Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
0fe23479 | 2 | * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> |
1da177e4 LT |
3 | * |
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> | |
2056a782 | 28 | #include <linux/blktrace_api.h> |
5f3ea37c | 29 | #include <trace/block.h> |
f1970baf | 30 | #include <scsi/sg.h> /* for struct sg_iovec */ |
1da177e4 | 31 | |
0bfc2455 IM |
32 | DEFINE_TRACE(block_split); |
33 | ||
e18b890b | 34 | static struct kmem_cache *bio_slab __read_mostly; |
1da177e4 | 35 | |
6feef531 | 36 | static mempool_t *bio_split_pool __read_mostly; |
1da177e4 | 37 | |
1da177e4 LT |
38 | /* |
39 | * if you change this list, also change bvec_alloc or things will | |
40 | * break badly! cannot be bigger than what you can fit into an | |
41 | * unsigned short | |
42 | */ | |
43 | ||
44 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } | |
6c036527 | 45 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { |
1da177e4 LT |
46 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
47 | }; | |
48 | #undef BV | |
49 | ||
1da177e4 LT |
50 | /* |
51 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
52 | * IO code that does not need private memory pools. | |
53 | */ | |
51d654e1 | 54 | struct bio_set *fs_bio_set; |
1da177e4 | 55 | |
7ba1ba12 MP |
56 | unsigned int bvec_nr_vecs(unsigned short idx) |
57 | { | |
58 | return bvec_slabs[idx].nr_vecs; | |
59 | } | |
60 | ||
51d654e1 | 61 | struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs) |
1da177e4 LT |
62 | { |
63 | struct bio_vec *bvl; | |
1da177e4 LT |
64 | |
65 | /* | |
0a0d96b0 JA |
66 | * If 'bs' is given, lookup the pool and do the mempool alloc. |
67 | * If not, this is a bio_kmalloc() allocation and just do a | |
68 | * kzalloc() for the exact number of vecs right away. | |
1da177e4 | 69 | */ |
0a0d96b0 JA |
70 | if (bs) { |
71 | /* | |
72 | * see comment near bvec_array define! | |
73 | */ | |
74 | switch (nr) { | |
75 | case 1: | |
76 | *idx = 0; | |
77 | break; | |
78 | case 2 ... 4: | |
79 | *idx = 1; | |
80 | break; | |
81 | case 5 ... 16: | |
82 | *idx = 2; | |
83 | break; | |
84 | case 17 ... 64: | |
85 | *idx = 3; | |
86 | break; | |
87 | case 65 ... 128: | |
88 | *idx = 4; | |
89 | break; | |
90 | case 129 ... BIO_MAX_PAGES: | |
91 | *idx = 5; | |
92 | break; | |
1da177e4 LT |
93 | default: |
94 | return NULL; | |
0a0d96b0 | 95 | } |
1da177e4 | 96 | |
0a0d96b0 JA |
97 | /* |
98 | * idx now points to the pool we want to allocate from | |
99 | */ | |
100 | bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask); | |
101 | if (bvl) | |
102 | memset(bvl, 0, | |
103 | bvec_nr_vecs(*idx) * sizeof(struct bio_vec)); | |
104 | } else | |
105 | bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask); | |
1da177e4 LT |
106 | |
107 | return bvl; | |
108 | } | |
109 | ||
3676347a | 110 | void bio_free(struct bio *bio, struct bio_set *bio_set) |
1da177e4 | 111 | { |
992c5dda JA |
112 | if (bio->bi_io_vec) { |
113 | const int pool_idx = BIO_POOL_IDX(bio); | |
1da177e4 | 114 | |
992c5dda JA |
115 | BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); |
116 | ||
117 | mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]); | |
118 | } | |
1da177e4 | 119 | |
7ba1ba12 MP |
120 | if (bio_integrity(bio)) |
121 | bio_integrity_free(bio, bio_set); | |
122 | ||
3676347a PO |
123 | mempool_free(bio, bio_set->bio_pool); |
124 | } | |
125 | ||
126 | /* | |
127 | * default destructor for a bio allocated with bio_alloc_bioset() | |
128 | */ | |
129 | static void bio_fs_destructor(struct bio *bio) | |
130 | { | |
131 | bio_free(bio, fs_bio_set); | |
1da177e4 LT |
132 | } |
133 | ||
0a0d96b0 JA |
134 | static void bio_kmalloc_destructor(struct bio *bio) |
135 | { | |
136 | kfree(bio->bi_io_vec); | |
137 | kfree(bio); | |
138 | } | |
139 | ||
858119e1 | 140 | void bio_init(struct bio *bio) |
1da177e4 | 141 | { |
2b94de55 | 142 | memset(bio, 0, sizeof(*bio)); |
1da177e4 | 143 | bio->bi_flags = 1 << BIO_UPTODATE; |
c7c22e4d | 144 | bio->bi_comp_cpu = -1; |
1da177e4 | 145 | atomic_set(&bio->bi_cnt, 1); |
1da177e4 LT |
146 | } |
147 | ||
148 | /** | |
149 | * bio_alloc_bioset - allocate a bio for I/O | |
150 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
151 | * @nr_iovecs: number of iovecs to pre-allocate | |
0a0d96b0 | 152 | * @bs: the bio_set to allocate from. If %NULL, just use kmalloc |
1da177e4 LT |
153 | * |
154 | * Description: | |
0a0d96b0 | 155 | * bio_alloc_bioset will first try its own mempool to satisfy the allocation. |
1da177e4 | 156 | * If %__GFP_WAIT is set then we will block on the internal pool waiting |
0a0d96b0 JA |
157 | * for a &struct bio to become free. If a %NULL @bs is passed in, we will |
158 | * fall back to just using @kmalloc to allocate the required memory. | |
1da177e4 LT |
159 | * |
160 | * allocate bio and iovecs from the memory pools specified by the | |
0a0d96b0 | 161 | * bio_set structure, or @kmalloc if none given. |
1da177e4 | 162 | **/ |
dd0fc66f | 163 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) |
1da177e4 | 164 | { |
0a0d96b0 JA |
165 | struct bio *bio; |
166 | ||
167 | if (bs) | |
168 | bio = mempool_alloc(bs->bio_pool, gfp_mask); | |
169 | else | |
170 | bio = kmalloc(sizeof(*bio), gfp_mask); | |
1da177e4 LT |
171 | |
172 | if (likely(bio)) { | |
173 | struct bio_vec *bvl = NULL; | |
174 | ||
175 | bio_init(bio); | |
176 | if (likely(nr_iovecs)) { | |
eeae1d48 | 177 | unsigned long uninitialized_var(idx); |
1da177e4 LT |
178 | |
179 | bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); | |
180 | if (unlikely(!bvl)) { | |
0a0d96b0 JA |
181 | if (bs) |
182 | mempool_free(bio, bs->bio_pool); | |
183 | else | |
184 | kfree(bio); | |
1da177e4 LT |
185 | bio = NULL; |
186 | goto out; | |
187 | } | |
188 | bio->bi_flags |= idx << BIO_POOL_OFFSET; | |
1ac0ae06 | 189 | bio->bi_max_vecs = bvec_nr_vecs(idx); |
1da177e4 LT |
190 | } |
191 | bio->bi_io_vec = bvl; | |
1da177e4 LT |
192 | } |
193 | out: | |
194 | return bio; | |
195 | } | |
196 | ||
dd0fc66f | 197 | struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs) |
1da177e4 | 198 | { |
3676347a PO |
199 | struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set); |
