blob: 3cba7ae34d758c860ecff09f2be059663d2de35b [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);
Denis ChengRq1ac0ae02008-08-04 11:56:30 +020080 if (bvl)
81 memset(bvl, 0, bvec_nr_vecs(*idx) * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83 return bvl;
84}
85
Peter Osterlund36763472005-09-06 15:16:42 -070086void bio_free(struct bio *bio, struct bio_set *bio_set)
Linus Torvalds1da177e2005-04-16 15:20:36 -070087{
Jens Axboe992c5dd2007-07-18 13:18:08 +020088 if (bio->bi_io_vec) {
89 const int pool_idx = BIO_POOL_IDX(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Jens Axboe992c5dd2007-07-18 13:18:08 +020091 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Jens Axboe992c5dd2007-07-18 13:18:08 +020093 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]);
94 }
95
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +020096 if (bio_integrity(bio))
97 bio_integrity_free(bio, bio_set);
98
Peter Osterlund36763472005-09-06 15:16:42 -070099 mempool_free(bio, bio_set->bio_pool);
100}
101
102/*
103 * default destructor for a bio allocated with bio_alloc_bioset()
104 */
105static void bio_fs_destructor(struct bio *bio)
106{
107 bio_free(bio, fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108}
109
Arjan van de Ven858119e2006-01-14 13:20:43 -0800110void bio_init(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111{
Jens Axboe2b94de52007-07-18 13:14:03 +0200112 memset(bio, 0, sizeof(*bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 bio->bi_flags = 1 << BIO_UPTODATE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 atomic_set(&bio->bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115}
116
117/**
118 * bio_alloc_bioset - allocate a bio for I/O
119 * @gfp_mask: the GFP_ mask given to the slab allocator
120 * @nr_iovecs: number of iovecs to pre-allocate
Martin Waitz67be2dd2005-05-01 08:59:26 -0700121 * @bs: the bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 *
123 * Description:
124 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
125 * If %__GFP_WAIT is set then we will block on the internal pool waiting
126 * for a &struct bio to become free.
127 *
128 * allocate bio and iovecs from the memory pools specified by the
129 * bio_set structure.
130 **/
Al Virodd0fc662005-10-07 07:46:04 +0100131struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
133 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask);
134
135 if (likely(bio)) {
136 struct bio_vec *bvl = NULL;
137
138 bio_init(bio);
139 if (likely(nr_iovecs)) {
Jens Axboeeeae1d42008-05-07 13:26:27 +0200140 unsigned long uninitialized_var(idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
142 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
143 if (unlikely(!bvl)) {
144 mempool_free(bio, bs->bio_pool);
145 bio = NULL;
146 goto out;
147 }
148 bio->bi_flags |= idx << BIO_POOL_OFFSET;
Denis ChengRq1ac0ae02008-08-04 11:56:30 +0200149 bio->bi_max_vecs = bvec_nr_vecs(idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 }
151 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 }
153out:
154 return bio;
155}
156
Al Virodd0fc662005-10-07 07:46:04 +0100157struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{
Peter Osterlund36763472005-09-06 15:16:42 -0700159 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
160
161 if (bio)
162 bio->bi_destructor = bio_fs_destructor;
163
164 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165}
166
167void zero_fill_bio(struct bio *bio)
168{
169 unsigned long flags;
170 struct bio_vec *bv;
171 int i;
172
173 bio_for_each_segment(bv, bio, i) {
174 char *data = bvec_kmap_irq(bv, &flags);
175 memset(data, 0, bv->bv_len);
176 flush_dcache_page(bv->bv_page);
177 bvec_kunmap_irq(data, &flags);
178 }
179}
180EXPORT_SYMBOL(zero_fill_bio);
181
182/**
183 * bio_put - release a reference to a bio
184 * @bio: bio to release reference to
185 *
186 * Description:
187 * Put a reference to a &struct bio, either one you have gotten with
188 * bio_alloc or bio_get. The last put of a bio will free it.
189 **/
190void bio_put(struct bio *bio)
191{
192 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
193
194 /*
195 * last put frees it
196 */
197 if (atomic_dec_and_test(&bio->bi_cnt)) {
198 bio->bi_next = NULL;
199 bio->bi_destructor(bio);
200 }
201}
202
Jens Axboe165125e2007-07-24 09:28:11 +0200203inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
205 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
206 blk_recount_segments(q, bio);
207
208 return bio->bi_phys_segments;
209}
210
Jens Axboe165125e2007-07-24 09:28:11 +0200211inline int bio_hw_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
213 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
214 blk_recount_segments(q, bio);
215
216 return bio->bi_hw_segments;
217}
218
219/**
220 * __bio_clone - clone a bio
221 * @bio: destination bio
222 * @bio_src: bio to clone
223 *
224 * Clone a &bio. Caller will own the returned bio, but not
225 * the actual data it points to. Reference count of returned
226 * bio will be one.
