blob: 88322b066acb68f8279e79eeca5842c6b02bd8ce [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
Christoph Lametere18b8902006-12-06 20:33:20 -080031static struct kmem_cache *bio_slab __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Eric Dumazetfa3536c2006-03-26 01:37:24 -080033mempool_t *bio_split_pool __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Linus Torvalds1da177e2005-04-16 15:20:36 -070035/*
36 * if you change this list, also change bvec_alloc or things will
37 * break badly! cannot be bigger than what you can fit into an
38 * unsigned short
39 */
40
41#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
Christoph Lameter6c036522005-07-07 17:56:59 -070042static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
44};
45#undef BV
46
47/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
49 * IO code that does not need private memory pools.
50 */
Martin K. Petersen51d654e2008-06-17 18:59:56 +020051struct bio_set *fs_bio_set;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +020053unsigned int bvec_nr_vecs(unsigned short idx)
54{
55 return bvec_slabs[idx].nr_vecs;
56}
57
Martin K. Petersen51d654e2008-06-17 18:59:56 +020058struct 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 -070059{
60 struct bio_vec *bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
62 /*
63 * see comment near bvec_array define!
64 */
65 switch (nr) {
66 case 1 : *idx = 0; break;
67 case 2 ... 4: *idx = 1; break;
68 case 5 ... 16: *idx = 2; break;
69 case 17 ... 64: *idx = 3; break;
70 case 65 ... 128: *idx = 4; break;
71 case 129 ... BIO_MAX_PAGES: *idx = 5; break;
72 default:
73 return NULL;
74 }
75 /*
76 * idx now points to the pool we want to allocate from
77 */
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
Andreas Mohrbf02c082006-10-11 01:22:24 -070080 if (bvl) {
81 struct biovec_slab *bp = bvec_slabs + *idx;
82
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
Andreas Mohrbf02c082006-10-11 01:22:24 -070084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86 return bvl;
87}
88
Peter Osterlund36763472005-09-06 15:16:42 -070089void bio_free(struct bio *bio, struct bio_set *bio_set)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090{
Jens Axboe992c5dd2007-07-18 13:18:08 +020091 if (bio->bi_io_vec) {
92 const int pool_idx = BIO_POOL_IDX(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Jens Axboe992c5dd2007-07-18 13:18:08 +020094 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
Jens Axboe992c5dd2007-07-18 13:18:08 +020096 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]);
97 }
98
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +020099 if (bio_integrity(bio))
100 bio_integrity_free(bio, bio_set);
101
Peter Osterlund36763472005-09-06 15:16:42 -0700102 mempool_free(bio, bio_set->bio_pool);
103}
104
105/*
106 * default destructor for a bio allocated with bio_alloc_bioset()
107 */
108static void bio_fs_destructor(struct bio *bio)
109{
110 bio_free(bio, fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111}
112
Arjan van de Ven858119e2006-01-14 13:20:43 -0800113void bio_init(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
Jens Axboe2b94de52007-07-18 13:14:03 +0200115 memset(bio, 0, sizeof(*bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 bio->bi_flags = 1 << BIO_UPTODATE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 atomic_set(&bio->bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118}
119
120/**
121 * bio_alloc_bioset - allocate a bio for I/O
122 * @gfp_mask: the GFP_ mask given to the slab allocator
123 * @nr_iovecs: number of iovecs to pre-allocate
Martin Waitz67be2dd2005-05-01 08:59:26 -0700124 * @bs: the bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 *
126 * Description:
127 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
128 * If %__GFP_WAIT is set then we will block on the internal pool waiting
129 * for a &struct bio to become free.
130 *
131 * allocate bio and iovecs from the memory pools specified by the
132 * bio_set structure.
133 **/
Al Virodd0fc662005-10-07 07:46:04 +0100134struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135{
136 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask);
137
138 if (likely(bio)) {
139 struct bio_vec *bvl = NULL;
140
141 bio_init(bio);
142 if (likely(nr_iovecs)) {
Jens Axboeeeae1d42008-05-07 13:26:27 +0200143 unsigned long uninitialized_var(idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
145 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
146 if (unlikely(!bvl)) {
147 mempool_free(bio, bs->bio_pool);
148 bio = NULL;
149 goto out;
150 }
151 bio->bi_flags |= idx << BIO_POOL_OFFSET;
152 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs;
153 }
154 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 }
156out:
157 return bio;
158}
159
Al Virodd0fc662005-10-07 07:46:04 +0100160struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161{
Peter Osterlund36763472005-09-06 15:16:42 -0700162 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
163
164 if (bio)
165 bio->bi_destructor = bio_fs_destructor;
166
167 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168}
169
170void zero_fill_bio(struct bio *bio)
171{
172 unsigned long flags;
173 struct bio_vec *bv;
174 int i;
175
176 bio_for_each_segment(bv, bio, i) {
177 char *data = bvec_kmap_irq(bv, &flags);
178 memset(data, 0, bv->bv_len);
179 flush_dcache_page(bv->bv_page);
180 bvec_kunmap_irq(data, &flags);
181 }
182}
183EXPORT_SYMBOL(zero_fill_bio);
184
185/**
186 * bio_put - release a reference to a bio
187 * @bio: bio to release reference to
188 *
189 * Description:
190 * Put a reference to a &struct bio, either one you have gotten with
191 * bio_alloc or bio_get. The last put of a bio will free it.