200 | ||
201 | if (bio) | |
202 | bio->bi_destructor = bio_fs_destructor; | |
203 | ||
204 | return bio; | |
1da177e4 LT |
205 | } |
206 | ||
0a0d96b0 JA |
207 | /* |
208 | * Like bio_alloc(), but doesn't use a mempool backing. This means that | |
209 | * it CAN fail, but while bio_alloc() can only be used for allocations | |
210 | * that have a short (finite) life span, bio_kmalloc() should be used | |
211 | * for more permanent bio allocations (like allocating some bio's for | |
212 | * initalization or setup purposes). | |
213 | */ | |
214 | struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs) | |
215 | { | |
216 | struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, NULL); | |
217 | ||
218 | if (bio) | |
219 | bio->bi_destructor = bio_kmalloc_destructor; | |
220 | ||
221 | return bio; | |
222 | } | |
223 | ||
1da177e4 LT |
224 | void zero_fill_bio(struct bio *bio) |
225 | { | |
226 | unsigned long flags; | |
227 | struct bio_vec *bv; | |
228 | int i; | |
229 | ||
230 | bio_for_each_segment(bv, bio, i) { | |
231 | char *data = bvec_kmap_irq(bv, &flags); | |
232 | memset(data, 0, bv->bv_len); | |
233 | flush_dcache_page(bv->bv_page); | |
234 | bvec_kunmap_irq(data, &flags); | |
235 | } | |
236 | } | |
237 | EXPORT_SYMBOL(zero_fill_bio); | |
238 | ||
239 | /** | |
240 | * bio_put - release a reference to a bio | |
241 | * @bio: bio to release reference to | |
242 | * | |
243 | * Description: | |
244 | * Put a reference to a &struct bio, either one you have gotten with | |
245 | * bio_alloc or bio_get. The last put of a bio will free it. | |
246 | **/ | |
247 | void bio_put(struct bio *bio) | |
248 | { | |
249 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); | |
250 | ||
251 | /* | |
252 | * last put frees it | |
253 | */ | |
254 | if (atomic_dec_and_test(&bio->bi_cnt)) { | |
255 | bio->bi_next = NULL; | |
256 | bio->bi_destructor(bio); | |
257 | } | |
258 | } | |
259 | ||
165125e1 | 260 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
261 | { |
262 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
263 | blk_recount_segments(q, bio); | |
264 | ||
265 | return bio->bi_phys_segments; | |
266 | } | |
267 | ||
1da177e4 LT |
268 | /** |
269 | * __bio_clone - clone a bio | |
270 | * @bio: destination bio | |
271 | * @bio_src: bio to clone | |
272 | * | |
273 | * Clone a &bio. Caller will own the returned bio, but not | |
274 | * the actual data it points to. Reference count of returned | |
275 | * bio will be one. | |
276 | */ | |
858119e1 | 277 | void __bio_clone(struct bio *bio, struct bio *bio_src) |
1da177e4 | 278 | { |
e525e153 AM |
279 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, |
280 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); | |
1da177e4 | 281 | |
5d84070e JA |
282 | /* |
283 | * most users will be overriding ->bi_bdev with a new target, | |
284 | * so we don't set nor calculate new physical/hw segment counts here | |
285 | */ | |
1da177e4 LT |
286 | bio->bi_sector = bio_src->bi_sector; |
287 | bio->bi_bdev = bio_src->bi_bdev; | |
288 | bio->bi_flags |= 1 << BIO_CLONED; | |
289 | bio->bi_rw = bio_src->bi_rw; | |
1da177e4 LT |
290 | bio->bi_vcnt = bio_src->bi_vcnt; |
291 | bio->bi_size = bio_src->bi_size; | |
a5453be4 | 292 | bio->bi_idx = bio_src->bi_idx; |
1da177e4 LT |
293 | } |
294 | ||
295 | /** | |
296 | * bio_clone - clone a bio | |
297 | * @bio: bio to clone | |
298 | * @gfp_mask: allocation priority | |
299 | * | |
300 | * Like __bio_clone, only also allocates the returned bio | |
301 | */ | |
dd0fc66f | 302 | struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
303 | { |
304 | struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); | |
305 | ||
7ba1ba12 MP |
306 | if (!b) |
307 | return NULL; | |
308 | ||
309 | b->bi_destructor = bio_fs_destructor; | |
310 | __bio_clone(b, bio); | |
311 | ||
312 | if (bio_integrity(bio)) { | |
313 | int ret; | |
314 | ||
315 | ret = bio_integrity_clone(b, bio, fs_bio_set); | |
316 | ||
317 | if (ret < 0) | |
318 | return NULL; | |
3676347a | 319 | } |
1da177e4 LT |
320 | |
321 | return b; | |
322 | } | |
323 | ||
324 | /** | |
325 | * bio_get_nr_vecs - return approx number of vecs | |
326 | * @bdev: I/O target | |
327 | * | |
328 | * Return the approximate number of pages we can send to this target. | |
329 | * There's no guarantee that you will be able to fit this number of pages | |
330 | * into a bio, it does not account for dynamic restrictions that vary | |
331 | * on offset. | |
332 | */ | |
333 | int bio_get_nr_vecs(struct block_device *bdev) | |
334 | { | |
165125e1 | 335 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
336 | int nr_pages; |
337 | ||
338 | nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
339 | if (nr_pages > q->max_phys_segments) | |
340 | nr_pages = q->max_phys_segments; | |
341 | if (nr_pages > q->max_hw_segments) | |
342 | nr_pages = q->max_hw_segments; | |
343 | ||
344 | return nr_pages; | |
345 | } | |
346 | ||
165125e1 | 347 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page |
defd94b7 MC |
348 | *page, unsigned int len, unsigned int offset, |
349 | unsigned short max_sectors) | |
1da177e4 LT |
350 | { |
351 | int retried_segments = 0; | |
352 | struct bio_vec *bvec; | |
353 | ||
354 | /* | |
355 | * cloned bio must not modify vec list | |
356 | */ | |
357 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
358 | return 0; | |
359 | ||
80cfd548 | 360 | if (((bio->bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
361 | return 0; |
362 | ||
80cfd548 JA |
363 | /* |
364 | * For filesystems with a blocksize smaller than the pagesize | |
365 | * we will often be called with the same page as last time and | |
366 | * a consecutive offset. Optimize this special case. | |
367 | */ | |
368 | if (bio->bi_vcnt > 0) { | |
369 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
370 | ||
371 | if (page == prev->bv_page && | |
372 | offset == prev->bv_offset + prev->bv_len) { | |
373 | prev->bv_len += len; | |
cc371e66 AK |
374 | |
375 | if (q->merge_bvec_fn) { | |
376 | struct bvec_merge_data bvm = { | |
377 | .bi_bdev = bio->bi_bdev, | |
378 | .bi_sector = bio->bi_sector, | |
379 | .bi_size = bio->bi_size, | |
