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 | |
51d654e1 | 53 | struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs) |
1da177e4 LT |
54 | { |
55 | struct bio_vec *bvl; | |
1da177e4 LT |
56 | |
57 | /* | |
58 | * see comment near bvec_array define! | |
59 | */ | |
60 | switch (nr) { | |
61 | case 1 : *idx = 0; break; | |
62 | case 2 ... 4: *idx = 1; break; | |
63 | case 5 ... 16: *idx = 2; break; | |
64 | case 17 ... 64: *idx = 3; break; | |
65 | case 65 ... 128: *idx = 4; break; | |
66 | case 129 ... BIO_MAX_PAGES: *idx = 5; break; | |
67 | default: | |
68 | return NULL; | |
69 | } | |
70 | /* | |
71 | * idx now points to the pool we want to allocate from | |
72 | */ | |
73 | ||
1da177e4 | 74 | bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask); |
bf02c082 AM |
75 | if (bvl) { |
76 | struct biovec_slab *bp = bvec_slabs + *idx; | |
77 | ||
1da177e4 | 78 | memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec)); |
bf02c082 | 79 | } |
1da177e4 LT |
80 | |
81 | return bvl; | |
82 | } | |
83 | ||
3676347a | 84 | void bio_free(struct bio *bio, struct bio_set *bio_set) |
1da177e4 | 85 | { |
992c5dda JA |
86 | if (bio->bi_io_vec) { |
87 | const int pool_idx = BIO_POOL_IDX(bio); | |
1da177e4 | 88 | |
992c5dda JA |
89 | BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); |
90 | ||
91 | mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]); | |
92 | } | |
1da177e4 | 93 | |
3676347a PO |
94 | mempool_free(bio, bio_set->bio_pool); |
95 | } | |
96 | ||
97 | /* | |
98 | * default destructor for a bio allocated with bio_alloc_bioset() | |
99 | */ | |
100 | static void bio_fs_destructor(struct bio *bio) | |
101 | { | |
102 | bio_free(bio, fs_bio_set); | |
1da177e4 LT |
103 | } |
104 | ||
858119e1 | 105 | void bio_init(struct bio *bio) |
1da177e4 | 106 | { |
2b94de55 | 107 | memset(bio, 0, sizeof(*bio)); |
1da177e4 | 108 | bio->bi_flags = 1 << BIO_UPTODATE; |
1da177e4 | 109 | atomic_set(&bio->bi_cnt, 1); |
1da177e4 LT |
110 | } |
111 | ||
112 | /** | |
113 | * bio_alloc_bioset - allocate a bio for I/O | |
114 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
115 | * @nr_iovecs: number of iovecs to pre-allocate | |
67be2dd1 | 116 | * @bs: the bio_set to allocate from |
1da177e4 LT |
117 | * |
118 | * Description: | |
119 | * bio_alloc_bioset will first try it's on mempool to satisfy the allocation. | |
120 | * If %__GFP_WAIT is set then we will block on the internal pool waiting | |
121 | * for a &struct bio to become free. | |
122 | * | |
123 | * allocate bio and iovecs from the memory pools specified by the | |
124 | * bio_set structure. | |
125 | **/ | |
dd0fc66f | 126 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) |
1da177e4 LT |
127 | { |
128 | struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask); | |
129 | ||
130 | if (likely(bio)) { | |
131 | struct bio_vec *bvl = NULL; | |
132 | ||
133 | bio_init(bio); | |
134 | if (likely(nr_iovecs)) { | |
eeae1d48 | 135 | unsigned long uninitialized_var(idx); |
1da177e4 LT |
136 | |
137 | bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); | |
138 | if (unlikely(!bvl)) { | |
139 | mempool_free(bio, bs->bio_pool); | |
140 | bio = NULL; | |
141 | goto out; | |
142 | } | |
143 | bio->bi_flags |= idx << BIO_POOL_OFFSET; | |
144 | bio->bi_max_vecs = bvec_slabs[idx].nr_vecs; | |
145 | } | |
146 | bio->bi_io_vec = bvl; | |
1da177e4 LT |
147 | } |
148 | out: | |
149 | return bio; | |
150 | } | |
151 | ||
dd0fc66f | 152 | struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs) |
1da177e4 | 153 | { |
3676347a PO |
154 | struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set); |
155 | ||
156 | if (bio) | |
157 | bio->bi_destructor = bio_fs_destructor; | |
158 | ||
159 | return bio; | |
1da177e4 LT |
160 | } |
161 | ||
162 | void zero_fill_bio(struct bio *bio) | |
163 | { | |
164 | unsigned long flags; | |
165 | struct bio_vec *bv; | |
166 | int i; | |
167 | ||
168 | bio_for_each_segment(bv, bio, i) { | |
169 | char *data = bvec_kmap_irq(bv, &flags); | |
170 | memset(data, 0, bv->bv_len); | |
171 | flush_dcache_page(bv->bv_page); | |
172 | bvec_kunmap_irq(data, &flags); | |
173 | } | |
174 | } | |
175 | EXPORT_SYMBOL(zero_fill_bio); | |
176 | ||
177 | /** | |
178 | * bio_put - release a reference to a bio | |
179 | * @bio: bio to release reference to | |
180 | * | |
181 | * Description: | |
182 | * Put a reference to a &struct bio, either one you have gotten with | |
183 | * bio_alloc or bio_get. The last put of a bio will free it. | |
184 | **/ | |
185 | void bio_put(struct bio *bio) | |
186 | { | |
187 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); | |
188 | ||
189 | /* | |
190 | * last put frees it | |
191 | */ | |
192 | if (atomic_dec_and_test(&bio->bi_cnt)) { | |
193 | bio->bi_next = NULL; | |
194 | bio->bi_destructor(bio); | |
195 | } | |
196 | } | |
197 | ||
165125e1 | 198 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
199 | { |
200 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
201 | blk_recount_segments(q, bio); | |
202 | ||
203 | return bio->bi_phys_segments; | |
204 | } | |
205 | ||
165125e1 | 206 | inline int bio_hw_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
207 | { |
208 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
209 | blk_recount_segments(q, bio); | |
210 | ||
211 | return bio->bi_hw_segments; | |
212 | } | |
213 | ||
214 | /** | |
215 | * __bio_clone - clone a bio | |
