]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
0fe23479 | 2 | * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> |
1da177e4 LT |
3 | * |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License version 2 as | |
6 | * published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
11 | * GNU General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public Licens | |
14 | * along with this program; if not, write to the Free Software | |
15 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- | |
16 | * | |
17 | */ | |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/bio.h> | |
21 | #include <linux/blkdev.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/kernel.h> | |
25 | #include <linux/module.h> | |
26 | #include <linux/mempool.h> | |
27 | #include <linux/workqueue.h> | |
2056a782 | 28 | #include <linux/blktrace_api.h> |
5f3ea37c | 29 | #include <trace/block.h> |
f1970baf | 30 | #include <scsi/sg.h> /* for struct sg_iovec */ |
1da177e4 | 31 | |
0bfc2455 IM |
32 | DEFINE_TRACE(block_split); |
33 | ||
e18b890b | 34 | static struct kmem_cache *bio_slab __read_mostly; |
1da177e4 | 35 | |
6feef531 | 36 | static mempool_t *bio_split_pool __read_mostly; |
1da177e4 | 37 | |
1da177e4 LT |
38 | /* |
39 | * if you change this list, also change bvec_alloc or things will | |
40 | * break badly! cannot be bigger than what you can fit into an | |
41 | * unsigned short | |
42 | */ | |
43 | ||
44 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } | |
6c036527 | 45 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { |
1da177e4 LT |
46 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
47 | }; | |
48 | #undef BV | |
49 | ||
1da177e4 LT |
50 | /* |
51 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
52 | * IO code that does not need private memory pools. | |
53 | */ | |
51d654e1 | 54 | struct bio_set *fs_bio_set; |
1da177e4 | 55 | |
7ba1ba12 MP |
56 | unsigned int bvec_nr_vecs(unsigned short idx) |
57 | { | |
58 | return bvec_slabs[idx].nr_vecs; | |
59 | } | |
60 | ||
7ff9345f JA |
61 | struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, |
62 | struct bio_set *bs) | |
1da177e4 LT |
63 | { |
64 | struct bio_vec *bvl; | |
1da177e4 LT |
65 | |
66 | /* | |
0a0d96b0 JA |
67 | * If 'bs' is given, lookup the pool and do the mempool alloc. |
68 | * If not, this is a bio_kmalloc() allocation and just do a | |
69 | * kzalloc() for the exact number of vecs right away. | |
1da177e4 | 70 | */ |
7ff9345f JA |
71 | if (!bs) |
72 | bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask); | |
73 | ||
74 | /* | |
75 | * see comment near bvec_array define! | |
76 | */ | |
77 | switch (nr) { | |
78 | case 1: | |
79 | *idx = 0; | |
80 | break; | |
81 | case 2 ... 4: | |
82 | *idx = 1; | |
83 | break; | |
84 | case 5 ... 16: | |
85 | *idx = 2; | |
86 | break; | |
87 | case 17 ... 64: | |
88 | *idx = 3; | |
89 | break; | |
90 | case 65 ... 128: | |
91 | *idx = 4; | |
92 | break; | |
93 | case 129 ... BIO_MAX_PAGES: | |
94 | *idx = 5; | |
95 | break; | |
96 | default: | |
97 | return NULL; | |
98 | } | |
99 | ||
100 | /* | |
101 | * idx now points to the pool we want to allocate from. only the | |
102 | * 1-vec entry pool is mempool backed. | |
103 | */ | |
104 | if (*idx == BIOVEC_MAX_IDX) { | |
105 | fallback: | |
106 | bvl = mempool_alloc(bs->bvec_pool, gfp_mask); | |
107 | } else { | |
108 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
109 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO); | |
110 | ||
0a0d96b0 | 111 | /* |
7ff9345f JA |
112 | * Make this allocation restricted and don't dump info on |
113 | * allocation failures, since we'll fallback to the mempool | |
114 | * in case of failure. | |
0a0d96b0 | 115 | */ |
7ff9345f | 116 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
1da177e4 | 117 | |
0a0d96b0 | 118 | /* |
7ff9345f JA |
119 | * Try a slab allocation. If this fails and __GFP_WAIT |
120 | * is set, retry with the 1-entry mempool | |
0a0d96b0 | 121 | */ |
7ff9345f JA |
122 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
123 | if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) { | |
124 | *idx = BIOVEC_MAX_IDX; | |
125 | goto fallback; | |
126 | } | |
127 | } | |
128 | ||
129 | if (bvl) | |
130 | memset(bvl, 0, bvec_nr_vecs(*idx) * sizeof(struct bio_vec)); | |
1da177e4 LT |
131 | |
132 | return bvl; | |
133 | } | |
134 | ||
7ff9345f | 135 | void bio_free(struct bio *bio, struct bio_set *bs) |
1da177e4 | 136 | { |
992c5dda JA |
137 | if (bio->bi_io_vec) { |
138 | const int pool_idx = BIO_POOL_IDX(bio); | |
1da177e4 | 139 | |
992c5dda JA |
140 | BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); |
141 | ||
7ff9345f JA |
142 | if (pool_idx == BIOVEC_MAX_IDX) |
143 | mempool_free(bio->bi_io_vec, bs->bvec_pool); | |
144 | else { | |
145 | struct biovec_slab *bvs = bvec_slabs + pool_idx; | |
146 | ||
147 | kmem_cache_free(bvs->slab, bio->bi_io_vec); | |
148 | } | |
992c5dda | 149 | } |
1da177e4 | 150 | |
7ba1ba12 | 151 | if (bio_integrity(bio)) |
7ff9345f | 152 | bio_integrity_free(bio, bs); |
7ba1ba12 | 153 | |
7ff9345f | 154 | mempool_free(bio, bs->bio_pool); |
3676347a PO |
155 | } |
156 | ||
157 | /* | |
158 | * default destructor for a bio allocated with bio_alloc_bioset() | |
159 | */ | |
160 | static void bio_fs_destructor(struct bio *bio) | |
161 | { | |
162 | bio_free(bio, fs_bio_set); | |
1da177e4 LT |
163 | } |
164 | ||
0a0d96b0 JA |
165 | static void bio_kmalloc_destructor(struct bio *bio) |
166 | { | |
167 | kfree(bio->bi_io_vec); | |
168 | kfree(bio); | |
169 | } | |
170 | ||
858119e1 | 171 | void bio_init(struct bio *bio) |
1da177e4 | 172 | { |
2b94de55 | 173 | memset(bio, 0, sizeof(*bio)); |
1da177e4 | 174 | bio->bi_flags = 1 << BIO_UPTODATE; |
c7c22e4d | 175 | bio->bi_comp_cpu = -1; |
1da177e4 | 176 | atomic_set(&bio->bi_cnt, 1); |
1da177e4 LT |
177 | } |
178 | ||
179 | /** | |
180 | * bio_alloc_bioset - allocate a bio for I/O | |
181 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
182 | * @nr_iovecs: number of iovecs to pre-allocate | |
0a0d96b0 | 183 | * @bs: the bio_set to allocate from. If %NULL, just use kmalloc |
1da177e4 LT |
184 | * |
185 | * Description: | |
0a0d96b0 | 186 | * bio_alloc_bioset will first try its own mempool to satisfy the allocation. |
1da177e4 | 187 | * If %__GFP_WAIT is set then we will block on the internal pool waiting |
0a0d96b0 JA |