192 **/
193void bio_put(struct bio *bio)
194{
195 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
196
197 /*
198 * last put frees it
199 */
200 if (atomic_dec_and_test(&bio->bi_cnt)) {
201 bio->bi_next = NULL;
202 bio->bi_destructor(bio);
203 }
204}
205
Jens Axboe165125e2007-07-24 09:28:11 +0200206inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
208 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
209 blk_recount_segments(q, bio);
210
211 return bio->bi_phys_segments;
212}
213
Jens Axboe165125e2007-07-24 09:28:11 +0200214inline int bio_hw_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215{
216 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
217 blk_recount_segments(q, bio);
218
219 return bio->bi_hw_segments;
220}
221
222/**
223 * __bio_clone - clone a bio
224 * @bio: destination bio
225 * @bio_src: bio to clone
226 *
227 * Clone a &bio. Caller will own the returned bio, but not
228 * the actual data it points to. Reference count of returned
229 * bio will be one.
230 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800231void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Andrew Mortone525e152005-08-07 09:42:12 -0700233 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
234 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
Jens Axboe5d840702008-01-25 12:44:44 +0100236 /*
237 * most users will be overriding ->bi_bdev with a new target,
238 * so we don't set nor calculate new physical/hw segment counts here
239 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 bio->bi_sector = bio_src->bi_sector;
241 bio->bi_bdev = bio_src->bi_bdev;
242 bio->bi_flags |= 1 << BIO_CLONED;
243 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 bio->bi_vcnt = bio_src->bi_vcnt;
245 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700246 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
249/**
250 * bio_clone - clone a bio
251 * @bio: bio to clone
252 * @gfp_mask: allocation priority
253 *
254 * Like __bio_clone, only also allocates the returned bio
255 */
Al Virodd0fc662005-10-07 07:46:04 +0100256struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
258 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
259
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200260 if (!b)
261 return NULL;
262
263 b->bi_destructor = bio_fs_destructor;
264 __bio_clone(b, bio);
265
266 if (bio_integrity(bio)) {
267 int ret;
268
269 ret = bio_integrity_clone(b, bio, fs_bio_set);
270
271 if (ret < 0)
272 return NULL;
Peter Osterlund36763472005-09-06 15:16:42 -0700273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 return b;
276}
277
278/**
279 * bio_get_nr_vecs - return approx number of vecs
280 * @bdev: I/O target
281 *
282 * Return the approximate number of pages we can send to this target.
283 * There's no guarantee that you will be able to fit this number of pages
284 * into a bio, it does not account for dynamic restrictions that vary
285 * on offset.
286 */
287int bio_get_nr_vecs(struct block_device *bdev)
288{
Jens Axboe165125e2007-07-24 09:28:11 +0200289 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 int nr_pages;
291
292 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
293 if (nr_pages > q->max_phys_segments)
294 nr_pages = q->max_phys_segments;
295 if (nr_pages > q->max_hw_segments)
296 nr_pages = q->max_hw_segments;
297
298 return nr_pages;
299}
300
Jens Axboe165125e2007-07-24 09:28:11 +0200301static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600302 *page, unsigned int len, unsigned int offset,
303 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
305 int retried_segments = 0;
306 struct bio_vec *bvec;
307
308 /*
309 * cloned bio must not modify vec list
310 */
311 if (unlikely(bio_flagged(bio, BIO_CLONED)))
312 return 0;
313
Jens Axboe80cfd542006-01-06 09:43:28 +0100314 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 return 0;
316
Jens Axboe80cfd542006-01-06 09:43:28 +0100317 /*
318 * For filesystems with a blocksize smaller than the pagesize
319 * we will often be called with the same page as last time and
320 * a consecutive offset. Optimize this special case.
321 */
322 if (bio->bi_vcnt > 0) {
323 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
324
325 if (page == prev->bv_page &&
326 offset == prev->bv_offset + prev->bv_len) {
327 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200328
329 if (q->merge_bvec_fn) {
330 struct bvec_merge_data bvm = {
331 .bi_bdev = bio->bi_bdev,
332 .bi_sector = bio->bi_sector,
333 .bi_size = bio->bi_size,
334 .bi_rw = bio->bi_rw,
335 };
336
337 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
338 prev->bv_len -= len;
339 return 0;
340 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100341 }
342
343 goto done;
344 }
345 }
346
347 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 return 0;
349
350 /*
351 * we might lose a segment or two here, but rather that than
352 * make this too complex.
