da177e4
LT
3259{
3260 struct bio *bio, *prevbio = NULL;
3261 int nr_phys_segs, nr_hw_segs;
3262 unsigned int phys_size, hw_size;
3263 request_queue_t *q = rq->q;
3264
3265 if (!rq->bio)
3266 return;
3267
3268 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
3269 rq_for_each_bio(bio, rq) {
3270 /* Force bio hw/phys segs to be recalculated. */
3271 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3272
3273 nr_phys_segs += bio_phys_segments(q, bio);
3274 nr_hw_segs += bio_hw_segments(q, bio);
3275 if (prevbio) {
3276 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3277 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3278
3279 if (blk_phys_contig_segment(q, prevbio, bio) &&
3280 pseg <= q->max_segment_size) {
3281 nr_phys_segs--;
3282 phys_size += prevbio->bi_size + bio->bi_size;
3283 } else
3284 phys_size = 0;
3285
3286 if (blk_hw_contig_segment(q, prevbio, bio) &&
3287 hseg <= q->max_segment_size) {
3288 nr_hw_segs--;
3289 hw_size += prevbio->bi_size + bio->bi_size;
3290 } else
3291 hw_size = 0;
3292 }
3293 prevbio = bio;
3294 }
3295
3296 rq->nr_phys_segments = nr_phys_segs;
3297 rq->nr_hw_segments = nr_hw_segs;
3298}
3299
93d17d3d 3300static void blk_recalc_rq_sectors(struct request *rq, int nsect)
1da177e4
LT
3301{
3302 if (blk_fs_request(rq)) {
3303 rq->hard_sector += nsect;
3304 rq->hard_nr_sectors -= nsect;
3305
3306 /*
3307 * Move the I/O submission pointers ahead if required.
3308 */
3309 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3310 (rq->sector <= rq->hard_sector)) {
3311 rq->sector = rq->hard_sector;
3312 rq->nr_sectors = rq->hard_nr_sectors;
3313 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3314 rq->current_nr_sectors = rq->hard_cur_sectors;
3315 rq->buffer = bio_data(rq->bio);
3316 }
3317
3318 /*
3319 * if total number of sectors is less than the first segment
3320 * size, something has gone terribly wrong
3321 */
3322 if (rq->nr_sectors < rq->current_nr_sectors) {
3323 printk("blk: request botched\n");
3324 rq->nr_sectors = rq->current_nr_sectors;
3325 }
3326 }
3327}
3328
3329static int __end_that_request_first(struct request *req, int uptodate,
3330 int nr_bytes)
3331{
3332 int total_bytes, bio_nbytes, error, next_idx = 0;
3333 struct bio *bio;
3334
2056a782
JA
3335 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3336
1da177e4
LT
3337 /*
3338 * extend uptodate bool to allow < 0 value to be direct io error
3339 */
3340 error = 0;
3341 if (end_io_error(uptodate))
3342 error = !uptodate ? -EIO : uptodate;
3343
3344 /*
3345 * for a REQ_BLOCK_PC request, we want to carry any eventual
3346 * sense key with us all the way through
3347 */
3348 if (!blk_pc_request(req))
3349 req->errors = 0;
3350
3351 if (!uptodate) {
4aff5e23 3352 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
1da177e4
LT
3353 printk("end_request: I/O error, dev %s, sector %llu\n",
3354 req->rq_disk ? req->rq_disk->disk_name : "?",
3355 (unsigned long long)req->sector);
3356 }
3357
d72d904a 3358 if (blk_fs_request(req) && req->rq_disk) {
a362357b
JA
