blob: 3312fcc3c0986401511c1fc7c3948a6af6c69876 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Jens Axboe0fe23472006-09-04 15:41:16 +02002 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
16 *
17 */
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/bio.h>
21#include <linux/blkdev.h>
22#include <linux/slab.h>
23#include <linux/init.h>
24#include <linux/kernel.h>
25#include <linux/module.h>
26#include <linux/mempool.h>
27#include <linux/workqueue.h>
Jens Axboe2056a782006-03-23 20:00:26 +010028#include <linux/blktrace_api.h>
James Bottomley f1970ba2005-06-20 14:06:52 +020029#include <scsi/sg.h> /* for struct sg_iovec */
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
Jens Axboe59725112007-04-02 10:06:42 +020031#define BIO_POOL_SIZE 2
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Christoph Lametere18b8902006-12-06 20:33:20 -080033static struct kmem_cache *bio_slab __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35#define BIOVEC_NR_POOLS 6
36
37/*
38 * a small number of entries is fine, not going to be performance critical.
39 * basically we just need to survive
40 */
Jens Axboe59725112007-04-02 10:06:42 +020041#define BIO_SPLIT_ENTRIES 2
Eric Dumazetfa3536c2006-03-26 01:37:24 -080042mempool_t *bio_split_pool __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44struct biovec_slab {
45 int nr_vecs;
46 char *name;
Christoph Lametere18b8902006-12-06 20:33:20 -080047 struct kmem_cache *slab;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048};
49
50/*
51 * if you change this list, also change bvec_alloc or things will
52 * break badly! cannot be bigger than what you can fit into an
53 * unsigned short
54 */
55
56#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
Christoph Lameter6c036522005-07-07 17:56:59 -070057static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
59};
60#undef BV
61
62/*
63 * bio_set is used to allow other portions of the IO system to
64 * allocate their own private memory pools for bio and iovec structures.
65 * These memory pools in turn all allocate from the bio_slab
66 * and the bvec_slabs[].
67 */
68struct bio_set {
69 mempool_t *bio_pool;
70 mempool_t *bvec_pools[BIOVEC_NR_POOLS];
71};
72
73/*
74 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
75 * IO code that does not need private memory pools.
76 */
77static struct bio_set *fs_bio_set;
78
Al Virodd0fc662005-10-07 07:46:04 +010079static inline struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080{
81 struct bio_vec *bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83 /*
84 * see comment near bvec_array define!
85 */
86 switch (nr) {
87 case 1 : *idx = 0; break;
88 case 2 ... 4: *idx = 1; break;
89 case 5 ... 16: *idx = 2; break;
90 case 17 ... 64: *idx = 3; break;
91 case 65 ... 128: *idx = 4; break;
92 case 129 ... BIO_MAX_PAGES: *idx = 5; break;
93 default:
94 return NULL;
95 }
96 /*
97 * idx now points to the pool we want to allocate from
98 */
99
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
Andreas Mohrbf02c082006-10-11 01:22:24 -0700101 if (bvl) {
102 struct biovec_slab *bp = bvec_slabs + *idx;
103
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
Andreas Mohrbf02c082006-10-11 01:22:24 -0700105 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
107 return bvl;
108}
109
Peter Osterlund36763472005-09-06 15:16:42 -0700110void bio_free(struct bio *bio, struct bio_set *bio_set)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111{
Jens Axboe992c5dd2007-07-18 13:18:08 +0200112 if (bio->bi_io_vec) {
113 const int pool_idx = BIO_POOL_IDX(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Jens Axboe992c5dd2007-07-18 13:18:08 +0200115 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Jens Axboe992c5dd2007-07-18 13:18:08 +0200117 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]);
118 }
119
Peter Osterlund36763472005-09-06 15:16:42 -0700120 mempool_free(bio, bio_set->bio_pool);
121}
122
123/*
124 * default destructor for a bio allocated with bio_alloc_bioset()
125 */
126static void bio_fs_destructor(struct bio *bio)
127{
128 bio_free(bio, fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129}
130
Arjan van de Ven858119e2006-01-14 13:20:43 -0800131void bio_init(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
Jens Axboe2b94de52007-07-18 13:14:03 +0200133 memset(bio, 0, sizeof(*bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 bio->bi_flags = 1 << BIO_UPTODATE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 atomic_set(&bio->bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136}
137
138/**
139 * bio_alloc_bioset - allocate a bio for I/O
140 * @gfp_mask: the GFP_ mask given to the slab allocator
141 * @nr_iovecs: number of iovecs to pre-allocate
Martin Waitz67be2dd2005-05-01 08:59:26 -0700142 * @bs: the bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 *
144 * Description:
145 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
146 * If %__GFP_WAIT is set then we will block on the internal pool waiting
147 * for a &struct bio to become free.