188 | * for a &struct bio to become free. If a %NULL @bs is passed in, we will |
189 | * fall back to just using @kmalloc to allocate the required memory. | |
1da177e4 LT |
190 | * |
191 | * allocate bio and iovecs from the memory pools specified by the | |
0a0d96b0 | 192 | * bio_set structure, or @kmalloc if none given. |
1da177e4 | 193 | **/ |
dd0fc66f | 194 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) |
1da177e4 | 195 | { |
0a0d96b0 JA |
196 | struct bio *bio; |
197 | ||
198 | if (bs) | |
199 | bio = mempool_alloc(bs->bio_pool, gfp_mask); | |
200 | else | |
201 | bio = kmalloc(sizeof(*bio), gfp_mask); | |
1da177e4 LT |
202 | |
203 | if (likely(bio)) { | |
204 | struct bio_vec *bvl = NULL; | |
205 | ||
206 | bio_init(bio); | |
207 | if (likely(nr_iovecs)) { | |
eeae1d48 | 208 | unsigned long uninitialized_var(idx); |
1da177e4 LT |
209 | |
210 | bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); | |
211 | if (unlikely(!bvl)) { | |
0a0d96b0 JA |
212 | if (bs) |
213 | mempool_free(bio, bs->bio_pool); | |
214 | else | |
215 | kfree(bio); | |
1da177e4 LT |
216 | bio = NULL; |
217 | goto out; | |
218 | } | |
219 | bio->bi_flags |= idx << BIO_POOL_OFFSET; | |
1ac0ae06 | 220 | bio->bi_max_vecs = bvec_nr_vecs(idx); |
1da177e4 LT |
221 | } |
222 | bio->bi_io_vec = bvl; | |
1da177e4 LT |
223 | } |
224 | out: | |
225 | return bio; | |
226 | } | |
227 | ||
dd0fc66f | 228 | struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs) |
1da177e4 | 229 | { |
3676347a PO |
230 | struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set); |
231 | ||
232 | if (bio) | |
233 | bio->bi_destructor = bio_fs_destructor; | |
234 | ||
235 | return bio; | |
1da177e4 LT |
236 | } |
237 | ||
0a0d96b0 JA |
238 | /* |
239 | * Like bio_alloc(), but doesn't use a mempool backing. This means that | |
240 | * it CAN fail, but while bio_alloc() can only be used for allocations | |
241 | * that have a short (finite) life span, bio_kmalloc() should be used | |
242 | * for more permanent bio allocations (like allocating some bio's for | |
243 | * initalization or setup purposes). | |
244 | */ | |
245 | struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs) | |
246 | { | |
247 | struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, NULL); | |
248 | ||
249 | if (bio) | |
250 | bio->bi_destructor = bio_kmalloc_destructor; | |
251 | ||
252 | return bio; | |
253 | } | |
254 | ||
1da177e4 LT |
255 | void zero_fill_bio(struct bio *bio) |
256 | { | |
257 | unsigned long flags; | |
258 | struct bio_vec *bv; | |
259 | int i; | |
260 | ||
261 | bio_for_each_segment(bv, bio, i) { | |
262 | char *data = bvec_kmap_irq(bv, &flags); | |
263 | memset(data, 0, bv->bv_len); | |
264 | flush_dcache_page(bv->bv_page); | |
265 | bvec_kunmap_irq(data, &flags); | |
266 | } | |
267 | } | |
268 | EXPORT_SYMBOL(zero_fill_bio); | |
269 | ||
270 | /** | |
271 | * bio_put - release a reference to a bio | |
272 | * @bio: bio to release reference to | |
273 | * | |
274 | * Description: | |
275 | * Put a reference to a &struct bio, either one you have gotten with | |
276 | * bio_alloc or bio_get. The last put of a bio will free it. | |
277 | **/ | |
278 | void bio_put(struct bio *bio) | |
279 | { | |
280 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); | |
281 | ||
282 | /* | |
283 | * last put frees it | |
284 | */ | |
285 | if (atomic_dec_and_test(&bio->bi_cnt)) { | |
286 | bio->bi_next = NULL; | |
287 | bio->bi_destructor(bio); | |
288 | } | |
289 | } | |
290 | ||
165125e1 | 291 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
292 | { |
293 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
294 | blk_recount_segments(q, bio); | |
295 | ||
296 | return bio->bi_phys_segments; | |
297 | } | |
298 | ||
1da177e4 LT |
299 | /** |
300 | * __bio_clone - clone a bio | |
301 | * @bio: destination bio | |
302 | * @bio_src: bio to clone | |
303 | * | |
304 | * Clone a &bio. Caller will own the returned bio, but not | |
305 | * the actual data it points to. Reference count of returned | |
306 | * bio will be one. | |
307 | */ | |
858119e1 | 308 | void __bio_clone(struct bio *bio, struct bio *bio_src) |
1da177e4 | 309 | { |
e525e153 AM |
310 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, |
311 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); | |
1da177e4 | 312 | |
5d84070e JA |
313 | /* |
314 | * most users will be overriding ->bi_bdev with a new target, | |
315 | * so we don't set nor calculate new physical/hw segment counts here | |
316 | */ | |
1da177e4 LT |
317 | bio->bi_sector = bio_src->bi_sector; |
318 | bio->bi_bdev = bio_src->bi_bdev; | |
319 | bio->bi_flags |= 1 << BIO_CLONED; | |
320 | bio->bi_rw = bio_src->bi_rw; | |
1da177e4 LT |
321 | bio->bi_vcnt = bio_src->bi_vcnt; |
322 | bio->bi_size = bio_src->bi_size; | |
a5453be4 | 323 | bio->bi_idx = bio_src->bi_idx; |
1da177e4 LT |
324 | } |
325 | ||
326 | /** | |
327 | * bio_clone - clone a bio | |
328 | * @bio: bio to clone | |
329 | * @gfp_mask: allocation priority | |
330 | * | |
331 | * Like __bio_clone, only also allocates the returned bio | |
332 | */ | |
dd0fc66f | 333 | struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
334 | { |
335 | struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); | |
336 | ||
7ba1ba12 MP |
337 | if (!b) |
338 | return NULL; | |
339 | ||
340 | b->bi_destructor = bio_fs_destructor; | |
341 | __bio_clone(b, bio); | |
342 | ||
343 | if (bio_integrity(bio)) { | |
344 | int ret; | |
345 | ||
346 | ret = bio_integrity_clone(b, bio, fs_bio_set); | |
347 | ||
348 | if (ret < 0) | |
349 | return NULL; | |
3676347a | 350 | } |
1da177e4 LT |
351 | |
352 | return b; | |
353 | } | |
354 | ||
355 | /** | |
356 | * bio_get_nr_vecs - return approx number of vecs | |
357 | * @bdev: I/O target | |
358 | * | |
359 | * Return the approximate number of pages we can send to this target. | |
360 | * There's no guarantee that you will be able to fit this number of pages | |
361 | * into a bio, it does not account for dynamic restrictions that vary | |
362 | * on offset. | |
363 | */ | |
364 | int bio_get_nr_vecs(struct block_device *bdev) | |
365 | { | |
165125e1 | 366 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
367 | int nr_pages; |
368 | ||
369 | nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
370 | if (nr_pages > q->max_phys_segments) | |
371 | nr_pages = q->max_phys_segments; | |