3359 const int rw = rq_data_dir(req);
3360
53e86061 3361 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
d72d904a
JA
3362 }
3363
1da177e4
LT
3364 total_bytes = bio_nbytes = 0;
3365 while ((bio = req->bio) != NULL) {
3366 int nbytes;
3367
3368 if (nr_bytes >= bio->bi_size) {
3369 req->bio = bio->bi_next;
3370 nbytes = bio->bi_size;
797e7dbb
TH
3371 if (!ordered_bio_endio(req, bio, nbytes, error))
3372 bio_endio(bio, nbytes, error);
1da177e4
LT
3373 next_idx = 0;
3374 bio_nbytes = 0;
3375 } else {
3376 int idx = bio->bi_idx + next_idx;
3377
3378 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3379 blk_dump_rq_flags(req, "__end_that");
3380 printk("%s: bio idx %d >= vcnt %d\n",
3381 __FUNCTION__,
3382 bio->bi_idx, bio->bi_vcnt);
3383 break;
3384 }
3385
3386 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3387 BIO_BUG_ON(nbytes > bio->bi_size);
3388
3389 /*
3390 * not a complete bvec done
3391 */
3392 if (unlikely(nbytes > nr_bytes)) {
3393 bio_nbytes += nr_bytes;
3394 total_bytes += nr_bytes;
3395 break;
3396 }
3397
3398 /*
3399 * advance to the next vector
3400 */
3401 next_idx++;
3402 bio_nbytes += nbytes;
3403 }
3404
3405 total_bytes += nbytes;
3406 nr_bytes -= nbytes;
3407
3408 if ((bio = req->bio)) {
3409 /*
3410 * end more in this run, or just return 'not-done'
3411 */
3412 if (unlikely(nr_bytes <= 0))
3413 break;
3414 }
3415 }
3416
3417 /*
3418 * completely done
3419 */
3420 if (!req->bio)
3421 return 0;
3422
3423 /*
3424 * if the request wasn't completed, update state
3425 */
3426 if (bio_nbytes) {
797e7dbb
TH
3427 if (!ordered_bio_endio(req, bio, bio_nbytes, error))
3428 bio_endio(bio, bio_nbytes, error);
1da177e4
LT
3429 bio->bi_idx += next_idx;
3430 bio_iovec(bio)->bv_offset += nr_bytes;
3431 bio_iovec(bio)->bv_len -= nr_bytes;
3432 }
3433
3434 blk_recalc_rq_sectors(req, total_bytes >> 9);
3435 blk_recalc_rq_segments(req);
3436 return 1;
3437}
3438
3439/**
3440 * end_that_request_first - end I/O on a request
3441 * @req: the request being processed
3442 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3443 * @nr_sectors: number of sectors to end I/O on
3444 *
3445 * Description:
3446 * Ends I/O on a number of sectors attached to @req, and sets it up
3447 * for the next range of segments (if any) in the cluster.
3448 *
3449 * Return:
3450 * 0 - we are done with this request, call end_that_request_last()
3451 * 1 - still buffers pending for this request
3452 **/
3453int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3454{
3455 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3456}
3457
3458EXPORT_SYMBOL(end_that_request_first);
3459
3460/**
3461 * end_that_request_chunk - end I/O on a request
3462 * @req: the request being processed
3463 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3464 * @nr_bytes: number of bytes to complete
3465 *
3466 * Description:
3467 * Ends I/O on a number of bytes attached to @req, and sets it up
3468 * for the next range of segments (if any). Like end_that_request_first(),
3469 * but deals with bytes instead of sectors.