380 | .bi_rw = bio->bi_rw, | |
381 | }; | |
382 | ||
383 | if (q->merge_bvec_fn(q, &bvm, prev) < len) { | |
384 | prev->bv_len -= len; | |
385 | return 0; | |
386 | } | |
80cfd548 JA |
387 | } |
388 | ||
389 | goto done; | |
390 | } | |
391 | } | |
392 | ||
393 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
1da177e4 LT |
394 | return 0; |
395 | ||
396 | /* | |
397 | * we might lose a segment or two here, but rather that than | |
398 | * make this too complex. | |
399 | */ | |
400 | ||
401 | while (bio->bi_phys_segments >= q->max_phys_segments | |
5df97b91 | 402 | || bio->bi_phys_segments >= q->max_hw_segments) { |
1da177e4 LT |
403 | |
404 | if (retried_segments) | |
405 | return 0; | |
406 | ||
407 | retried_segments = 1; | |
408 | blk_recount_segments(q, bio); | |
409 | } | |
410 | ||
411 | /* | |
412 | * setup the new entry, we might clear it again later if we | |
413 | * cannot add the page | |
414 | */ | |
415 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
416 | bvec->bv_page = page; | |
417 | bvec->bv_len = len; | |
418 | bvec->bv_offset = offset; | |
419 | ||
420 | /* | |
421 | * if queue has other restrictions (eg varying max sector size | |
422 | * depending on offset), it can specify a merge_bvec_fn in the | |
423 | * queue to get further control | |
424 | */ | |
425 | if (q->merge_bvec_fn) { | |
cc371e66 AK |
426 | struct bvec_merge_data bvm = { |
427 | .bi_bdev = bio->bi_bdev, | |
428 | .bi_sector = bio->bi_sector, | |
429 | .bi_size = bio->bi_size, | |
430 | .bi_rw = bio->bi_rw, | |
431 | }; | |
432 | ||
1da177e4 LT |
433 | /* |
434 | * merge_bvec_fn() returns number of bytes it can accept | |
435 | * at this offset | |
436 | */ | |
cc371e66 | 437 | if (q->merge_bvec_fn(q, &bvm, bvec) < len) { |
1da177e4 LT |
438 | bvec->bv_page = NULL; |
439 | bvec->bv_len = 0; | |
440 | bvec->bv_offset = 0; | |
441 | return 0; | |
442 | } | |
443 | } | |
444 | ||
445 | /* If we may be able to merge these biovecs, force a recount */ | |
b8b3e16c | 446 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) |
1da177e4 LT |
447 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
448 | ||
449 | bio->bi_vcnt++; | |
450 | bio->bi_phys_segments++; | |
80cfd548 | 451 | done: |
1da177e4 LT |
452 | bio->bi_size += len; |
453 | return len; | |
454 | } | |
455 | ||
6e68af66 MC |
456 | /** |
457 | * bio_add_pc_page - attempt to add page to bio | |
fddfdeaf | 458 | * @q: the target queue |
6e68af66 MC |
459 | * @bio: destination bio |
460 | * @page: page to add | |
461 | * @len: vec entry length | |
462 | * @offset: vec entry offset | |
463 | * | |
464 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
465 | * number of reasons, such as the bio being full or target block | |
466 | * device limitations. The target block device must allow bio's | |
467 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
468 | * page to an empty bio. This should only be used by REQ_PC bios. | |
469 | */ | |
165125e1 | 470 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
6e68af66 MC |
471 | unsigned int len, unsigned int offset) |
472 | { | |
defd94b7 | 473 | return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors); |
6e68af66 MC |
474 | } |
475 | ||
1da177e4 LT |
476 | /** |
477 | * bio_add_page - attempt to add page to bio | |
478 | * @bio: destination bio | |
479 | * @page: page to add | |
480 | * @len: vec entry length | |
481 | * @offset: vec entry offset | |
482 | * | |
483 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
484 | * number of reasons, such as the bio being full or target block | |
485 | * device limitations. The target block device must allow bio's | |
486 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
487 | * page to an empty bio. | |
488 | */ | |
489 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
490 | unsigned int offset) | |
491 | { | |
defd94b7 MC |
492 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
493 | return __bio_add_page(q, bio, page, len, offset, q->max_sectors); | |
1da177e4 LT |
494 | } |
495 | ||
496 | struct bio_map_data { | |
497 | struct bio_vec *iovecs; | |
c5dec1c3 | 498 | struct sg_iovec *sgvecs; |
152e283f FT |
499 | int nr_sgvecs; |
500 | int is_our_pages; | |
1da177e4 LT |
501 | }; |
502 | ||
c5dec1c3 | 503 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
152e283f FT |
504 | struct sg_iovec *iov, int iov_count, |
505 | int is_our_pages) | |
1da177e4 LT |
506 | { |
507 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); | |
c5dec1c3 FT |
508 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
509 | bmd->nr_sgvecs = iov_count; | |
152e283f | 510 | bmd->is_our_pages = is_our_pages; |
1da177e4 LT |
511 | bio->bi_private = bmd; |
512 | } | |
513 | ||
514 | static void bio_free_map_data(struct bio_map_data *bmd) | |
515 | { | |
516 | kfree(bmd->iovecs); | |
c5dec1c3 | 517 | kfree(bmd->sgvecs); |
1da177e4 LT |
518 | kfree(bmd); |
519 | } | |
520 | ||
76029ff3 FT |
521 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count, |
522 | gfp_t gfp_mask) | |
1da177e4 | 523 | { |
76029ff3 | 524 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask); |
1da177e4 LT |
525 | |
526 | if (!bmd) | |
527 | return NULL; | |
528 | ||
76029ff3 | 529 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask); |
c5dec1c3 FT |
530 | if (!bmd->iovecs) { |
531 | kfree(bmd); | |
532 | return NULL; | |
533 | } | |
534 | ||
76029ff3 | 535 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask); |
c5dec1c3 | 536 | if (bmd->sgvecs) |
1da177e4 LT |
537 | return bmd; |
538 | ||
c5dec1c3 | 539 | kfree(bmd->iovecs); |
1da177e4 LT |
540 | kfree(bmd); |
541 | return NULL; | |
542 | } | |
543 | ||
aefcc28a | 544 | static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs, |
152e283f FT |
545 | struct sg_iovec *iov, int iov_count, int uncopy, |
546 | int do_free_page) | |
c5dec1c3 FT |
547 | { |
548 | int ret = 0, i; | |
549 | struct bio_vec *bvec; | |
550 | int iov_idx = 0; | |
551 | unsigned int iov_off = 0; | |
552 | int read = bio_data_dir(bio) == READ; | |
553 | ||
554 | __bio_for_each_segment(bvec, bio, i, 0) { | |
555 | char *bv_addr = page_address(bvec->bv_page); | |
aefcc28a | 556 | unsigned int bv_len = iovecs[i].bv_len; |
