blob: 78562574cb524d87b06883bce5002542b77b5865 [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)) {
Jens Axboeeeae1d42008-05-07 13:26:27 +0200161 unsigned long uninitialized_var(idx);
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;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200447 int nr_sgvecs;
448 struct sg_iovec *sgvecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449};
450
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200451static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
452 struct sg_iovec *iov, int iov_count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
454 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200455 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
456 bmd->nr_sgvecs = iov_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 bio->bi_private = bmd;
458}
459
460static void bio_free_map_data(struct bio_map_data *bmd)
461{
462 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200463 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 kfree(bmd);
465}
466
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200467static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
469 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
470
471 if (!bmd)
472 return NULL;
473
474 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200475 if (!bmd->iovecs) {
476 kfree(bmd);
477 return NULL;
478 }
479
480 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, GFP_KERNEL);
481 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 return bmd;
483
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200484 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 kfree(bmd);
486 return NULL;
487}
488
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200489static int __bio_copy_iov(struct bio *bio, struct sg_iovec *iov, int iov_count,
490 int uncopy)
491{
492 int ret = 0, i;
493 struct bio_vec *bvec;
494 int iov_idx = 0;
495 unsigned int iov_off = 0;
496 int read = bio_data_dir(bio) == READ;
497
498 __bio_for_each_segment(bvec, bio, i, 0) {
499 char *bv_addr = page_address(bvec->bv_page);
500 unsigned int bv_len = bvec->bv_len;
501
502 while (bv_len && iov_idx < iov_count) {
503 unsigned int bytes;
504 char *iov_addr;
505
506 bytes = min_t(unsigned int,
507 iov[iov_idx].iov_len - iov_off, bv_len);
508 iov_addr = iov[iov_idx].iov_base + iov_off;
509
510 if (!ret) {
511 if (!read && !uncopy)
512 ret = copy_from_user(bv_addr, iov_addr,
513 bytes);
514 if (read && uncopy)
515 ret = copy_to_user(iov_addr, bv_addr,
516 bytes);
517
518 if (ret)
519 ret = -EFAULT;
520 }
521
522 bv_len -= bytes;
523 bv_addr += bytes;
524 iov_addr += bytes;
525 iov_off += bytes;
526
527 if (iov[iov_idx].iov_len == iov_off) {
528 iov_idx++;
529 iov_off = 0;
530 }
531 }
532
533 if (uncopy)
534 __free_page(bvec->bv_page);
535 }
536
537 return ret;
538}
539
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540/**
541 * bio_uncopy_user - finish previously mapped bio
542 * @bio: bio being terminated
543 *
544 * Free pages allocated from bio_copy_user() and write back data
545 * to user space in case of a read.
546 */
547int bio_uncopy_user(struct bio *bio)
548{
549 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200550 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200552 ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 bio_free_map_data(bmd);
555 bio_put(bio);
556 return ret;
557}
558
559/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200560 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 * @q: destination block queue
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200562 * @iov: the iovec.
563 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 * @write_to_vm: bool indicating writing to pages or not
565 *
566 * Prepares and returns a bio for indirect user io, bouncing data
567 * to/from kernel pages as necessary. Must be paired with
568 * call bio_uncopy_user() on io completion.