3470 *
3471 * Return:
3472 * 0 - we are done with this request, call end_that_request_last()
3473 * 1 - still buffers pending for this request
3474 **/
3475int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3476{
3477 return __end_that_request_first(req, uptodate, nr_bytes);
3478}
3479
3480EXPORT_SYMBOL(end_that_request_chunk);
3481
ff856bad
JA
3482/*
3483 * splice the completion data to a local structure and hand off to
3484 * process_completion_queue() to complete the requests
3485 */
3486static void blk_done_softirq(struct softirq_action *h)
3487{
626ab0e6 3488 struct list_head *cpu_list, local_list;
ff856bad
JA
3489
3490 local_irq_disable();
3491 cpu_list = &__get_cpu_var(blk_cpu_done);
626ab0e6 3492 list_replace_init(cpu_list, &local_list);
ff856bad
JA
3493 local_irq_enable();
3494
3495 while (!list_empty(&local_list)) {
3496 struct request *rq = list_entry(local_list.next, struct request, donelist);
3497
3498 list_del_init(&rq->donelist);
3499 rq->q->softirq_done_fn(rq);
3500 }
3501}
3502
ff856bad
JA
3503static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
3504 void *hcpu)
3505{
3506 /*
3507 * If a CPU goes away, splice its entries to the current CPU
3508 * and trigger a run of the softirq
3509 */
8bb78442 3510 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
ff856bad
JA
3511 int cpu = (unsigned long) hcpu;
3512
3513 local_irq_disable();
3514 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3515 &__get_cpu_var(blk_cpu_done));
3516 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3517 local_irq_enable();
3518 }
3519
3520 return NOTIFY_OK;
3521}
3522
3523
054cc8a2 3524static struct notifier_block __devinitdata blk_cpu_notifier = {
ff856bad
JA
3525 .notifier_call = blk_cpu_notify,
3526};
3527
ff856bad
JA
3528/**
3529 * blk_complete_request - end I/O on a request
3530 * @req: the request being processed
3531 *
3532 * Description:
3533 * Ends all I/O on a request. It does not handle partial completions,
d6e05edc 3534 * unless the driver actually implements this in its completion callback
ff856bad
JA
3535 * through requeueing. Theh actual completion happens out-of-order,
3536 * through a softirq handler. The user must have registered a completion
3537 * callback through blk_queue_softirq_done().
3538 **/
3539
3540void blk_complete_request(struct request *req)
3541{
3542 struct list_head *cpu_list;
3543 unsigned long flags;
3544
3545 BUG_ON(!req->q->softirq_done_fn);
3546
3547 local_irq_save(flags);
3548
3549 cpu_list = &__get_cpu_var(blk_cpu_done);
3550 list_add_tail(&req->donelist, cpu_list);
3551 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3552
3553 local_irq_restore(flags);
3554}
3555
3556EXPORT_SYMBOL(blk_complete_request);
3557
1da177e4
LT
3558/*
3559 * queue lock must be held
3560 */
8ffdc655 3561void end_that_request_last(struct request *req, int uptodate)
1da177e4
LT
3562{
3563 struct gendisk *disk = req->rq_disk;
8ffdc655
TH
3564 int error;
3565
3566 /*
3567 * extend uptodate bool to allow < 0 value to be direct io error
3568 */
3569 error = 0;
3570 if (end_io_error(uptodate))
3571 error = !uptodate ? -EIO : uptodate;
1da177e4
LT
3572
3573 if (unlikely(laptop_mode) && blk_fs_request(req))
3574 laptop_io_completion();
3575
fd0ff8aa
JA
3576 /*
3577 * Account IO completion. bar_rq isn't accounted as a normal
3578 * IO on queueing nor completion. Accounting the containing
3579 * request is enough.