148 *
149 * allocate bio and iovecs from the memory pools specified by the
150 * bio_set structure.
151 **/
Al Virodd0fc662005-10-07 07:46:04 +0100152struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153{
154 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask);
155
156 if (likely(bio)) {
157 struct bio_vec *bvl = NULL;
158
159 bio_init(bio);
160 if (likely(nr_iovecs)) {
Andreas Mohrbf02c082006-10-11 01:22:24 -0700161 unsigned long idx = 0; /* shut up gcc */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
163 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
164 if (unlikely(!bvl)) {
165 mempool_free(bio, bs->bio_pool);
166 bio = NULL;
167 goto out;
168 }
169 bio->bi_flags |= idx << BIO_POOL_OFFSET;
170 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs;
171 }
172 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 }
174out:
175 return bio;
176}
177
Al Virodd0fc662005-10-07 07:46:04 +0100178struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179{
Peter Osterlund36763472005-09-06 15:16:42 -0700180 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
181
182 if (bio)
183 bio->bi_destructor = bio_fs_destructor;
184
185 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186}
187
188void zero_fill_bio(struct bio *bio)
189{
190 unsigned long flags;
191 struct bio_vec *bv;
192 int i;
193
194 bio_for_each_segment(bv, bio, i) {
195 char *data = bvec_kmap_irq(bv, &flags);
196 memset(data, 0, bv->bv_len);
197 flush_dcache_page(bv->bv_page);
198 bvec_kunmap_irq(data, &flags);
199 }
200}
201EXPORT_SYMBOL(zero_fill_bio);
202
203/**
204 * bio_put - release a reference to a bio
205 * @bio: bio to release reference to
206 *
207 * Description:
208 * Put a reference to a &struct bio, either one you have gotten with
209 * bio_alloc or bio_get. The last put of a bio will free it.
210 **/
211void bio_put(struct bio *bio)
212{
213 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
214
215 /*
216 * last put frees it
217 */
218 if (atomic_dec_and_test(&bio->bi_cnt)) {
219 bio->bi_next = NULL;
220 bio->bi_destructor(bio);
221 }
222}
223
Jens Axboe165125e2007-07-24 09:28:11 +0200224inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
226 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
227 blk_recount_segments(q, bio);
228
229 return bio->bi_phys_segments;
230}
231
Jens Axboe165125e2007-07-24 09:28:11 +0200232inline int bio_hw_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
234 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
235 blk_recount_segments(q, bio);
236
237 return bio->bi_hw_segments;
238}
239
240/**
241 * __bio_clone - clone a bio
242 * @bio: destination bio
243 * @bio_src: bio to clone
244 *
245 * Clone a &bio. Caller will own the returned bio, but not
246 * the actual data it points to. Reference count of returned
247 * bio will be one.
248 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800249void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Andrew Mortone525e152005-08-07 09:42:12 -0700251 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
252 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Jens Axboe5d840702008-01-25 12:44:44 +0100254 /*
255 * most users will be overriding ->bi_bdev with a new target,
256 * so we don't set nor calculate new physical/hw segment counts here
257 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 bio->bi_sector = bio_src->bi_sector;
259 bio->bi_bdev = bio_src->bi_bdev;
260 bio->bi_flags |= 1 << BIO_CLONED;
261 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 bio->bi_vcnt = bio_src->bi_vcnt;
263 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700264 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
266
267/**
268 * bio_clone - clone a bio
269 * @bio: bio to clone
270 * @gfp_mask: allocation priority
271 *
272 * Like __bio_clone, only also allocates the returned bio
273 */
Al Virodd0fc662005-10-07 07:46:04 +0100274struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
276 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
277
Peter Osterlund36763472005-09-06 15:16:42 -0700278 if (b) {
279 b->bi_destructor = bio_fs_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 __bio_clone(b, bio);
Peter Osterlund36763472005-09-06 15:16:42 -0700281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
283 return b;
284}
285
286/**
287 * bio_get_nr_vecs - return approx number of vecs
288 * @bdev: I/O target
289 *
290 * Return the approximate number of pages we can send to this target.
291 * There's no guarantee that you will be able to fit this number of pages
292 * into a bio, it does not account for dynamic restrictions that vary
293 * on offset.