227 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800228void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
Andrew Mortone525e152005-08-07 09:42:12 -0700230 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
231 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Jens Axboe5d840702008-01-25 12:44:44 +0100233 /*
234 * most users will be overriding ->bi_bdev with a new target,
235 * so we don't set nor calculate new physical/hw segment counts here
236 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 bio->bi_sector = bio_src->bi_sector;
238 bio->bi_bdev = bio_src->bi_bdev;
239 bio->bi_flags |= 1 << BIO_CLONED;
240 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 bio->bi_vcnt = bio_src->bi_vcnt;
242 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700243 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
245
246/**
247 * bio_clone - clone a bio
248 * @bio: bio to clone
249 * @gfp_mask: allocation priority
250 *
251 * Like __bio_clone, only also allocates the returned bio
252 */
Al Virodd0fc662005-10-07 07:46:04 +0100253struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
255 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
256
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200257 if (!b)
258 return NULL;
259
260 b->bi_destructor = bio_fs_destructor;
261 __bio_clone(b, bio);
262
263 if (bio_integrity(bio)) {
264 int ret;
265
266 ret = bio_integrity_clone(b, bio, fs_bio_set);
267
268 if (ret < 0)
269 return NULL;
Peter Osterlund36763472005-09-06 15:16:42 -0700270 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
272 return b;
273}
274
275/**
276 * bio_get_nr_vecs - return approx number of vecs
277 * @bdev: I/O target
278 *
279 * Return the approximate number of pages we can send to this target.
280 * There's no guarantee that you will be able to fit this number of pages
281 * into a bio, it does not account for dynamic restrictions that vary
282 * on offset.
283 */
284int bio_get_nr_vecs(struct block_device *bdev)
285{
Jens Axboe165125e2007-07-24 09:28:11 +0200286 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 int nr_pages;
288
289 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
290 if (nr_pages > q->max_phys_segments)
291 nr_pages = q->max_phys_segments;
292 if (nr_pages > q->max_hw_segments)
293 nr_pages = q->max_hw_segments;
294
295 return nr_pages;
296}
297
Jens Axboe165125e2007-07-24 09:28:11 +0200298static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600299 *page, unsigned int len, unsigned int offset,
300 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
302 int retried_segments = 0;
303 struct bio_vec *bvec;
304
305 /*
306 * cloned bio must not modify vec list
307 */
308 if (unlikely(bio_flagged(bio, BIO_CLONED)))
309 return 0;
310
Jens Axboe80cfd542006-01-06 09:43:28 +0100311 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 return 0;
313
Jens Axboe80cfd542006-01-06 09:43:28 +0100314 /*
315 * For filesystems with a blocksize smaller than the pagesize
316 * we will often be called with the same page as last time and
317 * a consecutive offset. Optimize this special case.
318 */
319 if (bio->bi_vcnt > 0) {
320 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
321
322 if (page == prev->bv_page &&
323 offset == prev->bv_offset + prev->bv_len) {
324 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200325
326 if (q->merge_bvec_fn) {
327 struct bvec_merge_data bvm = {
328 .bi_bdev = bio->bi_bdev,
329 .bi_sector = bio->bi_sector,
330 .bi_size = bio->bi_size,
331 .bi_rw = bio->bi_rw,
332 };
333
334 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
335 prev->bv_len -= len;
336 return 0;
337 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100338 }
339
340 goto done;
341 }
342 }
343
344 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 return 0;
346
347 /*
348 * we might lose a segment or two here, but rather that than
349 * make this too complex.
350 */
351
352 while (bio->bi_phys_segments >= q->max_phys_segments
353 || bio->bi_hw_segments >= q->max_hw_segments
354 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
355
356 if (retried_segments)
357 return 0;
358
359 retried_segments = 1;
360 blk_recount_segments(q, bio);
361 }
362
363 /*
364 * setup the new entry, we might clear it again later if we
365 * cannot add the page
366 */
367 bvec = &bio->bi_io_vec[bio->bi_vcnt];
368 bvec->bv_page = page;
369 bvec->bv_len = len;
370 bvec->bv_offset = offset;
371
372 /*
373 * if queue has other restrictions (eg varying max sector size
374 * depending on offset), it can specify a merge_bvec_fn in the
375 * queue to get further control
376 */
377 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200378 struct bvec_merge_data bvm = {
379 .bi_bdev = bio->bi_bdev,
380 .bi_sector = bio->bi_sector,
381 .bi_size = bio->bi_size,
382 .bi_rw = bio->bi_rw,
383 };
384
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 /*
386 * merge_bvec_fn() returns number of bytes it can accept
387 * at this offset
388 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200389 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 bvec->bv_page = NULL;
391 bvec->bv_len = 0;
392 bvec->bv_offset = 0;
393 return 0;
394 }
395 }
396
397 /* If we may be able to merge these biovecs, force a recount */
398 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
399 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
400 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
401
402 bio->bi_vcnt++;
403 bio->bi_phys_segments++;
404 bio->bi_hw_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100405 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 bio->bi_size += len;
407 return len;
408}
409
410/**
Mike Christie6e68af62005-11-11 05:30:27 -0600411 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100412 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600413 * @bio: destination bio
414 * @page: page to add
415 * @len: vec entry length
416 * @offset: vec entry offset
417 *
418 * Attempt to add a page to the bio_vec maplist. This can fail for a
419 * number of reasons, such as the bio being full or target block
420 * device limitations. The target block device must allow bio's
421 * smaller than PAGE_SIZE, so it is always possible to add a single
422 * page to an empty bio. This should only be used by REQ_PC bios.