3580 */
3581 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
1da177e4 3582 unsigned long duration = jiffies - req->start_time;
a362357b
JA
3583 const int rw = rq_data_dir(req);
3584
3585 __disk_stat_inc(disk, ios[rw]);
3586 __disk_stat_add(disk, ticks[rw], duration);
1da177e4
LT
3587 disk_round_stats(disk);
3588 disk->in_flight--;
3589 }
3590 if (req->end_io)
8ffdc655 3591 req->end_io(req, error);
1da177e4
LT
3592 else
3593 __blk_put_request(req->q, req);
3594}
3595
3596EXPORT_SYMBOL(end_that_request_last);
3597
3598void end_request(struct request *req, int uptodate)
3599{
3600 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3601 add_disk_randomness(req->rq_disk);
3602 blkdev_dequeue_request(req);
8ffdc655 3603 end_that_request_last(req, uptodate);
1da177e4
LT
3604 }
3605}
3606
3607EXPORT_SYMBOL(end_request);
3608
3609void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3610{
4aff5e23
JA
3611 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3612 rq->cmd_flags |= (bio->bi_rw & 3);
1da177e4
LT
3613
3614 rq->nr_phys_segments = bio_phys_segments(q, bio);
3615 rq->nr_hw_segments = bio_hw_segments(q, bio);
3616 rq->current_nr_sectors = bio_cur_sectors(bio);
3617 rq->hard_cur_sectors = rq->current_nr_sectors;
3618 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3619 rq->buffer = bio_data(bio);
0e75f906 3620 rq->data_len = bio->bi_size;
1da177e4
LT
3621
3622 rq->bio = rq->biotail = bio;
3623}
3624
3625EXPORT_SYMBOL(blk_rq_bio_prep);
3626
3627int kblockd_schedule_work(struct work_struct *work)
3628{
3629 return queue_work(kblockd_workqueue, work);
3630}
3631
3632EXPORT_SYMBOL(kblockd_schedule_work);
3633
19a75d83 3634void kblockd_flush_work(struct work_struct *work)
1da177e4 3635{
28e53bdd 3636 cancel_work_sync(work);
1da177e4 3637}
19a75d83 3638EXPORT_SYMBOL(kblockd_flush_work);
1da177e4
LT
3639
3640int __init blk_dev_init(void)
3641{
ff856bad
JA
3642 int i;
3643
1da177e4
LT
3644 kblockd_workqueue = create_workqueue("kblockd");
3645 if (!kblockd_workqueue)
3646 panic("Failed to create kblockd\n");
3647
3648 request_cachep = kmem_cache_create("blkdev_requests",
3649 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3650
3651 requestq_cachep = kmem_cache_create("blkdev_queue",
3652 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3653
3654 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3655 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3656
0a945022 3657 for_each_possible_cpu(i)
ff856bad
JA
3658 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3659
3660 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
5a67e4c5 3661 register_hotcpu_notifier(&blk_cpu_notifier);
ff856bad 3662
f772b3d9
VT
3663 blk_max_low_pfn = max_low_pfn - 1;
3664 blk_max_pfn = max_pfn - 1;
1da177e4
LT
3665
3666 return 0;
3667}
3668
3669/*
3670 * IO Context helper functions
3671 */
3672void put_io_context(struct io_context *ioc)
3673{
3674 if (ioc == NULL)
3675 return;
3676
3677 BUG_ON(atomic_read(&ioc->refcount) == 0);
3678
3679 if (atomic_dec_and_test(&ioc->refcount)) {
e2d74ac0
JA
3680 struct cfq_io_context *cic;
3681
334e94de 3682 rcu_read_lock();
1da177e4
LT
3683 if (ioc->aic && ioc->aic->dtor)
3684 ioc->aic->dtor(ioc->aic);
e2d74ac0 3685 if (ioc->cic_root.rb_node != NULL) {
7143dd4b
JA
3686 struct rb_node *n = rb_first(&ioc->cic_root);
3687
3688 cic = rb_entry(n, struct cfq_io_context, rb_node);
e2d74ac0
JA
3689 cic->dtor(ioc);
3690 }
334e94de 3691 rcu_read_unlock();
1da177e4
LT
3692
3693 kmem_cache_free(iocontext_cachep, ioc);
3694 }
3695}
3696EXPORT_SYMBOL(put_io_context);
3697
3698/* Called by the exitting task */
3699void exit_io_context(void)
3700{
1da177e4 3701 struct io_context *ioc;
e2d74ac0 3702 struct cfq_io_context *cic;
1da177e4 3703
22e2c507 3704 task_lock(current);
1da177e4
LT
3705 ioc = current->io_context;
3706 current->io_context = NULL;
22e2c507 3707 task_unlock(current);
1da177e4 3708
25034d7a 3709 ioc->task = NULL;
1da177e4
LT
3710 if (ioc->aic && ioc->aic->exit)
3711 ioc->aic->exit(ioc->aic);
e2d74ac0
JA
3712 if (ioc->cic_root.rb_node != NULL) {
3713 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3714 cic->exit(ioc);
3715 }
25034d7a 3716
1da177e4
LT
3717 put_io_context(ioc);
3718}
3719
3720/*
3721 * If the current task has no IO context then create one and initialise it.
fb3cc432 3722 * Otherwise, return its existing IO context.