216 | * @bio: destination bio | |
217 | * @bio_src: bio to clone | |
218 | * | |
219 | * Clone a &bio. Caller will own the returned bio, but not | |
220 | * the actual data it points to. Reference count of returned | |
221 | * bio will be one. | |
222 | */ | |
858119e1 | 223 | void __bio_clone(struct bio *bio, struct bio *bio_src) |
1da177e4 | 224 | { |
e525e153 AM |
225 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, |
226 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); | |
1da177e4 | 227 | |
5d84070e JA |
228 | /* |
229 | * most users will be overriding ->bi_bdev with a new target, | |
230 | * so we don't set nor calculate new physical/hw segment counts here | |
231 | */ | |
1da177e4 LT |
232 | bio->bi_sector = bio_src->bi_sector; |
233 | bio->bi_bdev = bio_src->bi_bdev; | |
234 | bio->bi_flags |= 1 << BIO_CLONED; | |
235 | bio->bi_rw = bio_src->bi_rw; | |
1da177e4 LT |
236 | bio->bi_vcnt = bio_src->bi_vcnt; |
237 | bio->bi_size = bio_src->bi_size; | |
a5453be4 | 238 | bio->bi_idx = bio_src->bi_idx; |
1da177e4 LT |
239 | } |
240 | ||
241 | /** | |
242 | * bio_clone - clone a bio | |
243 | * @bio: bio to clone | |
244 | * @gfp_mask: allocation priority | |
245 | * | |
246 | * Like __bio_clone, only also allocates the returned bio | |
247 | */ | |
dd0fc66f | 248 | struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
249 | { |
250 | struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); | |
251 | ||
3676347a PO |
252 | if (b) { |
253 | b->bi_destructor = bio_fs_destructor; | |
1da177e4 | 254 | __bio_clone(b, bio); |
3676347a | 255 | } |
1da177e4 LT |
256 | |
257 | return b; | |
258 | } | |
259 | ||
260 | /** | |
261 | * bio_get_nr_vecs - return approx number of vecs | |
262 | * @bdev: I/O target | |
263 | * | |
264 | * Return the approximate number of pages we can send to this target. | |
265 | * There's no guarantee that you will be able to fit this number of pages | |
266 | * into a bio, it does not account for dynamic restrictions that vary | |
267 | * on offset. | |
268 | */ | |
269 | int bio_get_nr_vecs(struct block_device *bdev) | |
270 | { | |
165125e1 | 271 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
272 | int nr_pages; |
273 | ||
274 | nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
275 | if (nr_pages > q->max_phys_segments) | |
276 | nr_pages = q->max_phys_segments; | |
277 | if (nr_pages > q->max_hw_segments) | |
278 | nr_pages = q->max_hw_segments; | |
279 | ||
280 | return nr_pages; | |
281 | } | |
282 | ||
165125e1 | 283 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page |
defd94b7 MC |
284 | *page, unsigned int len, unsigned int offset, |
285 | unsigned short max_sectors) | |
1da177e4 LT |
286 | { |
287 | int retried_segments = 0; | |
288 | struct bio_vec *bvec; | |
289 | ||
290 | /* | |
291 | * cloned bio must not modify vec list | |
292 | */ | |
293 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
294 | return 0; | |
295 | ||
80cfd548 | 296 | if (((bio->bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
297 | return 0; |
298 | ||
80cfd548 JA |
299 | /* |
300 | * For filesystems with a blocksize smaller than the pagesize | |
301 | * we will often be called with the same page as last time and | |
302 | * a consecutive offset. Optimize this special case. | |
303 | */ | |
304 | if (bio->bi_vcnt > 0) { | |
305 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
306 | ||
307 | if (page == prev->bv_page && | |
308 | offset == prev->bv_offset + prev->bv_len) { | |
309 | prev->bv_len += len; | |
310 | if (q->merge_bvec_fn && | |
311 | q->merge_bvec_fn(q, bio, prev) < len) { | |
312 | prev->bv_len -= len; | |
313 | return 0; | |
314 | } | |
315 | ||
316 | goto done; | |
317 | } | |
318 | } | |
319 | ||
320 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
1da177e4 LT |
321 | return 0; |
322 | ||
323 | /* | |
324 | * we might lose a segment or two here, but rather that than | |
325 | * make this too complex. | |
326 | */ | |
327 | ||
328 | while (bio->bi_phys_segments >= q->max_phys_segments | |
329 | || bio->bi_hw_segments >= q->max_hw_segments | |
330 | || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) { | |
331 | ||
332 | if (retried_segments) | |
333 | return 0; | |
334 | ||
335 | retried_segments = 1; | |
336 | blk_recount_segments(q, bio); | |
337 | } | |
338 | ||
339 | /* | |
340 | * setup the new entry, we might clear it again later if we | |
341 | * cannot add the page | |
342 | */ | |
343 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
344 | bvec->bv_page = page; | |
345 | bvec->bv_len = len; | |
346 | bvec->bv_offset = offset; | |
347 | ||
348 | /* | |
349 | * if queue has other restrictions (eg varying max sector size | |
350 | * depending on offset), it can specify a merge_bvec_fn in the | |
351 | * queue to get further control | |
352 | */ | |
353 | if (q->merge_bvec_fn) { | |
354 | /* | |
355 | * merge_bvec_fn() returns number of bytes it can accept | |
356 | * at this offset | |
357 | */ | |
358 | if (q->merge_bvec_fn(q, bio, bvec) < len) { | |
359 | bvec->bv_page = NULL; | |
360 | bvec->bv_len = 0; | |
361 | bvec->bv_offset = 0; | |
362 | return 0; | |
363 | } | |
364 | } | |
365 | ||
366 | /* If we may be able to merge these biovecs, force a recount */ | |
367 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) || | |
368 | BIOVEC_VIRT_MERGEABLE(bvec-1, bvec))) | |