423 */
Jens Axboe165125e2007-07-24 09:28:11 +0200424int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600425 unsigned int len, unsigned int offset)
426{
Mike Christiedefd94b2005-12-05 02:37:06 -0600427 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
Mike Christie6e68af62005-11-11 05:30:27 -0600428}
429
430/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 * bio_add_page - attempt to add page to bio
432 * @bio: destination bio
433 * @page: page to add
434 * @len: vec entry length
435 * @offset: vec entry offset
436 *
437 * Attempt to add a page to the bio_vec maplist. This can fail for a
438 * number of reasons, such as the bio being full or target block
439 * device limitations. The target block device must allow bio's
440 * smaller than PAGE_SIZE, so it is always possible to add a single
441 * page to an empty bio.
442 */
443int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
444 unsigned int offset)
445{
Mike Christiedefd94b2005-12-05 02:37:06 -0600446 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
447 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
450struct bio_map_data {
451 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200452 int nr_sgvecs;
453 struct sg_iovec *sgvecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454};
455
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200456static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
457 struct sg_iovec *iov, int iov_count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
459 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200460 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
461 bmd->nr_sgvecs = iov_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 bio->bi_private = bmd;
463}
464
465static void bio_free_map_data(struct bio_map_data *bmd)
466{
467 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200468 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 kfree(bmd);
470}
471
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200472static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
473 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474{
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200475 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
477 if (!bmd)
478 return NULL;
479
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200480 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200481 if (!bmd->iovecs) {
482 kfree(bmd);
483 return NULL;
484 }
485
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200486 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200487 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 return bmd;
489
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200490 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 kfree(bmd);
492 return NULL;
493}
494
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200495static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
496 struct sg_iovec *iov, int iov_count, int uncopy)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200497{
498 int ret = 0, i;
499 struct bio_vec *bvec;
500 int iov_idx = 0;
501 unsigned int iov_off = 0;
502 int read = bio_data_dir(bio) == READ;
503
504 __bio_for_each_segment(bvec, bio, i, 0) {
505 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200506 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200507
508 while (bv_len && iov_idx < iov_count) {
509 unsigned int bytes;
510 char *iov_addr;
511
512 bytes = min_t(unsigned int,
513 iov[iov_idx].iov_len - iov_off, bv_len);
514 iov_addr = iov[iov_idx].iov_base + iov_off;
515
516 if (!ret) {
517 if (!read && !uncopy)
518 ret = copy_from_user(bv_addr, iov_addr,
519 bytes);
520 if (read && uncopy)
521 ret = copy_to_user(iov_addr, bv_addr,
522 bytes);
523
524 if (ret)
525 ret = -EFAULT;
526 }
527
528 bv_len -= bytes;
529 bv_addr += bytes;
530 iov_addr += bytes;
531 iov_off += bytes;
532
533 if (iov[iov_idx].iov_len == iov_off) {
534 iov_idx++;
535 iov_off = 0;
536 }
537 }
538
539 if (uncopy)
540 __free_page(bvec->bv_page);
541 }
542
543 return ret;
544}
545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546/**
547 * bio_uncopy_user - finish previously mapped bio
548 * @bio: bio being terminated
549 *
550 * Free pages allocated from bio_copy_user() and write back data
551 * to user space in case of a read.
552 */
553int bio_uncopy_user(struct bio *bio)
554{
555 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200556 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200558 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, bmd->nr_sgvecs, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 bio_free_map_data(bmd);
561 bio_put(bio);
562 return ret;
563}
564
565/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200566 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 * @q: destination block queue
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200568 * @iov: the iovec.
569 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 * @write_to_vm: bool indicating writing to pages or not
571 *
572 * Prepares and returns a bio for indirect user io, bouncing data
573 * to/from kernel pages as necessary. Must be paired with
574 * call bio_uncopy_user() on io completion.