294 */
295int bio_get_nr_vecs(struct block_device *bdev)
296{
Jens Axboe165125e2007-07-24 09:28:11 +0200297 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 int nr_pages;
299
300 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
301 if (nr_pages > q->max_phys_segments)
302 nr_pages = q->max_phys_segments;
303 if (nr_pages > q->max_hw_segments)
304 nr_pages = q->max_hw_segments;
305
306 return nr_pages;
307}
308
Jens Axboe165125e2007-07-24 09:28:11 +0200309static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600310 *page, unsigned int len, unsigned int offset,
311 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
313 int retried_segments = 0;
314 struct bio_vec *bvec;
315
316 /*
317 * cloned bio must not modify vec list
318 */
319 if (unlikely(bio_flagged(bio, BIO_CLONED)))
320 return 0;
321
Jens Axboe80cfd542006-01-06 09:43:28 +0100322 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 return 0;
324
Jens Axboe80cfd542006-01-06 09:43:28 +0100325 /*
326 * For filesystems with a blocksize smaller than the pagesize
327 * we will often be called with the same page as last time and
328 * a consecutive offset. Optimize this special case.
329 */
330 if (bio->bi_vcnt > 0) {
331 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
332
333 if (page == prev->bv_page &&
334 offset == prev->bv_offset + prev->bv_len) {
335 prev->bv_len += len;
336 if (q->merge_bvec_fn &&
337 q->merge_bvec_fn(q, bio, prev) < len) {
338 prev->bv_len -= len;
339 return 0;
340 }
341
342 goto done;
343 }
344 }
345
346 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 return 0;
348
349 /*
350 * we might lose a segment or two here, but rather that than
351 * make this too complex.
352 */
353
354 while (bio->bi_phys_segments >= q->max_phys_segments
355 || bio->bi_hw_segments >= q->max_hw_segments
356 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
357
358 if (retried_segments)
359 return 0;
360
361 retried_segments = 1;
362 blk_recount_segments(q, bio);
363 }
364
365 /*
366 * setup the new entry, we might clear it again later if we
367 * cannot add the page
368 */
369 bvec = &bio->bi_io_vec[bio->bi_vcnt];
370 bvec->bv_page = page;
371 bvec->bv_len = len;
372 bvec->bv_offset = offset;
373
374 /*
375 * if queue has other restrictions (eg varying max sector size
376 * depending on offset), it can specify a merge_bvec_fn in the
377 * queue to get further control
378 */
379 if (q->merge_bvec_fn) {
380 /*
381 * merge_bvec_fn() returns number of bytes it can accept
382 * at this offset
383 */
384 if (q->merge_bvec_fn(q, bio, bvec) < len) {
385 bvec->bv_page = NULL;
386 bvec->bv_len = 0;
387 bvec->bv_offset = 0;
388 return 0;
389 }
390 }
391
392 /* If we may be able to merge these biovecs, force a recount */
393 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
394 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
395 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
396
397 bio->bi_vcnt++;
398 bio->bi_phys_segments++;
399 bio->bi_hw_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100400 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 bio->bi_size += len;
402 return len;
403}
404
405/**
Mike Christie6e68af62005-11-11 05:30:27 -0600406 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100407 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600408 * @bio: destination bio
409 * @page: page to add
410 * @len: vec entry length
411 * @offset: vec entry offset
412 *
413 * Attempt to add a page to the bio_vec maplist. This can fail for a
414 * number of reasons, such as the bio being full or target block
415 * device limitations. The target block device must allow bio's
416 * smaller than PAGE_SIZE, so it is always possible to add a single
417 * page to an empty bio. This should only be used by REQ_PC bios.
418 */
Jens Axboe165125e2007-07-24 09:28:11 +0200419int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600420 unsigned int len, unsigned int offset)
421{
Mike Christiedefd94b2005-12-05 02:37:06 -0600422 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
Mike Christie6e68af62005-11-11 05:30:27 -0600423}
424
425/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 * bio_add_page - attempt to add page to bio
427 * @bio: destination bio
428 * @page: page to add
429 * @len: vec entry length
430 * @offset: vec entry offset
431 *
432 * Attempt to add a page to the bio_vec maplist. This can fail for a
433 * number of reasons, such as the bio being full or target block
434 * device limitations. The target block device must allow bio's
435 * smaller than PAGE_SIZE, so it is always possible to add a single
436 * page to an empty bio.