1da177e4 3723 *
fb3cc432
NP
3724 * This returned IO context doesn't have a specifically elevated refcount,
3725 * but since the current task itself holds a reference, the context can be
3726 * used in general code, so long as it stays within `current` context.
1da177e4 3727 */
b5deef90 3728static struct io_context *current_io_context(gfp_t gfp_flags, int node)
1da177e4
LT
3729{
3730 struct task_struct *tsk = current;
1da177e4
LT
3731 struct io_context *ret;
3732
1da177e4 3733 ret = tsk->io_context;
fb3cc432
NP
3734 if (likely(ret))
3735 return ret;
1da177e4 3736
b5deef90 3737 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
1da177e4
LT
3738 if (ret) {
3739 atomic_set(&ret->refcount, 1);
22e2c507 3740 ret->task = current;
fc46379d 3741 ret->ioprio_changed = 0;
1da177e4
LT
3742 ret->last_waited = jiffies; /* doesn't matter... */
3743 ret->nr_batch_requests = 0; /* because this is 0 */
3744 ret->aic = NULL;
e2d74ac0 3745 ret->cic_root.rb_node = NULL;
4e521c27 3746 ret->ioc_data = NULL;
9f83e45e
ON
3747 /* make sure set_task_ioprio() sees the settings above */
3748 smp_wmb();
fb3cc432
NP
3749 tsk->io_context = ret;
3750 }
1da177e4 3751
fb3cc432
NP
3752 return ret;
3753}
3754EXPORT_SYMBOL(current_io_context);
1da177e4 3755
fb3cc432
NP
3756/*
3757 * If the current task has no IO context then create one and initialise it.
3758 * If it does have a context, take a ref on it.
3759 *
3760 * This is always called in the context of the task which submitted the I/O.
3761 */
b5deef90 3762struct io_context *get_io_context(gfp_t gfp_flags, int node)
fb3cc432
NP
3763{
3764 struct io_context *ret;
b5deef90 3765 ret = current_io_context(gfp_flags, node);
fb3cc432 3766 if (likely(ret))
1da177e4 3767 atomic_inc(&ret->refcount);
1da177e4
LT
3768 return ret;
3769}
3770EXPORT_SYMBOL(get_io_context);
3771
3772void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3773{
3774 struct io_context *src = *psrc;
3775 struct io_context *dst = *pdst;
3776
3777 if (src) {
3778 BUG_ON(atomic_read(&src->refcount) == 0);
3779 atomic_inc(&src->refcount);
3780 put_io_context(dst);
3781 *pdst = src;
3782 }
3783}
3784EXPORT_SYMBOL(copy_io_context);
3785
3786void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3787{
3788 struct io_context *temp;
3789 temp = *ioc1;
3790 *ioc1 = *ioc2;
3791 *ioc2 = temp;
3792}
3793EXPORT_SYMBOL(swap_io_context);
3794
3795/*
3796 * sysfs parts below
3797 */
3798struct queue_sysfs_entry {
3799 struct attribute attr;
3800 ssize_t (*show)(struct request_queue *, char *);
3801 ssize_t (*store)(struct request_queue *, const char *, size_t);
3802};
3803
3804static ssize_t
3805queue_var_show(unsigned int var, char *page)
3806{
3807 return sprintf(page, "%d\n", var);
3808}
3809
3810static ssize_t
3811queue_var_store(unsigned long *var, const char *page, size_t count)
3812{
3813 char *p = (char *) page;
3814
3815 *var = simple_strtoul(p, &p, 10);
3816 return count;
3817}
3818
3819static ssize_t queue_requests_show(struct request_queue *q, char *page)
3820{
3821 return queue_var_show(q->nr_requests, (page));
3822}
3823
3824static ssize_t
3825queue_requests_store(struct request_queue *q, const char *page, size_t count)
3826{
3827 struct request_list *rl = &q->rq;
c981ff9f
AV
3828 unsigned long nr;
3829 int ret = queue_var_store(&nr, page, count);
3830 if (nr < BLKDEV_MIN_RQ)
3831 nr = BLKDEV_MIN_RQ;
1da177e4 3832
c981ff9f