575 */
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200576struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov,
577 int iov_count, int write_to_vm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 struct bio_map_data *bmd;
580 struct bio_vec *bvec;
581 struct page *page;
582 struct bio *bio;
583 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200584 int nr_pages = 0;
585 unsigned int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200587 for (i = 0; i < iov_count; i++) {
588 unsigned long uaddr;
589 unsigned long end;
590 unsigned long start;
591
592 uaddr = (unsigned long)iov[i].iov_base;
593 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
594 start = uaddr >> PAGE_SHIFT;
595
596 nr_pages += end - start;
597 len += iov[i].iov_len;
598 }
599
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200600 bmd = bio_alloc_map_data(nr_pages, iov_count, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 if (!bmd)
602 return ERR_PTR(-ENOMEM);
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 ret = -ENOMEM;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200605 bio = bio_alloc(GFP_KERNEL, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 if (!bio)
607 goto out_bmd;
608
609 bio->bi_rw |= (!write_to_vm << BIO_RW);
610
611 ret = 0;
612 while (len) {
613 unsigned int bytes = PAGE_SIZE;
614
615 if (bytes > len)
616 bytes = len;
617
618 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
619 if (!page) {
620 ret = -ENOMEM;
621 break;
622 }
623
Mike Christie0e75f902006-12-01 10:40:55 +0100624 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
627 len -= bytes;
628 }
629
630 if (ret)
631 goto cleanup;
632
633 /*
634 * success
635 */
636 if (!write_to_vm) {
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200637 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200638 if (ret)
639 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 }
641
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200642 bio_set_map_data(bmd, bio, iov, iov_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 return bio;
644cleanup:
645 bio_for_each_segment(bvec, bio, i)
646 __free_page(bvec->bv_page);
647
648 bio_put(bio);
649out_bmd:
650 bio_free_map_data(bmd);
651 return ERR_PTR(ret);
652}
653
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200654/**
655 * bio_copy_user - copy user data to bio
656 * @q: destination block queue
657 * @uaddr: start of user address
658 * @len: length in bytes
659 * @write_to_vm: bool indicating writing to pages or not
660 *
661 * Prepares and returns a bio for indirect user io, bouncing data
662 * to/from kernel pages as necessary. Must be paired with
663 * call bio_uncopy_user() on io completion.
664 */
665struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr,
666 unsigned int len, int write_to_vm)
667{
668 struct sg_iovec iov;
669
670 iov.iov_base = (void __user *)uaddr;
671 iov.iov_len = len;
672
673 return bio_copy_user_iov(q, &iov, 1, write_to_vm);
674}
675
Jens Axboe165125e2007-07-24 09:28:11 +0200676static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200677 struct block_device *bdev,
678 struct sg_iovec *iov, int iov_count,
679 int write_to_vm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
James Bottomley f1970ba2005-06-20 14:06:52 +0200681 int i, j;
682 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 struct page **pages;
684 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200685 int cur_page = 0;
686 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
James Bottomley f1970ba2005-06-20 14:06:52 +0200688 for (i = 0; i < iov_count; i++) {
689 unsigned long uaddr = (unsigned long)iov[i].iov_base;
690 unsigned long len = iov[i].iov_len;
691 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
692 unsigned long start = uaddr >> PAGE_SHIFT;
693
694 nr_pages += end - start;
695 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100696 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200697 */
Mike Christiead2d7222006-12-01 10:40:20 +0100698 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200699 return ERR_PTR(-EINVAL);
700 }
701
702 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 return ERR_PTR(-EINVAL);
704
705 bio = bio_alloc(GFP_KERNEL, nr_pages);
706 if (!bio)
707 return ERR_PTR(-ENOMEM);
708
709 ret = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800710 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 if (!pages)
712 goto out;
713
James Bottomley f1970ba2005-06-20 14:06:52 +0200714 for (i = 0; i < iov_count; i++) {
715 unsigned long uaddr = (unsigned long)iov[i].iov_base;
716 unsigned long len = iov[i].iov_len;
717 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
718 unsigned long start = uaddr >> PAGE_SHIFT;
719 const int local_nr_pages = end - start;
720 const int page_limit = cur_page + local_nr_pages;
721
Nick Pigginf5dd33c2008-07-25 19:45:25 -0700722 ret = get_user_pages_fast(uaddr, local_nr_pages,
723 write_to_vm, &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +0200724 if (ret < local_nr_pages) {
725 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200726 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200727 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
James Bottomley f1970ba2005-06-20 14:06:52 +0200729 offset = uaddr & ~PAGE_MASK;
730 for (j = cur_page; j < page_limit; j++) {
731 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
James Bottomley f1970ba2005-06-20 14:06:52 +0200733 if (len <= 0)
734 break;
735
736 if (bytes > len)
737 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
James Bottomley f1970ba2005-06-20 14:06:52 +0200739 /*
740 * sorry...