353 */
354
355 while (bio->bi_phys_segments >= q->max_phys_segments
356 || bio->bi_hw_segments >= q->max_hw_segments
357 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
358
359 if (retried_segments)
360 return 0;
361
362 retried_segments = 1;
363 blk_recount_segments(q, bio);
364 }
365
366 /*
367 * setup the new entry, we might clear it again later if we
368 * cannot add the page
369 */
370 bvec = &bio->bi_io_vec[bio->bi_vcnt];
371 bvec->bv_page = page;
372 bvec->bv_len = len;
373 bvec->bv_offset = offset;
374
375 /*
376 * if queue has other restrictions (eg varying max sector size
377 * depending on offset), it can specify a merge_bvec_fn in the
378 * queue to get further control
379 */
380 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200381 struct bvec_merge_data bvm = {
382 .bi_bdev = bio->bi_bdev,
383 .bi_sector = bio->bi_sector,
384 .bi_size = bio->bi_size,
385 .bi_rw = bio->bi_rw,
386 };
387
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 /*
389 * merge_bvec_fn() returns number of bytes it can accept
390 * at this offset
391 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200392 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 bvec->bv_page = NULL;
394 bvec->bv_len = 0;
395 bvec->bv_offset = 0;
396 return 0;
397 }
398 }
399
400 /* If we may be able to merge these biovecs, force a recount */
401 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
402 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
403 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
404
405 bio->bi_vcnt++;
406 bio->bi_phys_segments++;
407 bio->bi_hw_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100408 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 bio->bi_size += len;
410 return len;
411}
412
413/**
Mike Christie6e68af62005-11-11 05:30:27 -0600414 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100415 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600416 * @bio: destination bio
417 * @page: page to add
418 * @len: vec entry length
419 * @offset: vec entry offset
420 *
421 * Attempt to add a page to the bio_vec maplist. This can fail for a
422 * number of reasons, such as the bio being full or target block
423 * device limitations. The target block device must allow bio's
424 * smaller than PAGE_SIZE, so it is always possible to add a single
425 * page to an empty bio. This should only be used by REQ_PC bios.
426 */
Jens Axboe165125e2007-07-24 09:28:11 +0200427int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600428 unsigned int len, unsigned int offset)
429{
Mike Christiedefd94b2005-12-05 02:37:06 -0600430 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
Mike Christie6e68af62005-11-11 05:30:27 -0600431}
432
433/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 * bio_add_page - attempt to add page to bio
435 * @bio: destination bio
436 * @page: page to add
437 * @len: vec entry length
438 * @offset: vec entry offset
439 *
440 * Attempt to add a page to the bio_vec maplist. This can fail for a
441 * number of reasons, such as the bio being full or target block
442 * device limitations. The target block device must allow bio's
443 * smaller than PAGE_SIZE, so it is always possible to add a single
444 * page to an empty bio.
445 */
446int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
447 unsigned int offset)
448{
Mike Christiedefd94b2005-12-05 02:37:06 -0600449 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
450 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451}
452
453struct bio_map_data {
454 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200455 int nr_sgvecs;
456 struct sg_iovec *sgvecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457};
458
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200459static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
460 struct sg_iovec *iov, int iov_count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
462 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200463 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
464 bmd->nr_sgvecs = iov_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 bio->bi_private = bmd;
466}
467
468static void bio_free_map_data(struct bio_map_data *bmd)
469{
470 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200471 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 kfree(bmd);
473}
474
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200475static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
478
479 if (!bmd)
480 return NULL;
481
482 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200483 if (!bmd->iovecs) {
484 kfree(bmd);
485 return NULL;
486 }
487
488 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, GFP_KERNEL);
489 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 return bmd;
491
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200492 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 kfree(bmd);
494 return NULL;
495}
496
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200497static int __bio_copy_iov(struct bio *bio, struct sg_iovec *iov, int iov_count,
498 int uncopy)
499{
500 int ret = 0, i;
501 struct bio_vec *bvec;
502 int iov_idx = 0;
503 unsigned int iov_off = 0;
504 int read = bio_data_dir(bio) == READ;
505
506 __bio_for_each_segment(bvec, bio, i, 0) {
507 char *bv_addr = page_address(bvec->bv_page);
508 unsigned int bv_len = bvec->bv_len;
509
510 while (bv_len && iov_idx < iov_count) {
511 unsigned int bytes;
512 char *iov_addr;
513
514 bytes = min_t(unsigned int,
515 iov[iov_idx].iov_len - iov_off, bv_len);
516 iov_addr = iov[iov_idx].iov_base + iov_off;
517
518 if (!ret) {
519 if (!read && !uncopy)
520 ret = copy_from_user(bv_addr, iov_addr,
521 bytes);
522 if (read && uncopy)
523 ret = copy_to_user(iov_addr, bv_addr,
524 bytes);
525
526 if (ret)
527 ret = -EFAULT;
528 }
529
530 bv_len -= bytes;
531 bv_addr += bytes;
532 iov_addr += bytes;
533 iov_off += bytes;
534
535 if (iov[iov_idx].iov_len == iov_off) {
536 iov_idx++;
537 iov_off = 0;
538 }
539 }
540
541 if (uncopy)
542 __free_page(bvec->bv_page);
543 }
544
545 return ret;
546}
547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548/**
549 * bio_uncopy_user - finish previously mapped bio
550 * @bio: bio being terminated
551 *
552 * Free pages allocated from bio_copy_user() and write back data
553 * to user space in case of a read.