372 | if (nr_pages > q->max_hw_segments) | |
373 | nr_pages = q->max_hw_segments; | |
374 | ||
375 | return nr_pages; | |
376 | } | |
377 | ||
165125e1 | 378 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page |
defd94b7 MC |
379 | *page, unsigned int len, unsigned int offset, |
380 | unsigned short max_sectors) | |
1da177e4 LT |
381 | { |
382 | int retried_segments = 0; | |
383 | struct bio_vec *bvec; | |
384 | ||
385 | /* | |
386 | * cloned bio must not modify vec list | |
387 | */ | |
388 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
389 | return 0; | |
390 | ||
80cfd548 | 391 | if (((bio->bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
392 | return 0; |
393 | ||
80cfd548 JA |
394 | /* |
395 | * For filesystems with a blocksize smaller than the pagesize | |
396 | * we will often be called with the same page as last time and | |
397 | * a consecutive offset. Optimize this special case. | |
398 | */ | |
399 | if (bio->bi_vcnt > 0) { | |
400 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
401 | ||
402 | if (page == prev->bv_page && | |
403 | offset == prev->bv_offset + prev->bv_len) { | |
404 | prev->bv_len += len; | |
cc371e66 AK |
405 | |
406 | if (q->merge_bvec_fn) { | |
407 | struct bvec_merge_data bvm = { | |
408 | .bi_bdev = bio->bi_bdev, | |
409 | .bi_sector = bio->bi_sector, | |
410 | .bi_size = bio->bi_size, | |
411 | .bi_rw = bio->bi_rw, | |
412 | }; | |
413 | ||
414 | if (q->merge_bvec_fn(q, &bvm, prev) < len) { | |
415 | prev->bv_len -= len; | |
416 | return 0; | |
417 | } | |
80cfd548 JA |
418 | } |
419 | ||
420 | goto done; | |
421 | } | |
422 | } | |
423 | ||
424 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
1da177e4 LT |
425 | return 0; |
426 | ||
427 | /* | |
428 | * we might lose a segment or two here, but rather that than | |
429 | * make this too complex. | |
430 | */ | |
431 | ||
432 | while (bio->bi_phys_segments >= q->max_phys_segments | |
5df97b91 | 433 | || bio->bi_phys_segments >= q->max_hw_segments) { |
1da177e4 LT |
434 | |
435 | if (retried_segments) | |
436 | return 0; | |
437 | ||
438 | retried_segments = 1; | |
439 | blk_recount_segments(q, bio); | |
440 | } | |
441 | ||
442 | /* | |
443 | * setup the new entry, we might clear it again later if we | |
444 | * cannot add the page | |
445 | */ | |
446 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
447 | bvec->bv_page = page; | |
448 | bvec->bv_len = len; | |
449 | bvec->bv_offset = offset; | |
450 | ||
451 | /* | |
452 | * if queue has other restrictions (eg varying max sector size | |
453 | * depending on offset), it can specify a merge_bvec_fn in the | |
454 | * queue to get further control | |
455 | */ | |
456 | if (q->merge_bvec_fn) { | |
cc371e66 AK |
457 | struct bvec_merge_data bvm = { |
458 | .bi_bdev = bio->bi_bdev, | |
459 | .bi_sector = bio->bi_sector, | |
460 | .bi_size = bio->bi_size, | |
461 | .bi_rw = bio->bi_rw, | |
462 | }; | |
463 | ||
1da177e4 LT |
464 | /* |
465 | * merge_bvec_fn() returns number of bytes it can accept | |
466 | * at this offset | |
467 | */ | |
cc371e66 | 468 | if (q->merge_bvec_fn(q, &bvm, bvec) < len) { |
1da177e4 LT |
469 | bvec->bv_page = NULL; |
470 | bvec->bv_len = 0; | |
471 | bvec->bv_offset = 0; | |
472 | return 0; | |
473 | } | |
474 | } | |
475 | ||
476 | /* If we may be able to merge these biovecs, force a recount */ | |
b8b3e16c | 477 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) |
1da177e4 LT |
478 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
479 | ||
480 | bio->bi_vcnt++; | |
481 | bio->bi_phys_segments++; | |
80cfd548 | 482 | done: |
1da177e4 LT |
483 | bio->bi_size += len; |
484 | return len; | |
485 | } | |
486 | ||
6e68af66 MC |
487 | /** |
488 | * bio_add_pc_page - attempt to add page to bio | |
fddfdeaf | 489 | * @q: the target queue |
6e68af66 MC |
490 | * @bio: destination bio |
491 | * @page: page to add | |
492 | * @len: vec entry length | |
493 | * @offset: vec entry offset | |
494 | * | |
495 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
496 | * number of reasons, such as the bio being full or target block | |
497 | * device limitations. The target block device must allow bio's | |
498 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
499 | * page to an empty bio. This should only be used by REQ_PC bios. | |
500 | */ | |
165125e1 | 501 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
6e68af66 MC |
502 | unsigned int len, unsigned int offset) |
503 | { | |
defd94b7 | 504 | return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors); |
6e68af66 MC |
505 | } |
506 | ||
1da177e4 LT |
507 | /** |
508 | * bio_add_page - attempt to add page to bio | |
509 | * @bio: destination bio | |
510 | * @page: page to add | |
511 | * @len: vec entry length | |
512 | * @offset: vec entry offset | |
513 | * | |
514 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
515 | * number of reasons, such as the bio being full or target block | |
516 | * device limitations. The target block device must allow bio's | |
517 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
518 | * page to an empty bio. | |
519 | */ | |
520 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
521 | unsigned int offset) | |
522 | { | |
defd94b7 MC |
523 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
524 | return __bio_add_page(q, bio, page, len, offset, q->max_sectors); | |
1da177e4 LT |
525 | } |
526 | ||
527 | struct bio_map_data { | |
528 | struct bio_vec *iovecs; | |
c5dec1c3 | 529 | struct sg_iovec *sgvecs; |
152e283f FT |
530 | int nr_sgvecs; |
531 | int is_our_pages; | |
1da177e4 LT |
532 | }; |
533 | ||
c5dec1c3 | 534 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
152e283f FT |
535 | struct sg_iovec *iov, int iov_count, |
536 | int is_our_pages) | |
1da177e4 LT |
537 | { |
538 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); | |
c5dec1c3 FT |
539 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
540 | bmd->nr_sgvecs = iov_count; | |
152e283f | 541 | bmd->is_our_pages = is_our_pages; |
1da177e4 LT |
542 | bio->bi_private = bmd; |
543 | } | |
544 | ||
545 | static void bio_free_map_data(struct bio_map_data *bmd) | |
546 | { | |
547 | kfree(bmd->iovecs); | |
c5dec1c3 | 548 | kfree(bmd->sgvecs); |
1da177e4 LT |
549 | kfree(bmd); |
550 | } | |
551 | ||
76029ff3 FT |
552 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count, |
