]>
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> |
f1970baf | 29 | #include <scsi/sg.h> /* for struct sg_iovec */ |
1da177e4 | 30 | |
e18b890b | 31 | static struct kmem_cache *bio_slab __read_mostly; |
1da177e4 | 32 | |
fa3536cc | 33 | mempool_t *bio_split_pool __read_mostly; |
1da177e4 | 34 | |
1da177e4 LT |
35 | /* |
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) } | |
6c036527 | 42 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { |
1da177e4 LT |
43 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
44 | }; | |
45 | #undef BV | |
46 | ||
1da177e4 LT |
47 | /* |
48 | * 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 | */ | |
51d654e1 | 51 | struct bio_set *fs_bio_set; |
1da177e4 | 52 | |
7ba1ba12 MP |
53 | unsigned int bvec_nr_vecs(unsigned short idx) |
54 | { | |
55 | return bvec_slabs[idx].nr_vecs; | |
56 | } | |
57 | ||
51d654e1 | 58 | struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs) |
1da177e4 LT |
59 | { |
60 | struct bio_vec *bvl; | |
1da177e4 LT |
61 | |
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 | ||
1da177e4 | 79 | bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask); |
1ac0ae06 DC |
80 | if (bvl) |
81 | memset(bvl, 0, bvec_nr_vecs(*idx) * sizeof(struct bio_vec)); | |
1da177e4 LT |
82 | |
83 | return bvl; | |
84 | } | |
85 | ||
3676347a | 86 | void bio_free(struct bio *bio, struct bio_set *bio_set) |
1da177e4 | 87 | { |
992c5dda JA |
88 | if (bio->bi_io_vec) { |
89 | const int pool_idx = BIO_POOL_IDX(bio); | |
1da177e4 | 90 | |
992c5dda JA |
91 | BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); |
92 | ||
93 | mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]); | |
94 | } | |
1da177e4 | 95 | |
7ba1ba12 MP |
96 | if (bio_integrity(bio)) |
97 | bio_integrity_free(bio, bio_set); | |
98 | ||
3676347a PO |
99 | mempool_free(bio, bio_set->bio_pool); |
100 | } | |
101 | ||
102 | /* | |
103 | * default destructor for a bio allocated with bio_alloc_bioset() | |
104 | */ | |
105 | static void bio_fs_destructor(struct bio *bio) | |
106 | { | |
107 | bio_free(bio, fs_bio_set); | |
1da177e4 LT |
108 | } |
109 | ||
858119e1 | 110 | void bio_init(struct bio *bio) |
1da177e4 | 111 | { |
2b94de55 | 112 | memset(bio, 0, sizeof(*bio)); |
1da177e4 | 113 | bio->bi_flags = 1 << BIO_UPTODATE; |
1da177e4 | 114 | atomic_set(&bio->bi_cnt, 1); |
1da177e4 LT |
115 | } |
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 | |
67be2dd1 | 121 | * @bs: the bio_set to allocate from |
1da177e4 LT |
122 | * |
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 | **/ | |
dd0fc66f | 131 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) |
1da177e4 LT |
132 | { |
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)) { | |
eeae1d48 | 140 | unsigned long uninitialized_var(idx); |
1da177e4 LT |
141 | |
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; | |
1ac0ae06 | 149 | bio->bi_max_vecs = bvec_nr_vecs(idx); |
1da177e4 LT |
150 | } |
151 | bio->bi_io_vec = bvl; | |
1da177e4 LT |
152 | } |
153 | out: | |
154 | return bio; | |
155 | } | |
156 | ||
dd0fc66f | 157 | struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs) |
1da177e4 | 158 | { |
3676347a PO |
159 | 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; | |
1da177e4 LT |
165 | } |
166 | ||
167 | void 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 | } | |
180 | EXPORT_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 | **/ | |
190 | void 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 | ||
165125e1 | 203 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
204 | { |
205 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
206 | blk_recount_segments(q, bio); | |
207 | ||
208 | return bio->bi_phys_segments; | |
209 | } | |
210 | ||
165125e1 | 211 | inline int bio_hw_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
212 | { |
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 | */ | |
858119e1 | 228 | void __bio_clone(struct bio *bio, struct bio *bio_src) |
1da177e4 | 229 | { |
e525e153 AM |
230 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, |
231 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); | |
1da177e4 | 232 | |
5d84070e JA |
233 | /* |
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 | */ | |
1da177e4 LT |
237 | 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; | |
1da177e4 LT |
241 | bio->bi_vcnt = bio_src->bi_vcnt; |
242 | bio->bi_size = bio_src->bi_size; | |
a5453be4 | 243 | bio->bi_idx = bio_src->bi_idx; |
1da177e4 LT |
244 | } |
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 | */ | |
dd0fc66f | 253 | struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
254 | { |
255 | struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); | |
256 | ||
7ba1ba12 MP |
257 | 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; | |
3676347a | 270 | } |
1da177e4 LT |
271 | |
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 | */ | |
284 | int bio_get_nr_vecs(struct block_device *bdev) | |
285 | { | |
165125e1 | 286 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
287 | 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 | ||