741 */
Mike Christiedefd94b2005-12-05 02:37:06 -0600742 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
743 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +0200744 break;
745
746 len -= bytes;
747 offset = 0;
748 }
749
750 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 /*
James Bottomley f1970ba2005-06-20 14:06:52 +0200752 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 */
James Bottomley f1970ba2005-06-20 14:06:52 +0200754 while (j < page_limit)
755 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 }
757
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 kfree(pages);
759
760 /*
761 * set data direction, and check if mapped pages need bouncing
762 */
763 if (!write_to_vm)
764 bio->bi_rw |= (1 << BIO_RW);
765
James Bottomley f1970ba2005-06-20 14:06:52 +0200766 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 bio->bi_flags |= (1 << BIO_USER_MAPPED);
768 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200769
770 out_unmap:
771 for (i = 0; i < nr_pages; i++) {
772 if(!pages[i])
773 break;
774 page_cache_release(pages[i]);
775 }
776 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 kfree(pages);
778 bio_put(bio);
779 return ERR_PTR(ret);
780}
781
782/**
783 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200784 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 * @bdev: destination block device
786 * @uaddr: start of user address
787 * @len: length in bytes
788 * @write_to_vm: bool indicating writing to pages or not
789 *
790 * Map the user space address into a bio suitable for io to a block
791 * device. Returns an error pointer in case of error.
792 */
Jens Axboe165125e2007-07-24 09:28:11 +0200793struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 unsigned long uaddr, unsigned int len, int write_to_vm)
795{
James Bottomley f1970ba2005-06-20 14:06:52 +0200796 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +0100798 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +0200799 iov.iov_len = len;
800
801 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
802}
803
804/**
805 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200806 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +0200807 * @bdev: destination block device
808 * @iov: the iovec.
809 * @iov_count: number of elements in the iovec
810 * @write_to_vm: bool indicating writing to pages or not
811 *
812 * Map the user space address into a bio suitable for io to a block
813 * device. Returns an error pointer in case of error.
814 */
Jens Axboe165125e2007-07-24 09:28:11 +0200815struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +0200816 struct sg_iovec *iov, int iov_count,
817 int write_to_vm)
818{
819 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200820
821 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
823 if (IS_ERR(bio))
824 return bio;
825
826 /*
827 * subtle -- if __bio_map_user() ended up bouncing a bio,
828 * it would normally disappear when its bi_end_io is run.
829 * however, we need it for the unmap, so grab an extra
830 * reference to it
831 */
832 bio_get(bio);
833
Mike Christie0e75f902006-12-01 10:40:55 +0100834 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835}
836
837static void __bio_unmap_user(struct bio *bio)
838{
839 struct bio_vec *bvec;
840 int i;
841
842 /*
843 * make sure we dirty pages we wrote to
844 */
845 __bio_for_each_segment(bvec, bio, i, 0) {
846 if (bio_data_dir(bio) == READ)
847 set_page_dirty_lock(bvec->bv_page);
848
849 page_cache_release(bvec->bv_page);
850 }
851
852 bio_put(bio);
853}
854
855/**
856 * bio_unmap_user - unmap a bio
857 * @bio: the bio being unmapped
858 *
859 * Unmap a bio previously mapped by bio_map_user(). Must be called with
860 * a process context.
861 *
862 * bio_unmap_user() may sleep.
863 */
864void bio_unmap_user(struct bio *bio)
865{
866 __bio_unmap_user(bio);
867 bio_put(bio);
868}
869
NeilBrown6712ecf2007-09-27 12:47:43 +0200870static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +0200871{
Jens Axboeb8238252005-06-20 14:05:27 +0200872 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +0200873}
874
875
Jens Axboe165125e2007-07-24 09:28:11 +0200876static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -0400877 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +0200878{
879 unsigned long kaddr = (unsigned long)data;
880 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
881 unsigned long start = kaddr >> PAGE_SHIFT;
882 const int nr_pages = end - start;
883 int offset, i;
884 struct bio *bio;
885
886 bio = bio_alloc(gfp_mask, nr_pages);
887 if (!bio)
888 return ERR_PTR(-ENOMEM);
889
890 offset = offset_in_page(kaddr);
891 for (i = 0; i < nr_pages; i++) {
892 unsigned int bytes = PAGE_SIZE - offset;
893
894 if (len <= 0)
895 break;
896
897 if (bytes > len)
898 bytes = len;
899
Mike Christiedefd94b2005-12-05 02:37:06 -0600900 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
901 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +0200902 break;
903
904 data += bytes;
905 len -= bytes;
906 offset = 0;
907 }
908
Jens Axboeb8238252005-06-20 14:05:27 +0200909 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +0200910 return bio;
911}
912
913/**
914 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +0200915 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +0200916 * @data: pointer to buffer to map
917 * @len: length in bytes
918 * @gfp_mask: allocation flags for bio allocation
919 *
920 * Map the kernel address into a bio suitable for io to a block
921 * device. Returns an error pointer in case of error.
922 */
Jens Axboe165125e2007-07-24 09:28:11 +0200923struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -0400924 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +0200925{
926 struct bio *bio;
927
928 bio = __bio_map_kern(q, data, len, gfp_mask);
929 if (IS_ERR(bio))
930 return bio;
931
932 if (bio->bi_size == len)
933 return bio;
934
935 /*
936 * Don't support partial mappings.