553 | gfp_t gfp_mask) | |
1da177e4 | 554 | { |
76029ff3 | 555 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask); |
1da177e4 LT |
556 | |
557 | if (!bmd) | |
558 | return NULL; | |
559 | ||
76029ff3 | 560 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask); |
c5dec1c3 FT |
561 | if (!bmd->iovecs) { |
562 | kfree(bmd); | |
563 | return NULL; | |
564 | } | |
565 | ||
76029ff3 | 566 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask); |
c5dec1c3 | 567 | if (bmd->sgvecs) |
1da177e4 LT |
568 | return bmd; |
569 | ||
c5dec1c3 | 570 | kfree(bmd->iovecs); |
1da177e4 LT |
571 | kfree(bmd); |
572 | return NULL; | |
573 | } | |
574 | ||
aefcc28a | 575 | static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs, |
152e283f FT |
576 | struct sg_iovec *iov, int iov_count, int uncopy, |
577 | int do_free_page) | |
c5dec1c3 FT |
578 | { |
579 | int ret = 0, i; | |
580 | struct bio_vec *bvec; | |
581 | int iov_idx = 0; | |
582 | unsigned int iov_off = 0; | |
583 | int read = bio_data_dir(bio) == READ; | |
584 | ||
585 | __bio_for_each_segment(bvec, bio, i, 0) { | |
586 | char *bv_addr = page_address(bvec->bv_page); | |
aefcc28a | 587 | unsigned int bv_len = iovecs[i].bv_len; |
c5dec1c3 FT |
588 | |
589 | while (bv_len && iov_idx < iov_count) { | |
590 | unsigned int bytes; | |
591 | char *iov_addr; | |
592 | ||
593 | bytes = min_t(unsigned int, | |
594 | iov[iov_idx].iov_len - iov_off, bv_len); | |
595 | iov_addr = iov[iov_idx].iov_base + iov_off; | |
596 | ||
597 | if (!ret) { | |
598 | if (!read && !uncopy) | |
599 | ret = copy_from_user(bv_addr, iov_addr, | |
600 | bytes); | |
601 | if (read && uncopy) | |
602 | ret = copy_to_user(iov_addr, bv_addr, | |
603 | bytes); | |
604 | ||
605 | if (ret) | |
606 | ret = -EFAULT; | |
607 | } | |
608 | ||
609 | bv_len -= bytes; | |
610 | bv_addr += bytes; | |
611 | iov_addr += bytes; | |
612 | iov_off += bytes; | |
613 | ||
614 | if (iov[iov_idx].iov_len == iov_off) { | |
615 | iov_idx++; | |
616 | iov_off = 0; | |
617 | } | |
618 | } | |
619 | ||
152e283f | 620 | if (do_free_page) |
c5dec1c3 FT |
621 | __free_page(bvec->bv_page); |
622 | } | |
623 | ||
624 | return ret; | |
625 | } | |
626 | ||
1da177e4 LT |
627 | /** |
628 | * bio_uncopy_user - finish previously mapped bio | |
629 | * @bio: bio being terminated | |
630 | * | |
631 | * Free pages allocated from bio_copy_user() and write back data | |
632 | * to user space in case of a read. | |
633 | */ | |
634 | int bio_uncopy_user(struct bio *bio) | |
635 | { | |
636 | struct bio_map_data *bmd = bio->bi_private; | |
81882766 | 637 | int ret = 0; |
1da177e4 | 638 | |
81882766 FT |
639 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) |
640 | ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, | |
641 | bmd->nr_sgvecs, 1, bmd->is_our_pages); | |
1da177e4 LT |
642 | bio_free_map_data(bmd); |
643 | bio_put(bio); | |
644 | return ret; | |
645 | } | |
646 | ||
647 | /** | |
c5dec1c3 | 648 | * bio_copy_user_iov - copy user data to bio |
1da177e4 | 649 | * @q: destination block queue |
152e283f | 650 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
c5dec1c3 FT |
651 | * @iov: the iovec. |
652 | * @iov_count: number of elements in the iovec | |
1da177e4 | 653 | * @write_to_vm: bool indicating writing to pages or not |
a3bce90e | 654 | * @gfp_mask: memory allocation flags |
1da177e4 LT |
655 | * |
656 | * Prepares and returns a bio for indirect user io, bouncing data | |
657 | * to/from kernel pages as necessary. Must be paired with | |
658 | * call bio_uncopy_user() on io completion. | |
659 | */ | |
152e283f FT |
660 | struct bio *bio_copy_user_iov(struct request_queue *q, |
661 | struct rq_map_data *map_data, | |
662 | struct sg_iovec *iov, int iov_count, | |
663 | int write_to_vm, gfp_t gfp_mask) | |
1da177e4 | 664 | { |
1da177e4 LT |
665 | struct bio_map_data *bmd; |
666 | struct bio_vec *bvec; | |
667 | struct page *page; | |
668 | struct bio *bio; | |
669 | int i, ret; | |
c5dec1c3 FT |
670 | int nr_pages = 0; |
671 | unsigned int len = 0; | |
1da177e4 | 672 | |
c5dec1c3 FT |
673 | for (i = 0; i < iov_count; i++) { |
674 | unsigned long uaddr; | |
675 | unsigned long end; | |
676 | unsigned long start; | |
677 | ||
678 | uaddr = (unsigned long)iov[i].iov_base; | |
679 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
680 | start = uaddr >> PAGE_SHIFT; | |
681 | ||
682 | nr_pages += end - start; | |
683 | len += iov[i].iov_len; | |
684 | } | |
685 | ||
a3bce90e | 686 | bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask); |
1da177e4 LT |
687 | if (!bmd) |
688 | return ERR_PTR(-ENOMEM); | |
689 | ||
1da177e4 | 690 | ret = -ENOMEM; |
a3bce90e | 691 | bio = bio_alloc(gfp_mask, nr_pages); |
1da177e4 LT |
692 | if (!bio) |
693 | goto out_bmd; | |
694 | ||
695 | bio->bi_rw |= (!write_to_vm << BIO_RW); | |
696 | ||
697 | ret = 0; | |
152e283f | 698 | i = 0; |
1da177e4 | 699 | while (len) { |
152e283f FT |
700 | unsigned int bytes; |
701 | ||
702 | if (map_data) | |
703 | bytes = 1U << (PAGE_SHIFT + map_data->page_order); | |
704 | else | |
705 | bytes = PAGE_SIZE; | |
1da177e4 LT |
706 | |
707 | if (bytes > len) | |
708 | bytes = len; | |
709 | ||
152e283f FT |
710 | if (map_data) { |
711 | if (i == map_data->nr_entries) { | |
712 | ret = -ENOMEM; | |
713 | break; | |
714 | } | |
715 | page = map_data->pages[i++]; | |
716 | } else | |
717 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1da177e4 LT |
718 | if (!page) { |
719 | ret = -ENOMEM; | |
720 | break; | |
721 | } | |
722 | ||
0e75f906 | 723 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) |
1da177e4 | 724 | break; |
1da177e4 LT |
725 | |
726 | len -= bytes; | |
727 | } | |
728 | ||
729 | if (ret) | |
730 | goto cleanup; | |
731 | ||
732 | /* | |
733 | * success | |
734 | */ | |
735 | if (!write_to_vm) { | |
152e283f | 736 | ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0); |
c5dec1c3 FT |
737 | if (ret) |
738 | goto cleanup; | |
1da177e4 LT |
739 | } |
740 | ||
152e283f | 741 | bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1); |
1da177e4 LT |
742 | return bio; |
743 | cleanup: | |
152e283f FT |
744 | if (!map_data) |
745 | bio_for_each_segment(bvec, bio, i) | |
746 | __free_page(bvec->bv_page); | |
1da177e4 LT |
747 | |
748 | bio_put(bio); | |
749 | out_bmd: | |
750 | bio_free_map_data(bmd); | |
751 | return ERR_PTR(ret); | |
752 | } | |
753 | ||
c5dec1c3 FT |
754 | /** |