c5dec1c3 FT |
557 | |
558 | while (bv_len && iov_idx < iov_count) { | |
559 | unsigned int bytes; | |
560 | char *iov_addr; | |
561 | ||
562 | bytes = min_t(unsigned int, | |
563 | iov[iov_idx].iov_len - iov_off, bv_len); | |
564 | iov_addr = iov[iov_idx].iov_base + iov_off; | |
565 | ||
566 | if (!ret) { | |
567 | if (!read && !uncopy) | |
568 | ret = copy_from_user(bv_addr, iov_addr, | |
569 | bytes); | |
570 | if (read && uncopy) | |
571 | ret = copy_to_user(iov_addr, bv_addr, | |
572 | bytes); | |
573 | ||
574 | if (ret) | |
575 | ret = -EFAULT; | |
576 | } | |
577 | ||
578 | bv_len -= bytes; | |
579 | bv_addr += bytes; | |
580 | iov_addr += bytes; | |
581 | iov_off += bytes; | |
582 | ||
583 | if (iov[iov_idx].iov_len == iov_off) { | |
584 | iov_idx++; | |
585 | iov_off = 0; | |
586 | } | |
587 | } | |
588 | ||
152e283f | 589 | if (do_free_page) |
c5dec1c3 FT |
590 | __free_page(bvec->bv_page); |
591 | } | |
592 | ||
593 | return ret; | |
594 | } | |
595 | ||
1da177e4 LT |
596 | /** |
597 | * bio_uncopy_user - finish previously mapped bio | |
598 | * @bio: bio being terminated | |
599 | * | |
600 | * Free pages allocated from bio_copy_user() and write back data | |
601 | * to user space in case of a read. | |
602 | */ | |
603 | int bio_uncopy_user(struct bio *bio) | |
604 | { | |
605 | struct bio_map_data *bmd = bio->bi_private; | |
81882766 | 606 | int ret = 0; |
1da177e4 | 607 | |
81882766 FT |
608 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) |
609 | ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, | |
610 | bmd->nr_sgvecs, 1, bmd->is_our_pages); | |
1da177e4 LT |
611 | bio_free_map_data(bmd); |
612 | bio_put(bio); | |
613 | return ret; | |
614 | } | |
615 | ||
616 | /** | |
c5dec1c3 | 617 | * bio_copy_user_iov - copy user data to bio |
1da177e4 | 618 | * @q: destination block queue |
152e283f | 619 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
c5dec1c3 FT |
620 | * @iov: the iovec. |
621 | * @iov_count: number of elements in the iovec | |
1da177e4 | 622 | * @write_to_vm: bool indicating writing to pages or not |
a3bce90e | 623 | * @gfp_mask: memory allocation flags |
1da177e4 LT |
624 | * |
625 | * Prepares and returns a bio for indirect user io, bouncing data | |
626 | * to/from kernel pages as necessary. Must be paired with | |
627 | * call bio_uncopy_user() on io completion. | |
628 | */ | |
152e283f FT |
629 | struct bio *bio_copy_user_iov(struct request_queue *q, |
630 | struct rq_map_data *map_data, | |
631 | struct sg_iovec *iov, int iov_count, | |
632 | int write_to_vm, gfp_t gfp_mask) | |
1da177e4 | 633 | { |
1da177e4 LT |
634 | struct bio_map_data *bmd; |
635 | struct bio_vec *bvec; | |
636 | struct page *page; | |
637 | struct bio *bio; | |
638 | int i, ret; | |
c5dec1c3 FT |
639 | int nr_pages = 0; |
640 | unsigned int len = 0; | |
1da177e4 | 641 | |
c5dec1c3 FT |
642 | for (i = 0; i < iov_count; i++) { |
643 | unsigned long uaddr; | |
644 | unsigned long end; | |
645 | unsigned long start; | |
646 | ||
647 | uaddr = (unsigned long)iov[i].iov_base; | |
648 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
649 | start = uaddr >> PAGE_SHIFT; | |
650 | ||
651 | nr_pages += end - start; | |
652 | len += iov[i].iov_len; | |
653 | } | |
654 | ||
a3bce90e | 655 | bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask); |
1da177e4 LT |
656 | if (!bmd) |
657 | return ERR_PTR(-ENOMEM); | |
658 | ||
1da177e4 | 659 | ret = -ENOMEM; |
a3bce90e | 660 | bio = bio_alloc(gfp_mask, nr_pages); |
1da177e4 LT |
661 | if (!bio) |
662 | goto out_bmd; | |
663 | ||
664 | bio->bi_rw |= (!write_to_vm << BIO_RW); | |
665 | ||
666 | ret = 0; | |
152e283f | 667 | i = 0; |
1da177e4 | 668 | while (len) { |
152e283f FT |
669 | unsigned int bytes; |
670 | ||
671 | if (map_data) | |
672 | bytes = 1U << (PAGE_SHIFT + map_data->page_order); | |
673 | else | |
674 | bytes = PAGE_SIZE; | |
1da177e4 LT |
675 | |
676 | if (bytes > len) | |
677 | bytes = len; | |
678 | ||
152e283f FT |
679 | if (map_data) { |
680 | if (i == map_data->nr_entries) { | |
681 | ret = -ENOMEM; | |
682 | break; | |
683 | } | |
684 | page = map_data->pages[i++]; | |
685 | } else | |
686 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1da177e4 LT |
687 | if (!page) { |
688 | ret = -ENOMEM; | |
689 | break; | |
690 | } | |
691 | ||
0e75f906 | 692 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) |
1da177e4 | 693 | break; |
1da177e4 LT |
694 | |
695 | len -= bytes; | |
696 | } | |
697 | ||
698 | if (ret) | |
699 | goto cleanup; | |
700 | ||
701 | /* | |
702 | * success | |
703 | */ | |
704 | if (!write_to_vm) { | |
152e283f | 705 | ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0); |
c5dec1c3 FT |
706 | if (ret) |
707 | goto cleanup; | |
1da177e4 LT |
708 | } |
709 | ||
152e283f | 710 | bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1); |
1da177e4 LT |
711 | return bio; |
712 | cleanup: | |
152e283f FT |
713 | if (!map_data) |
714 | bio_for_each_segment(bvec, bio, i) | |
715 | __free_page(bvec->bv_page); | |
1da177e4 LT |
716 | |
717 | bio_put(bio); | |
718 | out_bmd: | |
719 | bio_free_map_data(bmd); | |
720 | return ERR_PTR(ret); | |
721 | } | |
722 | ||
c5dec1c3 FT |
723 | /** |
724 | * bio_copy_user - copy user data to bio | |
725 | * @q: destination block queue | |
152e283f | 726 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
c5dec1c3 FT |
727 | * @uaddr: start of user address |
728 | * @len: length in bytes | |
729 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 730 | * @gfp_mask: memory allocation flags |
c5dec1c3 FT |
731 | * |
732 | * Prepares and returns a bio for indirect user io, bouncing data | |
733 | * to/from kernel pages as necessary. Must be paired with | |
734 | * call bio_uncopy_user() on io completion. | |
735 | */ | |
152e283f FT |
736 | struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data, |
737 | unsigned long uaddr, unsigned int len, | |
738 | int write_to_vm, gfp_t gfp_mask) | |
c5dec1c3 FT |
739 | { |
740 | struct sg_iovec iov; | |
741 | ||
742 | iov.iov_base = (void __user *)uaddr; | |
743 | iov.iov_len = len; | |
744 | ||
152e283f | 745 | return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask); |