369 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
370 | ||
371 | bio->bi_vcnt++; | |
372 | bio->bi_phys_segments++; | |
373 | bio->bi_hw_segments++; | |
80cfd548 | 374 | done: |
1da177e4 LT |
375 | bio->bi_size += len; |
376 | return len; | |
377 | } | |
378 | ||
6e68af66 MC |
379 | /** |
380 | * bio_add_pc_page - attempt to add page to bio | |
fddfdeaf | 381 | * @q: the target queue |
6e68af66 MC |
382 | * @bio: destination bio |
383 | * @page: page to add | |
384 | * @len: vec entry length | |
385 | * @offset: vec entry offset | |
386 | * | |
387 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
388 | * number of reasons, such as the bio being full or target block | |
389 | * device limitations. The target block device must allow bio's | |
390 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
391 | * page to an empty bio. This should only be used by REQ_PC bios. | |
392 | */ | |
165125e1 | 393 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
6e68af66 MC |
394 | unsigned int len, unsigned int offset) |
395 | { | |
defd94b7 | 396 | return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors); |
6e68af66 MC |
397 | } |
398 | ||
1da177e4 LT |
399 | /** |
400 | * bio_add_page - attempt to add page to bio | |
401 | * @bio: destination bio | |
402 | * @page: page to add | |
403 | * @len: vec entry length | |
404 | * @offset: vec entry offset | |
405 | * | |
406 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
407 | * number of reasons, such as the bio being full or target block | |
408 | * device limitations. The target block device must allow bio's | |
409 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
410 | * page to an empty bio. | |
411 | */ | |
412 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
413 | unsigned int offset) | |
414 | { | |
defd94b7 MC |
415 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
416 | return __bio_add_page(q, bio, page, len, offset, q->max_sectors); | |
1da177e4 LT |
417 | } |
418 | ||
419 | struct bio_map_data { | |
420 | struct bio_vec *iovecs; | |
c5dec1c3 FT |
421 | int nr_sgvecs; |
422 | struct sg_iovec *sgvecs; | |
1da177e4 LT |
423 | }; |
424 | ||
c5dec1c3 FT |
425 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
426 | struct sg_iovec *iov, int iov_count) | |
1da177e4 LT |
427 | { |
428 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); | |
c5dec1c3 FT |
429 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
430 | bmd->nr_sgvecs = iov_count; | |
1da177e4 LT |
431 | bio->bi_private = bmd; |
432 | } | |
433 | ||
434 | static void bio_free_map_data(struct bio_map_data *bmd) | |
435 | { | |
436 | kfree(bmd->iovecs); | |
c5dec1c3 | 437 | kfree(bmd->sgvecs); |
1da177e4 LT |
438 | kfree(bmd); |
439 | } | |
440 | ||
c5dec1c3 | 441 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count) |
1da177e4 LT |
442 | { |
443 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL); | |
444 | ||
445 | if (!bmd) | |
446 | return NULL; | |
447 | ||
448 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL); | |
c5dec1c3 FT |
449 | if (!bmd->iovecs) { |
450 | kfree(bmd); | |
451 | return NULL; | |
452 | } | |
453 | ||
454 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, GFP_KERNEL); | |
455 | if (bmd->sgvecs) | |
1da177e4 LT |
456 | return bmd; |
457 | ||
c5dec1c3 | 458 | kfree(bmd->iovecs); |
1da177e4 LT |
459 | kfree(bmd); |
460 | return NULL; | |
461 | } | |
462 | ||
c5dec1c3 FT |
463 | static int __bio_copy_iov(struct bio *bio, struct sg_iovec *iov, int iov_count, |
464 | int uncopy) | |
465 | { | |
466 | int ret = 0, i; | |
467 | struct bio_vec *bvec; | |
468 | int iov_idx = 0; | |
469 | unsigned int iov_off = 0; | |
470 | int read = bio_data_dir(bio) == READ; | |
471 | ||
472 | __bio_for_each_segment(bvec, bio, i, 0) { | |
473 | char *bv_addr = page_address(bvec->bv_page); | |
474 | unsigned int bv_len = bvec->bv_len; | |
475 | ||
476 | while (bv_len && iov_idx < iov_count) { | |
477 | unsigned int bytes; | |
478 | char *iov_addr; | |
479 | ||
480 | bytes = min_t(unsigned int, | |
481 | iov[iov_idx].iov_len - iov_off, bv_len); | |
482 | iov_addr = iov[iov_idx].iov_base + iov_off; | |
483 | ||
484 | if (!ret) { | |
485 | if (!read && !uncopy) | |
486 | ret = copy_from_user(bv_addr, iov_addr, | |
487 | bytes); | |
488 | if (read && uncopy) | |
489 | ret = copy_to_user(iov_addr, bv_addr, | |
490 | bytes); | |
491 | ||
492 | if (ret) | |
493 | ret = -EFAULT; | |
494 | } | |
495 | ||
496 | bv_len -= bytes; | |
497 | bv_addr += bytes; | |
498 | iov_addr += bytes; | |
499 | iov_off += bytes; | |
500 | ||
501 | if (iov[iov_idx].iov_len == iov_off) { | |
502 | iov_idx++; | |
503 | iov_off = 0; | |
504 | } | |
505 | } | |
506 | ||
507 | if (uncopy) | |
508 | __free_page(bvec->bv_page); | |
509 | } | |
510 | ||
511 | return ret; | |
512 | } | |
513 | ||
1da177e4 LT |
514 | /** |
515 | * bio_uncopy_user - finish previously mapped bio | |
516 | * @bio: bio being terminated | |
517 | * | |
518 | * Free pages allocated from bio_copy_user() and write back data | |
519 | * to user space in case of a read. | |
520 | */ | |
521 | int bio_uncopy_user(struct bio *bio) | |
522 | { | |
523 | struct bio_map_data *bmd = bio->bi_private; | |
c5dec1c3 | 524 | int ret; |
1da177e4 | 525 | |
c5dec1c3 | 526 | ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs, 1); |
1da177e4 | 527 | |
1da177e4 LT |
528 | bio_free_map_data(bmd); |
529 | bio_put(bio); | |