755 | * bio_copy_user - copy user data to bio | |
756 | * @q: destination block queue | |
152e283f | 757 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
c5dec1c3 FT |
758 | * @uaddr: start of user address |
759 | * @len: length in bytes | |
760 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 761 | * @gfp_mask: memory allocation flags |
c5dec1c3 FT |
762 | * |
763 | * Prepares and returns a bio for indirect user io, bouncing data | |
764 | * to/from kernel pages as necessary. Must be paired with | |
765 | * call bio_uncopy_user() on io completion. | |
766 | */ | |
152e283f FT |
767 | struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data, |
768 | unsigned long uaddr, unsigned int len, | |
769 | int write_to_vm, gfp_t gfp_mask) | |
c5dec1c3 FT |
770 | { |
771 | struct sg_iovec iov; | |
772 | ||
773 | iov.iov_base = (void __user *)uaddr; | |
774 | iov.iov_len = len; | |
775 | ||
152e283f | 776 | return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask); |
c5dec1c3 FT |
777 | } |
778 | ||
165125e1 | 779 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
f1970baf JB |
780 | struct block_device *bdev, |
781 | struct sg_iovec *iov, int iov_count, | |
a3bce90e | 782 | int write_to_vm, gfp_t gfp_mask) |
1da177e4 | 783 | { |
f1970baf JB |
784 | int i, j; |
785 | int nr_pages = 0; | |
1da177e4 LT |
786 | struct page **pages; |
787 | struct bio *bio; | |
f1970baf JB |
788 | int cur_page = 0; |
789 | int ret, offset; | |
1da177e4 | 790 | |
f1970baf JB |
791 | for (i = 0; i < iov_count; i++) { |
792 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
793 | unsigned long len = iov[i].iov_len; | |
794 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
795 | unsigned long start = uaddr >> PAGE_SHIFT; | |
796 | ||
797 | nr_pages += end - start; | |
798 | /* | |
ad2d7225 | 799 | * buffer must be aligned to at least hardsector size for now |
f1970baf | 800 | */ |
ad2d7225 | 801 | if (uaddr & queue_dma_alignment(q)) |
f1970baf JB |
802 | return ERR_PTR(-EINVAL); |
803 | } | |
804 | ||
805 | if (!nr_pages) | |
1da177e4 LT |
806 | return ERR_PTR(-EINVAL); |
807 | ||
a3bce90e | 808 | bio = bio_alloc(gfp_mask, nr_pages); |
1da177e4 LT |
809 | if (!bio) |
810 | return ERR_PTR(-ENOMEM); | |
811 | ||
812 | ret = -ENOMEM; | |
a3bce90e | 813 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); |
1da177e4 LT |
814 | if (!pages) |
815 | goto out; | |
816 | ||
f1970baf JB |
817 | for (i = 0; i < iov_count; i++) { |
818 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
819 | unsigned long len = iov[i].iov_len; | |
820 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
821 | unsigned long start = uaddr >> PAGE_SHIFT; | |
822 | const int local_nr_pages = end - start; | |
823 | const int page_limit = cur_page + local_nr_pages; | |
824 | ||
f5dd33c4 NP |
825 | ret = get_user_pages_fast(uaddr, local_nr_pages, |
826 | write_to_vm, &pages[cur_page]); | |
99172157 JA |
827 | if (ret < local_nr_pages) { |
828 | ret = -EFAULT; | |
f1970baf | 829 | goto out_unmap; |
99172157 | 830 | } |
f1970baf JB |
831 | |
832 | offset = uaddr & ~PAGE_MASK; | |
833 | for (j = cur_page; j < page_limit; j++) { | |
834 | unsigned int bytes = PAGE_SIZE - offset; | |
835 | ||
836 | if (len <= 0) | |
837 | break; | |
838 | ||
839 | if (bytes > len) | |
840 | bytes = len; | |
841 | ||
842 | /* | |
843 | * sorry... | |
844 | */ | |
defd94b7 MC |
845 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
846 | bytes) | |
f1970baf JB |
847 | break; |
848 | ||
849 | len -= bytes; | |
850 | offset = 0; | |
851 | } | |
1da177e4 | 852 | |
f1970baf | 853 | cur_page = j; |
1da177e4 | 854 | /* |
f1970baf | 855 | * release the pages we didn't map into the bio, if any |
1da177e4 | 856 | */ |
f1970baf JB |
857 | while (j < page_limit) |
858 | page_cache_release(pages[j++]); | |
1da177e4 LT |
859 | } |
860 | ||
1da177e4 LT |
861 | kfree(pages); |
862 | ||
863 | /* | |
864 | * set data direction, and check if mapped pages need bouncing | |
865 | */ | |
866 | if (!write_to_vm) | |
867 | bio->bi_rw |= (1 << BIO_RW); | |
868 | ||
f1970baf | 869 | bio->bi_bdev = bdev; |
1da177e4 LT |
870 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
871 | return bio; | |
f1970baf JB |
872 | |
873 | out_unmap: | |
874 | for (i = 0; i < nr_pages; i++) { | |
875 | if(!pages[i]) | |
876 | break; | |
877 | page_cache_release(pages[i]); | |
878 | } | |
879 | out: | |
1da177e4 LT |
880 | kfree(pages); |
881 | bio_put(bio); | |
882 | return ERR_PTR(ret); | |
883 | } | |
884 | ||
885 | /** | |
886 | * bio_map_user - map user address into bio | |
165125e1 | 887 | * @q: the struct request_queue for the bio |
1da177e4 LT |
888 | * @bdev: destination block device |
889 | * @uaddr: start of user address | |
890 | * @len: length in bytes | |
891 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 892 | * @gfp_mask: memory allocation flags |
1da177e4 LT |
893 | * |
894 | * Map the user space address into a bio suitable for io to a block | |
895 | * device. Returns an error pointer in case of error. | |
896 | */ | |
165125e1 | 897 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
a3bce90e FT |
898 | unsigned long uaddr, unsigned int len, int write_to_vm, |
899 | gfp_t gfp_mask) | |
f1970baf JB |
900 | { |
901 | struct sg_iovec iov; | |
902 | ||
3f70353e | 903 | iov.iov_base = (void __user *)uaddr; |
f1970baf JB |
904 | iov.iov_len = len; |
905 | ||
a3bce90e | 906 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask); |
f1970baf JB |
907 | } |
908 | ||
909 | /** | |
910 | * bio_map_user_iov - map user sg_iovec table into bio | |
165125e1 | 911 | * @q: the struct request_queue for the bio |
f1970baf JB |
912 | * @bdev: destination block device |
913 | * @iov: the iovec. | |
914 | * @iov_count: number of elements in the iovec | |
915 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 916 | * @gfp_mask: memory allocation flags |
f1970baf JB |
917 | * |
918 | * Map the user space address into a bio suitable for io to a block | |
919 | * device. Returns an error pointer in case of error. | |
920 | */ | |
165125e1 | 921 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
f1970baf | 922 | struct sg_iovec *iov, int iov_count, |
a3bce90e | 923 | int write_to_vm, gfp_t gfp_mask) |
1da177e4 LT |
924 | { |
925 | struct bio *bio; | |
926 | ||
a3bce90e FT |
927 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm, |