c5dec1c3 FT |
746 | } |
747 | ||
165125e1 | 748 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
f1970baf JB |
749 | struct block_device *bdev, |
750 | struct sg_iovec *iov, int iov_count, | |
a3bce90e | 751 | int write_to_vm, gfp_t gfp_mask) |
1da177e4 | 752 | { |
f1970baf JB |
753 | int i, j; |
754 | int nr_pages = 0; | |
1da177e4 LT |
755 | struct page **pages; |
756 | struct bio *bio; | |
f1970baf JB |
757 | int cur_page = 0; |
758 | int ret, offset; | |
1da177e4 | 759 | |
f1970baf JB |
760 | for (i = 0; i < iov_count; i++) { |
761 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
762 | unsigned long len = iov[i].iov_len; | |
763 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
764 | unsigned long start = uaddr >> PAGE_SHIFT; | |
765 | ||
766 | nr_pages += end - start; | |
767 | /* | |
ad2d7225 | 768 | * buffer must be aligned to at least hardsector size for now |
f1970baf | 769 | */ |
ad2d7225 | 770 | if (uaddr & queue_dma_alignment(q)) |
f1970baf JB |
771 | return ERR_PTR(-EINVAL); |
772 | } | |
773 | ||
774 | if (!nr_pages) | |
1da177e4 LT |
775 | return ERR_PTR(-EINVAL); |
776 | ||
a3bce90e | 777 | bio = bio_alloc(gfp_mask, nr_pages); |
1da177e4 LT |
778 | if (!bio) |
779 | return ERR_PTR(-ENOMEM); | |
780 | ||
781 | ret = -ENOMEM; | |
a3bce90e | 782 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); |
1da177e4 LT |
783 | if (!pages) |
784 | goto out; | |
785 | ||
f1970baf JB |
786 | for (i = 0; i < iov_count; i++) { |
787 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
788 | unsigned long len = iov[i].iov_len; | |
789 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
790 | unsigned long start = uaddr >> PAGE_SHIFT; | |
791 | const int local_nr_pages = end - start; | |
792 | const int page_limit = cur_page + local_nr_pages; | |
793 | ||
f5dd33c4 NP |
794 | ret = get_user_pages_fast(uaddr, local_nr_pages, |
795 | write_to_vm, &pages[cur_page]); | |
99172157 JA |
796 | if (ret < local_nr_pages) { |
797 | ret = -EFAULT; | |
f1970baf | 798 | goto out_unmap; |
99172157 | 799 | } |
f1970baf JB |
800 | |
801 | offset = uaddr & ~PAGE_MASK; | |
802 | for (j = cur_page; j < page_limit; j++) { | |
803 | unsigned int bytes = PAGE_SIZE - offset; | |
804 | ||
805 | if (len <= 0) | |
806 | break; | |
807 | ||
808 | if (bytes > len) | |
809 | bytes = len; | |
810 | ||
811 | /* | |
812 | * sorry... | |
813 | */ | |
defd94b7 MC |
814 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
815 | bytes) | |
f1970baf JB |
816 | break; |
817 | ||
818 | len -= bytes; | |
819 | offset = 0; | |
820 | } | |
1da177e4 | 821 | |
f1970baf | 822 | cur_page = j; |
1da177e4 | 823 | /* |
f1970baf | 824 | * release the pages we didn't map into the bio, if any |
1da177e4 | 825 | */ |
f1970baf JB |
826 | while (j < page_limit) |
827 | page_cache_release(pages[j++]); | |
1da177e4 LT |
828 | } |
829 | ||
1da177e4 LT |
830 | kfree(pages); |
831 | ||
832 | /* | |
833 | * set data direction, and check if mapped pages need bouncing | |
834 | */ | |
835 | if (!write_to_vm) | |
836 | bio->bi_rw |= (1 << BIO_RW); | |
837 | ||
f1970baf | 838 | bio->bi_bdev = bdev; |
1da177e4 LT |
839 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
840 | return bio; | |
f1970baf JB |
841 | |
842 | out_unmap: | |
843 | for (i = 0; i < nr_pages; i++) { | |
844 | if(!pages[i]) | |
845 | break; | |
846 | page_cache_release(pages[i]); | |
847 | } | |
848 | out: | |
1da177e4 LT |
849 | kfree(pages); |
850 | bio_put(bio); | |
851 | return ERR_PTR(ret); | |
852 | } | |
853 | ||
854 | /** | |
855 | * bio_map_user - map user address into bio | |
165125e1 | 856 | * @q: the struct request_queue for the bio |
1da177e4 LT |
857 | * @bdev: destination block device |
858 | * @uaddr: start of user address | |
859 | * @len: length in bytes | |
860 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 861 | * @gfp_mask: memory allocation flags |
1da177e4 LT |
862 | * |
863 | * Map the user space address into a bio suitable for io to a block | |
864 | * device. Returns an error pointer in case of error. | |
865 | */ | |
165125e1 | 866 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
a3bce90e FT |
867 | unsigned long uaddr, unsigned int len, int write_to_vm, |
868 | gfp_t gfp_mask) | |
f1970baf JB |
869 | { |
870 | struct sg_iovec iov; | |
871 | ||
3f70353e | 872 | iov.iov_base = (void __user *)uaddr; |
f1970baf JB |
873 | iov.iov_len = len; |
874 | ||
a3bce90e | 875 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask); |
f1970baf JB |
876 | } |
877 | ||
878 | /** | |
879 | * bio_map_user_iov - map user sg_iovec table into bio | |
165125e1 | 880 | * @q: the struct request_queue for the bio |
f1970baf JB |
881 | * @bdev: destination block device |
882 | * @iov: the iovec. | |
883 | * @iov_count: number of elements in the iovec | |
884 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 885 | * @gfp_mask: memory allocation flags |
f1970baf JB |
886 | * |
887 | * Map the user space address into a bio suitable for io to a block | |
888 | * device. Returns an error pointer in case of error. | |
889 | */ | |
165125e1 | 890 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
f1970baf | 891 | struct sg_iovec *iov, int iov_count, |
a3bce90e | 892 | int write_to_vm, gfp_t gfp_mask) |
1da177e4 LT |
893 | { |
894 | struct bio *bio; | |
895 | ||
a3bce90e FT |
896 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm, |
897 | gfp_mask); | |
1da177e4 LT |
898 | if (IS_ERR(bio)) |
899 | return bio; | |
900 | ||
901 | /* | |
902 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
903 | * it would normally disappear when its bi_end_io is run. | |
904 | * however, we need it for the unmap, so grab an extra | |
905 | * reference to it | |
906 | */ | |
907 | bio_get(bio); | |
908 | ||
0e75f906 | 909 | return bio; |
1da177e4 LT |
910 | } |
911 | ||
912 | static void __bio_unmap_user(struct bio *bio) | |
913 | { | |
914 | struct bio_vec *bvec; | |
915 | int i; | |
916 | ||
917 | /* | |
918 | * make sure we dirty pages we wrote to | |
919 | */ | |
920 | __bio_for_each_segment(bvec, bio, i, 0) { | |
921 | if (bio_data_dir(bio) == READ) | |
922 | set_page_dirty_lock(bvec->bv_page); | |
923 | ||