937 */
938 bio_put(bio);
939 return ERR_PTR(-EINVAL);
940}
941
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200942static void bio_copy_kern_endio(struct bio *bio, int err)
943{
944 struct bio_vec *bvec;
945 const int read = bio_data_dir(bio) == READ;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200946 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200947 int i;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200948 char *p = bmd->sgvecs[0].iov_base;
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200949
950 __bio_for_each_segment(bvec, bio, i, 0) {
951 char *addr = page_address(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200952 int len = bmd->iovecs[i].bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200953
954 if (read && !err)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200955 memcpy(p, addr, len);
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200956
957 __free_page(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200958 p += len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200959 }
960
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200961 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200962 bio_put(bio);
963}
964
965/**
966 * bio_copy_kern - copy kernel address into bio
967 * @q: the struct request_queue for the bio
968 * @data: pointer to buffer to copy
969 * @len: length in bytes
970 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +0200971 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200972 *
973 * copy the kernel address into a bio suitable for io to a block
974 * device. Returns an error pointer in case of error.
975 */
976struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
977 gfp_t gfp_mask, int reading)
978{
979 unsigned long kaddr = (unsigned long)data;
980 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
981 unsigned long start = kaddr >> PAGE_SHIFT;
982 const int nr_pages = end - start;
983 struct bio *bio;
984 struct bio_vec *bvec;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200985 struct bio_map_data *bmd;
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200986 int i, ret;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200987 struct sg_iovec iov;
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200988
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200989 iov.iov_base = data;
990 iov.iov_len = len;
991
992 bmd = bio_alloc_map_data(nr_pages, 1, gfp_mask);
993 if (!bmd)
994 return ERR_PTR(-ENOMEM);
995
996 ret = -ENOMEM;
FUJITA Tomonori68154e92008-04-25 12:47:50 +0200997 bio = bio_alloc(gfp_mask, nr_pages);
998 if (!bio)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200999 goto out_bmd;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001000
1001 while (len) {
1002 struct page *page;
1003 unsigned int bytes = PAGE_SIZE;
1004
1005 if (bytes > len)
1006 bytes = len;
1007
1008 page = alloc_page(q->bounce_gfp | gfp_mask);
1009 if (!page) {
1010 ret = -ENOMEM;
1011 goto cleanup;
1012 }
1013
1014 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) {
1015 ret = -EINVAL;
1016 goto cleanup;
1017 }
1018
1019 len -= bytes;
1020 }
1021
1022 if (!reading) {
1023 void *p = data;
1024
1025 bio_for_each_segment(bvec, bio, i) {
1026 char *addr = page_address(bvec->bv_page);
1027
1028 memcpy(addr, p, bvec->bv_len);
1029 p += bvec->bv_len;
1030 }
1031 }
1032
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001033 bio->bi_private = bmd;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001034 bio->bi_end_io = bio_copy_kern_endio;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001035
1036 bio_set_map_data(bmd, bio, &iov, 1);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001037 return bio;
1038cleanup:
1039 bio_for_each_segment(bvec, bio, i)
1040 __free_page(bvec->bv_page);
1041
1042 bio_put(bio);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001043out_bmd:
1044 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001045
1046 return ERR_PTR(ret);
1047}
1048
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049/*
1050 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1051 * for performing direct-IO in BIOs.
1052 *
1053 * The problem is that we cannot run set_page_dirty() from interrupt context
1054 * because the required locks are not interrupt-safe. So what we can do is to
1055 * mark the pages dirty _before_ performing IO. And in interrupt context,
1056 * check that the pages are still dirty. If so, fine. If not, redirty them
1057 * in process context.
1058 *
1059 * We special-case compound pages here: normally this means reads into hugetlb
1060 * pages. The logic in here doesn't really work right for compound pages
1061 * because the VM does not uniformly chase down the head page in all cases.
1062 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1063 * handle them at all. So we skip compound pages here at an early stage.
1064 *
1065 * Note that this code is very hard to test under normal circumstances because
1066 * direct-io pins the pages with get_user_pages(). This makes
1067 * is_page_cache_freeable return false, and the VM will not clean the pages.
1068 * But other code (eg, pdflush) could clean the pages if they are mapped
1069 * pagecache.
1070 *
1071 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1072 * deferred bio dirtying paths.
1073 */
1074
1075/*
1076 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1077 */
1078void bio_set_pages_dirty(struct bio *bio)
1079{
1080 struct bio_vec *bvec = bio->bi_io_vec;
1081 int i;
1082
1083 for (i = 0; i < bio->bi_vcnt; i++) {
1084 struct page *page = bvec[i].bv_page;
1085
1086 if (page && !PageCompound(page))
1087 set_page_dirty_lock(page);
1088 }
1089}
1090
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001091static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092{
1093 struct bio_vec *bvec = bio->bi_io_vec;
1094 int i;
1095
1096 for (i = 0; i < bio->bi_vcnt; i++) {
1097 struct page *page = bvec[i].bv_page;
1098
1099 if (page)
1100 put_page(page);
1101 }
1102}
1103
1104/*
1105 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1106 * If they are, then fine. If, however, some pages are clean then they must
1107 * have been written out during the direct-IO read. So we take another ref on
1108 * the BIO and the offending pages and re-dirty the pages in process context.