437 */
438int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
439 unsigned int offset)
440{
Mike Christiedefd94b2005-12-05 02:37:06 -0600441 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
442 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
445struct bio_map_data {
446 struct bio_vec *iovecs;
447 void __user *userptr;
448};
449
450static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio)
451{
452 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
453 bio->bi_private = bmd;
454}
455
456static void bio_free_map_data(struct bio_map_data *bmd)
457{
458 kfree(bmd->iovecs);
459 kfree(bmd);
460}
461
462static struct bio_map_data *bio_alloc_map_data(int nr_segs)
463{
464 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
465
466 if (!bmd)
467 return NULL;
468
469 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
470 if (bmd->iovecs)
471 return bmd;
472
473 kfree(bmd);
474 return NULL;
475}
476
477/**
478 * bio_uncopy_user - finish previously mapped bio
479 * @bio: bio being terminated
480 *
481 * Free pages allocated from bio_copy_user() and write back data
482 * to user space in case of a read.
483 */
484int bio_uncopy_user(struct bio *bio)
485{
486 struct bio_map_data *bmd = bio->bi_private;
487 const int read = bio_data_dir(bio) == READ;
488 struct bio_vec *bvec;
489 int i, ret = 0;
490
491 __bio_for_each_segment(bvec, bio, i, 0) {
492 char *addr = page_address(bvec->bv_page);
493 unsigned int len = bmd->iovecs[i].bv_len;
494
495 if (read && !ret && copy_to_user(bmd->userptr, addr, len))
496 ret = -EFAULT;
497
498 __free_page(bvec->bv_page);
499 bmd->userptr += len;
500 }
501 bio_free_map_data(bmd);
502 bio_put(bio);
503 return ret;
504}
505
506/**
507 * bio_copy_user - copy user data to bio
508 * @q: destination block queue
509 * @uaddr: start of user address
510 * @len: length in bytes
511 * @write_to_vm: bool indicating writing to pages or not
512 *
513 * Prepares and returns a bio for indirect user io, bouncing data
514 * to/from kernel pages as necessary. Must be paired with
515 * call bio_uncopy_user() on io completion.
516 */
Jens Axboe165125e2007-07-24 09:28:11 +0200517struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 unsigned int len, int write_to_vm)
519{
520 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
521 unsigned long start = uaddr >> PAGE_SHIFT;
522 struct bio_map_data *bmd;
523 struct bio_vec *bvec;
524 struct page *page;
525 struct bio *bio;
526 int i, ret;
527
528 bmd = bio_alloc_map_data(end - start);
529 if (!bmd)
530 return ERR_PTR(-ENOMEM);
531
532 bmd->userptr = (void __user *) uaddr;
533
534 ret = -ENOMEM;
535 bio = bio_alloc(GFP_KERNEL, end - start);
536 if (!bio)
537 goto out_bmd;
538
539 bio->bi_rw |= (!write_to_vm << BIO_RW);
540
541 ret = 0;
542 while (len) {
543 unsigned int bytes = PAGE_SIZE;
544
545 if (bytes > len)
546 bytes = len;
547
548 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
549 if (!page) {
550 ret = -ENOMEM;
551 break;
552 }
553
Mike Christie0e75f902006-12-01 10:40:55 +0100554 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
557 len -= bytes;
558 }
559
560 if (ret)
561 goto cleanup;
562
563 /*
564 * success
565 */
566 if (!write_to_vm) {
567 char __user *p = (char __user *) uaddr;
568
569 /*
570 * for a write, copy in data to kernel pages
571 */
572 ret = -EFAULT;
573 bio_for_each_segment(bvec, bio, i) {
574 char *addr = page_address(bvec->bv_page);
575
576 if (copy_from_user(addr, p, bvec->bv_len))
577 goto cleanup;
578 p += bvec->bv_len;
579 }
580 }
581
582 bio_set_map_data(bmd, bio);
583 return bio;
584cleanup:
585 bio_for_each_segment(bvec, bio, i)
586 __free_page(bvec->bv_page);
587
588 bio_put(bio);
589out_bmd:
590 bio_free_map_data(bmd);
591 return ERR_PTR(ret);
592}
593
Jens Axboe165125e2007-07-24 09:28:11 +0200594static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200595 struct block_device *bdev,
596 struct sg_iovec *iov, int iov_count,
597 int write_to_vm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598{
James Bottomley f1970ba2005-06-20 14:06:52 +0200599 int i, j;
600 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 struct page **pages;
602 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200603 int cur_page = 0;
604 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
James Bottomley f1970ba2005-06-20 14:06:52 +0200606 for (i = 0; i < iov_count; i++) {
607 unsigned long uaddr = (unsigned long)iov[i].iov_base;
608 unsigned long len = iov[i].iov_len;
609 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
610 unsigned long start = uaddr >> PAGE_SHIFT;
611
612 nr_pages += end - start;
613 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100614 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200615 */