924 | page_cache_release(bvec->bv_page); | |
925 | } | |
926 | ||
927 | bio_put(bio); | |
928 | } | |
929 | ||
930 | /** | |
931 | * bio_unmap_user - unmap a bio | |
932 | * @bio: the bio being unmapped | |
933 | * | |
934 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
935 | * a process context. | |
936 | * | |
937 | * bio_unmap_user() may sleep. | |
938 | */ | |
939 | void bio_unmap_user(struct bio *bio) | |
940 | { | |
941 | __bio_unmap_user(bio); | |
942 | bio_put(bio); | |
943 | } | |
944 | ||
6712ecf8 | 945 | static void bio_map_kern_endio(struct bio *bio, int err) |
b823825e | 946 | { |
b823825e | 947 | bio_put(bio); |
b823825e JA |
948 | } |
949 | ||
950 | ||
165125e1 | 951 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
27496a8c | 952 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 MC |
953 | { |
954 | unsigned long kaddr = (unsigned long)data; | |
955 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
956 | unsigned long start = kaddr >> PAGE_SHIFT; | |
957 | const int nr_pages = end - start; | |
958 | int offset, i; | |
959 | struct bio *bio; | |
960 | ||
961 | bio = bio_alloc(gfp_mask, nr_pages); | |
962 | if (!bio) | |
963 | return ERR_PTR(-ENOMEM); | |
964 | ||
965 | offset = offset_in_page(kaddr); | |
966 | for (i = 0; i < nr_pages; i++) { | |
967 | unsigned int bytes = PAGE_SIZE - offset; | |
968 | ||
969 | if (len <= 0) | |
970 | break; | |
971 | ||
972 | if (bytes > len) | |
973 | bytes = len; | |
974 | ||
defd94b7 MC |
975 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
976 | offset) < bytes) | |
df46b9a4 MC |
977 | break; |
978 | ||
979 | data += bytes; | |
980 | len -= bytes; | |
981 | offset = 0; | |
982 | } | |
983 | ||
b823825e | 984 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
985 | return bio; |
986 | } | |
987 | ||
988 | /** | |
989 | * bio_map_kern - map kernel address into bio | |
165125e1 | 990 | * @q: the struct request_queue for the bio |
df46b9a4 MC |
991 | * @data: pointer to buffer to map |
992 | * @len: length in bytes | |
993 | * @gfp_mask: allocation flags for bio allocation | |
994 | * | |
995 | * Map the kernel address into a bio suitable for io to a block | |
996 | * device. Returns an error pointer in case of error. | |
997 | */ | |
165125e1 | 998 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
27496a8c | 999 | gfp_t gfp_mask) |
df46b9a4 MC |
1000 | { |
1001 | struct bio *bio; | |
1002 | ||
1003 | bio = __bio_map_kern(q, data, len, gfp_mask); | |
1004 | if (IS_ERR(bio)) | |
1005 | return bio; | |
1006 | ||
1007 | if (bio->bi_size == len) | |
1008 | return bio; | |
1009 | ||
1010 | /* | |
1011 | * Don't support partial mappings. | |
1012 | */ | |
1013 | bio_put(bio); | |
1014 | return ERR_PTR(-EINVAL); | |
1015 | } | |
1016 | ||
68154e90 FT |
1017 | static void bio_copy_kern_endio(struct bio *bio, int err) |
1018 | { | |
1019 | struct bio_vec *bvec; | |
1020 | const int read = bio_data_dir(bio) == READ; | |
76029ff3 | 1021 | struct bio_map_data *bmd = bio->bi_private; |
68154e90 | 1022 | int i; |
76029ff3 | 1023 | char *p = bmd->sgvecs[0].iov_base; |
68154e90 FT |
1024 | |
1025 | __bio_for_each_segment(bvec, bio, i, 0) { | |
1026 | char *addr = page_address(bvec->bv_page); | |
76029ff3 | 1027 | int len = bmd->iovecs[i].bv_len; |
68154e90 FT |
1028 | |
1029 | if (read && !err) | |
76029ff3 | 1030 | memcpy(p, addr, len); |
68154e90 FT |
1031 | |
1032 | __free_page(bvec->bv_page); | |
76029ff3 | 1033 | p += len; |
68154e90 FT |
1034 | } |
1035 | ||
76029ff3 | 1036 | bio_free_map_data(bmd); |
68154e90 FT |
1037 | bio_put(bio); |
1038 | } | |
1039 | ||
1040 | /** | |
1041 | * bio_copy_kern - copy kernel address into bio | |
1042 | * @q: the struct request_queue for the bio | |
1043 | * @data: pointer to buffer to copy | |
1044 | * @len: length in bytes | |
1045 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 1046 | * @reading: data direction is READ |
68154e90 FT |
1047 | * |
1048 | * copy the kernel address into a bio suitable for io to a block | |
1049 | * device. Returns an error pointer in case of error. | |
1050 | */ | |
1051 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1052 | gfp_t gfp_mask, int reading) | |
1053 | { | |
68154e90 FT |
1054 | struct bio *bio; |
1055 | struct bio_vec *bvec; | |
4d8ab62e | 1056 | int i; |
68154e90 | 1057 | |
4d8ab62e FT |
1058 | bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask); |
1059 | if (IS_ERR(bio)) | |
1060 | return bio; | |
68154e90 FT |
1061 | |
1062 | if (!reading) { | |
1063 | void *p = data; | |
1064 | ||
1065 | bio_for_each_segment(bvec, bio, i) { | |
1066 | char *addr = page_address(bvec->bv_page); | |
1067 | ||
1068 | memcpy(addr, p, bvec->bv_len); | |
1069 | p += bvec->bv_len; | |
1070 | } | |
1071 | } | |
1072 | ||
68154e90 | 1073 | bio->bi_end_io = bio_copy_kern_endio; |
76029ff3 | 1074 | |
68154e90 | 1075 | return bio; |
68154e90 FT |
1076 | } |
1077 | ||
1da177e4 LT |
1078 | /* |
1079 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1080 | * for performing direct-IO in BIOs. | |
1081 | * | |
1082 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1083 | * because the required locks are not interrupt-safe. So what we can do is to | |
1084 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1085 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1086 | * in process context. | |
1087 | * | |
1088 | * We special-case compound pages here: normally this means reads into hugetlb | |
1089 | * pages. The logic in here doesn't really work right for compound pages | |
1090 | * because the VM does not uniformly chase down the head page in all cases. | |
1091 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1092 | * handle them at all. So we skip compound pages here at an early stage. | |
1093 | * | |
1094 | * Note that this code is very hard to test under normal circumstances because | |
1095 | * direct-io pins the pages with get_user_pages(). This makes | |
1096 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
1097 | * But other code (eg, pdflush) could clean the pages if they are mapped | |
1098 | * pagecache. | |
1099 | * | |
1100 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1101 | * deferred bio dirtying paths. | |