1109 *
1110 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1111 * here on. It will run one page_cache_release() against each page and will
1112 * run one bio_put() against the BIO.
1113 */
1114
David Howells65f27f32006-11-22 14:55:48 +00001115static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
David Howells65f27f32006-11-22 14:55:48 +00001117static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118static DEFINE_SPINLOCK(bio_dirty_lock);
1119static struct bio *bio_dirty_list;
1120
1121/*
1122 * This runs in process context
1123 */
David Howells65f27f32006-11-22 14:55:48 +00001124static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125{
1126 unsigned long flags;
1127 struct bio *bio;
1128
1129 spin_lock_irqsave(&bio_dirty_lock, flags);
1130 bio = bio_dirty_list;
1131 bio_dirty_list = NULL;
1132 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1133
1134 while (bio) {
1135 struct bio *next = bio->bi_private;
1136
1137 bio_set_pages_dirty(bio);
1138 bio_release_pages(bio);
1139 bio_put(bio);
1140 bio = next;
1141 }
1142}
1143
1144void bio_check_pages_dirty(struct bio *bio)
1145{
1146 struct bio_vec *bvec = bio->bi_io_vec;
1147 int nr_clean_pages = 0;
1148 int i;
1149
1150 for (i = 0; i < bio->bi_vcnt; i++) {
1151 struct page *page = bvec[i].bv_page;
1152
1153 if (PageDirty(page) || PageCompound(page)) {
1154 page_cache_release(page);
1155 bvec[i].bv_page = NULL;
1156 } else {
1157 nr_clean_pages++;
1158 }
1159 }
1160
1161 if (nr_clean_pages) {
1162 unsigned long flags;
1163
1164 spin_lock_irqsave(&bio_dirty_lock, flags);
1165 bio->bi_private = bio_dirty_list;
1166 bio_dirty_list = bio;
1167 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1168 schedule_work(&bio_dirty_work);
1169 } else {
1170 bio_put(bio);
1171 }
1172}
1173
1174/**
1175 * bio_endio - end I/O on a bio
1176 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 * @error: error, if any
1178 *
1179 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001180 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001181 * preferred way to end I/O on a bio, it takes care of clearing
1182 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1183 * established -Exxxx (-EIO, for instance) error values in case
1184 * something went wrong. Noone should call bi_end_io() directly on a
1185 * bio unless they own it and thus know that it has an end_io
1186 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001188void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
1190 if (error)
1191 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001192 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1193 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
NeilBrown5bb23a62007-09-27 12:46:13 +02001195 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001196 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197}
1198
1199void bio_pair_release(struct bio_pair *bp)
1200{
1201 if (atomic_dec_and_test(&bp->cnt)) {
1202 struct bio *master = bp->bio1.bi_private;
1203
NeilBrown6712ecf2007-09-27 12:47:43 +02001204 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 mempool_free(bp, bp->bio2.bi_private);
1206 }
1207}
1208
NeilBrown6712ecf2007-09-27 12:47:43 +02001209static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210{
1211 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1212
1213 if (err)
1214 bp->error = err;
1215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217}
1218
NeilBrown6712ecf2007-09-27 12:47:43 +02001219static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220{
1221 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1222
1223 if (err)
1224 bp->error = err;
1225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227}
1228
1229/*
1230 * split a bio - only worry about a bio with a single page
1231 * in it's iovec
1232 */
1233struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1234{
1235 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1236
1237 if (!bp)
1238 return bp;
1239
Jens Axboe2056a782006-03-23 20:00:26 +01001240 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
1241 bi->bi_sector + first_sectors);
1242
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 BUG_ON(bi->bi_vcnt != 1);
1244 BUG_ON(bi->bi_idx != 0);
1245 atomic_set(&bp->cnt, 3);
1246 bp->error = 0;
1247 bp->bio1 = *bi;
1248 bp->bio2 = *bi;
1249 bp->bio2.bi_sector += first_sectors;
1250 bp->bio2.bi_size -= first_sectors << 9;
1251 bp->bio1.bi_size = first_sectors << 9;
1252
1253 bp->bv1 = bi->bi_io_vec[0];
1254 bp->bv2 = bi->bi_io_vec[0];
1255 bp->bv2.bv_offset += first_sectors << 9;
1256 bp->bv2.bv_len -= first_sectors << 9;
1257 bp->bv1.bv_len = first_sectors << 9;
1258
1259 bp->bio1.bi_io_vec = &bp->bv1;
1260 bp->bio2.bi_io_vec = &bp->bv2;
1261
NeilBrowna2eb0c12006-05-22 22:35:27 -07001262 bp->bio1.bi_max_vecs = 1;
1263 bp->bio2.bi_max_vecs = 1;
1264
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 bp->bio1.bi_end_io = bio_pair_end_1;
1266 bp->bio2.bi_end_io = bio_pair_end_2;
1267
1268 bp->bio1.bi_private = bi;
1269 bp->bio2.bi_private = pool;
1270
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001271 if (bio_integrity(bi))
1272 bio_integrity_split(bi, bp, first_sectors);
1273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 return bp;
1275}
1276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
1278/*
1279 * create memory pools for biovec's in a bio_set.