Mike Christiead2d7222006-12-01 10:40:20 +0100616 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200617 return ERR_PTR(-EINVAL);
618 }
619
620 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 return ERR_PTR(-EINVAL);
622
623 bio = bio_alloc(GFP_KERNEL, nr_pages);
624 if (!bio)
625 return ERR_PTR(-ENOMEM);
626
627 ret = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800628 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 if (!pages)
630 goto out;
631
James Bottomley f1970ba2005-06-20 14:06:52 +0200632 for (i = 0; i < iov_count; i++) {
633 unsigned long uaddr = (unsigned long)iov[i].iov_base;
634 unsigned long len = iov[i].iov_len;
635 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
636 unsigned long start = uaddr >> PAGE_SHIFT;
637 const int local_nr_pages = end - start;
638 const int page_limit = cur_page + local_nr_pages;
639
640 down_read(&current->mm->mmap_sem);
641 ret = get_user_pages(current, current->mm, uaddr,
642 local_nr_pages,
643 write_to_vm, 0, &pages[cur_page], NULL);
644 up_read(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Jens Axboe99172152006-06-16 13:02:29 +0200646 if (ret < local_nr_pages) {
647 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200648 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200649 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
James Bottomley f1970ba2005-06-20 14:06:52 +0200651 offset = uaddr & ~PAGE_MASK;
652 for (j = cur_page; j < page_limit; j++) {
653 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
James Bottomley f1970ba2005-06-20 14:06:52 +0200655 if (len <= 0)
656 break;
657
658 if (bytes > len)
659 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
James Bottomley f1970ba2005-06-20 14:06:52 +0200661 /*
662 * sorry...
663 */
Mike Christiedefd94b2005-12-05 02:37:06 -0600664 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
665 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +0200666 break;
667
668 len -= bytes;
669 offset = 0;
670 }
671
672 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 /*
James Bottomley f1970ba2005-06-20 14:06:52 +0200674 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 */
James Bottomley f1970ba2005-06-20 14:06:52 +0200676 while (j < page_limit)
677 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
679
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 kfree(pages);
681
682 /*
683 * set data direction, and check if mapped pages need bouncing
684 */
685 if (!write_to_vm)
686 bio->bi_rw |= (1 << BIO_RW);
687
James Bottomley f1970ba2005-06-20 14:06:52 +0200688 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 bio->bi_flags |= (1 << BIO_USER_MAPPED);
690 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200691
692 out_unmap:
693 for (i = 0; i < nr_pages; i++) {
694 if(!pages[i])
695 break;
696 page_cache_release(pages[i]);
697 }
698 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 kfree(pages);
700 bio_put(bio);
701 return ERR_PTR(ret);
702}
703
704/**
705 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200706 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 * @bdev: destination block device
708 * @uaddr: start of user address
709 * @len: length in bytes
710 * @write_to_vm: bool indicating writing to pages or not
711 *
712 * Map the user space address into a bio suitable for io to a block
713 * device. Returns an error pointer in case of error.
714 */
Jens Axboe165125e2007-07-24 09:28:11 +0200715struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 unsigned long uaddr, unsigned int len, int write_to_vm)
717{
James Bottomley f1970ba2005-06-20 14:06:52 +0200718 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +0100720 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +0200721 iov.iov_len = len;
722
723 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
724}
725
726/**
727 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200728 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +0200729 * @bdev: destination block device
730 * @iov: the iovec.
731 * @iov_count: number of elements in the iovec
732 * @write_to_vm: bool indicating writing to pages or not
733 *
734 * Map the user space address into a bio suitable for io to a block
735 * device. Returns an error pointer in case of error.
736 */
Jens Axboe165125e2007-07-24 09:28:11 +0200737struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +0200738 struct sg_iovec *iov, int iov_count,
739 int write_to_vm)
740{
741 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200742
743 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
745 if (IS_ERR(bio))
746 return bio;
747
748 /*
749 * subtle -- if __bio_map_user() ended up bouncing a bio,
750 * it would normally disappear when its bi_end_io is run.