554 */
555int bio_uncopy_user(struct bio *bio)
556{
557 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200558 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200560 ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 bio_free_map_data(bmd);
563 bio_put(bio);
564 return ret;
565}
566
567/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200568 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 * @q: destination block queue
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200570 * @iov: the iovec.
571 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 * @write_to_vm: bool indicating writing to pages or not
573 *
574 * Prepares and returns a bio for indirect user io, bouncing data
575 * to/from kernel pages as necessary. Must be paired with
576 * call bio_uncopy_user() on io completion.
577 */
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200578struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov,
579 int iov_count, int write_to_vm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 struct bio_map_data *bmd;
582 struct bio_vec *bvec;
583 struct page *page;
584 struct bio *bio;
585 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200586 int nr_pages = 0;
587 unsigned int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200589 for (i = 0; i < iov_count; i++) {
590 unsigned long uaddr;
591 unsigned long end;
592 unsigned long start;
593
594 uaddr = (unsigned long)iov[i].iov_base;
595 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
596 start = uaddr >> PAGE_SHIFT;
597
598 nr_pages += end - start;
599 len += iov[i].iov_len;
600 }
601
602 bmd = bio_alloc_map_data(nr_pages, iov_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 if (!bmd)
604 return ERR_PTR(-ENOMEM);
605
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 ret = -ENOMEM;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200607 bio = bio_alloc(GFP_KERNEL, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 if (!bio)
609 goto out_bmd;
610
611 bio->bi_rw |= (!write_to_vm << BIO_RW);
612
613 ret = 0;
614 while (len) {
615 unsigned int bytes = PAGE_SIZE;
616
617 if (bytes > len)
618 bytes = len;
619
620 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
621 if (!page) {
622 ret = -ENOMEM;
623 break;
624 }
625
Mike Christie0e75f902006-12-01 10:40:55 +0100626 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
629 len -= bytes;
630 }
631
632 if (ret)
633 goto cleanup;
634
635 /*
636 * success
637 */
638 if (!write_to_vm) {
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200639 ret = __bio_copy_iov(bio, iov, iov_count, 0);
640 if (ret)
641 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 }
643
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200644 bio_set_map_data(bmd, bio, iov, iov_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 return bio;
646cleanup:
647 bio_for_each_segment(bvec, bio, i)
648 __free_page(bvec->bv_page);
649
650 bio_put(bio);
651out_bmd:
652 bio_free_map_data(bmd);
653 return ERR_PTR(ret);
654}
655
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200656/**
657 * bio_copy_user - copy user data to bio
658 * @q: destination block queue
659 * @uaddr: start of user address
660 * @len: length in bytes
661 * @write_to_vm: bool indicating writing to pages or not
662 *
663 * Prepares and returns a bio for indirect user io, bouncing data
664 * to/from kernel pages as necessary. Must be paired with
665 * call bio_uncopy_user() on io completion.
666 */
667struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr,
668 unsigned int len, int write_to_vm)
669{
670 struct sg_iovec iov;
671
672 iov.iov_base = (void __user *)uaddr;
673 iov.iov_len = len;
674
675 return bio_copy_user_iov(q, &iov, 1, write_to_vm);
676}
677
Jens Axboe165125e2007-07-24 09:28:11 +0200678static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200679 struct block_device *bdev,
680 struct sg_iovec *iov, int iov_count,
681 int write_to_vm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
James Bottomley f1970ba2005-06-20 14:06:52 +0200683 int i, j;
684 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 struct page **pages;
686 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200687 int cur_page = 0;
688 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
James Bottomley f1970ba2005-06-20 14:06:52 +0200690 for (i = 0; i < iov_count; i++) {
691 unsigned long uaddr = (unsigned long)iov[i].iov_base;
692 unsigned long len = iov[i].iov_len;
693 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
694 unsigned long start = uaddr >> PAGE_SHIFT;
695
696 nr_pages += end - start;
697 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100698 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200699 */
Mike Christiead2d7222006-12-01 10:40:20 +0100700 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200701 return ERR_PTR(-EINVAL);
702 }
703
704 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 return ERR_PTR(-EINVAL);
706
707 bio = bio_alloc(GFP_KERNEL, nr_pages);
708 if (!bio)
709 return ERR_PTR(-ENOMEM);
710
711 ret = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800712 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (!pages)
714 goto out;
715
James Bottomley f1970ba2005-06-20 14:06:52 +0200716 for (i = 0; i < iov_count; i++) {
717 unsigned long uaddr = (unsigned long)iov[i].iov_base;
718 unsigned long len = iov[i].iov_len;
719 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
720 unsigned long start = uaddr >> PAGE_SHIFT;
721 const int local_nr_pages = end - start;
722 const int page_limit = cur_page + local_nr_pages;
723
724 down_read(&current->mm->mmap_sem);
725 ret = get_user_pages(current, current->mm, uaddr,
726 local_nr_pages,
727 write_to_vm, 0, &pages[cur_page], NULL);
728 up_read(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Jens Axboe99172152006-06-16 13:02:29 +0200730 if (ret < local_nr_pages) {
731 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200732 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200733 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
James Bottomley f1970ba2005-06-20 14:06:52 +0200735 offset = uaddr & ~PAGE_MASK;
736 for (j = cur_page; j < page_limit; j++) {
737 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
James Bottomley f1970ba2005-06-20 14:06:52 +0200739 if (len <= 0)
740 break;
741
742 if (bytes > len)
743 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
James Bottomley f1970ba2005-06-20 14:06:52 +0200745 /*
746 * sorry...