530 | return ret; | |
531 | } | |
532 | ||
533 | /** | |
c5dec1c3 | 534 | * bio_copy_user_iov - copy user data to bio |
1da177e4 | 535 | * @q: destination block queue |
c5dec1c3 FT |
536 | * @iov: the iovec. |
537 | * @iov_count: number of elements in the iovec | |
1da177e4 LT |
538 | * @write_to_vm: bool indicating writing to pages or not |
539 | * | |
540 | * Prepares and returns a bio for indirect user io, bouncing data | |
541 | * to/from kernel pages as necessary. Must be paired with | |
542 | * call bio_uncopy_user() on io completion. | |
543 | */ | |
c5dec1c3 FT |
544 | struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov, |
545 | int iov_count, int write_to_vm) | |
1da177e4 | 546 | { |
1da177e4 LT |
547 | struct bio_map_data *bmd; |
548 | struct bio_vec *bvec; | |
549 | struct page *page; | |
550 | struct bio *bio; | |
551 | int i, ret; | |
c5dec1c3 FT |
552 | int nr_pages = 0; |
553 | unsigned int len = 0; | |
1da177e4 | 554 | |
c5dec1c3 FT |
555 | for (i = 0; i < iov_count; i++) { |
556 | unsigned long uaddr; | |
557 | unsigned long end; | |
558 | unsigned long start; | |
559 | ||
560 | uaddr = (unsigned long)iov[i].iov_base; | |
561 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
562 | start = uaddr >> PAGE_SHIFT; | |
563 | ||
564 | nr_pages += end - start; | |
565 | len += iov[i].iov_len; | |
566 | } | |
567 | ||
568 | bmd = bio_alloc_map_data(nr_pages, iov_count); | |
1da177e4 LT |
569 | if (!bmd) |
570 | return ERR_PTR(-ENOMEM); | |
571 | ||
1da177e4 | 572 | ret = -ENOMEM; |
c5dec1c3 | 573 | bio = bio_alloc(GFP_KERNEL, nr_pages); |
1da177e4 LT |
574 | if (!bio) |
575 | goto out_bmd; | |
576 | ||
577 | bio->bi_rw |= (!write_to_vm << BIO_RW); | |
578 | ||
579 | ret = 0; | |
580 | while (len) { | |
581 | unsigned int bytes = PAGE_SIZE; | |
582 | ||
583 | if (bytes > len) | |
584 | bytes = len; | |
585 | ||
586 | page = alloc_page(q->bounce_gfp | GFP_KERNEL); | |
587 | if (!page) { | |
588 | ret = -ENOMEM; | |
589 | break; | |
590 | } | |
591 | ||
0e75f906 | 592 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) |
1da177e4 | 593 | break; |
1da177e4 LT |
594 | |
595 | len -= bytes; | |
596 | } | |
597 | ||
598 | if (ret) | |
599 | goto cleanup; | |
600 | ||
601 | /* | |
602 | * success | |
603 | */ | |
604 | if (!write_to_vm) { | |
c5dec1c3 FT |
605 | ret = __bio_copy_iov(bio, iov, iov_count, 0); |
606 | if (ret) | |
607 | goto cleanup; | |
1da177e4 LT |
608 | } |
609 | ||
c5dec1c3 | 610 | bio_set_map_data(bmd, bio, iov, iov_count); |
1da177e4 LT |
611 | return bio; |
612 | cleanup: | |
613 | bio_for_each_segment(bvec, bio, i) | |
614 | __free_page(bvec->bv_page); | |
615 | ||
616 | bio_put(bio); | |
617 | out_bmd: | |
618 | bio_free_map_data(bmd); | |
619 | return ERR_PTR(ret); | |
620 | } | |
621 | ||
c5dec1c3 FT |
622 | /** |
623 | * bio_copy_user - copy user data to bio | |
624 | * @q: destination block queue | |
625 | * @uaddr: start of user address | |
626 | * @len: length in bytes | |
627 | * @write_to_vm: bool indicating writing to pages or not | |
628 | * | |
629 | * Prepares and returns a bio for indirect user io, bouncing data | |
630 | * to/from kernel pages as necessary. Must be paired with | |
631 | * call bio_uncopy_user() on io completion. | |
632 | */ | |
633 | struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr, | |
634 | unsigned int len, int write_to_vm) | |
635 | { | |
636 | struct sg_iovec iov; | |
637 | ||
638 | iov.iov_base = (void __user *)uaddr; | |
639 | iov.iov_len = len; | |
640 | ||
641 | return bio_copy_user_iov(q, &iov, 1, write_to_vm); | |
642 | } | |
643 | ||
165125e1 | 644 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
f1970baf JB |
645 | struct block_device *bdev, |
646 | struct sg_iovec *iov, int iov_count, | |
647 | int write_to_vm) | |
1da177e4 | 648 | { |
f1970baf JB |
649 | int i, j; |
650 | int nr_pages = 0; | |
1da177e4 LT |
651 | struct page **pages; |
652 | struct bio *bio; | |
f1970baf JB |
653 | int cur_page = 0; |
654 | int ret, offset; | |
1da177e4 | 655 | |
f1970baf JB |
656 | for (i = 0; i < iov_count; i++) { |
657 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
658 | unsigned long len = iov[i].iov_len; | |
659 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
660 | unsigned long start = uaddr >> PAGE_SHIFT; | |
661 | ||
662 | nr_pages += end - start; | |
663 | /* | |
ad2d7225 | 664 | * buffer must be aligned to at least hardsector size for now |
f1970baf | 665 | */ |
ad2d7225 | 666 | if (uaddr & queue_dma_alignment(q)) |
f1970baf JB |
667 | return ERR_PTR(-EINVAL); |
668 | } | |
669 | ||
670 | if (!nr_pages) | |
1da177e4 LT |
671 | return ERR_PTR(-EINVAL); |
672 | ||
673 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
674 | if (!bio) | |
675 | return ERR_PTR(-ENOMEM); | |
676 | ||
677 | ret = -ENOMEM; | |
11b0b5ab | 678 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); |
1da177e4 LT |
679 | if (!pages) |
680 | goto out; | |
681 | ||
f1970baf JB |
682 | for (i = 0; i < iov_count; i++) { |
683 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
684 | unsigned long len = iov[i].iov_len; | |
685 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
686 | unsigned long start = uaddr >> PAGE_SHIFT; | |
687 | const int local_nr_pages = end - start; | |
688 | const int page_limit = cur_page + local_nr_pages; | |
689 | ||
690 | down_read(¤t->mm->mmap_sem); | |