751 * however, we need it for the unmap, so grab an extra
752 * reference to it
753 */
754 bio_get(bio);
755
Mike Christie0e75f902006-12-01 10:40:55 +0100756 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757}
758
759static void __bio_unmap_user(struct bio *bio)
760{
761 struct bio_vec *bvec;
762 int i;
763
764 /*
765 * make sure we dirty pages we wrote to
766 */
767 __bio_for_each_segment(bvec, bio, i, 0) {
768 if (bio_data_dir(bio) == READ)
769 set_page_dirty_lock(bvec->bv_page);
770
771 page_cache_release(bvec->bv_page);
772 }
773
774 bio_put(bio);
775}
776
777/**
778 * bio_unmap_user - unmap a bio
779 * @bio: the bio being unmapped
780 *
781 * Unmap a bio previously mapped by bio_map_user(). Must be called with
782 * a process context.
783 *
784 * bio_unmap_user() may sleep.
785 */
786void bio_unmap_user(struct bio *bio)
787{
788 __bio_unmap_user(bio);
789 bio_put(bio);
790}
791
NeilBrown6712ecf2007-09-27 12:47:43 +0200792static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +0200793{
Jens Axboeb8238252005-06-20 14:05:27 +0200794 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +0200795}
796
797
Jens Axboe165125e2007-07-24 09:28:11 +0200798static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -0400799 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +0200800{
801 unsigned long kaddr = (unsigned long)data;
802 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
803 unsigned long start = kaddr >> PAGE_SHIFT;
804 const int nr_pages = end - start;
805 int offset, i;
806 struct bio *bio;
807
808 bio = bio_alloc(gfp_mask, nr_pages);
809 if (!bio)
810 return ERR_PTR(-ENOMEM);
811
812 offset = offset_in_page(kaddr);
813 for (i = 0; i < nr_pages; i++) {
814 unsigned int bytes = PAGE_SIZE - offset;
815
816 if (len <= 0)
817 break;
818
819 if (bytes > len)
820 bytes = len;
821
Mike Christiedefd94b2005-12-05 02:37:06 -0600822 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
823 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +0200824 break;
825
826 data += bytes;
827 len -= bytes;
828 offset = 0;
829 }
830
Jens Axboeb8238252005-06-20 14:05:27 +0200831 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +0200832 return bio;
833}
834
835/**
836 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200837 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +0200838 * @data: pointer to buffer to map
839 * @len: length in bytes
840 * @gfp_mask: allocation flags for bio allocation
841 *
842 * Map the kernel address into a bio suitable for io to a block
843 * device. Returns an error pointer in case of error.
844 */
Jens Axboe165125e2007-07-24 09:28:11 +0200845struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -0400846 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +0200847{
848 struct bio *bio;
849
850 bio = __bio_map_kern(q, data, len, gfp_mask);
851 if (IS_ERR(bio))
852 return bio;
853
854 if (bio->bi_size == len)
855 return bio;
856
857 /*
858 * Don't support partial mappings.
859 */
860 bio_put(bio);
861 return ERR_PTR(-EINVAL);
862}
863
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864/*
865 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
866 * for performing direct-IO in BIOs.
867 *
868 * The problem is that we cannot run set_page_dirty() from interrupt context
869 * because the required locks are not interrupt-safe. So what we can do is to
870 * mark the pages dirty _before_ performing IO. And in interrupt context,
871 * check that the pages are still dirty. If so, fine. If not, redirty them
872 * in process context.
873 *
874 * We special-case compound pages here: normally this means reads into hugetlb
875 * pages. The logic in here doesn't really work right for compound pages
876 * because the VM does not uniformly chase down the head page in all cases.
877 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
878 * handle them at all. So we skip compound pages here at an early stage.
879 *
880 * Note that this code is very hard to test under normal circumstances because
881 * direct-io pins the pages with get_user_pages(). This makes
882 * is_page_cache_freeable return false, and the VM will not clean the pages.
883 * But other code (eg, pdflush) could clean the pages if they are mapped
884 * pagecache.
885 *
886 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
887 * deferred bio dirtying paths.
888 */
889
890/*
891 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
892 */
893void bio_set_pages_dirty(struct bio *bio)
894{
895 struct bio_vec *bvec = bio->bi_io_vec;
896 int i;
897
898 for (i = 0; i < bio->bi_vcnt; i++) {
899 struct page *page = bvec[i].bv_page;
900
901 if (page && !PageCompound(page))
902 set_page_dirty_lock(page);
903 }
904}
905
Adrian Bunk86b6c7a2008-02-18 13:48:32 +0100906static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
908 struct bio_vec *bvec = bio->bi_io_vec;
909 int i;
910
911 for (i = 0; i < bio->bi_vcnt; i++) {
912 struct page *page = bvec[i].bv_page;
913
914 if (page)
915 put_page(page);
916 }
917}
918
919/*
920 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
921 * If they are, then fine. If, however, some pages are clean then they must
922 * have been written out during the direct-IO read. So we take another ref on
923 * the BIO and the offending pages and re-dirty the pages in process context.