165125e1 | 298 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page |
defd94b7 MC |
299 | *page, unsigned int len, unsigned int offset, |
300 | unsigned short max_sectors) | |
1da177e4 LT |
301 | { |
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 | ||
80cfd548 | 311 | if (((bio->bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
312 | return 0; |
313 | ||
80cfd548 JA |
314 | /* |
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; | |
cc371e66 AK |
325 | |
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 | } | |
80cfd548 JA |
338 | } |
339 | ||
340 | goto done; | |
341 | } | |
342 | } | |
343 | ||
344 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
1da177e4 LT |
345 | 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 | |
b8b3e16c | 353 | || bio->bi_hw_segments >= q->max_hw_segments) { |
1da177e4 LT |
354 | |
355 | if (retried_segments) | |
356 | return 0; | |
357 | ||
358 | retried_segments = 1; | |
359 | blk_recount_segments(q, bio); | |
360 | } | |
361 | ||
362 | /* | |
363 | * setup the new entry, we might clear it again later if we | |
364 | * cannot add the page | |
365 | */ | |
366 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
367 | bvec->bv_page = page; | |
368 | bvec->bv_len = len; | |
369 | bvec->bv_offset = offset; | |
370 | ||
371 | /* | |
372 | * if queue has other restrictions (eg varying max sector size | |
373 | * depending on offset), it can specify a merge_bvec_fn in the | |
374 | * queue to get further control | |
375 | */ | |
376 | if (q->merge_bvec_fn) { | |
cc371e66 AK |
377 | struct bvec_merge_data bvm = { |
378 | .bi_bdev = bio->bi_bdev, | |
379 | .bi_sector = bio->bi_sector, | |
380 | .bi_size = bio->bi_size, | |
381 | .bi_rw = bio->bi_rw, | |
382 | }; | |
383 | ||
1da177e4 LT |
384 | /* |
385 | * merge_bvec_fn() returns number of bytes it can accept | |
386 | * at this offset | |
387 | */ | |
cc371e66 | 388 | if (q->merge_bvec_fn(q, &bvm, bvec) < len) { |
1da177e4 LT |
389 | bvec->bv_page = NULL; |
390 | bvec->bv_len = 0; | |
391 | bvec->bv_offset = 0; | |
392 | return 0; | |
393 | } | |
394 | } | |
395 | ||
396 | /* If we may be able to merge these biovecs, force a recount */ | |
b8b3e16c | 397 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) |
1da177e4 LT |
398 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
399 | ||
400 | bio->bi_vcnt++; | |
401 | bio->bi_phys_segments++; | |
402 | bio->bi_hw_segments++; | |
80cfd548 | 403 | done: |
1da177e4 LT |
404 | bio->bi_size += len; |
405 | return len; | |
406 | } | |
407 | ||
6e68af66 MC |
408 | /** |
409 | * bio_add_pc_page - attempt to add page to bio | |
fddfdeaf | 410 | * @q: the target queue |
6e68af66 MC |
411 | * @bio: destination bio |
412 | * @page: page to add | |
413 | * @len: vec entry length | |
414 | * @offset: vec entry offset | |
415 | * | |
416 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
417 | * number of reasons, such as the bio being full or target block | |
418 | * device limitations. The target block device must allow bio's | |
419 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
420 | * page to an empty bio. This should only be used by REQ_PC bios. | |
421 | */ | |
165125e1 | 422 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
6e68af66 MC |
423 | unsigned int len, unsigned int offset) |
424 | { | |
defd94b7 | 425 | return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors); |
6e68af66 MC |
426 | } |
427 | ||
1da177e4 LT |
428 | /** |
429 | * bio_add_page - attempt to add page to bio | |
430 | * @bio: destination bio | |
431 | * @page: page to add | |
432 | * @len: vec entry length | |
433 | * @offset: vec entry offset | |
434 | * | |
435 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
436 | * number of reasons, such as the bio being full or target block | |
437 | * device limitations. The target block device must allow bio's | |
438 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
439 | * page to an empty bio. | |
440 | */ | |
441 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
442 | unsigned int offset) | |
443 | { | |
defd94b7 MC |
444 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
445 | return __bio_add_page(q, bio, page, len, offset, q->max_sectors); | |
1da177e4 LT |
446 | } |
447 | ||
448 | struct bio_map_data { | |
449 | struct bio_vec *iovecs; | |
c5dec1c3 FT |
450 | int nr_sgvecs; |
451 | struct sg_iovec *sgvecs; | |
1da177e4 LT |
452 | }; |
453 | ||
c5dec1c3 FT |
454 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
455 | struct sg_iovec *iov, int iov_count) | |
1da177e4 LT |
456 | { |
457 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); | |
c5dec1c3 FT |
458 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
459 | bmd->nr_sgvecs = iov_count; | |
1da177e4 LT |
460 | bio->bi_private = bmd; |
461 | } | |
462 | ||
463 | static void bio_free_map_data(struct bio_map_data *bmd) | |
464 | { | |
465 | kfree(bmd->iovecs); | |
c5dec1c3 | 466 | kfree(bmd->sgvecs); |
1da177e4 LT |
467 | kfree(bmd); |
468 | } | |
469 | ||
76029ff3 FT |
470 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count, |
471 | gfp_t gfp_mask) | |
1da177e4 | 472 | { |