691 | ret = get_user_pages(current, current->mm, uaddr, | |
692 | local_nr_pages, | |
693 | write_to_vm, 0, &pages[cur_page], NULL); | |
694 | up_read(¤t->mm->mmap_sem); | |
695 | ||
99172157 JA |
696 | if (ret < local_nr_pages) { |
697 | ret = -EFAULT; | |
f1970baf | 698 | goto out_unmap; |
99172157 | 699 | } |
f1970baf JB |
700 | |
701 | offset = uaddr & ~PAGE_MASK; | |
702 | for (j = cur_page; j < page_limit; j++) { | |
703 | unsigned int bytes = PAGE_SIZE - offset; | |
704 | ||
705 | if (len <= 0) | |
706 | break; | |
707 | ||
708 | if (bytes > len) | |
709 | bytes = len; | |
710 | ||
711 | /* | |
712 | * sorry... | |
713 | */ | |
defd94b7 MC |
714 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
715 | bytes) | |
f1970baf JB |
716 | break; |
717 | ||
718 | len -= bytes; | |
719 | offset = 0; | |
720 | } | |
1da177e4 | 721 | |
f1970baf | 722 | cur_page = j; |
1da177e4 | 723 | /* |
f1970baf | 724 | * release the pages we didn't map into the bio, if any |
1da177e4 | 725 | */ |
f1970baf JB |
726 | while (j < page_limit) |
727 | page_cache_release(pages[j++]); | |
1da177e4 LT |
728 | } |
729 | ||
1da177e4 LT |
730 | kfree(pages); |
731 | ||
732 | /* | |
733 | * set data direction, and check if mapped pages need bouncing | |
734 | */ | |
735 | if (!write_to_vm) | |
736 | bio->bi_rw |= (1 << BIO_RW); | |
737 | ||
f1970baf | 738 | bio->bi_bdev = bdev; |
1da177e4 LT |
739 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
740 | return bio; | |
f1970baf JB |
741 | |
742 | out_unmap: | |
743 | for (i = 0; i < nr_pages; i++) { | |
744 | if(!pages[i]) | |
745 | break; | |
746 | page_cache_release(pages[i]); | |
747 | } | |
748 | out: | |
1da177e4 LT |
749 | kfree(pages); |
750 | bio_put(bio); | |
751 | return ERR_PTR(ret); | |
752 | } | |
753 | ||
754 | /** | |
755 | * bio_map_user - map user address into bio | |
165125e1 | 756 | * @q: the struct request_queue for the bio |
1da177e4 LT |
757 | * @bdev: destination block device |
758 | * @uaddr: start of user address | |
759 | * @len: length in bytes | |
760 | * @write_to_vm: bool indicating writing to pages or not | |
761 | * | |
762 | * Map the user space address into a bio suitable for io to a block | |
763 | * device. Returns an error pointer in case of error. | |
764 | */ | |
165125e1 | 765 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
1da177e4 | 766 | unsigned long uaddr, unsigned int len, int write_to_vm) |
f1970baf JB |
767 | { |
768 | struct sg_iovec iov; | |
769 | ||
3f70353e | 770 | iov.iov_base = (void __user *)uaddr; |
f1970baf JB |
771 | iov.iov_len = len; |
772 | ||
773 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm); | |
774 | } | |
775 | ||
776 | /** | |
777 | * bio_map_user_iov - map user sg_iovec table into bio | |
165125e1 | 778 | * @q: the struct request_queue for the bio |
f1970baf JB |
779 | * @bdev: destination block device |
780 | * @iov: the iovec. | |
781 | * @iov_count: number of elements in the iovec | |
782 | * @write_to_vm: bool indicating writing to pages or not | |
783 | * | |
784 | * Map the user space address into a bio suitable for io to a block | |
785 | * device. Returns an error pointer in case of error. | |
786 | */ | |
165125e1 | 787 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
f1970baf JB |
788 | struct sg_iovec *iov, int iov_count, |
789 | int write_to_vm) | |
1da177e4 LT |
790 | { |
791 | struct bio *bio; | |
792 | ||
f1970baf | 793 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm); |
1da177e4 LT |
794 | |
795 | if (IS_ERR(bio)) | |
796 | return bio; | |
797 | ||
798 | /* | |
799 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
800 | * it would normally disappear when its bi_end_io is run. | |
801 | * however, we need it for the unmap, so grab an extra | |
802 | * reference to it | |
803 | */ | |
804 | bio_get(bio); | |
805 | ||
0e75f906 | 806 | return bio; |
1da177e4 LT |
807 | } |
808 | ||
809 | static void __bio_unmap_user(struct bio *bio) | |
810 | { | |
811 | struct bio_vec *bvec; | |
812 | int i; | |
813 | ||
814 | /* | |
815 | * make sure we dirty pages we wrote to | |
816 | */ | |
817 | __bio_for_each_segment(bvec, bio, i, 0) { | |
818 | if (bio_data_dir(bio) == READ) | |
819 | set_page_dirty_lock(bvec->bv_page); | |
820 | ||
821 | page_cache_release(bvec->bv_page); | |
822 | } | |
823 | ||
824 | bio_put(bio); | |
825 | } | |
826 | ||
827 | /** | |
828 | * bio_unmap_user - unmap a bio | |
829 | * @bio: the bio being unmapped | |
830 | * | |
831 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
832 | * a process context. | |
833 | * | |
834 | * bio_unmap_user() may sleep. | |
835 | */ | |
836 | void bio_unmap_user(struct bio *bio) | |
837 | { | |
838 | __bio_unmap_user(bio); | |
839 | bio_put(bio); | |
840 | } | |
841 | ||
6712ecf8 | 842 | static void bio_map_kern_endio(struct bio *bio, int err) |
b823825e | 843 | { |
b823825e | 844 | bio_put(bio); |
b823825e JA |
845 | } |
846 | ||
847 | ||
165125e1 | 848 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
27496a8c | 849 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 MC |
850 | { |
851 | unsigned long kaddr = (unsigned long)data; | |
852 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
853 | unsigned long start = kaddr >> PAGE_SHIFT; | |
854 | const int nr_pages = end - start; | |
855 | int offset, i; | |
856 | struct bio *bio; | |
857 | ||
858 | bio = bio_alloc(gfp_mask, nr_pages); | |
859 | if (!bio) | |