569 */
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200570struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov,
571 int iov_count, int write_to_vm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 struct bio_map_data *bmd;
574 struct bio_vec *bvec;
575 struct page *page;
576 struct bio *bio;
577 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200578 int nr_pages = 0;
579 unsigned int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200581 for (i = 0; i < iov_count; i++) {
582 unsigned long uaddr;
583 unsigned long end;
584 unsigned long start;
585
586 uaddr = (unsigned long)iov[i].iov_base;
587 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
588 start = uaddr >> PAGE_SHIFT;
589
590 nr_pages += end - start;
591 len += iov[i].iov_len;
592 }
593
594 bmd = bio_alloc_map_data(nr_pages, iov_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 if (!bmd)
596 return ERR_PTR(-ENOMEM);
597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 ret = -ENOMEM;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200599 bio = bio_alloc(GFP_KERNEL, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 if (!bio)
601 goto out_bmd;
602
603 bio->bi_rw |= (!write_to_vm << BIO_RW);
604
605 ret = 0;
606 while (len) {
607 unsigned int bytes = PAGE_SIZE;
608
609 if (bytes > len)
610 bytes = len;
611
612 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
613 if (!page) {
614 ret = -ENOMEM;
615 break;
616 }
617
Mike Christie0e75f902006-12-01 10:40:55 +0100618 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 len -= bytes;
622 }
623
624 if (ret)
625 goto cleanup;
626
627 /*
628 * success
629 */
630 if (!write_to_vm) {
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200631 ret = __bio_copy_iov(bio, iov, iov_count, 0);
632 if (ret)
633 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
635
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200636 bio_set_map_data(bmd, bio, iov, iov_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 return bio;
638cleanup:
639 bio_for_each_segment(bvec, bio, i)
640 __free_page(bvec->bv_page);
641
642 bio_put(bio);
643out_bmd:
644 bio_free_map_data(bmd);
645 return ERR_PTR(ret);
646}
647
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200648/**
649 * bio_copy_user - copy user data to bio
650 * @q: destination block queue
651 * @uaddr: start of user address
652 * @len: length in bytes
653 * @write_to_vm: bool indicating writing to pages or not
654 *
655 * Prepares and returns a bio for indirect user io, bouncing data
656 * to/from kernel pages as necessary. Must be paired with
657 * call bio_uncopy_user() on io completion.
658 */
659struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr,
660 unsigned int len, int write_to_vm)
661{
662 struct sg_iovec iov;
663
664 iov.iov_base = (void __user *)uaddr;
665 iov.iov_len = len;
666
667 return bio_copy_user_iov(q, &iov, 1, write_to_vm);
668}
669
Jens Axboe165125e2007-07-24 09:28:11 +0200670static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200671 struct block_device *bdev,
672 struct sg_iovec *iov, int iov_count,
673 int write_to_vm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
James Bottomley f1970ba2005-06-20 14:06:52 +0200675 int i, j;
676 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 struct page **pages;
678 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200679 int cur_page = 0;
680 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
James Bottomley f1970ba2005-06-20 14:06:52 +0200682 for (i = 0; i < iov_count; i++) {
683 unsigned long uaddr = (unsigned long)iov[i].iov_base;
684 unsigned long len = iov[i].iov_len;
685 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
686 unsigned long start = uaddr >> PAGE_SHIFT;
687
688 nr_pages += end - start;
689 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100690 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200691 */
Mike Christiead2d7222006-12-01 10:40:20 +0100692 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200693 return ERR_PTR(-EINVAL);
694 }
695
696 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 return ERR_PTR(-EINVAL);
698
699 bio = bio_alloc(GFP_KERNEL, nr_pages);
700 if (!bio)
701 return ERR_PTR(-ENOMEM);
702
703 ret = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800704 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 if (!pages)
706 goto out;
707
James Bottomley f1970ba2005-06-20 14:06:52 +0200708 for (i = 0; i < iov_count; i++) {
709 unsigned long uaddr = (unsigned long)iov[i].iov_base;
710 unsigned long len = iov[i].iov_len;
711 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
712 unsigned long start = uaddr >> PAGE_SHIFT;
713 const int local_nr_pages = end - start;
714 const int page_limit = cur_page + local_nr_pages;
715
716 down_read(&current->mm->mmap_sem);
717 ret = get_user_pages(current, current->mm, uaddr,
718 local_nr_pages,
719 write_to_vm, 0, &pages[cur_page], NULL);
720 up_read(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Jens Axboe99172152006-06-16 13:02:29 +0200722 if (ret < local_nr_pages) {
723 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200724 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200725 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
James Bottomley f1970ba2005-06-20 14:06:52 +0200727 offset = uaddr & ~PAGE_MASK;
728 for (j = cur_page; j < page_limit; j++) {
729 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
James Bottomley f1970ba2005-06-20 14:06:52 +0200731 if (len <= 0)
732 break;
733
734 if (bytes > len)
735 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
James Bottomley f1970ba2005-06-20 14:06:52 +0200737 /*
738 * sorry...