928 | gfp_mask); | |
1da177e4 LT |
929 | if (IS_ERR(bio)) |
930 | return bio; | |
931 | ||
932 | /* | |
933 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
934 | * it would normally disappear when its bi_end_io is run. | |
935 | * however, we need it for the unmap, so grab an extra | |
936 | * reference to it | |
937 | */ | |
938 | bio_get(bio); | |
939 | ||
0e75f906 | 940 | return bio; |
1da177e4 LT |
941 | } |
942 | ||
943 | static void __bio_unmap_user(struct bio *bio) | |
944 | { | |
945 | struct bio_vec *bvec; | |
946 | int i; | |
947 | ||
948 | /* | |
949 | * make sure we dirty pages we wrote to | |
950 | */ | |
951 | __bio_for_each_segment(bvec, bio, i, 0) { | |
952 | if (bio_data_dir(bio) == READ) | |
953 | set_page_dirty_lock(bvec->bv_page); | |
954 | ||
955 | page_cache_release(bvec->bv_page); | |
956 | } | |
957 | ||
958 | bio_put(bio); | |
959 | } | |
960 | ||
961 | /** | |
962 | * bio_unmap_user - unmap a bio | |
963 | * @bio: the bio being unmapped | |
964 | * | |
965 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
966 | * a process context. | |
967 | * | |
968 | * bio_unmap_user() may sleep. | |
969 | */ | |
970 | void bio_unmap_user(struct bio *bio) | |
971 | { | |
972 | __bio_unmap_user(bio); | |
973 | bio_put(bio); | |
974 | } | |
975 | ||
6712ecf8 | 976 | static void bio_map_kern_endio(struct bio *bio, int err) |
b823825e | 977 | { |
b823825e | 978 | bio_put(bio); |
b823825e JA |
979 | } |
980 | ||
981 | ||
165125e1 | 982 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
27496a8c | 983 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 MC |
984 | { |
985 | unsigned long kaddr = (unsigned long)data; | |
986 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
987 | unsigned long start = kaddr >> PAGE_SHIFT; | |
988 | const int nr_pages = end - start; | |
989 | int offset, i; | |
990 | struct bio *bio; | |
991 | ||
992 | bio = bio_alloc(gfp_mask, nr_pages); | |
993 | if (!bio) | |
994 | return ERR_PTR(-ENOMEM); | |
995 | ||
996 | offset = offset_in_page(kaddr); | |
997 | for (i = 0; i < nr_pages; i++) { | |
998 | unsigned int bytes = PAGE_SIZE - offset; | |
999 | ||
1000 | if (len <= 0) | |
1001 | break; | |
1002 | ||
1003 | if (bytes > len) | |
1004 | bytes = len; | |
1005 | ||
defd94b7 MC |
1006 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
1007 | offset) < bytes) | |
df46b9a4 MC |
1008 | break; |
1009 | ||
1010 | data += bytes; | |
1011 | len -= bytes; | |
1012 | offset = 0; | |
1013 | } | |
1014 | ||
b823825e | 1015 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
1016 | return bio; |
1017 | } | |
1018 | ||
1019 | /** | |
1020 | * bio_map_kern - map kernel address into bio | |
165125e1 | 1021 | * @q: the struct request_queue for the bio |
df46b9a4 MC |
1022 | * @data: pointer to buffer to map |
1023 | * @len: length in bytes | |
1024 | * @gfp_mask: allocation flags for bio allocation | |
1025 | * | |
1026 | * Map the kernel address into a bio suitable for io to a block | |
1027 | * device. Returns an error pointer in case of error. | |
1028 | */ | |
165125e1 | 1029 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
27496a8c | 1030 | gfp_t gfp_mask) |
df46b9a4 MC |
1031 | { |
1032 | struct bio *bio; | |
1033 | ||
1034 | bio = __bio_map_kern(q, data, len, gfp_mask); | |
1035 | if (IS_ERR(bio)) | |
1036 | return bio; | |
1037 | ||
1038 | if (bio->bi_size == len) | |
1039 | return bio; | |
1040 | ||
1041 | /* | |
1042 | * Don't support partial mappings. | |
1043 | */ | |
1044 | bio_put(bio); | |
1045 | return ERR_PTR(-EINVAL); | |
1046 | } | |
1047 | ||
68154e90 FT |
1048 | static void bio_copy_kern_endio(struct bio *bio, int err) |
1049 | { | |
1050 | struct bio_vec *bvec; | |
1051 | const int read = bio_data_dir(bio) == READ; | |
76029ff3 | 1052 | struct bio_map_data *bmd = bio->bi_private; |
68154e90 | 1053 | int i; |
76029ff3 | 1054 | char *p = bmd->sgvecs[0].iov_base; |
68154e90 FT |
1055 | |
1056 | __bio_for_each_segment(bvec, bio, i, 0) { | |
1057 | char *addr = page_address(bvec->bv_page); | |
76029ff3 | 1058 | int len = bmd->iovecs[i].bv_len; |
68154e90 FT |
1059 | |
1060 | if (read && !err) | |
76029ff3 | 1061 | memcpy(p, addr, len); |
68154e90 FT |
1062 | |
1063 | __free_page(bvec->bv_page); | |
76029ff3 | 1064 | p += len; |
68154e90 FT |
1065 | } |
1066 | ||
76029ff3 | 1067 | bio_free_map_data(bmd); |
68154e90 FT |
1068 | bio_put(bio); |
1069 | } | |
1070 | ||
1071 | /** | |
1072 | * bio_copy_kern - copy kernel address into bio | |
1073 | * @q: the struct request_queue for the bio | |
1074 | * @data: pointer to buffer to copy | |
1075 | * @len: length in bytes | |
1076 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 1077 | * @reading: data direction is READ |
68154e90 FT |
1078 | * |
1079 | * copy the kernel address into a bio suitable for io to a block | |
1080 | * device. Returns an error pointer in case of error. | |
1081 | */ | |
1082 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1083 | gfp_t gfp_mask, int reading) | |
1084 | { | |
68154e90 FT |
1085 | struct bio *bio; |
1086 | struct bio_vec *bvec; | |
4d8ab62e | 1087 | int i; |
68154e90 | 1088 | |
4d8ab62e FT |
1089 | bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask); |
1090 | if (IS_ERR(bio)) | |
1091 | return bio; | |
68154e90 FT |
1092 | |
1093 | if (!reading) { | |
1094 | void *p = data; | |
1095 | ||
1096 | bio_for_each_segment(bvec, bio, i) { | |
1097 | char *addr = page_address(bvec->bv_page); | |
1098 | ||
1099 | memcpy(addr, p, bvec->bv_len); | |
1100 | p += bvec->bv_len; | |
1101 | } | |
1102 | } | |
1103 | ||
68154e90 | 1104 | bio->bi_end_io = bio_copy_kern_endio; |
76029ff3 | 1105 | |
68154e90 | 1106 | return bio; |
68154e90 FT |
1107 | } |
1108 | ||
1da177e4 LT |
1109 | /* |
1110 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1111 | * for performing direct-IO in BIOs. | |
1112 | * | |
1113 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1114 | * because the required locks are not interrupt-safe. So what we can do is to | |
1115 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1116 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1117 | * in process context. | |
1118 | * | |
1119 | * We special-case compound pages here: normally this means reads into hugetlb | |
1120 | * pages. The logic in here doesn't really work right for compound pages | |