76029ff3 | 473 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask); |
1da177e4 LT |
474 | |
475 | if (!bmd) | |
476 | return NULL; | |
477 | ||
76029ff3 | 478 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask); |
c5dec1c3 FT |
479 | if (!bmd->iovecs) { |
480 | kfree(bmd); | |
481 | return NULL; | |
482 | } | |
483 | ||
76029ff3 | 484 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask); |
c5dec1c3 | 485 | if (bmd->sgvecs) |
1da177e4 LT |
486 | return bmd; |
487 | ||
c5dec1c3 | 488 | kfree(bmd->iovecs); |
1da177e4 LT |
489 | kfree(bmd); |
490 | return NULL; | |
491 | } | |
492 | ||
aefcc28a FT |
493 | static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs, |
494 | struct sg_iovec *iov, int iov_count, int uncopy) | |
c5dec1c3 FT |
495 | { |
496 | int ret = 0, i; | |
497 | struct bio_vec *bvec; | |
498 | int iov_idx = 0; | |
499 | unsigned int iov_off = 0; | |
500 | int read = bio_data_dir(bio) == READ; | |
501 | ||
502 | __bio_for_each_segment(bvec, bio, i, 0) { | |
503 | char *bv_addr = page_address(bvec->bv_page); | |
aefcc28a | 504 | unsigned int bv_len = iovecs[i].bv_len; |
c5dec1c3 FT |
505 | |
506 | while (bv_len && iov_idx < iov_count) { | |
507 | unsigned int bytes; | |
508 | char *iov_addr; | |
509 | ||
510 | bytes = min_t(unsigned int, | |
511 | iov[iov_idx].iov_len - iov_off, bv_len); | |
512 | iov_addr = iov[iov_idx].iov_base + iov_off; | |
513 | ||
514 | if (!ret) { | |
515 | if (!read && !uncopy) | |
516 | ret = copy_from_user(bv_addr, iov_addr, | |
517 | bytes); | |
518 | if (read && uncopy) | |
519 | ret = copy_to_user(iov_addr, bv_addr, | |
520 | bytes); | |
521 | ||
522 | if (ret) | |
523 | ret = -EFAULT; | |
524 | } | |
525 | ||
526 | bv_len -= bytes; | |
527 | bv_addr += bytes; | |
528 | iov_addr += bytes; | |
529 | iov_off += bytes; | |
530 | ||
531 | if (iov[iov_idx].iov_len == iov_off) { | |
532 | iov_idx++; | |
533 | iov_off = 0; | |
534 | } | |
535 | } | |
536 | ||
537 | if (uncopy) | |
538 | __free_page(bvec->bv_page); | |
539 | } | |
540 | ||
541 | return ret; | |
542 | } | |
543 | ||
1da177e4 LT |
544 | /** |
545 | * bio_uncopy_user - finish previously mapped bio | |
546 | * @bio: bio being terminated | |
547 | * | |
548 | * Free pages allocated from bio_copy_user() and write back data | |
549 | * to user space in case of a read. | |
550 | */ | |
551 | int bio_uncopy_user(struct bio *bio) | |
552 | { | |
553 | struct bio_map_data *bmd = bio->bi_private; | |
c5dec1c3 | 554 | int ret; |
1da177e4 | 555 | |
aefcc28a | 556 | ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, bmd->nr_sgvecs, 1); |
1da177e4 | 557 | |
1da177e4 LT |
558 | bio_free_map_data(bmd); |
559 | bio_put(bio); | |
560 | return ret; | |
561 | } | |
562 | ||
563 | /** | |
c5dec1c3 | 564 | * bio_copy_user_iov - copy user data to bio |
1da177e4 | 565 | * @q: destination block queue |
c5dec1c3 FT |
566 | * @iov: the iovec. |
567 | * @iov_count: number of elements in the iovec | |
1da177e4 LT |
568 | * @write_to_vm: bool indicating writing to pages or not |
569 | * | |
570 | * Prepares and returns a bio for indirect user io, bouncing data | |
571 | * to/from kernel pages as necessary. Must be paired with | |
572 | * call bio_uncopy_user() on io completion. | |
573 | */ | |
c5dec1c3 FT |
574 | struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov, |
575 | int iov_count, int write_to_vm) | |
1da177e4 | 576 | { |
1da177e4 LT |
577 | struct bio_map_data *bmd; |
578 | struct bio_vec *bvec; | |
579 | struct page *page; | |
580 | struct bio *bio; | |
581 | int i, ret; | |
c5dec1c3 FT |
582 | int nr_pages = 0; |
583 | unsigned int len = 0; | |
1da177e4 | 584 | |
c5dec1c3 FT |
585 | for (i = 0; i < iov_count; i++) { |
586 | unsigned long uaddr; | |
587 | unsigned long end; | |
588 | unsigned long start; | |
589 | ||
590 | uaddr = (unsigned long)iov[i].iov_base; | |
591 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
592 | start = uaddr >> PAGE_SHIFT; | |
593 | ||
594 | nr_pages += end - start; | |
595 | len += iov[i].iov_len; | |
596 | } | |
597 | ||
76029ff3 | 598 | bmd = bio_alloc_map_data(nr_pages, iov_count, GFP_KERNEL); |
1da177e4 LT |
599 | if (!bmd) |
600 | return ERR_PTR(-ENOMEM); | |
601 | ||
1da177e4 | 602 | ret = -ENOMEM; |
c5dec1c3 | 603 | bio = bio_alloc(GFP_KERNEL, nr_pages); |
1da177e4 LT |
604 | if (!bio) |
605 | goto out_bmd; | |
606 | ||
607 | bio->bi_rw |= (!write_to_vm << BIO_RW); | |
608 | ||
609 | ret = 0; | |
610 | while (len) { | |
611 | unsigned int bytes = PAGE_SIZE; | |
612 | ||
613 | if (bytes > len) | |
614 | bytes = len; | |
615 | ||
616 | page = alloc_page(q->bounce_gfp | GFP_KERNEL); | |
617 | if (!page) { | |
618 | ret = -ENOMEM; | |
619 | break; | |
620 | } | |
621 | ||
0e75f906 | 622 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) |
1da177e4 | 623 | break; |
1da177e4 LT |
624 | |
625 | len -= bytes; | |
626 | } | |
627 | ||
628 | if (ret) | |
629 | goto cleanup; | |
630 | ||
631 | /* | |
632 | * success | |
633 | */ | |
634 | if (!write_to_vm) { | |
aefcc28a | 635 | ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0); |
c5dec1c3 FT |
636 | if (ret) |
637 | goto cleanup; | |
1da177e4 LT |