AV
3833 spin_lock_irq(q->queue_lock);
3834 q->nr_requests = nr;
1da177e4
LT
3835 blk_queue_congestion_threshold(q);
3836
3837 if (rl->count[READ] >= queue_congestion_on_threshold(q))
79e2de4b 3838 blk_set_queue_congested(q, READ);
1da177e4 3839 else if (rl->count[READ] < queue_congestion_off_threshold(q))
79e2de4b 3840 blk_clear_queue_congested(q, READ);
1da177e4
LT
3841
3842 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
79e2de4b 3843 blk_set_queue_congested(q, WRITE);
1da177e4 3844 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
79e2de4b 3845 blk_clear_queue_congested(q, WRITE);
1da177e4
LT
3846
3847 if (rl->count[READ] >= q->nr_requests) {
3848 blk_set_queue_full(q, READ);
3849 } else if (rl->count[READ]+1 <= q->nr_requests) {
3850 blk_clear_queue_full(q, READ);
3851 wake_up(&rl->wait[READ]);
3852 }
3853
3854 if (rl->count[WRITE] >= q->nr_requests) {
3855 blk_set_queue_full(q, WRITE);
3856 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3857 blk_clear_queue_full(q, WRITE);
3858 wake_up(&rl->wait[WRITE]);
3859 }
c981ff9f 3860 spin_unlock_irq(q->queue_lock);
1da177e4
LT
3861 return ret;
3862}
3863
3864static ssize_t queue_ra_show(struct request_queue *q, char *page)
3865{
3866 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3867
3868 return queue_var_show(ra_kb, (page));
3869}
3870
3871static ssize_t
3872queue_ra_store(struct request_queue *q, const char *page, size_t count)
3873{
3874 unsigned long ra_kb;
3875 ssize_t ret = queue_var_store(&ra_kb, page, count);
3876
3877 spin_lock_irq(q->queue_lock);
1da177e4
LT
3878 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3879 spin_unlock_irq(q->queue_lock);
3880
3881 return ret;
3882}
3883
3884static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3885{
3886 int max_sectors_kb = q->max_sectors >> 1;
3887
3888 return queue_var_show(max_sectors_kb, (page));
3889}
3890
3891static ssize_t
3892queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3893{
3894 unsigned long max_sectors_kb,
3895 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3896 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3897 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3898 int ra_kb;
3899
3900 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3901 return -EINVAL;
3902 /*
3903 * Take the queue lock to update the readahead and max_sectors
3904 * values synchronously:
3905 */
3906 spin_lock_irq(q->queue_lock);
3907 /*
3908 * Trim readahead window as well, if necessary:
3909 */
3910 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3911 if (ra_kb > max_sectors_kb)
3912 q->backing_dev_info.ra_pages =
3913 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3914
3915 q->max_sectors = max_sectors_kb << 1;
3916 spin_unlock_irq(q->queue_lock);
3917
3918 return ret;
3919}
3920
3921static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3922{
3923 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3924
3925 return queue_var_show(max_hw_sectors_kb, (page));
3926}
3927
3928
3929static struct queue_sysfs_entry queue_requests_entry = {
3930 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3931 .show = queue_requests_show,
3932 .store = queue_requests_store,
3933};
3934
3935static struct queue_sysfs_entry queue_ra_entry = {
3936 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3937 .show = queue_ra_show,
3938 .store = queue_ra_store,