739 */
Mike Christiedefd94b2005-12-05 02:37:06 -0600740 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
741 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +0200742 break;
743
744 len -= bytes;
745 offset = 0;
746 }
747
748 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 /*
James Bottomley f1970ba2005-06-20 14:06:52 +0200750 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 */
James Bottomley f1970ba2005-06-20 14:06:52 +0200752 while (j < page_limit)
753 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 }
755
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 kfree(pages);
757
758 /*
759 * set data direction, and check if mapped pages need bouncing
760 */
761 if (!write_to_vm)
762 bio->bi_rw |= (1 << BIO_RW);
763
James Bottomley f1970ba2005-06-20 14:06:52 +0200764 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 bio->bi_flags |= (1 << BIO_USER_MAPPED);
766 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200767
768 out_unmap:
769 for (i = 0; i < nr_pages; i++) {
770 if(!pages[i])
771 break;
772 page_cache_release(pages[i]);
773 }
774 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 kfree(pages);
776 bio_put(bio);
777 return ERR_PTR(ret);
778}
779
780/**
781 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200782 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 * @bdev: destination block device
784 * @uaddr: start of user address
785 * @len: length in bytes
786 * @write_to_vm: bool indicating writing to pages or not
787 *
788 * Map the user space address into a bio suitable for io to a block
789 * device. Returns an error pointer in case of error.
790 */
Jens Axboe165125e2007-07-24 09:28:11 +0200791struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 unsigned long uaddr, unsigned int len, int write_to_vm)
793{
James Bottomley f1970ba2005-06-20 14:06:52 +0200794 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +0100796 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +0200797 iov.iov_len = len;
798
799 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
800}
801
802/**
803 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200804 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +0200805 * @bdev: destination block device
806 * @iov: the iovec.
807 * @iov_count: number of elements in the iovec
808 * @write_to_vm: bool indicating writing to pages or not
809 *
810 * Map the user space address into a bio suitable for io to a block
811 * device. Returns an error pointer in case of error.
812 */
Jens Axboe165125e2007-07-24 09:28:11 +0200813struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +0200814 struct sg_iovec *iov, int iov_count,
815 int write_to_vm)
816{
817 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200818
819 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821 if (IS_ERR(bio))
822 return bio;
823
824 /*
825 * subtle -- if __bio_map_user() ended up bouncing a bio,
826 * it would normally disappear when its bi_end_io is run.
827 * however, we need it for the unmap, so grab an extra
828 * reference to it
829 */
830 bio_get(bio);
831
Mike Christie0e75f902006-12-01 10:40:55 +0100832 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
834
835static void __bio_unmap_user(struct bio *bio)
836{
837 struct bio_vec *bvec;
838 int i;
839
840 /*
841 * make sure we dirty pages we wrote to
842 */
843 __bio_for_each_segment(bvec, bio, i, 0) {
844 if (bio_data_dir(bio) == READ)
845 set_page_dirty_lock(bvec->bv_page);
846
847 page_cache_release(bvec->bv_page);
848 }
849
850 bio_put(bio);
851}
852
853/**
854 * bio_unmap_user - unmap a bio
855 * @bio: the bio being unmapped
856 *
857 * Unmap a bio previously mapped by bio_map_user(). Must be called with
858 * a process context.
859 *
860 * bio_unmap_user() may sleep.