638 | } |
639 | ||
c5dec1c3 | 640 | bio_set_map_data(bmd, bio, iov, iov_count); |
1da177e4 LT |
641 | return bio; |
642 | cleanup: | |
643 | bio_for_each_segment(bvec, bio, i) | |
644 | __free_page(bvec->bv_page); | |
645 | ||
646 | bio_put(bio); | |
647 | out_bmd: | |
648 | bio_free_map_data(bmd); | |
649 | return ERR_PTR(ret); | |
650 | } | |
651 | ||
c5dec1c3 FT |
652 | /** |
653 | * bio_copy_user - copy user data to bio | |
654 | * @q: destination block queue | |
655 | * @uaddr: start of user address | |
656 | * @len: length in bytes | |
657 | * @write_to_vm: bool indicating writing to pages or not | |
658 | * | |
659 | * Prepares and returns a bio for indirect user io, bouncing data | |
660 | * to/from kernel pages as necessary. Must be paired with | |
661 | * call bio_uncopy_user() on io completion. | |
662 | */ | |
663 | struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr, | |
664 | unsigned int len, int write_to_vm) | |
665 | { | |
666 | struct sg_iovec iov; | |
667 | ||
668 | iov.iov_base = (void __user *)uaddr; | |
669 | iov.iov_len = len; | |
670 | ||
671 | return bio_copy_user_iov(q, &iov, 1, write_to_vm); | |
672 | } | |
673 | ||
165125e1 | 674 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
f1970baf JB |
675 | struct block_device *bdev, |
676 | struct sg_iovec *iov, int iov_count, | |
677 | int write_to_vm) | |
1da177e4 | 678 | { |
f1970baf JB |
679 | int i, j; |
680 | int nr_pages = 0; | |
1da177e4 LT |
681 | struct page **pages; |
682 | struct bio *bio; | |
f1970baf JB |
683 | int cur_page = 0; |
684 | int ret, offset; | |
1da177e4 | 685 | |
f1970baf JB |
686 | for (i = 0; i < iov_count; i++) { |
687 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
688 | unsigned long len = iov[i].iov_len; | |
689 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
690 | unsigned long start = uaddr >> PAGE_SHIFT; | |
691 | ||
692 | nr_pages += end - start; | |
693 | /* | |
ad2d7225 | 694 | * buffer must be aligned to at least hardsector size for now |
f1970baf | 695 | */ |
ad2d7225 | 696 | if (uaddr & queue_dma_alignment(q)) |
f1970baf JB |
697 | return ERR_PTR(-EINVAL); |
698 | } | |
699 | ||
700 | if (!nr_pages) | |
1da177e4 LT |
701 | return ERR_PTR(-EINVAL); |
702 | ||
703 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
704 | if (!bio) | |
705 | return ERR_PTR(-ENOMEM); | |
706 | ||
707 | ret = -ENOMEM; | |
11b0b5ab | 708 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); |
1da177e4 LT |
709 | if (!pages) |
710 | goto out; | |
711 | ||
f1970baf JB |
712 | for (i = 0; i < iov_count; i++) { |
713 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
714 | unsigned long len = iov[i].iov_len; | |
715 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
716 | unsigned long start = uaddr >> PAGE_SHIFT; | |
717 | const int local_nr_pages = end - start; | |
718 | const int page_limit = cur_page + local_nr_pages; | |
719 | ||
f5dd33c4 NP |
720 | ret = get_user_pages_fast(uaddr, local_nr_pages, |
721 | write_to_vm, &pages[cur_page]); | |
99172157 JA |
722 | if (ret < local_nr_pages) { |
723 | ret = -EFAULT; | |
f1970baf | 724 | goto out_unmap; |
99172157 | 725 | } |
f1970baf JB |
726 | |
727 | offset = uaddr & ~PAGE_MASK; | |
728 | for (j = cur_page; j < page_limit; j++) { | |
729 | unsigned int bytes = PAGE_SIZE - offset; | |
730 | ||
731 | if (len <= 0) | |
732 | break; | |
733 | ||
734 | if (bytes > len) | |
735 | bytes = len; | |
736 | ||
737 | /* | |
738 | * sorry... | |
739 | */ | |
defd94b7 MC |
740 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
741 | bytes) | |
f1970baf JB |
742 | break; |
743 | ||
744 | len -= bytes; | |
745 | offset = 0; | |
746 | } | |
1da177e4 | 747 | |
f1970baf | 748 | cur_page = j; |
1da177e4 | 749 | /* |
f1970baf | 750 | * release the pages we didn't map into the bio, if any |
1da177e4 | 751 | */ |
f1970baf JB |
752 | while (j < page_limit) |
753 | page_cache_release(pages[j++]); | |
1da177e4 LT |
754 | } |
755 | ||
1da177e4 LT |
756 | kfree(pages); |
757 | ||
758 | /* | |
759 | * set data direction, and check if mapped pages need bouncing | |
760 | */ | |
761 | if (!write_to_vm) | |
762 | bio->bi_rw |= (1 << BIO_RW); | |
763 | ||
f1970baf | 764 | bio->bi_bdev = bdev; |
1da177e4 LT |
765 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
766 | return bio; | |
f1970baf JB |
767 | |
768 | out_unmap: | |
769 | for (i = 0; i < nr_pages; i++) { | |
770 | if(!pages[i]) | |
771 | break; | |
772 | page_cache_release(pages[i]); | |
773 | } | |
774 | out: | |
1da177e4 LT |
775 | kfree(pages); |
776 | bio_put(bio); | |
777 | return ERR_PTR(ret); | |
778 | } | |
779 | ||
780 | /** | |
781 | * bio_map_user - map user address into bio | |
165125e1 | 782 | * @q: the struct request_queue for the bio |
1da177e4 LT |
783 | * @bdev: destination block device |
784 | * @uaddr: start of user address | |
785 | * @len: length in bytes | |
786 | * @write_to_vm: bool indicating writing to pages or not | |
787 | * | |
788 | * Map the user space address into a bio suitable for io to a block | |
789 | * device. Returns an error pointer in case of error. | |
790 | */ | |