3939};
3940
3941static struct queue_sysfs_entry queue_max_sectors_entry = {
3942 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3943 .show = queue_max_sectors_show,
3944 .store = queue_max_sectors_store,
3945};
3946
3947static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3948 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3949 .show = queue_max_hw_sectors_show,
3950};
3951
3952static struct queue_sysfs_entry queue_iosched_entry = {
3953 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3954 .show = elv_iosched_show,
3955 .store = elv_iosched_store,
3956};
3957
3958static struct attribute *default_attrs[] = {
3959 &queue_requests_entry.attr,
3960 &queue_ra_entry.attr,
3961 &queue_max_hw_sectors_entry.attr,
3962 &queue_max_sectors_entry.attr,
3963 &queue_iosched_entry.attr,
3964 NULL,
3965};
3966
3967#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3968
3969static ssize_t
3970queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3971{
3972 struct queue_sysfs_entry *entry = to_queue(attr);
483f4afc
AV
3973 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
3974 ssize_t res;
1da177e4 3975
1da177e4 3976 if (!entry->show)
6c1852a0 3977 return -EIO;
483f4afc
AV
3978 mutex_lock(&q->sysfs_lock);
3979 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3980 mutex_unlock(&q->sysfs_lock);
3981 return -ENOENT;
3982 }
3983 res = entry->show(q, page);
3984 mutex_unlock(&q->sysfs_lock);
3985 return res;
1da177e4
LT
3986}
3987
3988static ssize_t
3989queue_attr_store(struct kobject *kobj, struct attribute *attr,
3990 const char *page, size_t length)
3991{
3992 struct queue_sysfs_entry *entry = to_queue(attr);
483f4afc
AV
3993 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
3994
3995 ssize_t res;
1da177e4 3996
1da177e4 3997 if (!entry->store)
6c1852a0 3998 return -EIO;
483f4afc
AV
3999 mutex_lock(&q->sysfs_lock);
4000 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4001 mutex_unlock(&q->sysfs_lock);
4002 return -ENOENT;
4003 }
4004 res = entry->store(q, page, length);
4005 mutex_unlock(&q->sysfs_lock);
4006 return res;
1da177e4
LT
4007}
4008
4009static struct sysfs_ops queue_sysfs_ops = {
4010 .show = queue_attr_show,
4011 .store = queue_attr_store,
4012};
4013
93d17d3d 4014static struct kobj_type queue_ktype = {
1da177e4
LT
4015 .sysfs_ops = &queue_sysfs_ops,
4016 .default_attrs = default_attrs,
483f4afc 4017 .release = blk_release_queue,
1da177e4
LT
4018};
4019
4020int blk_register_queue(struct gendisk *disk)
4021{
4022 int ret;
4023
4024 request_queue_t *q = disk->queue;
4025
4026 if (!q || !q->request_fn)
4027 return -ENXIO;
4028
4029 q->kobj.parent = kobject_get(&disk->kobj);
1da177e4 4030
483f4afc 4031 ret = kobject_add(&q->kobj);
1da177e4
LT
4032 if (ret < 0)
4033 return ret;
4034
483f4afc
AV
4035 kobject_uevent(&q->kobj, KOBJ_ADD);
4036
1da177e4
LT
4037 ret = elv_register_queue(q);
4038 if (ret) {
483f4afc
AV
4039 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4040 kobject_del(&q->kobj);
1da177e4
LT
4041 return ret;
4042 }
4043
4044 return 0;
4045}
4046
4047void blk_unregister_queue(struct gendisk *disk)
4048{
4049 request_queue_t *q = disk->queue;
4050
4051 if (q && q->request_fn) {
4052 elv_unregister_queue(q);
4053
483f4afc
AV
4054 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4055 kobject_del(&q->kobj);
1da177e4
LT
4056 kobject_put(&disk->kobj);
4057 }
4058}