861 */
862void bio_unmap_user(struct bio *bio)
863{
864 __bio_unmap_user(bio);
865 bio_put(bio);
866}
867
NeilBrown6712ecf2007-09-27 12:47:43 +0200868static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +0200869{
Jens Axboeb8238252005-06-20 14:05:27 +0200870 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +0200871}
872
873
Jens Axboe165125e2007-07-24 09:28:11 +0200874static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -0400875 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +0200876{
877 unsigned long kaddr = (unsigned long)data;
878 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
879 unsigned long start = kaddr >> PAGE_SHIFT;
880 const int nr_pages = end - start;
881 int offset, i;
882 struct bio *bio;
883
884 bio = bio_alloc(gfp_mask, nr_pages);
885 if (!bio)
886 return ERR_PTR(-ENOMEM);
887
888 offset = offset_in_page(kaddr);
889 for (i = 0; i < nr_pages; i++) {
890 unsigned int bytes = PAGE_SIZE - offset;
891
892 if (len <= 0)
893 break;
894
895 if (bytes > len)
896 bytes = len;
897
Mike Christiedefd94b2005-12-05 02:37:06 -0600898 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
899 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +0200900 break;
901
902 data += bytes;
903 len -= bytes;
904 offset = 0;
905 }
906
Jens Axboeb8238252005-06-20 14:05:27 +0200907 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +0200908 return bio;
909}
910
911/**
912 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200913 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +0200914 * @data: pointer to buffer to map
915 * @len: length in bytes
916 * @gfp_mask: allocation flags for bio allocation
917 *
918 * Map the kernel address into a bio suitable for io to a block
919 * device. Returns an error pointer in case of error.
920 */
Jens Axboe165125e2007-07-24 09:28:11 +0200921struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -0400922 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +0200923{
924 struct bio *bio;
925
926 bio = __bio_map_kern(q, data, len, gfp_mask);
927 if (IS_ERR(bio))
928 return bio;
929
930 if (bio->bi_size == len)
931 return bio;
932
933 /*
934 * Don't support partial mappings.
935 */
936 bio_put(bio);
937 return ERR_PTR(-EINVAL);
938}
939
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200940static void bio_copy_kern_endio(struct bio *bio, int err)
941{
942 struct bio_vec *bvec;
943 const int read = bio_data_dir(bio) == READ;
944 char *p = bio->bi_private;
945 int i;
946
947 __bio_for_each_segment(bvec, bio, i, 0) {
948 char *addr = page_address(bvec->bv_page);
949
950 if (read && !err)
951 memcpy(p, addr, bvec->bv_len);
952
953 __free_page(bvec->bv_page);
954 p += bvec->bv_len;
955 }
956
957 bio_put(bio);
958}
959
960/**
961 * bio_copy_kern - copy kernel address into bio
962 * @q: the struct request_queue for the bio
963 * @data: pointer to buffer to copy
964 * @len: length in bytes
965 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +0200966 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200967 *
968 * copy the kernel address into a bio suitable for io to a block
969 * device. Returns an error pointer in case of error.
970 */
971struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
972 gfp_t gfp_mask, int reading)
973{
974 unsigned long kaddr = (unsigned long)data;
975 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
976 unsigned long start = kaddr >> PAGE_SHIFT;
977 const int nr_pages = end - start;
978 struct bio *bio;
979 struct bio_vec *bvec;
980 int i, ret;
981
982 bio = bio_alloc(gfp_mask, nr_pages);
983 if (!bio)
984 return ERR_PTR(-ENOMEM);
985
986 while (len) {
987 struct page *page;
988 unsigned int bytes = PAGE_SIZE;
989
990 if (bytes > len)
991 bytes = len;
992
993 page = alloc_page(q->bounce_gfp | gfp_mask);
994 if (!page) {
995 ret = -ENOMEM;
996 goto cleanup;
997 }
998
999 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) {
1000 ret = -EINVAL;
1001 goto cleanup;
1002 }
1003
1004 len -= bytes;
1005 }
1006
1007 if (!reading) {
1008 void *p = data;
1009
1010 bio_for_each_segment(bvec, bio, i) {
1011 char *addr = page_address(bvec->bv_page);
1012
1013 memcpy(addr, p, bvec->bv_len);
1014 p += bvec->bv_len;
1015 }
1016 }
1017
1018 bio->bi_private = data;
1019 bio->bi_end_io = bio_copy_kern_endio;
1020 return bio;
1021cleanup:
1022 bio_for_each_segment(bvec, bio, i)
1023 __free_page(bvec->bv_page);
1024
1025 bio_put(bio);
1026
1027 return ERR_PTR(ret);
1028}
1029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030/*
1031 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1032 * for performing direct-IO in BIOs.