1102 | */ | |
1103 | ||
1104 | /* | |
1105 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1106 | */ | |
1107 | void bio_set_pages_dirty(struct bio *bio) | |
1108 | { | |
1109 | struct bio_vec *bvec = bio->bi_io_vec; | |
1110 | int i; | |
1111 | ||
1112 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1113 | struct page *page = bvec[i].bv_page; | |
1114 | ||
1115 | if (page && !PageCompound(page)) | |
1116 | set_page_dirty_lock(page); | |
1117 | } | |
1118 | } | |
1119 | ||
86b6c7a7 | 1120 | static void bio_release_pages(struct bio *bio) |
1da177e4 LT |
1121 | { |
1122 | struct bio_vec *bvec = bio->bi_io_vec; | |
1123 | int i; | |
1124 | ||
1125 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1126 | struct page *page = bvec[i].bv_page; | |
1127 | ||
1128 | if (page) | |
1129 | put_page(page); | |
1130 | } | |
1131 | } | |
1132 | ||
1133 | /* | |
1134 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1135 | * If they are, then fine. If, however, some pages are clean then they must | |
1136 | * have been written out during the direct-IO read. So we take another ref on | |
1137 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1138 | * | |
1139 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1140 | * here on. It will run one page_cache_release() against each page and will | |
1141 | * run one bio_put() against the BIO. | |
1142 | */ | |
1143 | ||
65f27f38 | 1144 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1145 | |
65f27f38 | 1146 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1147 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1148 | static struct bio *bio_dirty_list; | |
1149 | ||
1150 | /* | |
1151 | * This runs in process context | |
1152 | */ | |
65f27f38 | 1153 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 LT |
1154 | { |
1155 | unsigned long flags; | |
1156 | struct bio *bio; | |
1157 | ||
1158 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1159 | bio = bio_dirty_list; | |
1160 | bio_dirty_list = NULL; | |
1161 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1162 | ||
1163 | while (bio) { | |
1164 | struct bio *next = bio->bi_private; | |
1165 | ||
1166 | bio_set_pages_dirty(bio); | |
1167 | bio_release_pages(bio); | |
1168 | bio_put(bio); | |
1169 | bio = next; | |
1170 | } | |
1171 | } | |
1172 | ||
1173 | void bio_check_pages_dirty(struct bio *bio) | |
1174 | { | |
1175 | struct bio_vec *bvec = bio->bi_io_vec; | |
1176 | int nr_clean_pages = 0; | |
1177 | int i; | |
1178 | ||
1179 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1180 | struct page *page = bvec[i].bv_page; | |
1181 | ||
1182 | if (PageDirty(page) || PageCompound(page)) { | |
1183 | page_cache_release(page); | |
1184 | bvec[i].bv_page = NULL; | |
1185 | } else { | |
1186 | nr_clean_pages++; | |
1187 | } | |
1188 | } | |
1189 | ||
1190 | if (nr_clean_pages) { | |
1191 | unsigned long flags; | |
1192 | ||
1193 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1194 | bio->bi_private = bio_dirty_list; | |
1195 | bio_dirty_list = bio; | |
1196 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1197 | schedule_work(&bio_dirty_work); | |
1198 | } else { | |
1199 | bio_put(bio); | |
1200 | } | |
1201 | } | |
1202 | ||
1203 | /** | |
1204 | * bio_endio - end I/O on a bio | |
1205 | * @bio: bio | |
1da177e4 LT |
1206 | * @error: error, if any |
1207 | * | |
1208 | * Description: | |
6712ecf8 | 1209 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
5bb23a68 N |
1210 | * preferred way to end I/O on a bio, it takes care of clearing |
1211 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1212 | * established -Exxxx (-EIO, for instance) error values in case | |
1213 | * something went wrong. Noone should call bi_end_io() directly on a | |
1214 | * bio unless they own it and thus know that it has an end_io | |
1215 | * function. | |
1da177e4 | 1216 | **/ |
6712ecf8 | 1217 | void bio_endio(struct bio *bio, int error) |
1da177e4 LT |
1218 | { |
1219 | if (error) | |
1220 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
9cc54d40 N |
1221 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1222 | error = -EIO; | |
1da177e4 | 1223 | |
5bb23a68 | 1224 | if (bio->bi_end_io) |
6712ecf8 | 1225 | bio->bi_end_io(bio, error); |
1da177e4 LT |
1226 | } |
1227 | ||
1228 | void bio_pair_release(struct bio_pair *bp) | |
1229 | { | |
1230 | if (atomic_dec_and_test(&bp->cnt)) { | |
1231 | struct bio *master = bp->bio1.bi_private; | |
1232 | ||
6712ecf8 | 1233 | bio_endio(master, bp->error); |
1da177e4 LT |
1234 | mempool_free(bp, bp->bio2.bi_private); |
1235 | } | |
1236 | } | |
1237 | ||
6712ecf8 | 1238 | static void bio_pair_end_1(struct bio *bi, int err) |
1da177e4 LT |
1239 | { |
1240 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); | |
1241 | ||
1242 | if (err) | |
1243 | bp->error = err; | |
1244 | ||
1da177e4 | 1245 | bio_pair_release(bp); |
1da177e4 LT |
1246 | } |
1247 | ||
6712ecf8 | 1248 | static void bio_pair_end_2(struct bio *bi, int err) |
1da177e4 LT |
1249 | { |
1250 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); | |
1251 | ||
1252 | if (err) | |
1253 | bp->error = err; | |
1254 | ||
1da177e4 | 1255 | bio_pair_release(bp); |
1da177e4 LT |
1256 | } |
1257 | ||
1258 | /* | |
1259 | * split a bio - only worry about a bio with a single page | |
1260 | * in it's iovec | |
1261 | */ | |
6feef531 | 1262 | struct bio_pair *bio_split(struct bio *bi, int first_sectors) |
1da177e4 | 1263 | { |
6feef531 | 1264 | struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO); |
1da177e4 LT |
1265 | |
1266 | if (!bp) | |
1267 | return bp; | |
1268 | ||
5f3ea37c | 1269 | trace_block_split(bdev_get_queue(bi->bi_bdev), bi, |
2056a782 JA |
1270 | bi->bi_sector + first_sectors); |
1271 | ||
1da177e4 LT |
1272 | BUG_ON(bi->bi_vcnt != 1); |
1273 | BUG_ON(bi->bi_idx != 0); | |
1274 | atomic_set(&bp->cnt, 3); | |
1275 | bp->error = 0; | |
1276 | bp->bio1 = *bi; | |
1277 | bp->bio2 = *bi; | |
1278 | bp->bio2.bi_sector += first_sectors; | |
1279 | bp->bio2.bi_size -= first_sectors << 9; | |
1280 | bp->bio1.bi_size = first_sectors << 9; | |
1281 | ||
1282 | bp->bv1 = bi->bi_io_vec[0]; | |
1283 | bp->bv2 = bi->bi_io_vec[0]; | |