860 | return ERR_PTR(-ENOMEM); | |
861 | ||
862 | offset = offset_in_page(kaddr); | |
863 | for (i = 0; i < nr_pages; i++) { | |
864 | unsigned int bytes = PAGE_SIZE - offset; | |
865 | ||
866 | if (len <= 0) | |
867 | break; | |
868 | ||
869 | if (bytes > len) | |
870 | bytes = len; | |
871 | ||
defd94b7 MC |
872 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
873 | offset) < bytes) | |
df46b9a4 MC |
874 | break; |
875 | ||
876 | data += bytes; | |
877 | len -= bytes; | |
878 | offset = 0; | |
879 | } | |
880 | ||
b823825e | 881 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
882 | return bio; |
883 | } | |
884 | ||
885 | /** | |
886 | * bio_map_kern - map kernel address into bio | |
165125e1 | 887 | * @q: the struct request_queue for the bio |
df46b9a4 MC |
888 | * @data: pointer to buffer to map |
889 | * @len: length in bytes | |
890 | * @gfp_mask: allocation flags for bio allocation | |
891 | * | |
892 | * Map the kernel address into a bio suitable for io to a block | |
893 | * device. Returns an error pointer in case of error. | |
894 | */ | |
165125e1 | 895 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
27496a8c | 896 | gfp_t gfp_mask) |
df46b9a4 MC |
897 | { |
898 | struct bio *bio; | |
899 | ||
900 | bio = __bio_map_kern(q, data, len, gfp_mask); | |
901 | if (IS_ERR(bio)) | |
902 | return bio; | |
903 | ||
904 | if (bio->bi_size == len) | |
905 | return bio; | |
906 | ||
907 | /* | |
908 | * Don't support partial mappings. | |
909 | */ | |
910 | bio_put(bio); | |
911 | return ERR_PTR(-EINVAL); | |
912 | } | |
913 | ||
68154e90 FT |
914 | static void bio_copy_kern_endio(struct bio *bio, int err) |
915 | { | |
916 | struct bio_vec *bvec; | |
917 | const int read = bio_data_dir(bio) == READ; | |
918 | char *p = bio->bi_private; | |
919 | int i; | |
920 | ||
921 | __bio_for_each_segment(bvec, bio, i, 0) { | |
922 | char *addr = page_address(bvec->bv_page); | |
923 | ||
924 | if (read && !err) | |
925 | memcpy(p, addr, bvec->bv_len); | |
926 | ||
927 | __free_page(bvec->bv_page); | |
928 | p += bvec->bv_len; | |
929 | } | |
930 | ||
931 | bio_put(bio); | |
932 | } | |
933 | ||
934 | /** | |
935 | * bio_copy_kern - copy kernel address into bio | |
936 | * @q: the struct request_queue for the bio | |
937 | * @data: pointer to buffer to copy | |
938 | * @len: length in bytes | |
939 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 940 | * @reading: data direction is READ |
68154e90 FT |
941 | * |
942 | * copy the kernel address into a bio suitable for io to a block | |
943 | * device. Returns an error pointer in case of error. | |
944 | */ | |
945 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
946 | gfp_t gfp_mask, int reading) | |
947 | { | |
948 | unsigned long kaddr = (unsigned long)data; | |
949 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
950 | unsigned long start = kaddr >> PAGE_SHIFT; | |
951 | const int nr_pages = end - start; | |
952 | struct bio *bio; | |
953 | struct bio_vec *bvec; | |
954 | int i, ret; | |
955 | ||
956 | bio = bio_alloc(gfp_mask, nr_pages); | |
957 | if (!bio) | |
958 | return ERR_PTR(-ENOMEM); | |
959 | ||
960 | while (len) { | |
961 | struct page *page; | |
962 | unsigned int bytes = PAGE_SIZE; | |
963 | ||
964 | if (bytes > len) | |
965 | bytes = len; | |
966 | ||
967 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
968 | if (!page) { | |
969 | ret = -ENOMEM; | |
970 | goto cleanup; | |
971 | } | |
972 | ||
973 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) { | |
974 | ret = -EINVAL; | |
975 | goto cleanup; | |
976 | } | |
977 | ||
978 | len -= bytes; | |
979 | } | |
980 | ||
981 | if (!reading) { | |
982 | void *p = data; | |
983 | ||
984 | bio_for_each_segment(bvec, bio, i) { | |
985 | char *addr = page_address(bvec->bv_page); | |
986 | ||
987 | memcpy(addr, p, bvec->bv_len); | |
988 | p += bvec->bv_len; | |
989 | } | |
990 | } | |
991 | ||
992 | bio->bi_private = data; | |
993 | bio->bi_end_io = bio_copy_kern_endio; | |
994 | return bio; | |
995 | cleanup: | |
996 | bio_for_each_segment(bvec, bio, i) | |
997 | __free_page(bvec->bv_page); | |
998 | ||
999 | bio_put(bio); | |
1000 | ||
1001 | return ERR_PTR(ret); | |
1002 | } | |
1003 | ||
1da177e4 LT |
1004 | /* |
1005 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1006 | * for performing direct-IO in BIOs. | |
1007 | * | |
1008 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1009 | * because the required locks are not interrupt-safe. So what we can do is to | |
1010 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1011 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1012 | * in process context. | |
1013 | * | |
1014 | * We special-case compound pages here: normally this means reads into hugetlb | |
1015 | * pages. The logic in here doesn't really work right for compound pages | |
1016 | * because the VM does not uniformly chase down the head page in all cases. | |
1017 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1018 | * handle them at all. So we skip compound pages here at an early stage. | |
1019 | * | |
1020 | * Note that this code is very hard to test under normal circumstances because | |
1021 | * direct-io pins the pages with get_user_pages(). This makes | |
1022 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
1023 | * But other code (eg, pdflush) could clean the pages if they are mapped | |
1024 | * pagecache. | |
1025 | * | |