1033 *
1034 * The problem is that we cannot run set_page_dirty() from interrupt context
1035 * because the required locks are not interrupt-safe. So what we can do is to
1036 * mark the pages dirty _before_ performing IO. And in interrupt context,
1037 * check that the pages are still dirty. If so, fine. If not, redirty them
1038 * in process context.
1039 *
1040 * We special-case compound pages here: normally this means reads into hugetlb
1041 * pages. The logic in here doesn't really work right for compound pages
1042 * because the VM does not uniformly chase down the head page in all cases.
1043 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1044 * handle them at all. So we skip compound pages here at an early stage.
1045 *
1046 * Note that this code is very hard to test under normal circumstances because
1047 * direct-io pins the pages with get_user_pages(). This makes
1048 * is_page_cache_freeable return false, and the VM will not clean the pages.
1049 * But other code (eg, pdflush) could clean the pages if they are mapped
1050 * pagecache.
1051 *
1052 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1053 * deferred bio dirtying paths.
1054 */
1055
1056/*
1057 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1058 */
1059void bio_set_pages_dirty(struct bio *bio)
1060{
1061 struct bio_vec *bvec = bio->bi_io_vec;
1062 int i;
1063
1064 for (i = 0; i < bio->bi_vcnt; i++) {
1065 struct page *page = bvec[i].bv_page;
1066
1067 if (page && !PageCompound(page))
1068 set_page_dirty_lock(page);
1069 }
1070}
1071
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001072static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
1074 struct bio_vec *bvec = bio->bi_io_vec;
1075 int i;
1076
1077 for (i = 0; i < bio->bi_vcnt; i++) {
1078 struct page *page = bvec[i].bv_page;
1079
1080 if (page)
1081 put_page(page);
1082 }
1083}
1084
1085/*
1086 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1087 * If they are, then fine. If, however, some pages are clean then they must
1088 * have been written out during the direct-IO read. So we take another ref on
1089 * the BIO and the offending pages and re-dirty the pages in process context.
1090 *
1091 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1092 * here on. It will run one page_cache_release() against each page and will
1093 * run one bio_put() against the BIO.
1094 */
1095
David Howells65f27f32006-11-22 14:55:48 +00001096static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
David Howells65f27f32006-11-22 14:55:48 +00001098static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099static DEFINE_SPINLOCK(bio_dirty_lock);
1100static struct bio *bio_dirty_list;
1101
1102/*
1103 * This runs in process context
1104 */
David Howells65f27f32006-11-22 14:55:48 +00001105static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106{
1107 unsigned long flags;
1108 struct bio *bio;
1109
1110 spin_lock_irqsave(&bio_dirty_lock, flags);
1111 bio = bio_dirty_list;
1112 bio_dirty_list = NULL;
1113 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1114
1115 while (bio) {
1116 struct bio *next = bio->bi_private;
1117
1118 bio_set_pages_dirty(bio);
1119 bio_release_pages(bio);
1120 bio_put(bio);
1121 bio = next;
1122 }
1123}
1124
1125void bio_check_pages_dirty(struct bio *bio)
1126{
1127 struct bio_vec *bvec = bio->bi_io_vec;
1128 int nr_clean_pages = 0;
1129 int i;
1130
1131 for (i = 0; i < bio->bi_vcnt; i++) {
1132 struct page *page = bvec[i].bv_page;
1133
1134 if (PageDirty(page) || PageCompound(page)) {
1135 page_cache_release(page);
1136 bvec[i].bv_page = NULL;
1137 } else {
1138 nr_clean_pages++;
1139 }
1140 }
1141
1142 if (nr_clean_pages) {
1143 unsigned long flags;
1144
