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