165125e1 | 791 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
1da177e4 | 792 | unsigned long uaddr, unsigned int len, int write_to_vm) |
f1970baf JB |
793 | { |
794 | struct sg_iovec iov; | |
795 | ||
3f70353e | 796 | iov.iov_base = (void __user *)uaddr; |
f1970baf JB |
797 | iov.iov_len = len; |
798 | ||
799 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm); | |
800 | } | |
801 | ||
802 | /** | |
803 | * bio_map_user_iov - map user sg_iovec table into bio | |
165125e1 | 804 | * @q: the struct request_queue for the bio |
f1970baf JB |
805 | * @bdev: destination block device |
806 | * @iov: the iovec. | |
807 | * @iov_count: number of elements in the iovec | |
808 | * @write_to_vm: bool indicating writing to pages or not | |
809 | * | |
810 | * Map the user space address into a bio suitable for io to a block | |
811 | * device. Returns an error pointer in case of error. | |
812 | */ | |
165125e1 | 813 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
f1970baf JB |
814 | struct sg_iovec *iov, int iov_count, |
815 | int write_to_vm) | |
1da177e4 LT |
816 | { |
817 | struct bio *bio; | |
818 | ||
f1970baf | 819 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm); |
1da177e4 LT |
820 | |
821 | if (IS_ERR(bio)) | |
822 | return bio; | |
823 | ||
824 | /* | |
825 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
826 | * it would normally disappear when its bi_end_io is run. | |
827 | * however, we need it for the unmap, so grab an extra | |
828 | * reference to it | |
829 | */ | |
830 | bio_get(bio); | |
831 | ||
0e75f906 | 832 | return bio; |
1da177e4 LT |
833 | } |
834 | ||
835 | static void __bio_unmap_user(struct bio *bio) | |
836 | { | |
837 | struct bio_vec *bvec; | |
838 | int i; | |
839 | ||
840 | /* | |
841 | * make sure we dirty pages we wrote to | |
842 | */ | |
843 | __bio_for_each_segment(bvec, bio, i, 0) { | |
844 | if (bio_data_dir(bio) == READ) | |
845 | set_page_dirty_lock(bvec->bv_page); | |
846 | ||
847 | page_cache_release(bvec->bv_page); | |
848 | } | |
849 | ||
850 | bio_put(bio); | |
851 | } | |
852 | ||
853 | /** | |
854 | * bio_unmap_user - unmap a bio | |
855 | * @bio: the bio being unmapped | |
856 | * | |
857 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
858 | * a process context. | |
859 | * | |
860 | * bio_unmap_user() may sleep. | |
861 | */ | |
862 | void bio_unmap_user(struct bio *bio) | |
863 | { | |
864 | __bio_unmap_user(bio); | |
865 | bio_put(bio); | |
866 | } | |
867 | ||
6712ecf8 | 868 | static void bio_map_kern_endio(struct bio *bio, int err) |
b823825e | 869 | { |
b823825e | 870 | bio_put(bio); |
b823825e JA |
871 | } |
872 | ||
873 | ||
165125e1 | 874 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
27496a8c | 875 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 MC |
876 | { |
877 | unsigned long kaddr = (unsigned long)data; | |
878 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
879 | unsigned long start = kaddr >> PAGE_SHIFT; | |
880 | const int nr_pages = end - start; | |
881 | int offset, i; | |
882 | struct bio *bio; | |
883 | ||
884 | bio = bio_alloc(gfp_mask, nr_pages); | |
885 | if (!bio) | |
886 | return ERR_PTR(-ENOMEM); | |
887 | ||
888 | offset = offset_in_page(kaddr); | |
889 | for (i = 0; i < nr_pages; i++) { | |
890 | unsigned int bytes = PAGE_SIZE - offset; | |
891 | ||
892 | if (len <= 0) | |
893 | break; | |
894 | ||
895 | if (bytes > len) | |
896 | bytes = len; | |
897 | ||
defd94b7 MC |
898 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
899 | offset) < bytes) | |
df46b9a4 MC |
900 | break; |
901 | ||
902 | data += bytes; | |
903 | len -= bytes; | |
904 | offset = 0; | |
905 | } | |
906 | ||
b823825e | 907 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
908 | return bio; |
909 | } | |
910 | ||
911 | /** | |
912 | * bio_map_kern - map kernel address into bio | |
165125e1 | 913 | * @q: the struct request_queue for the bio |
df46b9a4 MC |
914 | * @data: pointer to buffer to map |
915 | * @len: length in bytes | |
916 | * @gfp_mask: allocation flags for bio allocation | |
917 | * | |
918 | * Map the kernel address into a bio suitable for io to a block | |
919 | * device. Returns an error pointer in case of error. | |
920 | */ | |
165125e1 | 921 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
27496a8c | 922 | gfp_t gfp_mask) |
df46b9a4 MC |
923 | { |
924 | struct bio *bio; | |
925 | ||
926 | bio = __bio_map_kern(q, data, len, gfp_mask); | |
927 | if (IS_ERR(bio)) | |
928 | return bio; | |
929 | ||
930 | if (bio->bi_size == len) | |
931 | return bio; | |
932 | ||
933 | /* | |
934 | * Don't support partial mappings. | |
935 | */ | |
936 | bio_put(bio); | |
937 | return ERR_PTR(-EINVAL); | |
938 | } | |
939 | ||
68154e90 FT |
940 | static void bio_copy_kern_endio(struct bio *bio, int err) |
941 | { | |
942 | struct bio_vec *bvec; | |
943 | const int read = bio_data_dir(bio) == READ; | |
76029ff3 | 944 | struct bio_map_data *bmd = bio->bi_private; |
68154e90 | 945 | int i; |
76029ff3 | 946 | char *p = bmd->sgvecs[0].iov_base; |
68154e90 FT |
947 | |
948 | __bio_for_each_segment(bvec, bio, i, 0) { | |
949 | char *addr = page_address(bvec->bv_page); | |