1145 spin_lock_irqsave(&bio_dirty_lock, flags);
1146 bio->bi_private = bio_dirty_list;
1147 bio_dirty_list = bio;
1148 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1149 schedule_work(&bio_dirty_work);
1150 } else {
1151 bio_put(bio);
1152 }
1153}
1154
1155/**
1156 * bio_endio - end I/O on a bio
1157 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 * @error: error, if any
1159 *
1160 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001161 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001162 * preferred way to end I/O on a bio, it takes care of clearing
1163 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1164 * established -Exxxx (-EIO, for instance) error values in case
1165 * something went wrong. Noone should call bi_end_io() directly on a
1166 * bio unless they own it and thus know that it has an end_io
1167 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001169void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170{
1171 if (error)
1172 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001173 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1174 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
NeilBrown5bb23a62007-09-27 12:46:13 +02001176 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001177 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178}
1179
1180void bio_pair_release(struct bio_pair *bp)
1181{
1182 if (atomic_dec_and_test(&bp->cnt)) {
1183 struct bio *master = bp->bio1.bi_private;
1184
NeilBrown6712ecf2007-09-27 12:47:43 +02001185 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 mempool_free(bp, bp->bio2.bi_private);
1187 }
1188}
1189
NeilBrown6712ecf2007-09-27 12:47:43 +02001190static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191{
1192 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1193
1194 if (err)
1195 bp->error = err;
1196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198}
1199
NeilBrown6712ecf2007-09-27 12:47:43 +02001200static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201{
1202 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1203
1204 if (err)
1205 bp->error = err;
1206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208}
1209
1210/*
1211 * split a bio - only worry about a bio with a single page
1212 * in it's iovec
1213 */
1214struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1215{
1216 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1217
1218 if (!bp)
1219 return bp;
1220
Jens Axboe2056a782006-03-23 20:00:26 +01001221 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
1222 bi->bi_sector + first_sectors);
1223
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 BUG_ON(bi->bi_vcnt != 1);
1225 BUG_ON(bi->bi_idx != 0);
1226 atomic_set(&bp->cnt, 3);
1227 bp->error = 0;
1228 bp->bio1 = *bi;
1229 bp->bio2 = *bi;
1230 bp->bio2.bi_sector += first_sectors;
1231 bp->bio2.bi_size -= first_sectors << 9;
1232 bp->bio1.bi_size = first_sectors << 9;
1233
1234 bp->bv1 = bi->bi_io_vec[0];
1235 bp->bv2 = bi->bi_io_vec[0];
1236 bp->bv2.bv_offset += first_sectors << 9;
1237 bp->bv2.bv_len -= first_sectors << 9;
1238 bp->bv1.bv_len = first_sectors << 9;
1239
1240 bp->bio1.bi_io_vec = &bp->bv1;
1241 bp->bio2.bi_io_vec = &bp->bv2;
1242
NeilBrowna2eb0c12006-05-22 22:35:27 -07001243 bp->bio1.bi_max_vecs = 1;
1244 bp->bio2.bi_max_vecs = 1;
1245
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 bp->bio1.bi_end_io = bio_pair_end_1;
1247 bp->bio2.bi_end_io = bio_pair_end_2;
1248
1249 bp->bio1.bi_private = bi;
1250 bp->bio2.bi_private = pool;
1251
1252 return bp;
1253}
1254
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
1256/*
1257 * create memory pools for biovec's in a bio_set.
1258 * use the global biovec slabs created for general use.