747 */
Mike Christiedefd94b2005-12-05 02:37:06 -0600748 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
749 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +0200750 break;
751
752 len -= bytes;
753 offset = 0;
754 }
755
756 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 /*
James Bottomley f1970ba2005-06-20 14:06:52 +0200758 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 */
James Bottomley f1970ba2005-06-20 14:06:52 +0200760 while (j < page_limit)
761 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 kfree(pages);
765
766 /*
767 * set data direction, and check if mapped pages need bouncing
768 */
769 if (!write_to_vm)
770 bio->bi_rw |= (1 << BIO_RW);
771
James Bottomley f1970ba2005-06-20 14:06:52 +0200772 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 bio->bi_flags |= (1 << BIO_USER_MAPPED);
774 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200775
776 out_unmap:
777 for (i = 0; i < nr_pages; i++) {
778 if(!pages[i])
779 break;
780 page_cache_release(pages[i]);
781 }
782 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 kfree(pages);
784 bio_put(bio);
785 return ERR_PTR(ret);
786}
787
788/**
789 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200790 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 * @bdev: destination block device
792 * @uaddr: start of user address
793 * @len: length in bytes
794 * @write_to_vm: bool indicating writing to pages or not
795 *
796 * Map the user space address into a bio suitable for io to a block
797 * device. Returns an error pointer in case of error.
798 */
Jens Axboe165125e2007-07-24 09:28:11 +0200799struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 unsigned long uaddr, unsigned int len, int write_to_vm)
801{
James Bottomley f1970ba2005-06-20 14:06:52 +0200802 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +0100804 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +0200805 iov.iov_len = len;
806
807 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
808}
809
810/**
811 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200812 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +0200813 * @bdev: destination block device
814 * @iov: the iovec.
815 * @iov_count: number of elements in the iovec
816 * @write_to_vm: bool indicating writing to pages or not
817 *
818 * Map the user space address into a bio suitable for io to a block
819 * device. Returns an error pointer in case of error.
820 */
Jens Axboe165125e2007-07-24 09:28:11 +0200821struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +0200822 struct sg_iovec *iov, int iov_count,
823 int write_to_vm)
824{
825 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200826
827 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
829 if (IS_ERR(bio))
830 return bio;
831
832 /*
833 * subtle -- if __bio_map_user() ended up bouncing a bio,
834 * it would normally disappear when its bi_end_io is run.
835 * however, we need it for the unmap, so grab an extra
836 * reference to it
837 */
838 bio_get(bio);
839
Mike Christie0e75f902006-12-01 10:40:55 +0100840 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841}
842
843static void __bio_unmap_user(struct bio *bio)
844{
845 struct bio_vec *bvec;
846 int i;
847
848 /*
849 * make sure we dirty pages we wrote to
850 */
851 __bio_for_each_segment(bvec, bio, i, 0) {
852 if (bio_data_dir(bio) == READ)
853 set_page_dirty_lock(bvec->bv_page);
854
855 page_cache_release(bvec->bv_page);
856 }
857
858 bio_put(bio);
859}
860
861/**
862 * bio_unmap_user - unmap a bio
863 * @bio: the bio being unmapped
864 *
865 * Unmap a bio previously mapped by bio_map_user(). Must be called with
866 * a process context.
867 *
868 * bio_unmap_user() may sleep.
869 */
870void bio_unmap_user(struct bio *bio)
871{
872 __bio_unmap_user(bio);
873 bio_put(bio);
874}
875
NeilBrown6712ecf2007-09-27 12:47:43 +0200876static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +0200877{
Jens Axboeb8238252005-06-20 14:05:27 +0200878 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +0200879}
880
881
Jens Axboe165125e2007-07-24 09:28:11 +0200882static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -0400883 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +0200884{
885 unsigned long kaddr = (unsigned long)data;
886 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
887 unsigned long start = kaddr >> PAGE_SHIFT;
888 const int nr_pages = end - start;
889 int offset, i;
890 struct bio *bio;
891
892 bio = bio_alloc(gfp_mask, nr_pages);
893 if (!bio)
894 return ERR_PTR(-ENOMEM);
895
896 offset = offset_in_page(kaddr);
897 for (i = 0; i < nr_pages; i++) {
898 unsigned int bytes = PAGE_SIZE - offset;
899
900 if (len <= 0)
901 break;
902
903 if (bytes > len)
904 bytes = len;
905
Mike Christiedefd94b2005-12-05 02:37:06 -0600906 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
907 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +0200908 break;
909
910 data += bytes;
911 len -= bytes;
912 offset = 0;
913 }
914
Jens Axboeb8238252005-06-20 14:05:27 +0200915 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +0200916 return bio;
917}
918
919/**
920 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200921 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +0200922 * @data: pointer to buffer to map
923 * @len: length in bytes
924 * @gfp_mask: allocation flags for bio allocation
925 *
926 * Map the kernel address into a bio suitable for io to a block
927 * device. Returns an error pointer in case of error.