924 *
925 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
926 * here on. It will run one page_cache_release() against each page and will
927 * run one bio_put() against the BIO.
928 */
929
David Howells65f27f32006-11-22 14:55:48 +0000930static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
David Howells65f27f32006-11-22 14:55:48 +0000932static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933static DEFINE_SPINLOCK(bio_dirty_lock);
934static struct bio *bio_dirty_list;
935
936/*
937 * This runs in process context
938 */
David Howells65f27f32006-11-22 14:55:48 +0000939static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
941 unsigned long flags;
942 struct bio *bio;
943
944 spin_lock_irqsave(&bio_dirty_lock, flags);
945 bio = bio_dirty_list;
946 bio_dirty_list = NULL;
947 spin_unlock_irqrestore(&bio_dirty_lock, flags);
948
949 while (bio) {
950 struct bio *next = bio->bi_private;
951
952 bio_set_pages_dirty(bio);
953 bio_release_pages(bio);
954 bio_put(bio);
955 bio = next;
956 }
957}
958
959void bio_check_pages_dirty(struct bio *bio)
960{
961 struct bio_vec *bvec = bio->bi_io_vec;
962 int nr_clean_pages = 0;
963 int i;
964
965 for (i = 0; i < bio->bi_vcnt; i++) {
966 struct page *page = bvec[i].bv_page;
967
968 if (PageDirty(page) || PageCompound(page)) {
969 page_cache_release(page);
970 bvec[i].bv_page = NULL;
971 } else {
972 nr_clean_pages++;
973 }
974 }
975
976 if (nr_clean_pages) {
977 unsigned long flags;
978
979 spin_lock_irqsave(&bio_dirty_lock, flags);
980 bio->bi_private = bio_dirty_list;
981 bio_dirty_list = bio;
982 spin_unlock_irqrestore(&bio_dirty_lock, flags);
983 schedule_work(&bio_dirty_work);
984 } else {
985 bio_put(bio);
986 }
987}
988
989/**
990 * bio_endio - end I/O on a bio
991 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 * @error: error, if any
993 *
994 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +0200995 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +0200996 * preferred way to end I/O on a bio, it takes care of clearing
997 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
998 * established -Exxxx (-EIO, for instance) error values in case
999 * something went wrong. Noone should call bi_end_io() directly on a
1000 * bio unless they own it and thus know that it has an end_io
1001 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001003void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
1005 if (error)
1006 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001007 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1008 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
NeilBrown5bb23a62007-09-27 12:46:13 +02001010 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001011 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012}
1013
1014void bio_pair_release(struct bio_pair *bp)
1015{
1016 if (atomic_dec_and_test(&bp->cnt)) {
1017 struct bio *master = bp->bio1.bi_private;
1018
NeilBrown6712ecf2007-09-27 12:47:43 +02001019 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 mempool_free(bp, bp->bio2.bi_private);
1021 }
1022}
1023
NeilBrown6712ecf2007-09-27 12:47:43 +02001024static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
1026 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1027
1028 if (err)
1029 bp->error = err;
1030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032}
1033
NeilBrown6712ecf2007-09-27 12:47:43 +02001034static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035{
1036 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1037
1038 if (err)
1039 bp->error = err;
1040
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042}
1043
1044/*
1045 * split a bio - only worry about a bio with a single page
1046 * in it's iovec
1047 */
1048struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1049{
1050 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1051
1052 if (!bp)
1053 return bp;
1054
Jens Axboe2056a782006-03-23 20:00:26 +01001055 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
1056 bi->bi_sector + first_sectors);
1057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 BUG_ON(bi->bi_vcnt != 1);
1059 BUG_ON(bi->bi_idx != 0);
1060 atomic_set(&bp->cnt, 3);
1061 bp->error = 0;
1062 bp->bio1 = *bi;
1063 bp->bio2 = *bi;
1064 bp->bio2.bi_sector += first_sectors;
1065 bp->bio2.bi_size -= first_sectors << 9;
1066 bp->bio1.bi_size = first_sectors << 9;
1067
1068 bp->bv1 = bi->bi_io_vec[0];
1069 bp->bv2 = bi->bi_io_vec[0];
1070 bp->bv2.bv_offset += first_sectors << 9;
1071 bp->bv2.bv_len -= first_sectors << 9;
1072 bp->bv1.bv_len = first_sectors << 9;
1073
1074 bp->bio1.bi_io_vec = &bp->bv1;
1075 bp->bio2.bi_io_vec = &bp->bv2;
1076
NeilBrowna2eb0c12006-05-22 22:35:27 -07001077 bp->bio1.bi_max_vecs = 1;
1078 bp->bio2.bi_max_vecs = 1;
1079
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 bp->bio1.bi_end_io = bio_pair_end_1;
1081 bp->bio2.bi_end_io = bio_pair_end_2;
1082
1083 bp->bio1.bi_private = bi;
1084 bp->bio2.bi_private = pool;
1085
1086 return bp;
1087}
1088
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
1090/*
1091 * create memory pools for biovec's in a bio_set.