1280 * use the global biovec slabs created for general use.
1281 */
Jens Axboe59725112007-04-02 10:06:42 +02001282static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283{
1284 int i;
1285
1286 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1287 struct biovec_slab *bp = bvec_slabs + i;
1288 mempool_t **bvp = bs->bvec_pools + i;
1289
Matthew Dobson93d23412006-03-26 01:37:50 -08001290 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 if (!*bvp)
1292 return -ENOMEM;
1293 }
1294 return 0;
1295}
1296
1297static void biovec_free_pools(struct bio_set *bs)
1298{
1299 int i;
1300
1301 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1302 mempool_t *bvp = bs->bvec_pools[i];
1303
1304 if (bvp)
1305 mempool_destroy(bvp);
1306 }
1307
1308}
1309
1310void bioset_free(struct bio_set *bs)
1311{
1312 if (bs->bio_pool)
1313 mempool_destroy(bs->bio_pool);
1314
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001315 bioset_integrity_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 biovec_free_pools(bs);
1317
1318 kfree(bs);
1319}
1320
Jens Axboe59725112007-04-02 10:06:42 +02001321struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001323 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 if (!bs)
1326 return NULL;
1327
Matthew Dobson93d23412006-03-26 01:37:50 -08001328 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 if (!bs->bio_pool)
1330 goto bad;
1331
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001332 if (bioset_integrity_create(bs, bio_pool_size))
1333 goto bad;
1334
Jens Axboe59725112007-04-02 10:06:42 +02001335 if (!biovec_create_pools(bs, bvec_pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 return bs;
1337
1338bad:
1339 bioset_free(bs);
1340 return NULL;
1341}
1342
1343static void __init biovec_init_slabs(void)
1344{
1345 int i;
1346
1347 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1348 int size;
1349 struct biovec_slab *bvs = bvec_slabs + i;
1350
1351 size = bvs->nr_vecs * sizeof(struct bio_vec);
1352 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001353 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 }
1355}
1356
1357static int __init init_bio(void)
1358{
Christoph Lameter0a31bd52007-05-06 14:49:57 -07001359 bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001361 bio_integrity_init_slab();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 biovec_init_slabs();
1363
Jens Axboe59725112007-04-02 10:06:42 +02001364 fs_bio_set = bioset_create(BIO_POOL_SIZE, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 if (!fs_bio_set)
1366 panic("bio: can't allocate bios\n");
1367
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001368 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1369 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 if (!bio_split_pool)
1371 panic("bio: can't create split pool\n");
1372
1373 return 0;
1374}
1375
1376subsys_initcall(init_bio);
1377
1378EXPORT_SYMBOL(bio_alloc);
1379EXPORT_SYMBOL(bio_put);
Peter Osterlund36763472005-09-06 15:16:42 -07001380EXPORT_SYMBOL(bio_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381EXPORT_SYMBOL(bio_endio);
1382EXPORT_SYMBOL(bio_init);
1383EXPORT_SYMBOL(__bio_clone);
1384EXPORT_SYMBOL(bio_clone);
1385EXPORT_SYMBOL(bio_phys_segments);
1386EXPORT_SYMBOL(bio_hw_segments);
1387EXPORT_SYMBOL(bio_add_page);
Mike Christie6e68af62005-11-11 05:30:27 -06001388EXPORT_SYMBOL(bio_add_pc_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389EXPORT_SYMBOL(bio_get_nr_vecs);
Jens Axboe40044ce2008-03-17 21:14:40 +01001390EXPORT_SYMBOL(bio_map_user);
1391EXPORT_SYMBOL(bio_unmap_user);
Mike Christie df46b9a2005-06-20 14:04:44 +02001392EXPORT_SYMBOL(bio_map_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001393EXPORT_SYMBOL(bio_copy_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394EXPORT_SYMBOL(bio_pair_release);
1395EXPORT_SYMBOL(bio_split);
1396EXPORT_SYMBOL(bio_split_pool);
1397EXPORT_SYMBOL(bio_copy_user);
1398EXPORT_SYMBOL(bio_uncopy_user);
1399EXPORT_SYMBOL(bioset_create);
1400EXPORT_SYMBOL(bioset_free);
1401EXPORT_SYMBOL(bio_alloc_bioset);