928 */
Jens Axboe165125e2007-07-24 09:28:11 +0200929struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -0400930 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +0200931{
932 struct bio *bio;
933
934 bio = __bio_map_kern(q, data, len, gfp_mask);
935 if (IS_ERR(bio))
936 return bio;
937
938 if (bio->bi_size == len)
939 return bio;
940
941 /*
942 * Don't support partial mappings.
943 */
944 bio_put(bio);
945 return ERR_PTR(-EINVAL);
946}
947
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200948static void bio_copy_kern_endio(struct bio *bio, int err)
949{
950 struct bio_vec *bvec;
951 const int read = bio_data_dir(bio) == READ;
952 char *p = bio->bi_private;
953 int i;
954
955 __bio_for_each_segment(bvec, bio, i, 0) {
956 char *addr = page_address(bvec->bv_page);
957
958 if (read && !err)
959 memcpy(p, addr, bvec->bv_len);
960
961 __free_page(bvec->bv_page);
962 p += bvec->bv_len;
963 }
964
965 bio_put(bio);
966}
967
968/**
969 * bio_copy_kern - copy kernel address into bio
970 * @q: the struct request_queue for the bio
971 * @data: pointer to buffer to copy
972 * @len: length in bytes
973 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +0200974 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200975 *
976 * copy the kernel address into a bio suitable for io to a block
977 * device. Returns an error pointer in case of error.
978 */
979struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
980 gfp_t gfp_mask, int reading)
981{
982 unsigned long kaddr = (unsigned long)data;
983 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
984 unsigned long start = kaddr >> PAGE_SHIFT;
985 const int nr_pages = end - start;
986 struct bio *bio;
987 struct bio_vec *bvec;
988 int i, ret;
989
990 bio = bio_alloc(gfp_mask, nr_pages);
991 if (!bio)
992 return ERR_PTR(-ENOMEM);
993
994 while (len) {
995 struct page *page;
996 unsigned int bytes = PAGE_SIZE;
997
998 if (bytes > len)
999 bytes = len;
1000
1001 page = alloc_page(q->bounce_gfp | gfp_mask);
1002 if (!page) {
1003 ret = -ENOMEM;
1004 goto cleanup;
1005 }
1006
1007 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) {
1008 ret = -EINVAL;
1009 goto cleanup;
1010 }
1011
1012 len -= bytes;
1013 }
1014
1015 if (!reading) {
1016 void *p = data;
1017
1018 bio_for_each_segment(bvec, bio, i) {
1019 char *addr = page_address(bvec->bv_page);
1020
1021 memcpy(addr, p, bvec->bv_len);
1022 p += bvec->bv_len;
1023 }
1024 }
1025
1026 bio->bi_private = data;
1027 bio->bi_end_io = bio_copy_kern_endio;
1028 return bio;
1029cleanup:
1030 bio_for_each_segment(bvec, bio, i)
1031 __free_page(bvec->bv_page);
1032
1033 bio_put(bio);
1034
1035 return ERR_PTR(ret);
1036}
1037
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038/*
1039 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1040 * for performing direct-IO in BIOs.
1041 *
1042 * The problem is that we cannot run set_page_dirty() from interrupt context
1043 * because the required locks are not interrupt-safe. So what we can do is to
1044 * mark the pages dirty _before_ performing IO. And in interrupt context,
1045 * check that the pages are still dirty. If so, fine. If not, redirty them
1046 * in process context.
1047 *
1048 * We special-case compound pages here: normally this means reads into hugetlb
1049 * pages. The logic in here doesn't really work right for compound pages
1050 * because the VM does not uniformly chase down the head page in all cases.
1051 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1052 * handle them at all. So we skip compound pages here at an early stage.
1053 *
1054 * Note that this code is very hard to test under normal circumstances because
1055 * direct-io pins the pages with get_user_pages(). This makes
1056 * is_page_cache_freeable return false, and the VM will not clean the pages.
1057 * But other code (eg, pdflush) could clean the pages if they are mapped
1058 * pagecache.
1059 *
1060 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1061 * deferred bio dirtying paths.