1259 */
Jens Axboe59725112007-04-02 10:06:42 +02001260static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
1262 int i;
1263
1264 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1265 struct biovec_slab *bp = bvec_slabs + i;
1266 mempool_t **bvp = bs->bvec_pools + i;
1267
Matthew Dobson93d23412006-03-26 01:37:50 -08001268 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 if (!*bvp)
1270 return -ENOMEM;
1271 }
1272 return 0;
1273}
1274
1275static void biovec_free_pools(struct bio_set *bs)
1276{
1277 int i;
1278
1279 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1280 mempool_t *bvp = bs->bvec_pools[i];
1281
1282 if (bvp)
1283 mempool_destroy(bvp);
1284 }
1285
1286}
1287
1288void bioset_free(struct bio_set *bs)
1289{
1290 if (bs->bio_pool)
1291 mempool_destroy(bs->bio_pool);
1292
1293 biovec_free_pools(bs);
1294
1295 kfree(bs);
1296}
1297
Jens Axboe59725112007-04-02 10:06:42 +02001298struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001300 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
1302 if (!bs)
1303 return NULL;
1304
Matthew Dobson93d23412006-03-26 01:37:50 -08001305 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 if (!bs->bio_pool)
1307 goto bad;
1308
Jens Axboe59725112007-04-02 10:06:42 +02001309 if (!biovec_create_pools(bs, bvec_pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 return bs;
1311
1312bad:
1313 bioset_free(bs);
1314 return NULL;
1315}
1316
1317static void __init biovec_init_slabs(void)
1318{
1319 int i;
1320
1321 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1322 int size;
1323 struct biovec_slab *bvs = bvec_slabs + i;
1324
1325 size = bvs->nr_vecs * sizeof(struct bio_vec);
1326 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001327 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 }
1329}
1330
1331static int __init init_bio(void)
1332{
Christoph Lameter0a31bd52007-05-06 14:49:57 -07001333 bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
1335 biovec_init_slabs();
1336
Jens Axboe59725112007-04-02 10:06:42 +02001337 fs_bio_set = bioset_create(BIO_POOL_SIZE, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 if (!fs_bio_set)
1339 panic("bio: can't allocate bios\n");
1340
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001341 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1342 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 if (!bio_split_pool)
1344 panic("bio: can't create split pool\n");
1345
1346 return 0;
1347}
1348
1349subsys_initcall(init_bio);
1350
1351EXPORT_SYMBOL(bio_alloc);
1352EXPORT_SYMBOL(bio_put);
Peter Osterlund36763472005-09-06 15:16:42 -07001353EXPORT_SYMBOL(bio_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354EXPORT_SYMBOL(bio_endio);
1355EXPORT_SYMBOL(bio_init);
1356EXPORT_SYMBOL(__bio_clone);
1357EXPORT_SYMBOL(bio_clone);
1358EXPORT_SYMBOL(bio_phys_segments);
1359EXPORT_SYMBOL(bio_hw_segments);
1360EXPORT_SYMBOL(bio_add_page);
Mike Christie6e68af62005-11-11 05:30:27 -06001361EXPORT_SYMBOL(bio_add_pc_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362EXPORT_SYMBOL(bio_get_nr_vecs);
Jens Axboe40044ce2008-03-17 21:14:40 +01001363EXPORT_SYMBOL(bio_map_user);
1364EXPORT_SYMBOL(bio_unmap_user);
Mike Christie df46b9a2005-06-20 14:04:44 +02001365EXPORT_SYMBOL(bio_map_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001366EXPORT_SYMBOL(bio_copy_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367EXPORT_SYMBOL(bio_pair_release);
1368EXPORT_SYMBOL(bio_split);
1369EXPORT_SYMBOL(bio_split_pool);
1370EXPORT_SYMBOL(bio_copy_user);
1371EXPORT_SYMBOL(bio_uncopy_user);
1372EXPORT_SYMBOL(bioset_create);
1373EXPORT_SYMBOL(bioset_free);
1374EXPORT_SYMBOL(bio_alloc_bioset);