76029ff3 | 950 | int len = bmd->iovecs[i].bv_len; |
68154e90 FT |
951 | |
952 | if (read && !err) | |
76029ff3 | 953 | memcpy(p, addr, len); |
68154e90 FT |
954 | |
955 | __free_page(bvec->bv_page); | |
76029ff3 | 956 | p += len; |
68154e90 FT |
957 | } |
958 | ||
76029ff3 | 959 | bio_free_map_data(bmd); |
68154e90 FT |
960 | bio_put(bio); |
961 | } | |
962 | ||
963 | /** | |
964 | * bio_copy_kern - copy kernel address into bio | |
965 | * @q: the struct request_queue for the bio | |
966 | * @data: pointer to buffer to copy | |
967 | * @len: length in bytes | |
968 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 969 | * @reading: data direction is READ |
68154e90 FT |
970 | * |
971 | * copy the kernel address into a bio suitable for io to a block | |
972 | * device. Returns an error pointer in case of error. | |
973 | */ | |
974 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
975 | gfp_t gfp_mask, int reading) | |
976 | { | |
977 | unsigned long kaddr = (unsigned long)data; | |
978 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
979 | unsigned long start = kaddr >> PAGE_SHIFT; | |
980 | const int nr_pages = end - start; | |
981 | struct bio *bio; | |
982 | struct bio_vec *bvec; | |
76029ff3 | 983 | struct bio_map_data *bmd; |
68154e90 | 984 | int i, ret; |
76029ff3 FT |
985 | struct sg_iovec iov; |
986 | ||
987 | iov.iov_base = data; | |
988 | iov.iov_len = len; | |
989 | ||
990 | bmd = bio_alloc_map_data(nr_pages, 1, gfp_mask); | |
991 | if (!bmd) | |
992 | return ERR_PTR(-ENOMEM); | |
68154e90 | 993 | |
76029ff3 | 994 | ret = -ENOMEM; |
68154e90 FT |
995 | bio = bio_alloc(gfp_mask, nr_pages); |
996 | if (!bio) | |
76029ff3 | 997 | goto out_bmd; |
68154e90 FT |
998 | |
999 | while (len) { | |
1000 | struct page *page; | |
1001 | unsigned int bytes = PAGE_SIZE; | |
1002 | ||
1003 | if (bytes > len) | |
1004 | bytes = len; | |
1005 | ||
1006 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1007 | if (!page) { | |
1008 | ret = -ENOMEM; | |
1009 | goto cleanup; | |
1010 | } | |
1011 | ||
1012 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) { | |
1013 | ret = -EINVAL; | |
1014 | goto cleanup; | |
1015 | } | |
1016 | ||
1017 | len -= bytes; | |
1018 | } | |
1019 | ||
1020 | if (!reading) { | |
1021 | void *p = data; | |
1022 | ||
1023 | bio_for_each_segment(bvec, bio, i) { | |
1024 | char *addr = page_address(bvec->bv_page); | |
1025 | ||
1026 | memcpy(addr, p, bvec->bv_len); | |
1027 | p += bvec->bv_len; | |
1028 | } | |
1029 | } | |
1030 | ||
76029ff3 | 1031 | bio->bi_private = bmd; |
68154e90 | 1032 | bio->bi_end_io = bio_copy_kern_endio; |
76029ff3 FT |
1033 | |
1034 | bio_set_map_data(bmd, bio, &iov, 1); | |
68154e90 FT |
1035 | return bio; |
1036 | cleanup: | |
1037 | bio_for_each_segment(bvec, bio, i) | |
1038 | __free_page(bvec->bv_page); | |
1039 | ||
1040 | bio_put(bio); | |
76029ff3 FT |
1041 | out_bmd: |
1042 | bio_free_map_data(bmd); | |
68154e90 FT |
1043 | |
1044 | return ERR_PTR(ret); | |
1045 | } | |
1046 | ||
1da177e4 LT |
1047 | /* |
1048 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1049 | * for performing direct-IO in BIOs. | |
1050 | * | |
1051 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1052 | * because the required locks are not interrupt-safe. So what we can do is to | |
1053 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1054 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1055 | * in process context. | |
1056 | * | |
1057 | * We special-case compound pages here: normally this means reads into hugetlb | |
1058 | * pages. The logic in here doesn't really work right for compound pages | |
1059 | * because the VM does not uniformly chase down the head page in all cases. | |
1060 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1061 | * handle them at all. So we skip compound pages here at an early stage. | |
1062 | * | |
1063 | * Note that this code is very hard to test under normal circumstances because | |
1064 | * direct-io pins the pages with get_user_pages(). This makes | |
1065 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
1066 | * But other code (eg, pdflush) could clean the pages if they are mapped | |
1067 | * pagecache. | |
1068 | * | |
1069 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1070 | * deferred bio dirtying paths. | |
1071 | */ | |
1072 | ||
1073 | /* | |
1074 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1075 | */ | |
1076 | void bio_set_pages_dirty(struct bio *bio) | |
1077 | { | |
1078 | struct bio_vec *bvec = bio->bi_io_vec; | |
1079 | int i; | |
1080 | ||
1081 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1082 | struct page *page = bvec[i].bv_page; | |
1083 | ||
1084 | if (page && !PageCompound(page)) | |
1085 | set_page_dirty_lock(page); | |
1086 | } | |
1087 | } | |
1088 | ||
86b6c7a7 | 1089 | static void bio_release_pages(struct bio *bio) |
1da177e4 LT |
1090 | { |
1091 | struct bio_vec *bvec = bio->bi_io_vec; | |
1092 | int i; | |
1093 | ||
1094 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1095 | struct page *page = bvec[i].bv_page; | |
1096 | ||
1097 | if (page) | |
1098 | put_page(page); | |