1062 */
1063
1064/*
1065 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1066 */
1067void bio_set_pages_dirty(struct bio *bio)
1068{
1069 struct bio_vec *bvec = bio->bi_io_vec;
1070 int i;
1071
1072 for (i = 0; i < bio->bi_vcnt; i++) {
1073 struct page *page = bvec[i].bv_page;
1074
1075 if (page && !PageCompound(page))
1076 set_page_dirty_lock(page);
1077 }
1078}
1079
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001080static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081{
1082 struct bio_vec *bvec = bio->bi_io_vec;
1083 int i;
1084
1085 for (i = 0; i < bio->bi_vcnt; i++) {
1086 struct page *page = bvec[i].bv_page;
1087
1088 if (page)
1089 put_page(page);
1090 }
1091}
1092
1093/*
1094 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1095 * If they are, then fine. If, however, some pages are clean then they must
1096 * have been written out during the direct-IO read. So we take another ref on
1097 * the BIO and the offending pages and re-dirty the pages in process context.
1098 *
1099 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1100 * here on. It will run one page_cache_release() against each page and will
1101 * run one bio_put() against the BIO.
1102 */
1103
David Howells65f27f32006-11-22 14:55:48 +00001104static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
David Howells65f27f32006-11-22 14:55:48 +00001106static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107static DEFINE_SPINLOCK(bio_dirty_lock);
1108static struct bio *bio_dirty_list;
1109
1110/*
1111 * This runs in process context
1112 */
David Howells65f27f32006-11-22 14:55:48 +00001113static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114{
1115 unsigned long flags;
1116 struct bio *bio;
1117
1118 spin_lock_irqsave(&bio_dirty_lock, flags);
1119 bio = bio_dirty_list;
1120 bio_dirty_list = NULL;
1121 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1122
1123 while (bio) {
1124 struct bio *next = bio->bi_private;
1125
1126 bio_set_pages_dirty(bio);
1127 bio_release_pages(bio);
1128 bio_put(bio);
1129 bio = next;
1130 }
1131}
1132
1133void bio_check_pages_dirty(struct bio *bio)
1134{
1135 struct bio_vec *bvec = bio->bi_io_vec;
1136 int nr_clean_pages = 0;
1137 int i;
1138
1139 for (i = 0; i < bio->bi_vcnt; i++) {
1140 struct page *page = bvec[i].bv_page;
1141
1142 if (PageDirty(page) || PageCompound(page)) {
1143 page_cache_release(page);
1144 bvec[i].bv_page = NULL;
1145 } else {
1146 nr_clean_pages++;
1147 }
1148 }
1149
1150 if (nr_clean_pages) {
1151 unsigned long flags;
1152
1153 spin_lock_irqsave(&bio_dirty_lock, flags);
1154 bio->bi_private = bio_dirty_list;
1155 bio_dirty_list = bio;
1156 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1157 schedule_work(&bio_dirty_work);
1158 } else {
1159 bio_put(bio);
1160 }
1161}
1162
1163/**
1164 * bio_endio - end I/O on a bio
1165 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 * @error: error, if any
1167 *
1168 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001169 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001170 * preferred way to end I/O on a bio, it takes care of clearing
1171 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1172 * established -Exxxx (-EIO, for instance) error values in case
1173 * something went wrong. Noone should call bi_end_io() directly on a
1174 * bio unless they own it and thus know that it has an end_io
1175 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001177void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178{
1179 if (error)
1180 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001181 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1182 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
NeilBrown5bb23a62007-09-27 12:46:13 +02001184 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001185 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186}
1187
1188void bio_pair_release(struct bio_pair *bp)
1189{
1190 if (atomic_dec_and_test(&bp->cnt)) {
1191 struct bio *master = bp->bio1.bi_private;
1192
NeilBrown6712ecf2007-09-27 12:47:43 +02001193 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 mempool_free(bp, bp->bio2.bi_private);
1195 }
1196}
1197
NeilBrown6712ecf2007-09-27 12:47:43 +02001198static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199{
1200 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1201
1202 if (err)
1203 bp->error = err;
1204
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206}
1207
NeilBrown6712ecf2007-09-27 12:47:43 +02001208static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209{
1210 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1211
1212 if (err)
1213 bp->error = err;
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216}
1217
1218/*
1219 * split a bio - only worry about a bio with a single page
1220 * in it's iovec
1221 */
1222struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1223{
1224 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1225
1226 if (!bp)
1227 return bp;
1228
Jens Axboe2056a782006-03-23 20:00:26 +01001229 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
1230 bi->bi_sector + first_sectors);
1231
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 BUG_ON(bi->bi_vcnt != 1);
1233 BUG_ON(bi->bi_idx != 0);
1234 atomic_set(&bp->cnt, 3);
1235 bp->error = 0;
1236 bp->bio1 = *bi;
1237 bp->bio2 = *bi;
1238 bp->bio2.bi_sector += first_sectors;
1239 bp->bio2.bi_size -= first_sectors << 9;
1240 bp->bio1.bi_size = first_sectors << 9;
1241
1242 bp->bv1 = bi->bi_io_vec[0];
1243 bp->bv2 = bi->bi_io_vec[0];
1244 bp->bv2.bv_offset += first_sectors << 9;
1245 bp->bv2.bv_len -= first_sectors << 9;
1246 bp->bv1.bv_len = first_sectors << 9;
1247
1248 bp->bio1.bi_io_vec = &bp->bv1;
1249 bp->bio2.bi_io_vec = &bp->bv2;
1250
NeilBrowna2eb0c12006-05-22 22:35:27 -07001251 bp->bio1.bi_max_vecs = 1;
1252 bp->bio2.bi_max_vecs = 1;
1253
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 bp->bio1.bi_end_io = bio_pair_end_1;
1255 bp->bio2.bi_end_io = bio_pair_end_2;
1256
1257 bp->bio1.bi_private = bi;
1258 bp->bio2.bi_private = pool;
1259
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001260 if (bio_integrity(bi))
1261 bio_integrity_split(bi, bp, first_sectors);
1262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 return bp;
1264}
1265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
1267/*
1268 * create memory pools for biovec's in a bio_set.