1121 | * because the VM does not uniformly chase down the head page in all cases. | |
1122 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1123 | * handle them at all. So we skip compound pages here at an early stage. | |
1124 | * | |
1125 | * Note that this code is very hard to test under normal circumstances because | |
1126 | * direct-io pins the pages with get_user_pages(). This makes | |
1127 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
1128 | * But other code (eg, pdflush) could clean the pages if they are mapped | |
1129 | * pagecache. | |
1130 | * | |
1131 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1132 | * deferred bio dirtying paths. | |
1133 | */ | |
1134 | ||
1135 | /* | |
1136 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1137 | */ | |
1138 | void bio_set_pages_dirty(struct bio *bio) | |
1139 | { | |
1140 | struct bio_vec *bvec = bio->bi_io_vec; | |
1141 | int i; | |
1142 | ||
1143 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1144 | struct page *page = bvec[i].bv_page; | |
1145 | ||
1146 | if (page && !PageCompound(page)) | |
1147 | set_page_dirty_lock(page); | |
1148 | } | |
1149 | } | |
1150 | ||
86b6c7a7 | 1151 | static void bio_release_pages(struct bio *bio) |
1da177e4 LT |
1152 | { |
1153 | struct bio_vec *bvec = bio->bi_io_vec; | |
1154 | int i; | |
1155 | ||
1156 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1157 | struct page *page = bvec[i].bv_page; | |
1158 | ||
1159 | if (page) | |
1160 | put_page(page); | |
1161 | } | |
1162 | } | |
1163 | ||
1164 | /* | |
1165 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1166 | * If they are, then fine. If, however, some pages are clean then they must | |
1167 | * have been written out during the direct-IO read. So we take another ref on | |
1168 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1169 | * | |
1170 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1171 | * here on. It will run one page_cache_release() against each page and will | |
1172 | * run one bio_put() against the BIO. | |
1173 | */ | |
1174 | ||
65f27f38 | 1175 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1176 | |
65f27f38 | 1177 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1178 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1179 | static struct bio *bio_dirty_list; | |
1180 | ||
1181 | /* | |
1182 | * This runs in process context | |
1183 | */ | |
65f27f38 | 1184 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 LT |
1185 | { |
1186 | unsigned long flags; | |
1187 | struct bio *bio; | |
1188 | ||
1189 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1190 | bio = bio_dirty_list; | |
1191 | bio_dirty_list = NULL; | |
1192 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1193 | ||
1194 | while (bio) { | |
1195 | struct bio *next = bio->bi_private; | |
1196 | ||
1197 | bio_set_pages_dirty(bio); | |
1198 | bio_release_pages(bio); | |
1199 | bio_put(bio); | |
1200 | bio = next; | |
1201 | } | |
1202 | } | |
1203 | ||
1204 | void bio_check_pages_dirty(struct bio *bio) | |
1205 | { | |
1206 | struct bio_vec *bvec = bio->bi_io_vec; | |
1207 | int nr_clean_pages = 0; | |
1208 | int i; | |
1209 | ||
1210 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1211 | struct page *page = bvec[i].bv_page; | |
1212 | ||
1213 | if (PageDirty(page) || PageCompound(page)) { | |
1214 | page_cache_release(page); | |
1215 | bvec[i].bv_page = NULL; | |
1216 | } else { | |
1217 | nr_clean_pages++; | |
1218 | } | |
1219 | } | |
1220 | ||
1221 | if (nr_clean_pages) { | |
1222 | unsigned long flags; | |
1223 | ||
1224 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1225 | bio->bi_private = bio_dirty_list; | |
1226 | bio_dirty_list = bio; | |
1227 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1228 | schedule_work(&bio_dirty_work); | |
1229 | } else { | |
1230 | bio_put(bio); | |
1231 | } | |
1232 | } | |
1233 | ||
1234 | /** | |
1235 | * bio_endio - end I/O on a bio | |
1236 | * @bio: bio | |
1da177e4 LT |
1237 | * @error: error, if any |
1238 | * | |
1239 | * Description: | |
6712ecf8 | 1240 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
5bb23a68 N |
1241 | * preferred way to end I/O on a bio, it takes care of clearing |
1242 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1243 | * established -Exxxx (-EIO, for instance) error values in case | |
1244 | * something went wrong. Noone should call bi_end_io() directly on a | |
1245 | * bio unless they own it and thus know that it has an end_io | |
1246 | * function. | |
1da177e4 | 1247 | **/ |
6712ecf8 | 1248 | void bio_endio(struct bio *bio, int error) |
1da177e4 LT |
1249 | { |
1250 | if (error) | |
1251 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
9cc54d40 N |
1252 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1253 | error = -EIO; | |
1da177e4 | 1254 | |
5bb23a68 | 1255 | if (bio->bi_end_io) |
6712ecf8 | 1256 | bio->bi_end_io(bio, error); |
1da177e4 LT |
1257 | } |
1258 | ||
1259 | void bio_pair_release(struct bio_pair *bp) | |
1260 | { | |
1261 | if (atomic_dec_and_test(&bp->cnt)) { | |
1262 | struct bio *master = bp->bio1.bi_private; | |
1263 | ||
6712ecf8 | 1264 | bio_endio(master, bp->error); |
1da177e4 LT |
1265 | mempool_free(bp, bp->bio2.bi_private); |
1266 | } | |
1267 | } | |
1268 | ||
6712ecf8 | 1269 | static void bio_pair_end_1(struct bio *bi, int err) |
1da177e4 LT |
1270 | { |
1271 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); | |
1272 | ||
1273 | if (err) | |
1274 | bp->error = err; | |
1275 | ||
1da177e4 | 1276 | bio_pair_release(bp); |
1da177e4 LT |
1277 | } |
1278 | ||
6712ecf8 | 1279 | static void bio_pair_end_2(struct bio *bi, int err) |
1da177e4 LT |
1280 | { |
1281 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); | |
1282 | ||
1283 | if (err) | |
1284 | bp->error = err; | |
1285 | ||
1da177e4 | 1286 | bio_pair_release(bp); |
1da177e4 LT |
1287 | } |
1288 | ||
1289 | /* | |
1290 | * split a bio - only worry about a bio with a single page | |
1291 | * in it's iovec | |
1292 | */ | |
6feef531 | 1293 | struct bio_pair *bio_split(struct bio *bi, int first_sectors) |
1da177e4 | 1294 | { |
6feef531 | 1295 | struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO); |
1da177e4 LT |
1296 | |
1297 | if (!bp) | |
1298 | return bp; | |
1299 | ||
5f3ea37c | 1300 | trace_block_split(bdev_get_queue(bi->bi_bdev), bi, |
2056a782 JA |
1301 | bi->bi_sector + first_sectors); |