1099 | } | |
1100 | } | |
1101 | ||
1102 | /* | |
1103 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1104 | * If they are, then fine. If, however, some pages are clean then they must | |
1105 | * have been written out during the direct-IO read. So we take another ref on | |
1106 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1107 | * | |
1108 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1109 | * here on. It will run one page_cache_release() against each page and will | |
1110 | * run one bio_put() against the BIO. | |
1111 | */ | |
1112 | ||
65f27f38 | 1113 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1114 | |
65f27f38 | 1115 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1116 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1117 | static struct bio *bio_dirty_list; | |
1118 | ||
1119 | /* | |
1120 | * This runs in process context | |
1121 | */ | |
65f27f38 | 1122 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 LT |
1123 | { |
1124 | unsigned long flags; | |
1125 | struct bio *bio; | |
1126 | ||
1127 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1128 | bio = bio_dirty_list; | |
1129 | bio_dirty_list = NULL; | |
1130 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1131 | ||
1132 | while (bio) { | |
1133 | struct bio *next = bio->bi_private; | |
1134 | ||
1135 | bio_set_pages_dirty(bio); | |
1136 | bio_release_pages(bio); | |
1137 | bio_put(bio); | |
1138 | bio = next; | |
1139 | } | |
1140 | } | |
1141 | ||
1142 | void bio_check_pages_dirty(struct bio *bio) | |
1143 | { | |
1144 | struct bio_vec *bvec = bio->bi_io_vec; | |
1145 | int nr_clean_pages = 0; | |
1146 | int i; | |
1147 | ||
1148 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1149 | struct page *page = bvec[i].bv_page; | |
1150 | ||
1151 | if (PageDirty(page) || PageCompound(page)) { | |
1152 | page_cache_release(page); | |
1153 | bvec[i].bv_page = NULL; | |
1154 | } else { | |
1155 | nr_clean_pages++; | |
1156 | } | |
1157 | } | |
1158 | ||
1159 | if (nr_clean_pages) { | |
1160 | unsigned long flags; | |
1161 | ||
1162 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1163 | bio->bi_private = bio_dirty_list; | |
1164 | bio_dirty_list = bio; | |
1165 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1166 | schedule_work(&bio_dirty_work); | |
1167 | } else { | |
1168 | bio_put(bio); | |
1169 | } | |
1170 | } | |
1171 | ||
1172 | /** | |
1173 | * bio_endio - end I/O on a bio | |
1174 | * @bio: bio | |
1da177e4 LT |
1175 | * @error: error, if any |
1176 | * | |
1177 | * Description: | |
6712ecf8 | 1178 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
5bb23a68 N |
1179 | * preferred way to end I/O on a bio, it takes care of clearing |
1180 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1181 | * established -Exxxx (-EIO, for instance) error values in case | |
1182 | * something went wrong. Noone should call bi_end_io() directly on a | |
1183 | * bio unless they own it and thus know that it has an end_io | |
1184 | * function. | |
1da177e4 | 1185 | **/ |
6712ecf8 | 1186 | void bio_endio(struct bio *bio, int error) |
1da177e4 LT |
1187 | { |
1188 | if (error) | |
1189 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
9cc54d40 N |
1190 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1191 | error = -EIO; | |
1da177e4 | 1192 | |
5bb23a68 | 1193 | if (bio->bi_end_io) |
6712ecf8 | 1194 | bio->bi_end_io(bio, error); |
1da177e4 LT |
1195 | } |
1196 | ||
1197 | void bio_pair_release(struct bio_pair *bp) | |
1198 | { | |
1199 | if (atomic_dec_and_test(&bp->cnt)) { | |
1200 | struct bio *master = bp->bio1.bi_private; | |
1201 | ||
6712ecf8 | 1202 | bio_endio(master, bp->error); |
1da177e4 LT |
1203 | mempool_free(bp, bp->bio2.bi_private); |
1204 | } | |
1205 | } | |
1206 | ||
6712ecf8 | 1207 | static void bio_pair_end_1(struct bio *bi, int err) |
1da177e4 LT |
1208 | { |
1209 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); | |
1210 | ||
1211 | if (err) | |
1212 | bp->error = err; | |
1213 | ||
1da177e4 | 1214 | bio_pair_release(bp); |
1da177e4 LT |
1215 | } |
1216 | ||
6712ecf8 | 1217 | static void bio_pair_end_2(struct bio *bi, int err) |
1da177e4 LT |
1218 | { |
1219 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); | |
1220 | ||
1221 | if (err) | |
1222 | bp->error = err; | |
1223 | ||
1da177e4 | 1224 | bio_pair_release(bp); |
1da177e4 LT |
1225 | } |
1226 | ||
1227 | /* | |
1228 | * split a bio - only worry about a bio with a single page | |
1229 | * in it's iovec | |
1230 | */ | |
1231 | struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors) | |
1232 | { | |
1233 | struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO); | |
1234 | ||
1235 | if (!bp) | |
1236 | return bp; | |
1237 | ||
2056a782 JA |
1238 | blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi, |
1239 | bi->bi_sector + first_sectors); | |
1240 | ||
1da177e4 LT |
1241 | BUG_ON(bi->bi_vcnt != 1); |
1242 | BUG_ON(bi->bi_idx != 0); | |
1243 | atomic_set(&bp->cnt, 3); | |
1244 | bp->error = 0; | |
1245 | bp->bio1 = *bi; | |
1246 | bp->bio2 = *bi; | |
1247 | bp->bio2.bi_sector += first_sectors; | |