1026 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1027 | * deferred bio dirtying paths. | |
1028 | */ | |
1029 | ||
1030 | /* | |
1031 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1032 | */ | |
1033 | void bio_set_pages_dirty(struct bio *bio) | |
1034 | { | |
1035 | struct bio_vec *bvec = bio->bi_io_vec; | |
1036 | int i; | |
1037 | ||
1038 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1039 | struct page *page = bvec[i].bv_page; | |
1040 | ||
1041 | if (page && !PageCompound(page)) | |
1042 | set_page_dirty_lock(page); | |
1043 | } | |
1044 | } | |
1045 | ||
86b6c7a7 | 1046 | static void bio_release_pages(struct bio *bio) |
1da177e4 LT |
1047 | { |
1048 | struct bio_vec *bvec = bio->bi_io_vec; | |
1049 | int i; | |
1050 | ||
1051 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1052 | struct page *page = bvec[i].bv_page; | |
1053 | ||
1054 | if (page) | |
1055 | put_page(page); | |
1056 | } | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1061 | * If they are, then fine. If, however, some pages are clean then they must | |
1062 | * have been written out during the direct-IO read. So we take another ref on | |
1063 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1064 | * | |
1065 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1066 | * here on. It will run one page_cache_release() against each page and will | |
1067 | * run one bio_put() against the BIO. | |
1068 | */ | |
1069 | ||
65f27f38 | 1070 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1071 | |
65f27f38 | 1072 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1073 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1074 | static struct bio *bio_dirty_list; | |
1075 | ||
1076 | /* | |
1077 | * This runs in process context | |
1078 | */ | |
65f27f38 | 1079 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 LT |
1080 | { |
1081 | unsigned long flags; | |
1082 | struct bio *bio; | |
1083 | ||
1084 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1085 | bio = bio_dirty_list; | |
1086 | bio_dirty_list = NULL; | |
1087 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1088 | ||
1089 | while (bio) { | |
1090 | struct bio *next = bio->bi_private; | |
1091 | ||
1092 | bio_set_pages_dirty(bio); | |
1093 | bio_release_pages(bio); | |
1094 | bio_put(bio); | |
1095 | bio = next; | |
1096 | } | |
1097 | } | |
1098 | ||
1099 | void bio_check_pages_dirty(struct bio *bio) | |
1100 | { | |
1101 | struct bio_vec *bvec = bio->bi_io_vec; | |
1102 | int nr_clean_pages = 0; | |
1103 | int i; | |
1104 | ||
1105 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1106 | struct page *page = bvec[i].bv_page; | |
1107 | ||
1108 | if (PageDirty(page) || PageCompound(page)) { | |
1109 | page_cache_release(page); | |
1110 | bvec[i].bv_page = NULL; | |
1111 | } else { | |
1112 | nr_clean_pages++; | |
1113 | } | |
1114 | } | |
1115 | ||
1116 | if (nr_clean_pages) { | |
1117 | unsigned long flags; | |
1118 | ||
1119 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1120 | bio->bi_private = bio_dirty_list; | |
1121 | bio_dirty_list = bio; | |
1122 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1123 | schedule_work(&bio_dirty_work); | |
1124 | } else { | |
1125 | bio_put(bio); | |
1126 | } | |
1127 | } | |
1128 | ||
1129 | /** | |
1130 | * bio_endio - end I/O on a bio | |
1131 | * @bio: bio | |
1da177e4 LT |
1132 | * @error: error, if any |
1133 | * | |
1134 | * Description: | |
6712ecf8 | 1135 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
5bb23a68 N |
1136 | * preferred way to end I/O on a bio, it takes care of clearing |
1137 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1138 | * established -Exxxx (-EIO, for instance) error values in case | |
1139 | * something went wrong. Noone should call bi_end_io() directly on a | |
1140 | * bio unless they own it and thus know that it has an end_io | |
1141 | * function. | |
1da177e4 | 1142 | **/ |
6712ecf8 | 1143 | void bio_endio(struct bio *bio, int error) |
1da177e4 LT |
1144 | { |
1145 | if (error) | |
1146 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
9cc54d40 N |
1147 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1148 | error = -EIO; | |
1da177e4 | 1149 | |
5bb23a68 | 1150 | if (bio->bi_end_io) |
6712ecf8 | 1151 | bio->bi_end_io(bio, error); |
1da177e4 LT |
1152 | } |
1153 | ||
1154 | void bio_pair_release(struct bio_pair *bp) | |
1155 | { | |
1156 | if (atomic_dec_and_test(&bp->cnt)) { | |
1157 | struct bio *master = bp->bio1.bi_private; | |
1158 | ||
6712ecf8 | 1159 | bio_endio(master, bp->error); |
1da177e4 LT |
1160 | mempool_free(bp, bp->bio2.bi_private); |
1161 | } | |
1162 | } | |
1163 | ||
6712ecf8 | 1164 | static void bio_pair_end_1(struct bio *bi, int err) |
1da177e4 LT |
1165 | { |
1166 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); | |
1167 | ||
1168 | if (err) | |
1169 | bp->error = err; | |
1170 | ||
1da177e4 | 1171 | bio_pair_release(bp); |
1da177e4 LT |
1172 | } |
1173 | ||
6712ecf8 | 1174 | static void bio_pair_end_2(struct bio *bi, int err) |
1da177e4 LT |
1175 | { |
1176 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); | |
1177 | ||
1178 | if (err) | |
1179 | bp->error = err; | |
1180 | ||
1da177e4 | 1181 | bio_pair_release(bp); |
1da177e4 LT |
1182 | } |
1183 | ||
1184 | /* | |
1185 | * split a bio - only worry about a bio with a single page | |
1186 | * in it's iovec | |
1187 | */ | |