1284 | bp->bv2.bv_offset += first_sectors << 9; | |
1285 | bp->bv2.bv_len -= first_sectors << 9; | |
1286 | bp->bv1.bv_len = first_sectors << 9; | |
1287 | ||
1288 | bp->bio1.bi_io_vec = &bp->bv1; | |
1289 | bp->bio2.bi_io_vec = &bp->bv2; | |
1290 | ||
a2eb0c10 N |
1291 | bp->bio1.bi_max_vecs = 1; |
1292 | bp->bio2.bi_max_vecs = 1; | |
1293 | ||
1da177e4 LT |
1294 | bp->bio1.bi_end_io = bio_pair_end_1; |
1295 | bp->bio2.bi_end_io = bio_pair_end_2; | |
1296 | ||
1297 | bp->bio1.bi_private = bi; | |
6feef531 | 1298 | bp->bio2.bi_private = bio_split_pool; |
1da177e4 | 1299 | |
7ba1ba12 MP |
1300 | if (bio_integrity(bi)) |
1301 | bio_integrity_split(bi, bp, first_sectors); | |
1302 | ||
1da177e4 LT |
1303 | return bp; |
1304 | } | |
1305 | ||
ad3316bf MP |
1306 | /** |
1307 | * bio_sector_offset - Find hardware sector offset in bio | |
1308 | * @bio: bio to inspect | |
1309 | * @index: bio_vec index | |
1310 | * @offset: offset in bv_page | |
1311 | * | |
1312 | * Return the number of hardware sectors between beginning of bio | |
1313 | * and an end point indicated by a bio_vec index and an offset | |
1314 | * within that vector's page. | |
1315 | */ | |
1316 | sector_t bio_sector_offset(struct bio *bio, unsigned short index, | |
1317 | unsigned int offset) | |
1318 | { | |
1319 | unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue); | |
1320 | struct bio_vec *bv; | |
1321 | sector_t sectors; | |
1322 | int i; | |
1323 | ||
1324 | sectors = 0; | |
1325 | ||
1326 | if (index >= bio->bi_idx) | |
1327 | index = bio->bi_vcnt - 1; | |
1328 | ||
1329 | __bio_for_each_segment(bv, bio, i, 0) { | |
1330 | if (i == index) { | |
1331 | if (offset > bv->bv_offset) | |
1332 | sectors += (offset - bv->bv_offset) / sector_sz; | |
1333 | break; | |
1334 | } | |
1335 | ||
1336 | sectors += bv->bv_len / sector_sz; | |
1337 | } | |
1338 | ||
1339 | return sectors; | |
1340 | } | |
1341 | EXPORT_SYMBOL(bio_sector_offset); | |
1da177e4 LT |
1342 | |
1343 | /* | |
1344 | * create memory pools for biovec's in a bio_set. | |
1345 | * use the global biovec slabs created for general use. | |
1346 | */ | |
5972511b | 1347 | static int biovec_create_pools(struct bio_set *bs, int pool_entries) |
1da177e4 LT |
1348 | { |
1349 | int i; | |
1350 | ||
1351 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1352 | struct biovec_slab *bp = bvec_slabs + i; | |
1353 | mempool_t **bvp = bs->bvec_pools + i; | |
1354 | ||
93d2341c | 1355 | *bvp = mempool_create_slab_pool(pool_entries, bp->slab); |
1da177e4 LT |
1356 | if (!*bvp) |
1357 | return -ENOMEM; | |
1358 | } | |
1359 | return 0; | |
1360 | } | |
1361 | ||
1362 | static void biovec_free_pools(struct bio_set *bs) | |
1363 | { | |
1364 | int i; | |
1365 | ||
1366 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1367 | mempool_t *bvp = bs->bvec_pools[i]; | |
1368 | ||
1369 | if (bvp) | |
1370 | mempool_destroy(bvp); | |
1371 | } | |
1372 | ||
1373 | } | |
1374 | ||
1375 | void bioset_free(struct bio_set *bs) | |
1376 | { | |
1377 | if (bs->bio_pool) | |
1378 | mempool_destroy(bs->bio_pool); | |
1379 | ||
7ba1ba12 | 1380 | bioset_integrity_free(bs); |
1da177e4 LT |
1381 | biovec_free_pools(bs); |
1382 | ||
1383 | kfree(bs); | |
1384 | } | |
1385 | ||
5972511b | 1386 | struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size) |
1da177e4 | 1387 | { |
11b0b5ab | 1388 | struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1da177e4 LT |
1389 | |
1390 | if (!bs) | |
1391 | return NULL; | |
1392 | ||
93d2341c | 1393 | bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab); |
1da177e4 LT |
1394 | if (!bs->bio_pool) |
1395 | goto bad; | |
1396 | ||
7ba1ba12 MP |
1397 | if (bioset_integrity_create(bs, bio_pool_size)) |
1398 | goto bad; | |
1399 | ||
5972511b | 1400 | if (!biovec_create_pools(bs, bvec_pool_size)) |
1da177e4 LT |
1401 | return bs; |
1402 | ||
1403 | bad: | |
1404 | bioset_free(bs); | |
1405 | return NULL; | |
1406 | } | |
1407 | ||
1408 | static void __init biovec_init_slabs(void) | |
1409 | { | |
1410 | int i; | |
1411 | ||
1412 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1413 | int size; | |
1414 | struct biovec_slab *bvs = bvec_slabs + i; | |
1415 | ||
1416 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
1417 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 1418 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
1419 | } |
1420 | } | |
1421 | ||
1422 | static int __init init_bio(void) | |
1423 | { | |
0a31bd5f | 1424 | bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
1da177e4 | 1425 | |
7ba1ba12 | 1426 | bio_integrity_init_slab(); |
1da177e4 LT |
1427 | biovec_init_slabs(); |
1428 | ||
5972511b | 1429 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 2); |
1da177e4 LT |
1430 | if (!fs_bio_set) |
1431 | panic("bio: can't allocate bios\n"); | |
1432 | ||
0eaae62a MD |
1433 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
1434 | sizeof(struct bio_pair)); | |
1da177e4 LT |
1435 | if (!bio_split_pool) |
1436 | panic("bio: can't create split pool\n"); | |
1437 | ||
1438 | return 0; | |
1439 | } | |
1440 | ||
1441 | subsys_initcall(init_bio); | |
1442 | ||
1443 | EXPORT_SYMBOL(bio_alloc); | |
0a0d96b0 | 1444 | EXPORT_SYMBOL(bio_kmalloc); |
1da177e4 | 1445 | EXPORT_SYMBOL(bio_put); |
3676347a | 1446 | EXPORT_SYMBOL(bio_free); |
1da177e4 LT |
1447 | EXPORT_SYMBOL(bio_endio); |
1448 | EXPORT_SYMBOL(bio_init); | |
1449 | EXPORT_SYMBOL(__bio_clone); | |
1450 | EXPORT_SYMBOL(bio_clone); | |
1451 | EXPORT_SYMBOL(bio_phys_segments); | |
1da177e4 | 1452 | EXPORT_SYMBOL(bio_add_page); |
6e68af66 | 1453 | EXPORT_SYMBOL(bio_add_pc_page); |
1da177e4 | 1454 | EXPORT_SYMBOL(bio_get_nr_vecs); |
40044ce0 JA |
1455 | EXPORT_SYMBOL(bio_map_user); |
1456 | EXPORT_SYMBOL(bio_unmap_user); | |
df46b9a4 | 1457 | EXPORT_SYMBOL(bio_map_kern); |
68154e90 | 1458 | EXPORT_SYMBOL(bio_copy_kern); |
1da177e4 LT |
1459 | EXPORT_SYMBOL(bio_pair_release); |
1460 | EXPORT_SYMBOL(bio_split); | |
1da177e4 LT |
1461 | EXPORT_SYMBOL(bio_copy_user); |
1462 | EXPORT_SYMBOL(bio_uncopy_user); | |
1463 | EXPORT_SYMBOL(bioset_create); | |
1464 | EXPORT_SYMBOL(bioset_free); | |
1465 | EXPORT_SYMBOL(bio_alloc_bioset); |