1269 * use the global biovec slabs created for general use.
1270 */
Jens Axboe59725112007-04-02 10:06:42 +02001271static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272{
1273 int i;
1274
1275 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1276 struct biovec_slab *bp = bvec_slabs + i;
1277 mempool_t **bvp = bs->bvec_pools + i;
1278
Matthew Dobson93d23412006-03-26 01:37:50 -08001279 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 if (!*bvp)
1281 return -ENOMEM;
1282 }
1283 return 0;
1284}
1285
1286static void biovec_free_pools(struct bio_set *bs)
1287{
1288 int i;
1289
1290 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1291 mempool_t *bvp = bs->bvec_pools[i];
1292
1293 if (bvp)
1294 mempool_destroy(bvp);
1295 }
1296
1297}
1298
1299void bioset_free(struct bio_set *bs)
1300{
1301 if (bs->bio_pool)
1302 mempool_destroy(bs->bio_pool);
1303
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001304 bioset_integrity_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 biovec_free_pools(bs);
1306
1307 kfree(bs);
1308}
1309
Jens Axboe59725112007-04-02 10:06:42 +02001310struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001312 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
1314 if (!bs)
1315 return NULL;
1316
Matthew Dobson93d23412006-03-26 01:37:50 -08001317 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 if (!bs->bio_pool)
1319 goto bad;
1320
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001321 if (bioset_integrity_create(bs, bio_pool_size))
1322 goto bad;
1323
Jens Axboe59725112007-04-02 10:06:42 +02001324 if (!biovec_create_pools(bs, bvec_pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 return bs;
1326
1327bad:
1328 bioset_free(bs);
1329 return NULL;
1330}
1331
1332static void __init biovec_init_slabs(void)
1333{
1334 int i;
1335
1336 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1337 int size;
1338 struct biovec_slab *bvs = bvec_slabs + i;
1339
1340 size = bvs->nr_vecs * sizeof(struct bio_vec);
1341 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001342 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 }
1344}
1345
1346static int __init init_bio(void)
1347{
Christoph Lameter0a31bd52007-05-06 14:49:57 -07001348 bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001350 bio_integrity_init_slab();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 biovec_init_slabs();
1352
Jens Axboe59725112007-04-02 10:06:42 +02001353 fs_bio_set = bioset_create(BIO_POOL_SIZE, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 if (!fs_bio_set)
1355 panic("bio: can't allocate bios\n");
1356
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001357 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1358 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 if (!bio_split_pool)
1360 panic("bio: can't create split pool\n");
1361
1362 return 0;
1363}
1364
1365subsys_initcall(init_bio);
1366
1367EXPORT_SYMBOL(bio_alloc);
1368EXPORT_SYMBOL(bio_put);
Peter Osterlund36763472005-09-06 15:16:42 -07001369EXPORT_SYMBOL(bio_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370EXPORT_SYMBOL(bio_endio);
1371EXPORT_SYMBOL(bio_init);
1372EXPORT_SYMBOL(__bio_clone);
1373EXPORT_SYMBOL(bio_clone);
1374EXPORT_SYMBOL(bio_phys_segments);
1375EXPORT_SYMBOL(bio_hw_segments);
1376EXPORT_SYMBOL(bio_add_page);
Mike Christie6e68af62005-11-11 05:30:27 -06001377EXPORT_SYMBOL(bio_add_pc_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378EXPORT_SYMBOL(bio_get_nr_vecs);
Jens Axboe40044ce2008-03-17 21:14:40 +01001379EXPORT_SYMBOL(bio_map_user);
1380EXPORT_SYMBOL(bio_unmap_user);
Mike Christie df46b9a2005-06-20 14:04:44 +02001381EXPORT_SYMBOL(bio_map_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001382EXPORT_SYMBOL(bio_copy_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383EXPORT_SYMBOL(bio_pair_release);
1384EXPORT_SYMBOL(bio_split);
1385EXPORT_SYMBOL(bio_split_pool);
1386EXPORT_SYMBOL(bio_copy_user);
1387EXPORT_SYMBOL(bio_uncopy_user);
1388EXPORT_SYMBOL(bioset_create);
1389EXPORT_SYMBOL(bioset_free);
1390EXPORT_SYMBOL(bio_alloc_bioset);