1188 | struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors) | |
1189 | { | |
1190 | struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO); | |
1191 | ||
1192 | if (!bp) | |
1193 | return bp; | |
1194 | ||
2056a782 JA |
1195 | blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi, |
1196 | bi->bi_sector + first_sectors); | |
1197 | ||
1da177e4 LT |
1198 | BUG_ON(bi->bi_vcnt != 1); |
1199 | BUG_ON(bi->bi_idx != 0); | |
1200 | atomic_set(&bp->cnt, 3); | |
1201 | bp->error = 0; | |
1202 | bp->bio1 = *bi; | |
1203 | bp->bio2 = *bi; | |
1204 | bp->bio2.bi_sector += first_sectors; | |
1205 | bp->bio2.bi_size -= first_sectors << 9; | |
1206 | bp->bio1.bi_size = first_sectors << 9; | |
1207 | ||
1208 | bp->bv1 = bi->bi_io_vec[0]; | |
1209 | bp->bv2 = bi->bi_io_vec[0]; | |
1210 | bp->bv2.bv_offset += first_sectors << 9; | |
1211 | bp->bv2.bv_len -= first_sectors << 9; | |
1212 | bp->bv1.bv_len = first_sectors << 9; | |
1213 | ||
1214 | bp->bio1.bi_io_vec = &bp->bv1; | |
1215 | bp->bio2.bi_io_vec = &bp->bv2; | |
1216 | ||
a2eb0c10 N |
1217 | bp->bio1.bi_max_vecs = 1; |
1218 | bp->bio2.bi_max_vecs = 1; | |
1219 | ||
1da177e4 LT |
1220 | bp->bio1.bi_end_io = bio_pair_end_1; |
1221 | bp->bio2.bi_end_io = bio_pair_end_2; | |
1222 | ||
1223 | bp->bio1.bi_private = bi; | |
1224 | bp->bio2.bi_private = pool; | |
1225 | ||
1226 | return bp; | |
1227 | } | |
1228 | ||
1da177e4 LT |
1229 | |
1230 | /* | |
1231 | * create memory pools for biovec's in a bio_set. | |
1232 | * use the global biovec slabs created for general use. | |
1233 | */ | |
5972511b | 1234 | static int biovec_create_pools(struct bio_set *bs, int pool_entries) |
1da177e4 LT |
1235 | { |
1236 | int i; | |
1237 | ||
1238 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1239 | struct biovec_slab *bp = bvec_slabs + i; | |
1240 | mempool_t **bvp = bs->bvec_pools + i; | |
1241 | ||
93d2341c | 1242 | *bvp = mempool_create_slab_pool(pool_entries, bp->slab); |
1da177e4 LT |
1243 | if (!*bvp) |
1244 | return -ENOMEM; | |
1245 | } | |
1246 | return 0; | |
1247 | } | |
1248 | ||
1249 | static void biovec_free_pools(struct bio_set *bs) | |
1250 | { | |
1251 | int i; | |
1252 | ||
1253 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1254 | mempool_t *bvp = bs->bvec_pools[i]; | |
1255 | ||
1256 | if (bvp) | |
1257 | mempool_destroy(bvp); | |
1258 | } | |
1259 | ||
1260 | } | |
1261 | ||
1262 | void bioset_free(struct bio_set *bs) | |
1263 | { | |
1264 | if (bs->bio_pool) | |
1265 | mempool_destroy(bs->bio_pool); | |
1266 | ||
1267 | biovec_free_pools(bs); | |
1268 | ||
1269 | kfree(bs); | |
1270 | } | |
1271 | ||
5972511b | 1272 | struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size) |
1da177e4 | 1273 | { |
11b0b5ab | 1274 | struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1da177e4 LT |
1275 | |
1276 | if (!bs) | |
1277 | return NULL; | |
1278 | ||
93d2341c | 1279 | bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab); |
1da177e4 LT |
1280 | if (!bs->bio_pool) |
1281 | goto bad; | |
1282 | ||
5972511b | 1283 | if (!biovec_create_pools(bs, bvec_pool_size)) |
1da177e4 LT |
1284 | return bs; |
1285 | ||
1286 | bad: | |
1287 | bioset_free(bs); | |
1288 | return NULL; | |
1289 | } | |
1290 | ||
1291 | static void __init biovec_init_slabs(void) | |
1292 | { | |
1293 | int i; | |
1294 | ||
1295 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1296 | int size; | |
1297 | struct biovec_slab *bvs = bvec_slabs + i; | |
1298 | ||
1299 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
1300 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 1301 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
1302 | } |
1303 | } | |
1304 | ||
1305 | static int __init init_bio(void) | |
1306 | { | |
0a31bd5f | 1307 | bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
1da177e4 LT |
1308 | |
1309 | biovec_init_slabs(); | |
1310 | ||
5972511b | 1311 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 2); |
1da177e4 LT |
1312 | if (!fs_bio_set) |
1313 | panic("bio: can't allocate bios\n"); | |
1314 | ||
0eaae62a MD |
1315 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
1316 | sizeof(struct bio_pair)); | |
1da177e4 LT |
1317 | if (!bio_split_pool) |
1318 | panic("bio: can't create split pool\n"); | |
1319 | ||
1320 | return 0; | |
1321 | } | |
1322 | ||
1323 | subsys_initcall(init_bio); | |
1324 | ||
1325 | EXPORT_SYMBOL(bio_alloc); | |
1326 | EXPORT_SYMBOL(bio_put); | |
3676347a | 1327 | EXPORT_SYMBOL(bio_free); |
1da177e4 LT |
1328 | EXPORT_SYMBOL(bio_endio); |
1329 | EXPORT_SYMBOL(bio_init); | |
1330 | EXPORT_SYMBOL(__bio_clone); | |
1331 | EXPORT_SYMBOL(bio_clone); | |
1332 | EXPORT_SYMBOL(bio_phys_segments); | |
1333 | EXPORT_SYMBOL(bio_hw_segments); | |
1334 | EXPORT_SYMBOL(bio_add_page); | |
6e68af66 | 1335 | EXPORT_SYMBOL(bio_add_pc_page); |
1da177e4 | 1336 | EXPORT_SYMBOL(bio_get_nr_vecs); |
40044ce0 JA |
1337 | EXPORT_SYMBOL(bio_map_user); |
1338 | EXPORT_SYMBOL(bio_unmap_user); | |
df46b9a4 | 1339 | EXPORT_SYMBOL(bio_map_kern); |
68154e90 | 1340 | EXPORT_SYMBOL(bio_copy_kern); |
1da177e4 LT |
1341 | EXPORT_SYMBOL(bio_pair_release); |
1342 | EXPORT_SYMBOL(bio_split); | |
1343 | EXPORT_SYMBOL(bio_split_pool); | |
1344 | EXPORT_SYMBOL(bio_copy_user); | |
1345 | EXPORT_SYMBOL(bio_uncopy_user); | |
1346 | EXPORT_SYMBOL(bioset_create); | |
1347 | EXPORT_SYMBOL(bioset_free); | |
1348 | EXPORT_SYMBOL(bio_alloc_bioset); |