1302 | ||
1da177e4 LT |
1303 | BUG_ON(bi->bi_vcnt != 1); |
1304 | BUG_ON(bi->bi_idx != 0); | |
1305 | atomic_set(&bp->cnt, 3); | |
1306 | bp->error = 0; | |
1307 | bp->bio1 = *bi; | |
1308 | bp->bio2 = *bi; | |
1309 | bp->bio2.bi_sector += first_sectors; | |
1310 | bp->bio2.bi_size -= first_sectors << 9; | |
1311 | bp->bio1.bi_size = first_sectors << 9; | |
1312 | ||
1313 | bp->bv1 = bi->bi_io_vec[0]; | |
1314 | bp->bv2 = bi->bi_io_vec[0]; | |
1315 | bp->bv2.bv_offset += first_sectors << 9; | |
1316 | bp->bv2.bv_len -= first_sectors << 9; | |
1317 | bp->bv1.bv_len = first_sectors << 9; | |
1318 | ||
1319 | bp->bio1.bi_io_vec = &bp->bv1; | |
1320 | bp->bio2.bi_io_vec = &bp->bv2; | |
1321 | ||
a2eb0c10 N |
1322 | bp->bio1.bi_max_vecs = 1; |
1323 | bp->bio2.bi_max_vecs = 1; | |
1324 | ||
1da177e4 LT |
1325 | bp->bio1.bi_end_io = bio_pair_end_1; |
1326 | bp->bio2.bi_end_io = bio_pair_end_2; | |
1327 | ||
1328 | bp->bio1.bi_private = bi; | |
6feef531 | 1329 | bp->bio2.bi_private = bio_split_pool; |
1da177e4 | 1330 | |
7ba1ba12 MP |
1331 | if (bio_integrity(bi)) |
1332 | bio_integrity_split(bi, bp, first_sectors); | |
1333 | ||
1da177e4 LT |
1334 | return bp; |
1335 | } | |
1336 | ||
ad3316bf MP |
1337 | /** |
1338 | * bio_sector_offset - Find hardware sector offset in bio | |
1339 | * @bio: bio to inspect | |
1340 | * @index: bio_vec index | |
1341 | * @offset: offset in bv_page | |
1342 | * | |
1343 | * Return the number of hardware sectors between beginning of bio | |
1344 | * and an end point indicated by a bio_vec index and an offset | |
1345 | * within that vector's page. | |
1346 | */ | |
1347 | sector_t bio_sector_offset(struct bio *bio, unsigned short index, | |
1348 | unsigned int offset) | |
1349 | { | |
1350 | unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue); | |
1351 | struct bio_vec *bv; | |
1352 | sector_t sectors; | |
1353 | int i; | |
1354 | ||
1355 | sectors = 0; | |
1356 | ||
1357 | if (index >= bio->bi_idx) | |
1358 | index = bio->bi_vcnt - 1; | |
1359 | ||
1360 | __bio_for_each_segment(bv, bio, i, 0) { | |
1361 | if (i == index) { | |
1362 | if (offset > bv->bv_offset) | |
1363 | sectors += (offset - bv->bv_offset) / sector_sz; | |
1364 | break; | |
1365 | } | |
1366 | ||
1367 | sectors += bv->bv_len / sector_sz; | |
1368 | } | |
1369 | ||
1370 | return sectors; | |
1371 | } | |
1372 | EXPORT_SYMBOL(bio_sector_offset); | |
1da177e4 LT |
1373 | |
1374 | /* | |
1375 | * create memory pools for biovec's in a bio_set. | |
1376 | * use the global biovec slabs created for general use. | |
1377 | */ | |
5972511b | 1378 | static int biovec_create_pools(struct bio_set *bs, int pool_entries) |
1da177e4 | 1379 | { |
7ff9345f | 1380 | struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; |
1da177e4 | 1381 | |
7ff9345f JA |
1382 | bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab); |
1383 | if (!bs->bvec_pool) | |
1384 | return -ENOMEM; | |
1da177e4 | 1385 | |
1da177e4 LT |
1386 | return 0; |
1387 | } | |
1388 | ||
1389 | static void biovec_free_pools(struct bio_set *bs) | |
1390 | { | |
7ff9345f | 1391 | mempool_destroy(bs->bvec_pool); |
1da177e4 LT |
1392 | } |
1393 | ||
1394 | void bioset_free(struct bio_set *bs) | |
1395 | { | |
1396 | if (bs->bio_pool) | |
1397 | mempool_destroy(bs->bio_pool); | |
1398 | ||
7ba1ba12 | 1399 | bioset_integrity_free(bs); |
1da177e4 LT |
1400 | biovec_free_pools(bs); |
1401 | ||
1402 | kfree(bs); | |
1403 | } | |
1404 | ||
5972511b | 1405 | struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size) |
1da177e4 | 1406 | { |
11b0b5ab | 1407 | struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1da177e4 LT |
1408 | |
1409 | if (!bs) | |
1410 | return NULL; | |
1411 | ||
93d2341c | 1412 | bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab); |
1da177e4 LT |
1413 | if (!bs->bio_pool) |
1414 | goto bad; | |
1415 | ||
7ba1ba12 MP |
1416 | if (bioset_integrity_create(bs, bio_pool_size)) |
1417 | goto bad; | |
1418 | ||
5972511b | 1419 | if (!biovec_create_pools(bs, bvec_pool_size)) |
1da177e4 LT |
1420 | return bs; |
1421 | ||
1422 | bad: | |
1423 | bioset_free(bs); | |
1424 | return NULL; | |
1425 | } | |
1426 | ||
1427 | static void __init biovec_init_slabs(void) | |
1428 | { | |
1429 | int i; | |
1430 | ||
1431 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1432 | int size; | |
1433 | struct biovec_slab *bvs = bvec_slabs + i; | |
1434 | ||
1435 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
1436 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 1437 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
1438 | } |
1439 | } | |
1440 | ||
1441 | static int __init init_bio(void) | |
1442 | { | |
0a31bd5f | 1443 | bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
1da177e4 | 1444 | |
7ba1ba12 | 1445 | bio_integrity_init_slab(); |
1da177e4 LT |
1446 | biovec_init_slabs(); |
1447 | ||
5972511b | 1448 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 2); |
1da177e4 LT |
1449 | if (!fs_bio_set) |
1450 | panic("bio: can't allocate bios\n"); | |
1451 | ||
0eaae62a MD |
1452 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
1453 | sizeof(struct bio_pair)); | |
1da177e4 LT |
1454 | if (!bio_split_pool) |
1455 | panic("bio: can't create split pool\n"); | |
1456 | ||
1457 | return 0; | |
1458 | } | |
1459 | ||
1460 | subsys_initcall(init_bio); | |
1461 | ||
1462 | EXPORT_SYMBOL(bio_alloc); | |
0a0d96b0 | 1463 | EXPORT_SYMBOL(bio_kmalloc); |
1da177e4 | 1464 | EXPORT_SYMBOL(bio_put); |
3676347a | 1465 | EXPORT_SYMBOL(bio_free); |
1da177e4 LT |
1466 | EXPORT_SYMBOL(bio_endio); |
1467 | EXPORT_SYMBOL(bio_init); | |
1468 | EXPORT_SYMBOL(__bio_clone); | |
1469 | EXPORT_SYMBOL(bio_clone); | |
1470 | EXPORT_SYMBOL(bio_phys_segments); | |
1da177e4 | 1471 | EXPORT_SYMBOL(bio_add_page); |
6e68af66 | 1472 | EXPORT_SYMBOL(bio_add_pc_page); |
1da177e4 | 1473 | EXPORT_SYMBOL(bio_get_nr_vecs); |
40044ce0 JA |
1474 | EXPORT_SYMBOL(bio_map_user); |
1475 | EXPORT_SYMBOL(bio_unmap_user); | |
df46b9a4 | 1476 | EXPORT_SYMBOL(bio_map_kern); |
68154e90 | 1477 | EXPORT_SYMBOL(bio_copy_kern); |
1da177e4 LT |
1478 | EXPORT_SYMBOL(bio_pair_release); |
1479 | EXPORT_SYMBOL(bio_split); | |
1da177e4 LT |
1480 | EXPORT_SYMBOL(bio_copy_user); |
1481 | EXPORT_SYMBOL(bio_uncopy_user); | |
1482 | EXPORT_SYMBOL(bioset_create); | |
1483 | EXPORT_SYMBOL(bioset_free); | |
1484 | EXPORT_SYMBOL(bio_alloc_bioset); |