1248 | bp->bio2.bi_size -= first_sectors << 9; | |
1249 | bp->bio1.bi_size = first_sectors << 9; | |
1250 | ||
1251 | bp->bv1 = bi->bi_io_vec[0]; | |
1252 | bp->bv2 = bi->bi_io_vec[0]; | |
1253 | bp->bv2.bv_offset += first_sectors << 9; | |
1254 | bp->bv2.bv_len -= first_sectors << 9; | |
1255 | bp->bv1.bv_len = first_sectors << 9; | |
1256 | ||
1257 | bp->bio1.bi_io_vec = &bp->bv1; | |
1258 | bp->bio2.bi_io_vec = &bp->bv2; | |
1259 | ||
a2eb0c10 N |
1260 | bp->bio1.bi_max_vecs = 1; |
1261 | bp->bio2.bi_max_vecs = 1; | |
1262 | ||
1da177e4 LT |
1263 | bp->bio1.bi_end_io = bio_pair_end_1; |
1264 | bp->bio2.bi_end_io = bio_pair_end_2; | |
1265 | ||
1266 | bp->bio1.bi_private = bi; | |
1267 | bp->bio2.bi_private = pool; | |
1268 | ||
7ba1ba12 MP |
1269 | if (bio_integrity(bi)) |
1270 | bio_integrity_split(bi, bp, first_sectors); | |
1271 | ||
1da177e4 LT |
1272 | return bp; |
1273 | } | |
1274 | ||
1da177e4 LT |
1275 | |
1276 | /* | |
1277 | * create memory pools for biovec's in a bio_set. | |
1278 | * use the global biovec slabs created for general use. | |
1279 | */ | |
5972511b | 1280 | static int biovec_create_pools(struct bio_set *bs, int pool_entries) |
1da177e4 LT |
1281 | { |
1282 | int i; | |
1283 | ||
1284 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1285 | struct biovec_slab *bp = bvec_slabs + i; | |
1286 | mempool_t **bvp = bs->bvec_pools + i; | |
1287 | ||
93d2341c | 1288 | *bvp = mempool_create_slab_pool(pool_entries, bp->slab); |
1da177e4 LT |
1289 | if (!*bvp) |
1290 | return -ENOMEM; | |
1291 | } | |
1292 | return 0; | |
1293 | } | |
1294 | ||
1295 | static void biovec_free_pools(struct bio_set *bs) | |
1296 | { | |
1297 | int i; | |
1298 | ||
1299 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1300 | mempool_t *bvp = bs->bvec_pools[i]; | |
1301 | ||
1302 | if (bvp) | |
1303 | mempool_destroy(bvp); | |
1304 | } | |
1305 | ||
1306 | } | |
1307 | ||
1308 | void bioset_free(struct bio_set *bs) | |
1309 | { | |
1310 | if (bs->bio_pool) | |
1311 | mempool_destroy(bs->bio_pool); | |
1312 | ||
7ba1ba12 | 1313 | bioset_integrity_free(bs); |
1da177e4 LT |
1314 | biovec_free_pools(bs); |
1315 | ||
1316 | kfree(bs); | |
1317 | } | |
1318 | ||
5972511b | 1319 | struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size) |
1da177e4 | 1320 | { |
11b0b5ab | 1321 | struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1da177e4 LT |
1322 | |
1323 | if (!bs) | |
1324 | return NULL; | |
1325 | ||
93d2341c | 1326 | bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab); |
1da177e4 LT |
1327 | if (!bs->bio_pool) |
1328 | goto bad; | |
1329 | ||
7ba1ba12 MP |
1330 | if (bioset_integrity_create(bs, bio_pool_size)) |
1331 | goto bad; | |
1332 | ||
5972511b | 1333 | if (!biovec_create_pools(bs, bvec_pool_size)) |
1da177e4 LT |
1334 | return bs; |
1335 | ||
1336 | bad: | |
1337 | bioset_free(bs); | |
1338 | return NULL; | |
1339 | } | |
1340 | ||
1341 | static void __init biovec_init_slabs(void) | |
1342 | { | |
1343 | int i; | |
1344 | ||
1345 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1346 | int size; | |
1347 | struct biovec_slab *bvs = bvec_slabs + i; | |
1348 | ||
1349 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
1350 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 1351 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
1352 | } |
1353 | } | |
1354 | ||
1355 | static int __init init_bio(void) | |
1356 | { | |
0a31bd5f | 1357 | bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
1da177e4 | 1358 | |
7ba1ba12 | 1359 | bio_integrity_init_slab(); |
1da177e4 LT |
1360 | biovec_init_slabs(); |
1361 | ||
5972511b | 1362 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 2); |
1da177e4 LT |
1363 | if (!fs_bio_set) |
1364 | panic("bio: can't allocate bios\n"); | |
1365 | ||
0eaae62a MD |
1366 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
1367 | sizeof(struct bio_pair)); | |
1da177e4 LT |
1368 | if (!bio_split_pool) |
1369 | panic("bio: can't create split pool\n"); | |
1370 | ||
1371 | return 0; | |
1372 | } | |
1373 | ||
1374 | subsys_initcall(init_bio); | |
1375 | ||
1376 | EXPORT_SYMBOL(bio_alloc); | |
1377 | EXPORT_SYMBOL(bio_put); | |
3676347a | 1378 | EXPORT_SYMBOL(bio_free); |
1da177e4 LT |
1379 | EXPORT_SYMBOL(bio_endio); |
1380 | EXPORT_SYMBOL(bio_init); | |
1381 | EXPORT_SYMBOL(__bio_clone); | |
1382 | EXPORT_SYMBOL(bio_clone); | |
1383 | EXPORT_SYMBOL(bio_phys_segments); | |
1384 | EXPORT_SYMBOL(bio_hw_segments); | |
1385 | EXPORT_SYMBOL(bio_add_page); | |
6e68af66 | 1386 | EXPORT_SYMBOL(bio_add_pc_page); |
1da177e4 | 1387 | EXPORT_SYMBOL(bio_get_nr_vecs); |
40044ce0 JA |
1388 | EXPORT_SYMBOL(bio_map_user); |
1389 | EXPORT_SYMBOL(bio_unmap_user); | |
df46b9a4 | 1390 | EXPORT_SYMBOL(bio_map_kern); |
68154e90 | 1391 | EXPORT_SYMBOL(bio_copy_kern); |
1da177e4 LT |
1392 | EXPORT_SYMBOL(bio_pair_release); |
1393 | EXPORT_SYMBOL(bio_split); | |
1394 | EXPORT_SYMBOL(bio_split_pool); | |
1395 | EXPORT_SYMBOL(bio_copy_user); | |
1396 | EXPORT_SYMBOL(bio_uncopy_user); | |
1397 | EXPORT_SYMBOL(bioset_create); | |
1398 | EXPORT_SYMBOL(bioset_free); | |
1399 | EXPORT_SYMBOL(bio_alloc_bioset); |