1092 * use the global biovec slabs created for general use.
1093 */
Jens Axboe59725112007-04-02 10:06:42 +02001094static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095{
1096 int i;
1097
1098 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1099 struct biovec_slab *bp = bvec_slabs + i;
1100 mempool_t **bvp = bs->bvec_pools + i;
1101
Matthew Dobson93d23412006-03-26 01:37:50 -08001102 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 if (!*bvp)
1104 return -ENOMEM;
1105 }
1106 return 0;
1107}
1108
1109static void biovec_free_pools(struct bio_set *bs)
1110{
1111 int i;
1112
1113 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1114 mempool_t *bvp = bs->bvec_pools[i];
1115
1116 if (bvp)
1117 mempool_destroy(bvp);
1118 }
1119
1120}
1121
1122void bioset_free(struct bio_set *bs)
1123{
1124 if (bs->bio_pool)
1125 mempool_destroy(bs->bio_pool);
1126
1127 biovec_free_pools(bs);
1128
1129 kfree(bs);
1130}
1131
Jens Axboe59725112007-04-02 10:06:42 +02001132struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001134 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 if (!bs)
1137 return NULL;
1138
Matthew Dobson93d23412006-03-26 01:37:50 -08001139 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 if (!bs->bio_pool)
1141 goto bad;
1142
Jens Axboe59725112007-04-02 10:06:42 +02001143 if (!biovec_create_pools(bs, bvec_pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 return bs;
1145
1146bad:
1147 bioset_free(bs);
1148 return NULL;
1149}
1150
1151static void __init biovec_init_slabs(void)
1152{
1153 int i;
1154
1155 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1156 int size;
1157 struct biovec_slab *bvs = bvec_slabs + i;
1158
1159 size = bvs->nr_vecs * sizeof(struct bio_vec);
1160 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001161 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 }
1163}
1164
1165static int __init init_bio(void)
1166{
Christoph Lameter0a31bd52007-05-06 14:49:57 -07001167 bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
1169 biovec_init_slabs();
1170
Jens Axboe59725112007-04-02 10:06:42 +02001171 fs_bio_set = bioset_create(BIO_POOL_SIZE, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 if (!fs_bio_set)
1173 panic("bio: can't allocate bios\n");
1174
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001175 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1176 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 if (!bio_split_pool)
1178 panic("bio: can't create split pool\n");
1179
1180 return 0;
1181}
1182
1183subsys_initcall(init_bio);
1184
1185EXPORT_SYMBOL(bio_alloc);
1186EXPORT_SYMBOL(bio_put);
Peter Osterlund36763472005-09-06 15:16:42 -07001187EXPORT_SYMBOL(bio_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188EXPORT_SYMBOL(bio_endio);
1189EXPORT_SYMBOL(bio_init);
1190EXPORT_SYMBOL(__bio_clone);
1191EXPORT_SYMBOL(bio_clone);
1192EXPORT_SYMBOL(bio_phys_segments);
1193EXPORT_SYMBOL(bio_hw_segments);
1194EXPORT_SYMBOL(bio_add_page);
Mike Christie6e68af62005-11-11 05:30:27 -06001195EXPORT_SYMBOL(bio_add_pc_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196EXPORT_SYMBOL(bio_get_nr_vecs);
Mike Christie df46b9a2005-06-20 14:04:44 +02001197EXPORT_SYMBOL(bio_map_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198EXPORT_SYMBOL(bio_pair_release);
1199EXPORT_SYMBOL(bio_split);
1200EXPORT_SYMBOL(bio_split_pool);
1201EXPORT_SYMBOL(bio_copy_user);
1202EXPORT_SYMBOL(bio_uncopy_user);
1203EXPORT_SYMBOL(bioset_create);
1204EXPORT_SYMBOL(bioset_free);
1205EXPORT_SYMBOL(bio_alloc_bioset);